Datasets:

Hendrik-a

/

arxiv

like 0

we analyze cross - correlations between price fluctuations of different stocks using methods of random matrix theory ( rmt ) . using two large databases , we calculate cross - correlation matrices * c * of returns constructed from ( i ) 30-min returns of 1000 us stocks for the 2-yr period 199495 ( ii ) 30-min returns of 881 us stocks for the 2-yr period 199697 , and ( iii ) 1-day returns of 422 us stocks for the 35-yr period 196296 . we test the statistics of the eigenvalues of * c * against a `` null hypothesis '' a random correlation matrix constructed from mutually uncorrelated time series . we find that a majority of the eigenvalues of * c * fall within the rmt bounds $ ] for the eigenvalues of random correlation matrices . we test the eigenvalues of * c * within the rmt bound for universal properties of random matrices and find good agreement with the results for the gaussian orthogonal ensemble of random matrices implying a large degree of randomness in the measured cross - correlation coefficients . further , we find that the distribution of eigenvector components for the eigenvectors corresponding to the eigenvalues outside the rmt bound display systematic deviations from the rmt prediction . in addition , we find that these `` deviating eigenvectors '' are stable in time . we analyze the components of the deviating eigenvectors and find that the largest eigenvalue corresponds to an influence common to all stocks . our analysis of the remaining deviating eigenvectors shows distinct groups , whose identities correspond to conventionally - identified business sectors . finally , we discuss applications to the construction of portfolios of stocks that have a stable ratio of risk to return . 2

cond-mat0108023

quantifying correlations between different stocks is a topic of interest not only for scientific reasons of understanding the economy as a complex dynamical system , but also for practical reasons such as asset allocation and portfolio - risk estimation @xcite . unlike most physical systems , where one relates correlations between subunits to basic interactions , the underlying `` interactions '' for the stock market problem are not known . here , we analyze cross - correlations between stocks by applying concepts and methods of random matrix theory , developed in the context of complex quantum systems where the precise nature of the interactions between subunits are not known . in order to quantify correlations , we first calculate the price change ( `` return '' ) of stock @xmath2 over a time scale @xmath3 @xmath4 where @xmath5 denotes the price of stock @xmath6 . since different stocks have varying levels of volatility ( standard deviation ) , we define a normalized return @xmath7 where @xmath8 is the standard deviation of @xmath9 , and @xmath10 denotes a time average over the period studied . we then compute the equal - time cross - correlation matrix * c * with elements @xmath11 by construction , the elements @xmath12 are restricted to the domain @xmath13 , where @xmath14 corresponds to perfect correlations , @xmath15 corresponds to perfect anti - correlations , and @xmath16 corresponds to uncorrelated pairs of stocks . the difficulties in analyzing the significance and meaning of the empirical cross - correlation coefficients @xmath12 are due to several reasons , which include the following : \(i ) market conditions change with time and the cross - correlations that exist between any pair of stocks may not be stationary . \(ii ) the finite length of time series available to estimate cross - correlations introduces `` measurement noise '' . if we use a long time series to circumvent the problem of finite length , our estimates will be affected by the non - stationarity of cross - correlations . for these reasons , the empirically - measured cross - correlations will contain `` random '' contributions , and it is a difficult problem in general to estimate from * c * the cross - correlations that are not a result of randomness . how can we identify from @xmath12 , those stocks that remained correlated ( on the average ) in the time period studied ? to answer this question , we test the statistics of * c * against the `` null hypothesis '' of a random correlation matrix a correlation matrix constructed from mutually uncorrelated time series . if the properties of * c * conform to those of a random correlation matrix , then it follows that the contents of the empirically - measured * c * are random . conversely , deviations of the properties of * c * from those of a random correlation matrix convey information about `` genuine '' correlations . thus , our goal shall be to compare the properties of * c * with those of a random correlation matrix and separate the content of * c * into two groups : ( a ) the part of * c * that conforms to the properties of random correlation matrices ( `` noise '' ) and ( b ) the part of * c * that deviates ( `` information '' ) . the study of statistical properties of matrices with independent random elements _ random matrices _ has a rich history originating in nuclear physics @xcite . in nuclear physics , the problem of interest 50 years ago was to understand the energy levels of complex nuclei , which the existing models failed to explain . rmt was developed in this context by wigner , dyson , mehta , and others in order to explain the statistics of energy levels of complex quantum systems . they postulated that the hamiltonian describing a heavy nucleus can be described by a matrix * h * with independent random elements @xmath17 drawn from a probability distribution @xcite . based on this assumption , a series of remarkable predictions were made which are found to be in agreement with the experimental data @xcite . for complex quantum systems , rmt predictions represent an average over all possible interactions @xcite . deviations from the _ universal _ predictions of rmt identify system - specific , non - random properties of the system under consideration , providing clues about the underlying interactions @xcite . recent studies @xcite applying rmt methods to analyze the properties of * c * show that @xmath18% of the eigenvalues of * c * agree with rmt predictions , suggesting a considerable degree of randomness in the measured cross - correlations . it is also found that there are deviations from rmt predictions for @xmath19% of the largest eigenvalues . these results prompt the following questions : * what is a possible interpretation for the deviations from rmt ? * are the deviations from rmt stable in time ? * what can we infer about the structure of * c * from these results ? * what are the practical implications of these results ? in the following , we address these questions in detail . we find that the largest eigenvalue of * c * represents the influence of the entire market that is common to all stocks . our analysis of the contents of the remaining eigenvalues that deviate from rmt shows the existence of cross - correlations between stocks of the same type of industry , stocks having large market capitalization , and stocks of firms having business in certain geographical areas @xcite . by calculating the scalar product of the eigenvectors from one time period to the next , we find that the `` deviating eigenvectors '' have varying degrees of time stability , quantified by the magnitude of the scalar product . the largest 2 - 3 eigenvectors are stable for extended periods of time , while for the rest of the deviating eigenvectors , the time stability decreases as the the corresponding eigenvalues are closer to the rmt upper bound . to test that the deviating eigenvalues are the only `` genuine '' information contained in * c * , we compare the eigenvalue statistics of * c * with the known universal properties of real symmetric random matrices , and we find good agreement with the rmt results . using the notion of the inverse participation ratio , we analyze the eigenvectors of * c * and find large values of inverse participation ratio at both edges of the eigenvalue spectrum suggesting a `` random band '' matrix structure for * c*. lastly , we discuss applications to the practical goal of finding an investment that provides a given return without exposure to unnecessary risk . in addition , it is possible that our methods can also be applied for filtering out ` noise ' in empirically - measured cross - correlation matrices in a wide variety of applications . this paper is organized as follows . section ii contains a brief description of the data analyzed . section iii discusses the statistics of cross - correlation coefficients . section iv discusses the eigenvalue distribution of * c * and compares with rmt results . section v tests the eigenvalue statistics * c * for universal properties of real symmetric random matrices and section vi contains a detailed analysis of the contents of eigenvectors that deviate from rmt . section vii discusses the time stability of the deviating eigenvectors . section viii contains applications of rmt methods to construct ` optimal ' portfolios that have a stable ratio of risk to return . finally , section ix contains some concluding remarks . we analyze two different databases covering securities from the three major us stock exchanges , namely the new york stock exchange ( nyse ) , the american stock exchange ( amex ) , and the national association of securities dealers automated quotation ( nasdaq ) . @xmath20 * database i : * we analyze the trades and quotes database , that documents all transactions for all major securities listed in all the three stock exchanges . we extract from this database time series of prices @xcite of the 1000 largest stocks by market capitalization on the starting date january 3 , 1994 . we analyze this database for the 2-yr period 199495 @xcite . from this database , we form @xmath21 records of 30-min returns of @xmath22 us stocks for the 2-yr period 199495 . we also analyze the prices of a subset comprising 881 stocks ( of those 1000 we analyze for 199495 ) that survived through two additional years 199697 . from this data , we extract @xmath21 records of 30-min returns of @xmath23 us stocks for the 2-yr period 199697 . @xmath20 * database ii : * we analyze the center for research in security prices ( crsp ) database . the crsp stock files cover common stocks listed on nyse beginning in 1925 , the amex beginning in 1962 , and the nasdaq beginning in 1972 . the files provide complete historical descriptive information and market data including comprehensive distribution information , high , low and closing prices , trading volumes , shares outstanding , and total returns . we analyze daily returns for the stocks that survive for the 35-yr period 196296 and extract @xmath24 records of 1-day returns for @xmath25 stocks . we analyze the distribution @xmath26 of the elements @xmath27 of the cross - correlation matrix * c * . we first examine @xmath26 for 30-min returns from the taq database for the 2-yr periods 199495 and 199697 [ fig . [ distcij](a ) ] . first , we note that @xmath26 is asymmetric and centered around a positive mean value ( @xmath28 ) , implying that positively - correlated behavior is more prevalent than negatively - correlated ( anti - correlated ) behavior . secondly , we find that @xmath29 depends on time , e.g. , the period 199697 shows a larger @xmath30 than the period 199495 . we contrast @xmath26 with a control a correlation matrix * r * with elements @xmath31 constructed from @xmath22 mutually - uncorrelated time series , each of length @xmath21 , generated using the empirically - found distribution of stock returns @xcite . figure [ distcij](a ) shows that @xmath32 is consistent with a gaussian with zero mean , in contrast to @xmath26 . in addition , we see that the part of @xmath26 for @xmath33 ( which corresponds to anti - correlations ) is within the gaussian curve for the control , suggesting the possibility that the observed negative cross - correlations in * c * may be an effect of randomness . figure [ distcij](b ) shows @xmath26 for daily returns from the crsp database for five non - overlapping 7-yr sub - periods in the 35-yr period 196296 . we see that the time dependence of @xmath30 is more pronounced in this plot . in particular , the period containing the market crash of october 19 , 1987 has the largest average value @xmath29 , suggesting the existence of cross - correlations that are more pronounced in volatile periods than in calm periods . we test this possibility by comparing @xmath30 with the average volatility of the market ( measured using the s&p 500 index ) , which shows large values of @xmath29 during periods of large volatility [ fig . [ volat ] ] . as stated above , our aim is to extract information about cross - correlations from * c*. so , we compare the properties of * c * with those of a random cross - correlation matrix @xcite . in matrix notation , the correlation matrix can be expressed as @xmath34 where * g * is an @xmath35 matrix with elements @xmath36 , and * g@xmath37 * denotes the transpose of * g*. therefore , we consider a `` random '' correlation matrix @xmath38 where * a * is an @xmath35 matrix containing @xmath39 time series of @xmath40 random elements with zero mean and unit variance , that are mutually uncorrelated . by construction * r * belongs to the type of matrices often referred to as wishart matrices in multivariate statistics @xcite . statistical properties of random matrices such as * r * are known @xcite . particularly , in the limit @xmath41 , such that @xmath42 is fixed , it was shown analytically @xcite that the distribution @xmath43 of eigenvalues @xmath44 of the random correlation matrix * r * is given by @xmath45 for @xmath44 within the bounds @xmath46 , where @xmath47 and @xmath48 are the minimum and maximum eigenvalues of * r * respectively , given by @xmath49 for finite @xmath40 and @xmath39 , the abrupt cut - off of @xmath43 is replaced by a rapidly - decaying edge @xcite . we next compare the eigenvalue distribution @xmath50 of * c * with @xmath51 @xcite . we examine @xmath52 min returns for @xmath22 stocks , each containing @xmath21 records . thus @xmath53 , and we obtain @xmath54 and @xmath55 from eq . ( [ lambdamaxmin ] ) . we compute the eigenvalues @xmath0 of * c * , where @xmath0 are rank ordered ( @xmath56 ) . figure [ evdist](a ) compares the probability distribution @xmath50 with @xmath51 calculated for @xmath53 . we note the presence of a well - defined `` bulk '' of eigenvalues which fall within the bounds @xmath57 $ ] for @xmath51 . we also note deviations for a few ( @xmath58 ) largest and smallest eigenvalues . in particular , the largest eigenvalue @xmath59 for the 2-yr period , which is @xmath60 times larger than @xmath61 . since eq . ( [ densuncorr ] ) is strictly valid only for @xmath62 and @xmath63 , we must test that the deviations that we find in fig . [ evdist](a ) for the largest few eigenvalues are not an effect of finite values of @xmath40 and @xmath39 . to this end , we contrast @xmath50 with the rmt result @xmath43 for the random correlation matrix of eq . ( [ matrixa ] ) , constructed from @xmath22 separate uncorrelated time series , each of the same length @xmath21 . we find good agreement with eq . ( [ densuncorr ] ) [ fig . [ evdist](b ) ] , thus showing that the deviations from rmt found for the largest few eigenvalues in fig . [ evdist](a ) are not a result of the fact that @xmath40 and @xmath39 are finite . figure [ evdist - daily ] compares @xmath50 for * c * calculated using @xmath64 daily returns of 422 stocks for the 7-yr period 199096 . we find a well - defined bulk of eigenvalues that fall within @xmath51 , and deviations from @xmath65 for large eigenvalues similar to what we found for @xmath66 min [ fig . [ evdist](a ) ] . thus , a comparison of @xmath50 with the rmt result @xmath43 allows us to distinguish the _ bulk _ of the eigenvalue spectrum of * c * that agrees with rmt ( random correlations ) from the deviations ( genuine correlations ) . the presence of a well - defined bulk of eigenvalues that agree with @xmath51 suggests that the contents of * c * are mostly random except for the eigenvalues that deviate . our conclusion was based on the comparison of the eigenvalue distribution @xmath50 of * c * with that of random matrices of the type * r = @xmath67 a a@xmath68*. quite generally , comparison of the eigenvalue distribution with @xmath51 alone is not sufficient to support the possibility that the bulk of the eigenvalue spectrum of * c * is random . random matrices that have drastically different @xmath50 share similar correlation structures in their eigenvalues universal properties that depend only on the general symmetries of the matrix @xcite . conversely , matrices that have the same eigenvalue distribution can have drastically different eigenvalue correlations . therefore , a test of randomness of * c * involves the investigation of correlations in the eigenvalues @xmath0 . since by definition * c * is a real symmetric matrix , we shall test the eigenvalue statistics * c * for universal features of eigenvalue correlations displayed by real symmetric random matrices . consider a @xmath69 real symmetric random matrix * s * with off - diagonal elements @xmath70 , which for @xmath71 are independent and identically distributed with zero mean @xmath72 and variance @xmath73 . it is conjectured based on analytical @xcite and extensive numerical evidence @xcite that in the limit @xmath74 , regardless of the distribution of elements @xmath70 , this class of matrices , on the scale of local mean eigenvalue spacing , display the universal properties ( eigenvalue correlation functions ) of the ensemble of matrices whose elements are distributed according to a gaussian probability measure called the gaussian orthogonal ensemble ( goe ) @xcite . formally , goe is defined on the space of real symmetric matrices by two requirements @xcite . the first is that the ensemble is invariant under orthogonal transformations , i.e. , for any goe matrix * z * , the transformation * z*@xmath75z@xmath76*w@xmath37 z w * , where * w * is any real orthogonal matrix ( * w w@xmath37=i * ) , leaves the joint probability @xmath77 of elements @xmath78 unchanged : @xmath79 . the second requirement is that the elements @xmath80 are statistically independent @xcite . by definition , random cross - correlation matrices * r * ( eq . ( [ matrixa ] ) ) that we are interested in are not strictly goe - type matrices , but rather belong to a special ensemble called the `` chiral '' goe @xcite . this can be seen by the following argument . define a matrix * b * @xmath81 \,.\end{aligned}\ ] ] the eigenvalues @xmath82 of * b * are given by @xmath83 and similarly , the eigenvalues @xmath44 of * r * are given by @xmath84 . thus , all non - zero eigenvalues of * b * occur in pairs , i.e. , for every eigenvalue @xmath44 of * r * , @xmath85 are eigenvalues of * b*. since the eigenvalues occur pairwise , the eigenvalue spectra of both * b * and * r * have special properties in the neighborhood of zero that are different from the standard goe @xcite . as these special properties decay rapidly as one goes further from zero , the eigenvalue correlations of * r * in the bulk of the spectrum are still consistent with those of the standard goe . therefore , our goal shall be to test the bulk of the eigenvalue spectrum of the empirically - measured cross - correlation matrix * c * with the known universal features of standard goe - type matrices . in the following , we test the statistical properties of the eigenvalues of * c * for three known universal properties @xcite displayed by goe matrices : ( i ) the distribution of nearest - neighbor eigenvalue spacings @xmath86 , ( ii ) the distribution of next - nearest - neighbor eigenvalue spacings @xmath87 , and ( iii ) the `` number variance '' statistic @xmath88 . the analytical results for the three properties listed above hold if the spacings between adjacent eigenvalues ( rank - ordered ) are expressed in units of _ average _ eigenvalue spacing . quite generally , the average eigenvalue spacing changes from one part of the eigenvalue spectrum to the next . so , in order to ensure that the eigenvalue spacing has a uniform _ average _ value throughout the spectrum , we must find a transformation called `` unfolding , '' which maps the eigenvalues @xmath0 to new variables called `` unfolded eigenvalues '' @xmath89 , whose distribution is uniform @xcite . unfolding ensures that the distances between eigenvalues are expressed in units of local mean eigenvalue spacing @xcite , and thus facilitates comparison with theoretical results . the procedures that we use for unfolding the eigenvalue spectrum are discussed in appendix a. we first consider the eigenvalue spacing distribution , which reflects two - point as well as eigenvalue correlation functions of all orders . we compare the eigenvalue spacing distribution of * c * with that of goe random matrices . for goe matrices , the distribution of `` nearest - neighbor '' eigenvalue spacings @xmath90 is given by @xcite @xmath91 often referred to as the `` wigner surmise '' @xcite . the gaussian decay of @xmath92 for large @xmath93 [ bold curve in fig . [ spacdist](a ) ] implies that @xmath92 `` probes '' scales only of the order of one eigenvalue spacing . thus , the spacing distribution is known to be robust across different unfolding procedures @xcite . we first calculate the distribution of the `` nearest - neighbor spacings '' @xmath94 of the unfolded eigenvalues obtained using the gaussian broadening procedure . figure [ spacdist](a ) shows that the distribution @xmath95 of nearest - neighbor eigenvalue spacings for * c * constructed from 30-min returns for the 2-yr period 199495 agrees well with the rmt result @xmath92 for goe matrices . identical results are obtained when we use the alternative unfolding procedure of fitting the eigenvalue distribution . in addition , we test the agreement of @xmath95 with rmt results by fitting @xmath96 to the one - parameter brody distribution @xcite @xmath97 where @xmath98^{1+\beta}$ ] . the case @xmath99 corresponds to the goe and @xmath100 corresponds to uncorrelated eigenvalues ( poisson - distributed spacings ) . we obtain @xmath101 , in good agreement with the goe prediction @xmath99 . to test non - parametrically that @xmath92 is the correct description for @xmath95 , we perform the kolmogorov - smirnov test . we find that at the 60@xmath102 confidence level , a kolmogorov - smirnov test can not reject the hypothesis that the goe is the correct description for @xmath95 . next , we analyze the nearest - neighbor spacing distribution @xmath96 for * c * constructed from daily returns for four 7-yr periods [ fig . [ spacdist - daily ] ] . we find good agreement with the goe result of eq . ( [ eq.3 ] ) , similar to what we find for * c * constructed from 30-min returns . we also test that both of the unfolding procedures discussed in appendix a yield consistent results . thus , we have seen that the eigenvalue - spacing distribution of empirically - measured cross - correlation matrices * c * is consistent with the rmt result for real symmetric random matrices . a second independent test for goe is the distribution @xmath103 of _ next_-nearest - neighbor spacings @xmath104 between the unfolded eigenvalues . for matrices of the goe type , according to a theorem due to ref . @xcite , the next - nearest neighbor spacings follow the statistics of the gaussian symplectic ensemble ( gse ) @xcite . in particular , the distribution of next - nearest - neighbor spacings @xmath105 for a goe matrix is identical to the distribution of nearest - neighbor spacings of the gaussian symplectic ensemble ( gse ) @xcite . figure [ spacdist](b ) shows that @xmath105 for the same data as fig . [ spacdist](a ) agrees well with the rmt result for the distribution of nearest - neighbor spacings of gse matrices , @xmath106 to probe for larger scales , pair correlations ( `` two - point '' correlations ) in the eigenvalues , we use the statistic @xmath88 often called the `` number variance , '' which is defined as the variance of the number of unfolded eigenvalues in intervals of length @xmath107 around each @xmath89 @xcite , @xmath108 ^ 2 \rangle_{\xi}\ , , \label{sigma2}\ ] ] where @xmath109 is the number of unfolded eigenvalues in the interval @xmath110 $ ] and @xmath111 denotes an average over all @xmath112 . if the eigenvalues are uncorrelated , @xmath113 . for the opposite extreme of a `` rigid '' eigenvalue spectrum ( e.g. simple harmonic oscillator ) , @xmath88 is a constant . quite generally , the number variance @xmath88 can be expressed as @xmath114 where @xmath115 ( called `` two - level cluster function '' ) is related to the two - point correlation function [ c.f . , @xcite , pp.79 ] . for the goe case , @xmath115 is explicitly given by @xmath116 where @xmath117 for large values of @xmath107 , the number variance @xmath88 for goe has the `` intermediate '' behavior @xmath118 figure [ numbervar ] shows that @xmath119 for * c * calculated using 30-min returns for 199495 agrees well with the rmt result of eq . ( [ defsigmagoe ] ) . for the range of @xmath107 shown in fig . [ numbervar ] , both unfolding procedures yield similar results . consistent results are obtained for * c * constructed from daily returns . to summarize this section , we have tested the statistics of * c * for universal features of eigenvalue correlations displayed by goe matrices . we have seen that the distribution of the nearest - neighbor spacings @xmath95 is in good agreement with the goe result . to test whether the eigenvalues of * c * display the rmt results for long - range two - point eigenvalue correlations , we analyzed the number variance @xmath88 and found good agreement with goe results . moreover , we also find that the statistics of next - nearest neighbor spacings conform to the predictions of rmt . these findings show that the statistics of the _ bulk _ of the eigenvalues of the empirical cross - correlation matrix * c * is consistent with those of a real symmetric random matrix . thus , information about genuine correlations are contained in the deviations from rmt , which we analyze below . the deviations of @xmath50 from the rmt result @xmath65 suggests that these deviations should also be displayed in the statistics of the corresponding eigenvector components @xcite . accordingly , in this section , we analyze the distribution of eigenvector components . the distribution of the components @xmath120 of eigenvector * u@xmath121 * of a random correlation matrix * r * should conform to a gaussian distribution with mean zero and unit variance @xcite , @xmath122 first , we compare the distribution of eigenvector components of * c * with eq . ( [ port - thomas ] ) . we analyze @xmath123 for * c * computed using 30-min returns for 199495 . we choose one typical eigenvalue @xmath124 from the bulk ( @xmath125 ) defined by @xmath51 of eq . ( [ densuncorr ] ) . figure [ distevec](a ) shows that @xmath123 for a typical * u@xmath121 * from the bulk shows good agreement with the rmt result @xmath126 . similar analysis on the other eigenvectors belonging to eigenvalues within the bulk yields consistent results , in agreement with the results of the previous sections that the bulk agrees with random matrix predictions . we test the agreement of the distribution @xmath123 with @xmath126 by calculating the kurtosis , which for a gaussian has the value @xmath127 . we find significant deviations from @xmath126 for @xmath128 largest and smallest eigenvalues . the remaining eigenvectors have values of kurtosis that are consistent with the gaussian value @xmath127 . consider next the `` deviating '' eigenvalues @xmath0 , larger than the rmt upper bound , @xmath129 . figure [ distevec](b ) and ( c ) show that , for deviating eigenvalues , the distribution of eigenvector components @xmath123 deviates systematically from the rmt result @xmath126 . finally , we examine the distribution of the components of the eigenvector * u@xmath130 * corresponding to the largest eigenvalue @xmath131 . figure [ distevec](d ) shows that @xmath132 deviates remarkably from a gaussian , and is approximately uniform , suggesting that all stocks participate . in addition , we find that almost all components of * u@xmath130 * have the same sign , thus causing @xmath123 to shift to one side . this suggests that the significant participants of eigenvector * u@xmath121 * have a common component that affects all of them with the same bias . since all components participate in the eigenvector corresponding to the largest eigenvalue , it represents an influence that is common to all stocks . thus , the largest eigenvector quantifies the qualitative notion that certain newsbreaks ( e.g. , an interest rate increase ) affect all stocks alike @xcite . one can also interpret the largest eigenvalue and its corresponding eigenvector as the collective ` response ' of the entire market to stimuli . we quantitatively investigate this notion by comparing the projection ( scalar product ) of the time series * g * on the eigenvector * u@xmath130 * , with a standard measure of us stock market performance the returns @xmath133 of the s&p 500 index . we calculate the projection @xmath134 of the time series @xmath135 on the eigenvector * u@xmath130 * , @xmath136 by definition , @xmath134 shows the return of the portfolio defined by * u@xmath130*. we compare @xmath134 with @xmath137 , and find remarkably similar behavior for the two , indicated by a large value of the correlation coefficient @xmath138 . figure [ sp - market ] shows @xmath134 regressed against @xmath139 , which shows relatively narrow scatter around a linear fit . thus , we interpret the eigenvector * u@xmath130 * as quantifying market - wide influences on all stocks @xcite . we analyze * c * at larger time scales of @xmath140 day and find similar results as above , suggesting that similar correlation structures exist for quite different time scales . our results for the distribution of eigenvector components agree with those reported in ref . @xcite , where @xmath141 day returns are analyzed . we next investigate how the largest eigenvalue changes as a function of time . figure [ volat ] shows the time dependence @xcite of the largest eigenvalue ( @xmath142 ) for the 35-yr period 196296 . we find large values of the largest eigenvalue during periods of high market volatility , which suggests strong collective behavior in regimes of high volatility . one way of statistically modeling an influence that is common to all stocks is to express the return @xmath9 of stock @xmath6 as @xmath143 where @xmath144 is an additive term that is the same for all stocks , @xmath145 , @xmath146 and @xmath147 are stock - specific constants , and @xmath148 . this common term @xmath144 gives rise to correlations between any pair of stocks . the decomposition of eq . ( [ marketm ] ) forms the basis of widely - used economic models , such as multi - factor models and the capital asset pricing model @xcite . since * u@xmath130 * represents an influence that is common to all stocks , we can approximate the term @xmath144 with @xmath134 . the parameters @xmath146 and @xmath147 can therefore be estimated by an ordinary least squares regression . next , we remove the contribution of @xmath134 to each time series @xmath149 , and construct * c * from the residuals @xmath150 of eq . ( [ marketm ] ) . figure [ corr - ev1000 ] shows that the distribution @xmath26 thus obtained has significantly smaller average value @xmath29 , showing that a large degree of cross - correlations contained in * c * can be attributed to the influence of the largest eigenvalue ( and its corresponding eigenvector ) @xcite . having studied the interpretation of the largest eigenvalue which deviates significantly from rmt results , we next focus on the remaining eigenvalues . the deviations of the distribution of components of an eigenvector * u@xmath121 * from the rmt prediction of a gaussian is more pronounced as the separation from the rmt upper bound @xmath151 increases . since proximity to @xmath48 increases the effects of randomness , we quantify the number of components that participate significantly in each eigenvector , which in turn reflects the degree of deviation from rmt result for the distribution of eigenvector components . to this end , we use the notion of the inverse participation ratio ( ipr ) , often applied in localization theory @xcite . the ipr of the eigenvector * u@xmath121 * is defined as @xmath152~^4\ , , \label{defipr}\ ] ] where @xmath153 , @xmath154 are the components of eigenvector * u@xmath121*. the meaning of @xmath155 can be illustrated by two limiting cases : ( i ) a vector with identical components @xmath156 has @xmath157 , whereas ( ii ) a vector with one component @xmath158 and the remainder zero has @xmath159 . thus , the ipr quantifies the reciprocal of the number of eigenvector components that contribute significantly . figure [ ipr](a ) shows @xmath155 for the case of the control of eq . ( [ matrixa ] ) using time series with the empirically - found distribution of returns @xcite . the average value of @xmath155 is @xmath160 with a narrow spread , indicating that the vectors are _ extended _ @xcite i.e . , almost all components contribute to them . fluctuations around this average value are confined to a narrow range ( standard deviation of @xmath161 ) . figure [ ipr](b ) shows that @xmath155 for * c * constructed from 30-min returns from the period 199495 , agrees with @xmath155 of the random control in the bulk ( @xmath162 ) . in contrast , the edges of the eigenvalue spectrum of * c * show significant deviations of @xmath155 from @xmath163 . the largest eigenvalue has @xmath164 for the 30-min data [ fig . [ ipr](b ) ] and @xmath165 for the 1-day data [ fig . [ ipr](c ) and ( d ) ] , showing that almost all stocks participate in the largest eigenvector . for the rest of the large eigenvalues which deviate from the rmt upper bound , @xmath155 values are approximately 4 - 5 times larger than @xmath163 , showing that there are varying numbers of stocks contributing to these eigenvectors . in addition , we also find that there are large @xmath155 values for vectors corresponding to few of the small eigenvalues @xmath166 . the deviations at both edges of the eigenvalue spectrum are considerably larger than @xmath167 , which suggests that the vectors are _ localized _ @xcite i.e . , only a few stocks contribute to them . the presence of vectors with large values of @xmath155 also arises in the theory of anderson localization@xcite . in the context of localization theory , one frequently finds `` random band matrices''@xcite containing extended states with small @xmath155 in the bulk of the eigenvalue spectrum , whereas edge states are localized and have large @xmath155 . our finding of localized states for small and large eigenvalues of the cross - correlation matrix * c * is reminiscent of anderson localization and suggests that * c * may have a random band matrix structure . a random band matrix * b * has elements @xmath168 independently drawn from different probability distributions . these distributions are often taken to be gaussian parameterized by their variance , which depends on @xmath6 and @xmath169 . although such matrices are random , they still contain probabilistic information arising from the fact that a metric can be defined on their set of indices @xmath6 . a related , but distinct way of analyzing cross - correlations by defining ` ultra - metric ' distances has been studied in ref . @xcite . we quantify the number of significant participants of an eigenvector using the ipr , and we examine the @xmath172 components of eigenvector * u@xmath121 * for common features @xcite . a direct examination of these eigenvectors , however , does not yield a straightforward interpretation of their economic relevance . to interpret their meaning , we note that the largest eigenvalue is an order of magnitude larger than the others , which constrains the remaining @xmath173 eigenvalues since tr * c * @xmath174 . thus , in order to analyze the deviating eigenvectors , we must remove the effect of the largest eigenvalue @xmath131 . in order to avoid the effect of @xmath131 , and thus @xmath134 , on the returns of each stock @xmath149 , we perform the regression of eq . ( [ marketm ] ) , and compute the residuals @xmath150 . we then calculate the correlation matrix * c * using @xmath150 in eq . ( [ norm - ret ] ) and eq . ( [ eq.2 ] ) . next , we compute the eigenvectors * u@xmath121 * of * c * thus obtained , and analyze their significant participants . the eigenvector * u@xmath171 * contains approximately @xmath175 significant participants , which are all stocks with large values of market capitalization . figure [ figmc ] shows that the magnitude of the eigenvector components of * u@xmath171 * shows an approximately logarithmic dependence on the market capitalizations of the corresponding stocks . we next analyze the significant contributors of the rest of the eigenvectors . we find that each of these deviating eigenvectors contains stocks belonging to similar or related industries as significant contributors . table [ industry.tab ] shows the ticker symbols and industry groups ( standard industry classification ( sic ) code ) for stocks corresponding to the ten largest eigenvector components of each eigenvector . we find that these eigenvectors partition the set of all stocks into distinct groups which contain stocks with large market capitalization ( * u@xmath171 * ) , stocks of firms in the electronics and computer industry ( * u@xmath176 * ) , a combination of gold mining and investment firms ( * u@xmath177 * and * u@xmath178 * ) , banking firms ( * u@xmath179 * ) , oil and gas refining and equipment ( * u@xmath180 * ) , auto manufacturing firms ( * u@xmath181 * ) , drug manufacturing firms ( * u@xmath182 * ) , and paper manufacturing ( * u@xmath170 * ) . one eigenvector ( * u@xmath183 * ) displays a mixture of three industry groups telecommunications , metal mining , and banking . an examination of these firms shows significant business activity in latin america . our results are also represented schematically in fig . [ evindustry ] . a similar classification of stocks into sectors using different methods is obtained in ref . @xcite . instead of performing the regression of eq ( [ marketm ] ) , one can remove the u - shaped intra - daily pattern using the procedure of ref @xcite and compute * c*. the results thus obtained are consistent with those obtained using the procedure of using the residuals of the regression of eq . ( [ marketm ] ) to compute * c * ( table [ industry.tab ] ) . often * c * is constructed from returns at longer time scales of @xmath140 week or 1 month to avoid short time scale effects @xcite . having examined the largest eigenvalues , we next focus on the smallest eigenvalues which show large values of @xmath155 [ fig . [ ipr ] ] . we find that the eigenvectors corresponding to the smallest eigenvalues contain as significant participants , pairs of stocks which have the largest values of @xmath12 in our sample . for example , the two largest components of * u@xmath184 * correspond to the stocks of texas instruments ( txn ) and micron technology ( mu ) with @xmath185 , the largest correlation coefficient in our sample . the largest components of * u@xmath186 * are telefonos de mexico ( tmx ) and grupo televisa ( tv ) with @xmath187 ( second largest correlation coefficient ) . the eigenvector * u@xmath188 * shows newmont gold company ( ngc ) and newmont mining corporation ( nem ) with @xmath189 ( third largest correlation coefficient ) as largest components . in all three eigenvectors , the relative sign of the two largest components is _ negative_. thus pairs of stocks with a correlation coefficient much larger than the average @xmath29 effectively `` decouple '' from other stocks . the appearance of strongly correlated pairs of stocks in the eigenvectors corresponding to the smallest eigenvalues of * c * can be qualitatively understood by considering the example of a @xmath190 cross - correlation matrix @xmath191 \ \ .\end{aligned}\ ] ] the eigenvalues of * c*@xmath192 are @xmath193 . the smaller eigenvalue @xmath194 decreases monotonically with increasing cross - correlation coefficient @xmath195 . the corresponding eigenvector is the anti - symmetric linear combination of the basis vectors @xmath196 and @xmath197 , in agreement with our empirical finding that the relative sign of largest components of eigenvectors corresponding to the smallest eigenvalues is negative . in this simple example , the symmetric linear combination of the two basis vectors appears as the eigenvector of the large eigenvalue @xmath198 . indeed , we find that txn and mu are the largest components of * u@xmath176 * , tmx and tv are the largest components of * u@xmath183 * , and nem and ngc are the largest and third largest components of * u@xmath178*. we next investigate the degree of stability in time of the eigenvectors corresponding to the eigenvalues that deviate from rmt results . since deviations from rmt results imply genuine correlations which remain stable in the period used to compute * c * , we expect the deviating eigenvectors to show some degree of time stability . we first identify the @xmath199 eigenvectors corresponding to the @xmath199 largest eigenvalues which deviate from the rmt upper bound @xmath200 . we then construct a @xmath201 matrix * d * with elements @xmath202 . next , we compute a @xmath203 `` overlap matrix '' * o(@xmath204 ) = d@xmath205 d@xmath206 * , with elements @xmath207 defined as the scalar product of eigenvector * u@xmath208 * of period a ( starting at time @xmath209 ) with * u@xmath210 * of period b at a later time @xmath211 , @xmath212 if all the @xmath199 eigenvectors are `` perfectly '' non - random and stable in time @xmath213 . we study the overlap matrices * o * using both high - frequency and daily data . for high - frequency data ( @xmath21 records at 30-min intervals ) , we use a moving window of length @xmath214 , and slide it through the entire 2-yr period using discrete time steps @xmath215 . we first identify the eigenvectors of the correlation matrices for each of these time periods . we then calculate overlap matrices * o(@xmath216 ) * , where @xmath217 , between the eigenvectors for @xmath218 and for @xmath219 . figure [ overlap - taq ] shows a grey scale pixel - representation of the matrix * o ( @xmath220 ) * , for different @xmath221 . first , we note that the eigenvectors that deviate from rmt bounds show varying degrees of stability ( @xmath222 ) in time . in particular , the stability in time is largest for * u@xmath130*. even at lags of @xmath223 yr the corresponding overlap @xmath224 . the remaining eigenvectors show decreasing amounts of stability as the rmt upper bound @xmath48 is approached . in particular , the 3 - 4 largest eigenvectors show large values of @xmath207 for up to @xmath225 yr . next , we repeat our analysis for daily returns of 422 stocks using @xmath226 records of 1-day returns , and a sliding window of length @xmath227 with discrete time steps @xmath228 days . instead of calculating * o(@xmath220 ) * for all starting points @xmath229 , we calculate * o(@xmath221)*@xmath230 * o(@xmath204 ) * @xmath231 , averaged over all @xmath232 , where @xmath233 . figure [ overlap - daily ] shows grey scale representations of * o ( @xmath221 ) * for increasing @xmath221 . we find similar results as found for shorter time scales , and find that eigenvectors corresponding to the largest 2 eigenvalues are stable for time scales as large as @xmath23420 yr . in particular , the eigenvector * u@xmath235 * shows an overlap of @xmath236 even over time scales of @xmath23430 yr . the randomness of the `` bulk '' seen in the previous sections has implications in optimal portfolio selection @xcite . we illustrate these using the markowitz theory of optimal portfolio selection @xcite . consider a portfolio @xmath237 of stocks with prices @xmath238 . the return on @xmath237 is given by @xmath239 where @xmath149 is the return on stock @xmath6 and @xmath240 is the fraction of wealth invested in stock @xmath6 . the fractions @xmath240 are normalized such that @xmath241 . the risk in holding the portfolio @xmath237 can be quantified by the variance @xmath242 where @xmath243 is the standard deviation ( average volatility ) of @xmath9 , and @xmath12 are elements of the cross - correlation matrix * c*. in order to find an optimal portfolio , we must minimize @xmath244 under the constraint that the return on the portfolio is some fixed value @xmath245 . in addition , we also have the constraint that @xmath246 . minimizing @xmath244 subject to these two constraints can be implemented by using two lagrange multipliers , which yields a system of linear equations for @xmath240 , which can then be solved . the optimal portfolios thus chosen can be represented as a plot of the return @xmath245 as a function of risk @xmath244 [ fig . [ rr ] ] . to find the effect of randomness of * c * on the selected optimal portfolio , we first partition the time period 199495 into two one - year periods . using the cross - correlation matrix * c@xmath247 * for 1994 , and @xmath9 for 1995 , we construct a family of optimal portfolios , and plot @xmath245 as a function of the predicted risk @xmath248 for 1995 [ fig . [ rr](a ) ] . for this family of portfolios , we also compute the risk @xmath249 realized during 1995 using * c@xmath250 * [ fig . [ rr](a ) ] . we find that the predicted risk is significantly smaller when compared to the realized risk , @xmath251 since the meaningful information in * c * is contained in the deviating eigenvectors ( whose eigenvalues are outside the rmt bounds ) , we must construct a ` filtered ' correlation matrix * c@xmath252 * , by retaining only the deviating eigenvectors . to this end , we first construct a diagonal matrix @xmath253 , with elements @xmath254 . we then transform @xmath253 to the basis of * c * , thus obtaining the ` filtered ' cross - correlation matrix * c@xmath252*. in addition , we set the diagonal elements @xmath255 , to preserve @xmath256(*c * ) @xmath257 @xmath258(*c*@xmath252 ) @xmath174 . we repeat the above calculations for finding the optimal portfolio using * c@xmath252 * instead of * c * in eq . ( [ varport ] ) . figure [ rr](b ) shows that the realized risk is now much closer to the predicted risk @xmath259 thus , the optimal portfolios constructed using * c@xmath252 * are significantly more stable in time . how can we understand the deviating eigenvalues i.e. , correlations that are stable in time ? one approach is to postulate that returns can be separated into idiosyncratic and common components i.e. , that returns can be separated into different additive `` factors '' , which represent various economic influences that are common to a set of stocks such as the type of industry , or the effect of news @xcite . on the other hand , in physical systems one starts from the interactions between the constituents , and then relates interactions to correlated `` modes '' of the system . in economic systems , we ask if a similar mechanism can give rise to the correlated behavior . in order to answer this question , we model stock price dynamics by a family of stochastic differential equations @xcite , which describe the ` instantaneous' returns @xmath260 as a random walk with couplings @xmath261 @xmath262 here , @xmath263 are gaussian random variables with correlation function @xmath264 , and @xmath265 sets the time scale of the problem . in the context of a soft spin model , the first two terms in the _ rhs _ of eq . ( [ langevin ] ) arise from the derivative of a double - well potential , enforcing the soft spin constraint . the interaction among soft - spins is given by the couplings @xmath261 . in the absence of the cubic term , and without interactions , @xmath266 are relaxation times of the @xmath267 correlation function . the return @xmath9 at a finite time interval @xmath3 is given by the integral of @xmath268 over @xmath3 . equation ( [ langevin ] ) is similar to the linearized description of interacting `` soft spins '' @xcite and is a generalized case of the models of refs . @xcite . without interactions , the variance of price changes on a scale @xmath269 is given by @xmath270 , in agreement with recent studies @xcite , where stock price changes are described by an anomalous diffusion and the variance of price changes is decomposed into a product of trading frequency ( analog of @xmath271 ) and the square of an `` impact parameter '' which is related to liquidity ( analog of @xmath272 ) . as the coupling strengths increase , the soft - spin system undergoes a transition to an ordered state with permanent local magnetizations . at the transition point , the spin dynamics are very `` slow '' as reflected in a power law decay of the spin autocorrelation function in time . to test whether this signature of strong interactions is present for the stock market problem , we analyze the correlation functions @xmath273 , where @xmath274 is the time series defined by eigenvector * u@xmath121*. instead of analyzing @xmath275 directly , we apply the detrended fluctuation analysis ( dfa ) method @xcite . figure [ autocorr ] shows that the correlation functions @xmath275 indeed decay as power laws @xcite for the deviating eigenvectors * u@xmath121 * in sharp contrast to the behavior of @xmath275 for the rest of the eigenvectors and the autocorrelation functions of individual stocks , which show only short - ranged correlations . we interpret this as evidence for strong interactions @xcite . in the absence of the non - linearities ( cubic term ) , we obtain only exponentially - decaying correlation functions for the `` modes '' corresponding to the large eigenvalues , which is inconsistent with our finding of power - law correlations . to summarize , we have tested the eigenvalue statistics of the empirically - measured correlation matrix * c * against the null hypothesis of a random correlation matrix . this allows us to distinguish genuine correlations from `` apparent '' correlations that are present even for random matrices . we find that the bulk of the eigenvalue spectrum of * c * shares universal properties with the gaussian orthogonal ensemble of random matrices . further , we analyze the deviations from rmt , and find that ( i ) the largest eigenvalue and its corresponding eigenvector represent the influence of the entire market on all stocks , and ( ii ) using the rest of the deviating eigenvectors , we can partition the set of all stocks studied into distinct subsets whose identity corresponds to conventionally - identified business sectors . these sectors are stable in time , in some cases for as many as 30 years . finally , we have seen that the deviating eigenvectors are useful for the construction of optimal portfolios which have a stable ratio of risk to return . we thank j - p . bouchaud , s. v. buldyrev , p. cizeau , e. derman , x. gabaix , j. hill , m. janjusevic , l. viciera , and j. zou for helpful discussions . we thank o. bohigas for pointing out ref . @xcite to us . br thanks dfg grant ro1 - 1/2447 for financial support . tg thanks boston university for warm hospitality . the center for polymer studies is supported by the nsf , british petroleum , the nih , and the nrcps ( ps1 rr13622 ) . as discussed in section v , random matrices display _ universal _ functional forms for eigenvalue correlations that depend only on the general symmetries of the matrix . a first step to test the data for such universal properties is to find a transformation called `` unfolding , '' which maps the eigenvalues @xmath0 to new variables called `` unfolded eigenvalues '' @xmath89 , whose distribution is uniform @xcite . unfolding ensures that the distances between eigenvalues are expressed in units of _ local _ mean eigenvalue spacing @xcite , and thus facilitates comparison with analytical results . we first define the cumulative distribution function of eigenvalues , which counts the number of eigenvalues in the interval @xmath276 , @xmath277 where @xmath278 denotes the probability density of eigenvalues and @xmath39 is the total number of eigenvalues . the function @xmath279 can be decomposed into an average and a fluctuating part , @xmath280 since @xmath281 on average , @xmath282 is the averaged eigenvalue density . the dimensionless , unfolded eigenvalues are then given by @xmath283 thus , the problem is to find @xmath284 . we follow two procedures for obtaining the unfolded eigenvalues @xmath89 : ( i ) a phenomenological procedure referred to as gaussian broadening @xcite , and ( ii ) fitting the cumulative distribution function @xmath279 of eq . ( [ staircase ] ) with the analytical expression for @xmath279 using eq . ( [ densuncorr ] ) . these procedures are discussed below . gaussian broadening @xcite is a phenomenological procedure that aims at approximating the function @xmath284 defined in eq . [ stair.dec ] using a series of gaussian functions . consider the eigenvalue distribution @xmath50 , which can be expressed as @xmath285 the @xmath286-functions about each eigenvalue are approximated by choosing a gaussian distribution centered around each eigenvalue with standard deviation @xmath287 , where @xmath288 is the size of the window used for broadening @xcite . integrating eq . ( [ pdfdelta ] ) provides an approximation to the function @xmath289 in the form of a series of error functions , which using eq . ( [ defunfold ] ) yields the unfolded eigenvalues . phenomenological procedures are likely to contain artificial scales , which can lead to an `` over - fitting '' of the smooth part @xmath289 by adding contributions from the fluctuating part @xmath290 . the second procedure for unfolding aims at circumventing this problem by fitting the cumulative distribution of eigenvalues @xmath279 ( eq . ( [ staircase ] ) ) with the analytical expression for @xmath291 where @xmath51 is the probability density of eigenvalues from eq . ( [ densuncorr ] ) . the fit is performed with @xmath47 , @xmath48 , and @xmath39 as free parameters . the fitted function is an estimate for @xmath284 , whereby we obtain the unfolded eigenvalues @xmath89 . one difficulty with this method is that the deviations of the spectrum of * c * from eq . ( [ densuncorr ] ) can be quite pronounced in certain periods , and it is difficult to find a good fit of the cumulative distribution of eigenvalues to eq . ( [ sengupta - stair ] ) . r. n. mantegna , eur . j. b * 11 * , 193 ( 1999 ) ; l. kullmann , j. kertsz , and r. n. mantegna , e - print cond - mat/0002238 ; an interesting analysis of cross - correlation between stock market indices can be found in g. bonanno , n. vandewalle , r. n. mantegna , e - print cond - mat/0001268 . v. plerou , p. gopikrishnan , l. a. n. amaral , m. meyer , and h. e. stanley , phys . e * 60 * , 6519 ( 1999 ) ; p. gopikrishnan , v. plerou , l. a. n. amaral , m. meyer , and h. e. stanley , phys . e * 60 * , 5305 ( 1999 ) ; p. gopikrishnan , m. meyer , l.a.n . amaral , and h. e. stanley , eur . j. b * 3 * , 139 ( 1998 ) . for goe matrices , the wigner surmise ( eq . ( [ eq.3 ] ) ) is not exact @xcite . despite this fact , the wigner surmise is widely used because the difference between the exact form of @xmath292 and eq . ( [ eq.3 ] ) is almost negligible @xcite . analogous to the goe , which is defined on the space of real symmetric matrices , one can define two other ensembles @xcite : ( i ) the gaussian unitary ensemble ( gue ) , which is defined on the space of hermitian matrices , with the requirement that the joint probability of elements is invariant under unitary transformations , and ( ii ) the gaussian symplectic ensemble ( gse ) , which is defined on the space of hermitian `` self - dual '' matrices with the requirement that the joint probability of elements is invariant under symplectic transformations . formal definitions can be found in ref . @xcite . y. v. fyodorov and a. d. mirlin , phys . 69 * , 1093 ( 1992 ) ; * 71 * , 412 ( 1993 ) ; int b * 8 * , 3795 ( 1994 ) ; a. d. mirlin and y. v. fyodorov , j. phys . a : math . gen . * 26 * , l551 ( 1993 ) ; e. p. wigner , ann . math . * 62 * , 548 ( 1955 ) . metals or semiconductors with impurities can be described by hamiltonians with random - hopping integrals [ f. wegner and r. oppermann , z. physik * b34 * , 327 ( 1979 ) ] . electron - hopping between neighboring sites is more probable than hopping over large distances , leading to a hamiltonian that is a random band matrix . y. liu , p. gopikrishnan , p. cizeau , c .- k . peng , m. meyer , and h. e. stanley , phys . e * 60 * , 1390 ( 1999 ) ; y. liu , p. cizeau , m. meyer , c .- k . peng , and h. e. stanley , physica a * 245 * , 437 ( 1997 ) ; p. cizeau , y. liu , m. meyer , c .- k . peng , and h. e. stanley , physica a * 245 * , 441 ( 1997 ) . in contrast , the autocorrelation function for the s&p 500 index returns @xmath293 displays only correlations on short - time scales of @xmath294 min , beyond which the autocorrelation function is at the level of noise @xcite . on the other hand , the returns for individual stocks have pronounced anti - correlations on short time scales ( @xmath295 min ) , which is an effect of the bid - ask bounce @xcite . for certain portfolios of stocks , returns are found to have long memory [ a. lo , econometrica * 59 * , 1279 ( 1991 ) ] . for the case of predominantly `` ferromagnetic '' couplings ( @xmath296 ) within disjoint groups of stocks , a factor model ( such as eq . ( [ marketm ] ) ) can be derived from the model of `` interacting stocks '' in eq . ( [ langevin ] ) . in the spirit of a mean - field approximation , the influence of the price changes of all other stocks in a group on the price of a given stock can be modeled by an effective field , which has to be calculated self - consistently . this effective field would then play a similar role as a factor in standard economic models . llc & & + + + xon & oil & gas equipment / services & 2911 + pg & cleaning products & 2840 + jnj & drug manufacturers / major & 2834 + ko & beverages - soft drinks & 2080 + pfe & drug manufacturers / major & 2834 + bel & telecom services / domestic & 4813 + mob & oil & gas equipment / services & 2911 + ben & asset management & 6282 + un & food - major diversified & 2000 + aig & property / casualty insurance & 6331 + + + txn & semiconductor - broad line & 3674 + mu & semiconductor - memory chips & 3674 + lsi & semiconductor - specialized & 3674 + mot & electronic equipment & 3663 + cpq & personal computers & 3571 + cy & semiconductor - broad line & 3674 + ter & semiconductor equip / materials & 3825 + nsm & semiconductor - broad line & 3674 + hwp & diversified computer systems & 3570 + ibm & diversified computer systems & 3570 + + + pdg & gold & 1040 + nem & gold & 1040 + ngc & gold & 1040 + abx & gold & 1040 + asa & closed - end fund - ( gold ) & 6799 + hm & gold & 1040 + bmg & gold & 1040 + au & gold & 1040 + hsm & general building materials & 5210 + mu & semiconductor - memory chips & 3674 + + + nem & gold & 1040 + pdg & gold & 1040 + abx & gold & 1040 + hm & gold & 1040 + ngc & gold & 1040 + asa & closed - end fund - ( gold ) & 6799 + bmg & gold & 1040 + chl & wireless communications & 4813 + cmb & money center banks & 6021 + cci & money center banks & 6021 + + + tmx & telecommunication services / foreign & 4813 + tv & broadcasting - television & 4833 + mxf & closed - end fund - foreign & 6726 + ica & heavy construction & 1600 + gtr & heavy construction & 1600 + ctc & telecom services / foreign & 4813 + pb & beverages - soft drinks & 2086 + ypf & independent oil & gas & 2911 + txn & semiconductor - broad line & 3674 + mu & semiconductor - memory chips & 3674 + + + bac & money center banks & 6021 + chl & wireless communications & 4813 + bk & money center banks & 6022 + cci & money center banks & 6021 + cmb & money center banks & 6021 + bt & money center banks & 6022 + jpm & money center banks & 6022 + mel & regional - northeast banks & 6021 + nb & money center banks & 6021 + wfc & money center banks & 6021 + + + bp & oil & gas equipment / services & 2911 + mob & oil & gas equipment / services & 2911 + slb & oil & gas equipment / services & 1389 + tx & major integrated oil / gas & 2911 + ucl & oil & gas refining / marketing & 1311 + arc & oil & gas equipment / services & 2911 + bhi & oil & gas equipment / services & 3533 + chv & major integrated oil / gas & 2911 + apc & independent oil & gas & 1311 + an & auto dealerships & 2911 + + + fpr & auto manufacturers / major & 3711 + f & auto manufacturers / major & 3711 + c & auto manufacturers / major & 3711 + gm & auto manufacturers / major & 3711 + txn & semiconductor - broad line & 3674 + adi & semiconductor - broad line & 3674 + cy & semiconductor - broad line & 3674 + ter & semiconductor equip / materials & 3825 + mga & auto parts & 3714 + lsi & semiconductor - specialized & 3674 + + + abt & drug manufacturers / major & 2834 + pfe & drug manufacturers / major & 2834 + sgp & drug manufacturers / major & 2834 + lly & drug manufacturers / major & 2834 + jnj & drug manufacturers / major & 2834 + ahc & oil & gas refining / marketing & 2911 + bmy & drug manufacturers / major & 2834 + hal & oil & gas equipment / services & 1600 + wla & drug manufacturers / major & 2834 + bhi & oil & gas equipment / services & 3533 +

we study stochastic particle systems made up of heterogeneous units . we introduce a general framework suitable to analytically study this kind of systems and apply it to two particular models of interest in economy and epidemiology . we show that particle heterogeneity can enhance or decrease the size of the collective fluctuations depending on the system , and that it is possible to infer the degree and the form of the heterogeneity distribution in the system by measuring only global variables and their fluctuations . our work shows that , in some cases , heterogeneity among the units composing a systems can be fully taken into account without losing analytical tractability most real systems are made up of heterogeneous units . whether considering a population of cells , a group of people or an array of lasers ( to name just a few examples ) , one never finds two units which behave exactly in the same way . despite this general fact , quantitative modeling most often assumes identical units , since this condition seems necessary for having analytically tractable models . moreover , in the general framework of complexity science , systems very often can be modeled only at a stochastic level , since a complete knowledge of all the variables , the precise dynamics of the units and the interaction with the environment is not available . one way to include system heterogeneity is to consider that the interactions between the units are not homogeneous but mediated by some complex network , an approach that has attracted enormous attention in the last years . an issue that has been less studied , beyond the role of particle heterogeneity in deterministic systems , is the heterogeneity in the behavior of the particles themselves in stochastic models . some exceptions include the recent reference , where the authors analyze the effect of heterogeneous transition rates on consensus times in the voter model , and works considering the effect of a few `` committed '' individuals in this and related models . in the context of statistical physics , the combined effects of stochasticity and heterogeneity have been considered , for example , in random - field ising models and spin glasses or in diffusion in disordered media . we aim here at developing a general framework for the analytical study of stochastic systems made up of heterogeneous units , applicable beyond equilibrium models or hamiltonian systems and suitable for a general class of complex systems of recent interest and at identifying some generic effects of particle heterogeneity on the macroscopic fluctuations . in this work we will show that the combined effect of stochasticity and heterogeneity can give rise to unexpected , non - trivial , results . while , based on nave arguments , one should conclude that global fluctuations increase in heterogeneous systems , we will show that in some systems of stochastic interacting particles fluctuations actually decrease with the degree of heterogeneity . moreover , we will see that it is possible to infer the degree of particle heterogeneity ( or `` diversity '' ) by measuring only global variables . this is an issue of great interest when one has access only to information at the macroscopic , population level , since it allows one to determine if heterogeneity is a relevant ingredient that needs to be included in the modeling . in this way , heterogeneity can be included when its presence is implied by the data and it does not enter as an extra free parameter . we will study first the simple case of independent particles ; then we will consider the general case of interacting particles and develop an approximated method of general validity to analytically study these systems ; next , as a way of example , this method will be applied to two particular models of interest in economy and epidemiology . our starting point is a stochastic description of a system composed by non - identical units , which we call generically `` particles '' or agents " . each particle is characterized by a constant parameter ( ) ; the value of this parameter differs among the particles and it is the source of heterogeneity considered . although there are more general ways of including heterogeneity , we will stick to this type of parametric heterogeneity because it is simple yet rather general . for simplicity , we assume that each particle can be in one of two possible states and define as the variable describing the state of particle at time ( the two - states assumption will be relaxed later ) . the collective state of the system is given by the total number of particles in state . sometimes , one does not have access to the individual dynamics and can only access experimentally the value of . we are interested in the statistical properties of this global variable and how do they depend on the degree of heterogeneity in the system . we will often refer to as the _ macroscopic _ variable and to the s as the _ microscopic _ ones .

1206.3547

we study first the case in which particles jump independently from state @xmath10 to @xmath7 and vice - versa , schematically : @xmath11 with rates that depend on the value of the heterogeneity parameter , @xmath12 . the probability @xmath13 for particle @xmath4 to be in state @xmath7 at time @xmath5 obeys the linear rate equation @xmath14 . in the case of constant rates , the solution is : @xmath15 , with @xmath16 . the results derived below apply equally if the rates depend on time or on the time that the particle has been in its current state ( if the rate depends on the time @xmath17 that the particle has been on its current state , the steady - state probability of finding the particle at state @xmath7 is @xmath18 with @xmath19 ) . using particle independence and that the moments with respect to realizations of the stochastic process of the random variable @xmath20 are given by @xmath21 , one obtains that the average and variance of the global variable @xmath22 are : @xmath23&=&\sum_{i=1}^n \left(p_i(t)-p_i(t)^2\right)= n\left(\overline{p(t)}-\overline{p(t)^2}\right),\label{varparticular}\end{aligned}\ ] ] where the overline denotes an average over the population , @xmath24 . if we consider a system where all particles are identical ( i.e. have the same values for the internal parameter @xmath25 ) , and keep the same average value @xmath26 for the global variable at time @xmath5 , the variance would be @xmath27=n\overline{p(t)}\left(1-\overline{p(t)}\right)\ge \sigma^2[n(t)]$ ] . we obtain the somehow counterintuitive result that a system of heterogeneous independent particles displays smaller fluctuations in its collective variable than another system with identical particles . this effect is illustrated in figure [ traject ] . the reduction in the variance of the collective variable is @xmath0 times the variance of @xmath28 over the population : @xmath29-\sigma^2[n(t)]=n\left(\overline{p(t)^2}-\overline{p(t)}^2\right),\ ] ] which is of the same order , @xmath30 , as the variance itself , giving a non - negligible correction . reading the previous formula backwards , one realizes that the moments of the collective variable give information about the degree of heterogeneity in the system : @xmath31}{n}.\ ] ] this expression is general , regardless the specific form in which @xmath28 is distributed over the population . higher moments of the heterogeneity distribution are also related to higher moments of the collective variable . this allows to infer the skewness , kurtosis and higher order characteristics of the heterogeneity distribution by measuring only global variables and their fluctuations . in the supplementary information it is shown that an equivalent result is obtained generically for @xmath32-state systems for @xmath33 . of a system of identical ( left panel ) and heterogeneous ( right panel ) particles , for a system of @xmath34 particles . the parameters were set as @xmath35 , with @xmath36 in the case of identical particles ( left panel ) and @xmath28 chosen from a symmetric beta distribution @xmath37^{\alpha-1}$ ] , with @xmath38 , being the sample mean and variance equal to @xmath39 , @xmath40=0.23 $ ] , respectively . note that the fluctuations of the average state are larger in the case of identical particles . [ traject ] ] besides the moments , one can derive the full probability distribution of the global variable . the generating function for the one particle probability distribution is @xmath41 and the generating function @xmath42 of the sum of independent random variables is the product of their generating functions , @xmath43 . expanding in powers of @xmath44 we can obtain the probability distribution for @xmath22 : @xmath45 , being @xmath46 and @xmath47 the group of permutations of @xmath0 elements . the model studied in this subsection , despite its simplicity , offers a reduced description of generic systems of non - interacting multi - stable units subject to fluctuations . the results obtained here are directly relevant if one is interested in the collective properties of one such system when the units are non - identical . one can reasonably argue that the independence property is too unrealistic for the study to be of any practical interest . we will be considering more complicated cases including non - independent units in the rest of the paper . however , this simple model presents in isolation a mechanism , spontaneous transitions , that can play a role in more complicated and relevant systems ( we will see this later ) . the simplicity of the model allows us to understand the effect of heterogeneity in this mechanism , and will give us insight in the role of heterogeneity in the behavior of more complicated systems . we will now discuss the situation in which the particular values of the parameter of each particle are not known . this introduces an additional source of uncertainty . for simplicity , we will focus on the @xmath48-states independent particles system considered before , but the discussion applies as well to general systems of interacting particles . this discussion will also allow us to take a closer look at the results obtained and clarify their meaning and relevance in different settings . often , one does not know the value of the parameter @xmath1 of each individual particle , but has some idea about how this parameter is distributed on the population , perhaps its probability distribution ( obtained for example by measuring individual behavior in an equivalent system ) . here , we will assume that the @xmath1 s are independent and identically distributed random variables with a given probability density @xmath49 . in this case , @xmath50 and @xmath51 $ ] are themselves random variables that , as shown above , depend on the particular realization of the @xmath1 s . the expected values of these quantities are obtained by averaging eqs.([medparticular],[varparticular ] ) over the distribution of the individual parameters : @xmath52}=n\left(\widehat{p(t)}-\widehat{p(t)^2}\right)\label{varmed},\ ] ] where the hat denotes an average with respect to @xmath49 , @xmath53 . again the variance is smaller than for a system of identical particles with the same mean value , namely , @xmath54-\widehat{\sigma^2[n(t)]}=n\left(\widehat{p(t)^2}-\widehat{p(t)}^2\right)$ ] . if we average the probability @xmath55 over the distribution of parameters we obtain a simple form for the probability of the global variable @xmath22 : @xmath56 a binomial distribution with parameter the average @xmath57 over the distribution @xmath58 . the variance of this distribution is @xmath59_\text{tot}=n\left(\widehat{p(t)}-\widehat{p(t)}^2\right)\label{vartotal},\ ] ] equal to the variance one would obtain in a system of identical particles with the same average , @xmath60 , a result in apparent contradiction with ( [ varmed ] ) . however , we should note that they refer to different things : expression ( [ varmed ] ) gives the average variance when the parameter values are given , so measuring the average uncertainty in @xmath22 due to the stochastic nature of the process . ( [ vartotal ] ) , in addition to the uncertainty coming from the stochasticity of the process , also includes the uncertainty on the parameter values . the two expressions are related by the law of total variance : @xmath61_{tot}&=&\widehat{\sigma^2[n]}+\sigma^2[\langle n|\lambda_1,\dots,\lambda_n\rangle]\nonumber\\ & = & n(\widehat{p}-\widehat{p^2})+\sigma^2\left[\sum_i p_i\right ] = n(\widehat{p}-\widehat{p}^2).\end{aligned}\ ] ] in @xmath62 $ ] , the variances are taken over the distribution of the @xmath1 s . if we are considering a particular system , the temporal fluctuations ( all the systems considered in this paper are ergodic , so we can think on averages over time or over the realization of the stochastic process interchangeably ) in @xmath22 will come only from the intrinsic stochasticity , and expressions ( [ varparticular],[varmed ] ) are the ones that measure it . expressions ( [ averagedistribindep],[vartotal ] ) are appropriate only if we are considering an ensemble of systems with a distribution of parameters and our different measurements may come from different systems in the ensemble . let us now consider a general system of interacting heterogeneous particles . the stochastic description now starts from a master equation for the @xmath0-particle probability distribution : @xmath63\nonumber\\ & & + \sum_{i=1}^{n}(e_i^{-1}-1)\left[(1-s_i)r_i^{+}p(s_1,\dots , s_n)\right]\label{mastereqgen},\end{aligned}\ ] ] with step operators defined as @xmath64 . the transition rates @xmath65 might now depend on the state of any other particle ( this is how interactions enter in the model ) . from eq.([mastereqgen ] ) one can derive for the moments and correlations : @xmath66 . \label{correlationsgeneral}\end{aligned}\ ] ] with @xmath67 and @xmath68 in the second equation ( recall that @xmath69 ) . in general , if the transition rates depend on the state variables @xmath20 , these equations are not closed since they involve higher order moments , and some approximation method is needed to proceed . systematic expansions in @xmath70 , including van kampen s @xmath71-expansion @xcite , are not applicable , since variables @xmath72 are not extensive . in the following , we introduce an approximation suitable for the analytical treatment of systems of globally coupled heterogeneous particles . we assume that the @xmath73-particle correlations @xmath74 with @xmath75 scale with system size as @xmath76 using this ansatz one can close the system of equations ( [ momentsgeneral],[correlationsgeneral ] ) for the mean values and the correlations . this is shown in the supplementary information for general transition rates of the form @xmath77 . while the resulting equations for the average values @xmath78 coincide with the mean - field rate equations usually formulated in a phenomenological way@xcite , our formulation allows us to compute the correlations and include , if needed , higher order corrections in a systematic way . assumption ( [ ansatz ] ) can be justified noting that it is consistent with @xmath79 which follows from van kampen s splitting of the global variable @xmath80 , with @xmath81 deterministic and @xmath82 stochastic . details are given in the supplementary information . the global variable @xmath22 is extensive and it is expected to follow van kampen s ansatz in many cases of interest . note , however , that since there is not a closed description for the macroscopic variable @xmath22 , one can not use van kampen s expansion , and our approach extends the implications of this splitting of the macroscopic variable to the correlations of the microscopic state variables . for simplicity , we have focused on @xmath48-states systems and assumed a constant number of particles . systems with @xmath32 states are also expected to follow ansatz ( [ ansatz ] ) , since the scaling of the global variable is not limited to @xmath48-sates systems . the case of variable , but bounded , number of particles can be included straightforwardly by considering an extra state . the unbounded case can also be considered performing an appropriate limit . if the system has some spatial structure , the ansatz ( [ ansatz ] ) is not expected to be valid , and some decay of the correlations with the distance is expected instead ; this interesting situation is left for future work . we will proceed by applying the presented method to analyze the role of heterogeneity in two models previously considered in the literature that apply to contexts in which the assumption of identical agents can hardly be justified : stock markets and disease spreading . we will focus on the steady - state properties of both models , skipping transient dynamics . kirman s model @xcite was proposed to study herding behavior in the context of stock markets and collective dynamics on ant colonies . in the stock market context , agent @xmath4 can be in two possible states ( e.g. @xmath83``pessimistic '' -with regard to future market price- and @xmath84``optimistic '' ) and it can switch from one to the other through two mechanisms : spontaneous transitions at a rate @xmath85 , and induced transitions at a rate @xmath86 , being @xmath87 the influence " of agent @xmath88 on other agents . the case @xmath89 corresponds to the voter model@xcite . in the original formulation , all agents have the same influence , i.e. @xmath90 . we generalize the model allowing the parameter @xmath1 to vary between agents . in @xcite , the effect of heterogeneity was explored numerically , but not in a systematic way . this model is interesting for us because it incorporates in a simple way two basic processes : spontaneous transitions and induced transitions . as we will see , due to its simplicity , a full analytical treatment is possible that will , in turn , allow us to obtain a deeper insight into the general effect of heterogeneity in systems of interacting particles . the master equation for the process is of the form ( [ mastereqgen ] ) , with rates given by : @xmath91 from ( [ momentsgeneral ] ) the averages and correlations obey : @xmath92 for @xmath68 and @xmath93 . note that , due to the particular form of the rates , these equations do not involve higher - order moments . this is a simplifying feature of this model that allows one to obtain exact expressions . the first equation leads to a steady state value @xmath94 ( a property that comes from the symmetry @xmath95 ) . using the relation @xmath51=\sum_{i , j}\sigma_{i , j}$ ] we obtain ( see supplementary information ) that the variance in the steady state is : @xmath96=\frac{n}{4}\left[1 + \frac{2\overline{\lambda}(1-n^{-1})}{4\epsilon+\overline{\lambda } } + \frac { ( n-3 + 2n^{-1})\overline{a}}{2\epsilon+\overline{a}}\right]\label{kirmanexact}\ ] ] with @xmath97 . the leading - order term , @xmath98=\frac{n}{4}\left[1+\frac{\overline{\lambda}}{2\epsilon}+\frac{\sigma^2[\lambda]}{2\epsilon\left(4\epsilon+\overline{\lambda}\right)}\right]+o(n^0 ) $ ] , with @xmath99=\overline { \lambda^2}-\overline\lambda^2 $ ] , can also be readily obtained using the ansatz ( [ ansatz ] ) . note that the presence of heterogeneity increases collective fluctuations . in fig.[kirmanaverage ] we compare expression ( [ kirmanexact ] ) with results coming from numerical simulations . as a function of the variance of the influence parameter @xmath100 in kirman s model with distributed influence . numerical simulations ( symbols ) and theoretical analysis ( lines ) , eq.([kirmanexact ] ) , for different number of agents @xmath0 and @xmath101 . @xmath1 are independent random variables distributed according to a lognormal or a gamma distribution with mean @xmath102 and variance @xmath103 . the results have been averaged over @xmath104 for @xmath105 and @xmath106 for @xmath34 realizations of the distribution of parameters . [ kirmanaverage ] ] in this case , the knowledge of @xmath107 and @xmath108 $ ] alone does not allow to infer the degree of heterogeneity present in the system , unless one knows from other sources @xmath109 and @xmath85 and it is not possible to conclude whether the observed fluctuations have a contribution due to the heterogeneity of the agents . however , the steady - state correlation function @xmath110(t)\equiv\langle n(t)n(0)\rangle_\textrm{st}-\langle n\rangle_\textrm{st}^2 $ ] , does include a term that allows to infer the possible heterogeneity . @xmath110(t)$ ] is obtained integrating eq.([kirmanaverage ] ) and performing the appropriate conditional averages ( see supplementary information ) : @xmath111(t)=\left(\sigma^2_\text{st}[n]-u\right)e^{-(2\epsilon+\overline{\lambda})t}+ue^{-2\epsilon t } , \label{kirmancorrfunc}\ ] ] with @xmath112-n/4)$ ] . the departure from a pure exponential decay signals the presence of heterogeneity ( for identical particles @xmath113 $ ] ) . fitting this expression to data one can obtain @xmath108 $ ] and the parameters @xmath114 . then , the use of expression ( [ kirmanexact ] ) would yield @xmath99 $ ] . in fig.[kirmancorrf ] we show that the numerical simulations indeed support the existence of two exponential decays for the correlation function . ) , solid lines ) . note that when heterogeneity is present ( @xmath115 ) the correlation function departs from purely exponential decay ( displayed as a dashed line ) . data for @xmath116 have been moved up @xmath117 units vertically for better visualization . parameters values are @xmath101 , @xmath34 . @xmath1 are independent random variable distributed according to a gamma with mean @xmath118 and variance , @xmath103 , indicated in the figure . a simple fit of expression ( [ kirmancorrfunc ] ) to the @xmath119 data gives @xmath120 . [ kirmancorrf ] ] interestingly , other ways to introduce heterogeneity in the system have different effects :- if the heterogeneity is introduced in the spontaneous transition rate , @xmath121 , making some particles more prone to spontaneous transitions that others ( but keeping @xmath122 , @xmath123 to isolate effects ) , collective fluctuations again increase with respect to the case of identical particles . -next , we can assume that the rate of induced change is different for different agents , even if all have the same influence . measuring this difference in susceptibility " ( to induced change ) with a parameter @xmath124 , we would have that the rate of induced change in agent @xmath4 is @xmath125 . the effect of heterogeneity in @xmath124 ( keeping again @xmath122 , @xmath123 ) is that the collective fluctuations decrease with the degree of heterogeneity in the susceptibility @xmath124.-setting some heterogeneous preference for the states among the particles , i.e. making @xmath126 , the spontaneous rate from @xmath10 to @xmath7 of particle @xmath4 , different from @xmath127 , the spontaneous rate from @xmath7 to @xmath10 of the same particle , decreases global fluctuations . in order to vary the preference for one state keeping constant the global intrinsic noise " of this particle ( note that the correlation time of particle @xmath4 , when isolated , is given by @xmath128 ) , we set @xmath129 and generate @xmath130 as i.i.d . random variables with a distribution with support contained in the interval @xmath131 $ ] . exact explicit expressions for the first moments of the global variables are ( see supplementary information):@xmath132_{st}&=&\frac{n}{4(\epsilon+\frac{2\lambda}{n})}\left[\overline{\epsilon^+}(1+\frac{\lambda}{\epsilon})-\frac{\overline{\epsilon^+}^2}{\epsilon}\left(\frac{\lambda}{2\epsilon}+1\right)-2\frac{\sigma^{2}[\epsilon]}{2\epsilon+\lambda}\right]\label{varkirmanprefer},\end{aligned}\ ] ] in figure ( [ fig : kirmanpreference ] ) the exact expressions ( [ medkirmanprefer ] , [ varkirmanprefer ] ) are compared with results coming from numerical simulations . in this case , the correlation function decays exponentially , @xmath111(t)=\sigma^2[n]_{\textrm{st}}e^{-2\epsilon t},\ ] ] independently of the degree of heterogeneity , so this form of heterogeneity can not be inferred by measuring the correlation function . numerical simulations confirm this result . as a function of the variance of the spontaneous transition rate to state @xmath7 , @xmath133 , in kirman s model . results coming from numerical simulations ( symbols ) and theoretical analysis ( solid lines , eqs.([medkirmanprefer ] , [ varkirmanprefer ] ) ) , for @xmath105 agents , @xmath134 and @xmath135 . @xmath136 are independent random variables distributed according to a symmetric beta distribution in the interval @xmath137 with mean @xmath138 and variance , @xmath139 . [ fig : kirmanpreference ] ] we have seen that heterogeneity can have an ambivalent effect over the size of the fluctuations , depending on the particular form it appears . we now provide intuitive arguments to understand these different effects . when the influence parameter , @xmath1 , varies from one unit to the other , there will be some largely influential agents and others with little influence . in the limit of very large heterogeneity we can think of a situation with a single agent with an extremely large influence and the others having a negligible one ( we are keeping a constant average influence ) . in this case , the highly influential agent drifts from one state to the other , essentially independently ( since other agents have negligible influence ) , but , due to its large influence , all the agents are attracted to its current state . in this follow the leader " regime , we obtain macroscopic transitions from one state to the other , corresponding to very large global fluctuations . the situation is the opposite for a non - identical susceptibility parameter @xmath124 where global fluctuations decrease as the diversity is increased . again , we can understand this in the limit of very large heterogeneity where a single agent ( or a small number of them ) has large susceptibility while all the others have a negligible one ( in order to keep average susceptibility constant ) . then , agents with small susceptibility change essentially independently , in an uncorrelated fashion , resulting in low global fluctuations ( note that in order to have large global fluctuations , the fluctuations in the state of the single agents should be correlated ) . in the case of diverse spontaneous transition rates , @xmath140 , global fluctuations increase with the degree of heterogeneity . in the limit of large heterogeneity , we would have a small number of agents with very large spontaneous transition rate , whose state would fluctuate in an uncorrelated fashion , and a large number of agents with low spontaneous transition rate , that essentially would only change state through induced transitions , giving rise to correlated fluctuations , resulting in large variance for the global variable . in the case in which agents display an intrinsic heterogeneous preference for one of the two states , the global fluctuations decrease with heterogeneity degree . we saw this already in the first section for non - interacting agents . here we see the same effect , suggesting that the phenomenon is robust and still plays a role when interaction is added . the coexistence of a small number of agents with large value for the parameter and a large number with a small value assumed in the previous arguments arises from the fact that an unbounded ( from above ) distribution of positive - defined parameters ( e.g. rates ) is skewed . however , all the effects of diversity commented are still present if the distribution is symmetric . in this case , nevertheless , the maximum degree of heterogeneity ( for a constant mean value ) is bounded , sometimes greatly limiting the maximum possible value of diversity . for symmetric distributions , a simple explanation is not so clear , but an asymmetry in the effect of increasing and decreasing the value of the parameter seems to be at the heart of the phenomenon . the previous example could be treated exactly because , due to symmetry , the interaction , non - linear terms , cancel out in the equations for the moments . in general , however , this is not the case , and the analytical treatment is more involved . here we consider an example of such case . the stochastic susceptible - infected - susceptible ( sis ) model is a paradigmatic model for the study of spreading of infectious disease @xcite as well as the diffusion of innovation @xcite and other types of social influence . despite its simplicity , it captures interesting phenomenology . the process is schematically described by : @xmath141 where @xmath142 ( resp . @xmath143 denotes agent @xmath4 being susceptible ( resp . infected ) . there are 3 basic elementary processes : ( i ) infected agent @xmath88 infects susceptible agent @xmath4 at a rate @xmath144 , being @xmath87 the infectivity parameter of agent @xmath88 ; ( ii ) infected agent @xmath88 becomes susceptible a rate @xmath145 ; ( iii ) susceptible agent @xmath88 gets infected spontaneously ( due to interactions with agents not considered in the system or other causes ) at a rate @xmath85 . this corresponds to the sis model with spontaneous contagions and distributed infectivity . in the absence of spontaneous infections @xmath89 , the system has a trivial steady state with zero infected agents . with @xmath146 the system has a non - trivial steady state whose properties we analyze in the following . as in the previous case , heterogeneity could appear in any parameter of the agents ( for example , in the recovery rate , in a``susceptibility '' parameter , etc . ) . we study first the case in which only the infectivity , @xmath1 , can vary from agent to agent . the effect of heterogeneity in the deterministic version of related models was studied recently @xcite . the master equation is of the form ( [ mastereqgen ] ) with rates @xmath147 . equations ( [ momentsgeneral]-[correlationsgeneral ] ) for the first moments can be closed in the steady state , using our main ansatz , to obtain explicit formulas for @xmath148 and @xmath51_\textrm{st}$ ] to any desired order in @xmath149 . in this case , however , the expressions are rather cumbersome and we skip them here . the results are plotted in figure ( [ sismed ] ) , where we compare the approximation to order @xmath150 with results coming from numerical simulations , showing good agreement . here both the average value and the variance are modified by the presence of heterogeneity ( the dependence of the average is , however , only in second order in @xmath70 , almost unnoticeable in the figure ) . as in the kirman model , the size of the fluctuations increase markedly with the amount of heterogeneity in the `` influence '' ( now influence to infecting others ) of the agents . , @xmath151 , @xmath152 , @xmath102 . [ sismed ] ] in this case , other ways to introduce heterogeneity also have different effects . when heterogeneity appears in the recovery rate @xmath145 , the mean number of infected agent increases , with a moderate effect over the variance ( resulting in smaller relative fluctuations ) . heterogeneity in the susceptibility to infection ( which would be introduced with the change @xmath153 , with @xmath124 distributed over the population ) decreases the fluctuations , with little effect over the mean value . heterogeneity in the spontaneous infection rate @xmath85 has almost no effect . in a real situation , one expects to find heterogeneity simultaneously in several of the parameters defining the model . when heterogeneity is present both in the infectivity and in the susceptibility , the effects of both types of heterogeneity essentially add up , with the size of the fluctuations increasing with the heterogeneity in the infectivity for a given level of heterogeneity in the susceptibility and fluctuations decreasing with the level of heterogeneity in the susceptibility for a given level of heterogeneity in the infectivity . the effects of heterogeneity in the infectivity and in the susceptibility are equivalent to those found in the kirman model , and can be intuitively understood in the same terms . heterogeneity in the recovery rate is similar to assigning an heterogeneous preference for the state @xmath10 ( recovery ) and its effect in the ( relative ) fluctuations is again the same as that in the case of the kirman model . this suggests that the effects of the heterogeneity found are generic and can be useful to understand the behavior of other systems . in this work , we have analyzed the combined effect of stochasticity and heterogeneity in interacting - particle systems . we have presented a formulation of the problem in terms of master equations for the individual units , but extracted conclusions about the fluctuations of collective variables . we have developed an approximation suitable for the analytical study of this general type of systems . we have shown that the heterogeneity can have an ambivalent effect on the fluctuations , enhancing or decreasing them depending on the form of the system and the way heterogeneity is introduced . in the case of independent particles , heterogeneity in the parameters always decreases the size of the global fluctuations . we have also demonstrated that it is possible to obtain precise information about the degree and the form of the heterogeneity present in the system by measuring only global variables and their fluctuations , provided that the underlying dynamical equations are known . in this way stochastic modeling allows to obtain information not accessible from a purely deterministic approach . we have also demonstrated that , in some cases , one can account for the heterogeneity of the particles without losing analytical tractability . heterogeneity among the constituent units of a system is a very generic feature , present in many different contexts and this work provides a framework for the systematic study of the effect of heterogeneity in stochastic systems , having thus a wide range of potential applicability . more research in this direction would be welcomed . we have developed and used analytical tools based on an extension of van kampen s ansatz on the relative weight of the fluctuations compared to the mean value , suitable for systems with particle heterogeneity . we have included the details of this method in the supplementary information . in some cases , and in order to compare with the analytical expressions , we have generated data from numerical simulations using a particular form of gillespie s algorithm @xcite that takes into account the heterogeneity in the population . we now explain this algorithm using , for the sake of concreteness , the specific case of the kirman model with distributed susceptibility . the parameters of the system are : the spontaneous transition rate @xmath85 , the influence parameter @xmath100 , the susceptibility parameter of each agent @xmath124 , and the total number of agents @xmath0 . in this case , the influence parameter can be reabsorbed rescaling @xmath124 , so we set @xmath154 without loss of generality . the variables of the system are the state of each agent @xmath72 . we will also use the total number of agents in state @xmath7 , @xmath155 , the total susceptibility of agents in state @xmath7 , @xmath156 , and the average susceptibility @xmath157 . at any given instant , two events can happen:(i ) an agent in state @xmath7 changes to state @xmath10 . this can happen due to a spontaneous transition , at a total rate @xmath158 , or due to an induced transition , at a total rate @xmath159.(i ) an agent in state @xmath10 changes to state @xmath7 . this can happen due to a spontaneous transition , at a total rate @xmath160 , or due to an induced transition , at a total rate @xmath161 . according to the gillespie method , that considers the continuous - time process , the time at which the next transition will take place is exponentially distributed , with average the inverse of the total rate . the probability that a given transition is realized is proportional to its rate . if the realized transition is a spontaneous one , the agent that actually undergoes it is selected at random ( since , in this case , they all have the same rate @xmath85 ) . if the transition is induced , the agent that undergoes it is selected with probability proportional to its susceptibility . it can be easily seen that this principles lead to an exact ( up to numerical precision ) simulation of sample paths of the stochastic process @xcite . the algorithm , then , proceeds as follows : + ( 0 ) evaluate the total number of particles in state @xmath7 , @xmath22 , and the total susceptibility of particles in state @xmath7 , @xmath71.(1 ) evaluate the total transition rate @xmath162.(2 ) generate the time for the next reaction , @xmath163 , as an exponential random variable with average @xmath164 . this can be done by setting @xmath165 , with @xmath166 a uniform random variable in the range @xmath167.(3 ) select which reaction takes place . for this , generate a uniform random variable , @xmath168 , in the range @xmath169 . + -if @xmath170 the transition will be a spontaneous transition from @xmath7 to @xmath10 ; select an agent , @xmath88 , at random among those at state @xmath7 . set @xmath171 , @xmath172.-if @xmath173 , the transition will be a spontaneous transition form @xmath10 to @xmath7 ; select an agent , @xmath88 , at random among those at state @xmath10 . set @xmath174 , @xmath175.-if @xmath176 , the transition will be an induced transition from @xmath7 to @xmath10 ; select an agent , @xmath88 , among those at state @xmath7 with probability proportional to the value of its susceptibility parameter @xmath177 . set @xmath171 , @xmath172.-if @xmath178 , the transition an induced transition from @xmath10 to @xmath7 ; select an agent , @xmath88 , among those at state @xmath10 with probability proportional to the value of its susceptibility parameter @xmath177 . set @xmath174 , @xmath175.(4 ) set @xmath179 . go to ( 1 ) . we thank e. hernandez - garcia for useful discussions . this work was supported by mineco ( spain ) , comunitat autnoma de les illes balears , feder , and the european commission under project fis2007 - 60327 . is supported by the jaepredoc program of csic . spudich , j. l. , koshland , jr . d. e. non - genetic individuality : chance in the single cell . _ nature _ * 262 , * 467 - 471 ( 1976 ) . snijder , b. , pelkmans , l. origins of regulated cell - to - cell variability . cell biol _ * 12 , * 119 - 125 ( 2011 ) . granovetter , m. threshold models of collective behavior . j. sociol . _ * 83 , * 1420 - 1443 ( 1978 ) . kirman , a. whom or what does the representative individual represent ? . perspect _ * 6 , * 117 - 136 ( 1992 ) . braiman , y. , kennedy , t. a. b. , wiesenfeld , k. and khibnik , a. entrainment of solid - state laser arrays . a _ * 52 , * 1500 - 1506 ( 1995 ) . oliva , r. a. and strogatz , s. h. dynamics of a large array of globally coupled lasers with distributed frequencies . . chaos _ * 11 , * 2359 - 2374 ( 2001 ) . albert , r. and barabsi , a .- statistical mechanics of complex networks . _ rev . mod . phys . _ * 74 , * 47 - 97 ( 2002 ) . boccaletti , s. , latora , v. , moreno , y. , chavez , m. , hwang , d .- u . complex networks : structure and dynamics _ phys . rep . _ * 424 , * 175 - 308 ( 2006 ) . tessone , c. j. , mirasso , c. r. , toral . r. and gunton , j. d. diversity - induced resonance . lett . _ * 97 , * 194101 1 - 4 ( 2006 ) . komin , n. , lacasa , l. and toral , r. critical behavior of a ginzburg landau model with additive quenched noise . _ j. stat . _ * p12008 , * 1 - 19 ( 2010 ) . peyton young , h. innovation diffusion in heterogeneous populations : contagion , social influence , and social learning . rev . _ * 99 , * 1899 - 1924 ( 2009 ) . novozhilov , a. s. epidemiological models with parametric heterogeneity : deterministic theory for closed populations _ math model nat pheno _ * 7 , * 147 - 167 ( 2012 ) . masuda , n. , gibert , n. and redner , s. heterogeneous voter models . e _ * 82 , * ( 010103r ) 1 - 4 ( 2010 ) . mobilia , m. and georgiev , i. t. , voting and catalytic processes with inhomogeneities . e _ * 71 , * 046102 1 - 17 ( 2005 ) . xie , j. and sreenivasan , s. social consensus through the influence of committed minorities . e _ * 84 , * 011130 1 - 8 ( 2011 ) . fisher , d. s. critical - behavior of random transverse - field ising sping chains _ phys . b _ * 51 , * 6411 - 6461 ( 1995 ) . young , a. p. ( ed . ) _ spin glasses and random fields _ world scientific , singapore ( 1998 ) . mzard , m. , parisi , g. and virasoro , m. a. _ spin glass theory and beyond _ world scientific , singapore ( 1987 ) . bouchaud , j-.p . and georges , a. , anomalous diffusion in disordered media : statistical mechanisms , models and physical applications . _ phys . rep . _ * 195 , * 127 - 293 ( 1990 ) . ben - avraham , d. and havlin , s. _ diffusion and reactions in fractals and disordered systems _ cambridge university press , cambridge ( 2000 ) . dushoff , j. host heterogeneity and disease endemicity : a moment - based approach . theor popul biol * 56 , * 325335 ( 1999 ) . van kampen , n. g. _ stochastic processes in physics and chemistry _ , north - holland , amsterdam ( 2004 ) . kirman , a. ants , rationality , and recruitment . _ q. j. econ . _ * 108 , * 137 - 156 ( 1993 ) . liggett , t. m. _ interacting particle systems _ , springer - verlag , new york ( 1985 ) . alfarano , s. and milakovi , m. network structure and n - dependence in agent - based herding models . _ j. econ . control _ * 33 , * 78 - 92 ( 2009 ) . anderson , r. m. _ population dynamics of infectious diseases : theory and applications _ , chapman and hall , london - new york ( 1982 ) . gillespie , d. t. _ exact stochastic simulation of coupled chemical reactions _ * 81,*(25 ) 2340 - 2361 ( 1977 ) . we consider here the case in which each particle can be in one of @xmath180 ( instead of @xmath48 ) possible states . we will show that the results obtained in the main text for @xmath181state systems also hold in this more general case . we label the states with the subscript @xmath182 , so in this case the variable describing the state of particle @xmath4 can take @xmath180 possible values , @xmath183 ( we start the labeling from @xmath10 to be consistent with the previous case , that would correspond to @xmath184 ) . let @xmath185 the probability that particle @xmath4 , with heterogeneity parameter @xmath1 , be on state @xmath186 . it satisfies the evolution equation : @xmath187 with @xmath188 a general transition matrix ( satisfying @xmath189 ) , that may depend in principle on time and on the time that the particle has been on its current state . to isolate the role of parameter heterogeneity , we assume that the initial condition is the same for all the particles ( or that the initial condition is determined by the value of @xmath1 ) such that the solution @xmath190 is the same for all particles sharing the same value of the parameter . the macroscopic state of the system will be described by the set of variables @xmath191 , that is , the number of particles in each state . the averages and variances of this variables are given by : @xmath192&=&\sum_{i=1}^n\left [ p(\lambda_i,\alpha , t)-p(\lambda_i,\alpha , t)^2\right].\end{aligned}\ ] ] this variance is again smaller that tat of a system of identical particles with same average , the difference given by : @xmath193_\text{id}-\sigma^2[n_\alpha(t)]=n\overline{p(\alpha , t)^2}-\overline{p(\alpha , t)}^2,\ ] ] a result exactly analogous to the one obtained in the previous case . the heterogeneity among the particles on the probability of occupation of level @xmath186 can be derived from the first moments of the occupation number of the level : @xmath194}{n}.\ ] ] note that , when focusing on the number of particles on state @xmath186 , the system effectively reduces to a @xmath181level one , with states @xmath186 and no-@xmath186 , so the results of the previous section can be translated directly . a different and some times relevant question can be considered when the labeling of the states is such that the order is well defined ( for example each state corresponds to an energy level or a distance from a reference ) . then the average state is meaningful and we can study its statistical properties . below we show that the variance of this mean level is again always smaller if heterogeneity is present . the average state of the system is given by @xmath195 . it is a random variable whose average and variance are given by : @xmath196&=&\sum_{\alpha,\beta=0}^{m-1}\frac{\alpha\beta}{n^2}(\langle n_\alpha n_\beta\rangle-\langle n_\alpha\rangle\langle n_\beta\rangle)=\frac{1}{n^2}\sum_{i=1}^n\left[\sum_{\alpha=0}^{m-1}\alpha^2p(\alpha,\lambda_i)-\sum_{\alpha,\beta=0}^{m-1}\alpha p(\alpha,\lambda_i)\beta p(\beta,\lambda_i)\right]\label{meanlevelhet}.\end{aligned}\ ] ] we have used @xmath197 and @xmath198 $ ] . a system of identical particles that had the same average occupation of the different levels i.e. @xmath199 @xmath200 , would have and average and variance of the mean level given by : @xmath201_\text{id}&=&\frac{1}{n}\sum_{\alpha=0}^{m-1}\alpha^2\frac{\langle n_\alpha\rangle}{n}-\frac{1}{n}\sum_{\alpha,\beta=0}^{m-1}\alpha\beta\frac{\langle n_\alpha\rangle}{n}\frac{\langle n_\beta\rangle}{n}\label{meanlevelid}.\end{aligned}\ ] ] we now define @xmath202 ( the average level of particle @xmath4 ) , and note that the first terms in the right - hand side of ( [ meanlevelhet ] ) and ( [ meanlevelid ] ) are equal , while the second terms can be written as : @xmath203 ^ 2=\frac{1}{n}\overline{g}^2,\end{aligned}\ ] ] which implies that @xmath204_\text{id}\ge \sigma^2[l]$ ] , i.e. the variance of the mean level is always smaller in a system of heterogeneous particles , the difference with respect to the case of identical ones being : @xmath205_\text{id}-\sigma^2[l]=\frac{1}{n}\left(\overline{g^2}-\overline{g}^2\right)=\frac{1}{n}\sum_{\alpha,\beta=0}^{m-1}\alpha\beta\left[\sum_{i=1}^n\frac{p(\alpha,\lambda_i)p(\beta,\lambda_i)}{n}-\sum_{i , j=1}^n\frac{p(\alpha,\lambda_i)p(\beta,\lambda_j)}{n^2}\right]\geq0.\ ] ] the correction to the variance in this case scales as @xmath70 , but again is of the same order as the variance itself , indicating a non - negligible correction . in this case to derive the heterogeneity of @xmath206 over the population one needs to know the average occupation level of each state @xmath207 and use : @xmath208.\ ] ] this can be written in terms of the variance of @xmath209 in an equivalent system of identical particles , @xmath204_\text{id}$ ] . if this is known , one can directly use @xmath210_\text{id}-\sigma^2[l]\right).\ ] ] note that , contrary to the two - level case , now the value of @xmath211 does not determine @xmath204_\text{id}$ ] . [ [ intuitive - origin - of - the - decrease - of - fluctuations - for - independent - units ] ] intuitive origin of the decrease of fluctuations for independent units ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ we have shown that a system of independent heterogeneous particles has smaller fluctuations for the collective variable than an equivalent system of identical ones . the origin of this result is the following ( for simplicity we refer to the case of @xmath48-state system ) : the average of the global variable is determined by the concentration of the states of the particles around state @xmath7 ( @xmath212 ) . the fluctuations ( measured by the variance ) of the global variable are determined by the stochastic fluctuations of the individual particles alone ( @xmath51=\sum_i\sigma^2[s_i]$ ] , since the particles are independent).in a system of heterogeneous particles , the dispersion of the states of the particles is due to the heterogeneity ( some prefer to be around sate @xmath10 , others prefer to be around sate @xmath7 ) plus their intrinsic stochasticity . in a system of identical particles , the dispersion comes from the stochasticity alone , so for a system of identical particles to have the same concentration in the states of the particles ( global average ) than a heterogeneous system , the intrinsic stochasticity has to be larger . this will give rise to larger fluctuations for the global variable.in particular , any given rational value of @xmath213 can be obtained with zero fluctuations , taking @xmath214 particles that are always at state @xmath7 and @xmath215 particles that are always at state @xmath10 . this explanation is illustrated in figure ( [ fig : trajectories ] ) . in the identical - particles system both particles fluctuate between @xmath7 and @xmath10 . in the heterogeneous case , one particle spends most of the time at @xmath7 and the other spends most of the time at @xmath10 . the probability of finding a given particle at @xmath7 is the same in both cases ( @xmath216 ) but in the heterogeneous case most of the time there is one particle at @xmath7 and one particle at @xmath10 , resulting on a value of the average state most often equal to @xmath216 , and so with smaller fluctuations . the situation is similar for a larger number of particles , as shown in figure 1 of the main text . an analogous picture emerges when one considers more that @xmath48 states . note that in every case we compare a system of heterogeneous particles with another of identical ones that has the same one - particle distribution i.e. @xmath217 , @xmath200 . the general evolution equations for the first moments are of the form : @xmath218 our main ansatz is that the @xmath73-particle correlations @xmath74 with @xmath219 scale with system size as @xmath76 we first show how the ansatz ( [ ansatz ] ) allows to close the system ( [ momentsgeneral ] , [ correlationsgeneral]).we assume that functional dependence of the rates on the sate variables is of the form @xmath77 . this includes , for example , rates of the form @xmath220 like the ones used in the examples analyzed . we further assume that the rates can be expanded as a power series : @xmath221 there are @xmath222 terms in the @xmath32th summand , @xmath223 , giving a total contribution of order @xmath224 . the terms in the right hand side of ( [ momentsgeneral ] ) are of the form : @xmath225 where @xmath226 corresponds to a term of the form @xmath227 , @xmath228 corresponds to @xmath229 and the last equality holds due to our ansatz . we see that the dominant terms are those with @xmath230 , which correspond to products of mean values of the form @xmath231 . we conclude that the ansatz allows to do the substitution @xmath232 in the evolution equations for the mean values . the evolution equations for the correlations read : @xmath233 in this case , the terms are of the form @xmath234 with @xmath235 . due to the presence of @xmath236 , the term in which only averages appears vanishes . reasoning as before , we see that the dominant terms are those proportional to @xmath237 , while those proportional to higher - order correlations can be neglected . in this case , the ansatz allows to do the substitution @xmath238 . in this way , the evolution equation for the correlations depend , at first order , only on averages and correlations and not on higher order moments . the validity of the ansatz ( [ ansatz ] ) itself can be established a posteriori by checking that the results obtained using the ansatz are consistent with it . in this section , we will link its validity with the well - known van kampen s ansatz @xcite that is the basis for the systematic system - size expansion . van kampen s ansatz consists on assuming that the variable of interest has a deterministic part of order @xmath71 plus a stochastic part of order @xmath239 , i.e. @xmath240 , where @xmath71 is a parameter of the system that controls the relative size of the changes due to elementary processes , typically the system size . in our system the role of the parameter @xmath71 is played by the total number of particles @xmath0 . as briefly stated in the main text , we can not expect that the single - particle variables that we are considering obey van kampen s ansatz , since they are not extensive . our variables @xmath72 have a deterministic and stochastic part that are both of order zero respect to @xmath0 ( note that @xmath241=\langle s_i\rangle(1-\langle s_i\rangle)$ ] ) . however , the macroscopic variable @xmath155 is indeed extensive and we can expect that it will follow van kampen s ansatz : @xmath242 . this implies that the @xmath73-th central moment of @xmath22 will scale as @xmath243 , i.e : @xmath244 now , assuming that @xmath245 for @xmath246 , with @xmath247 independent of @xmath0 i.e. the @xmath73-particle correlations are all or the same order in @xmath0 , so that @xmath248 scales as @xmath249 ( note that there are of the order of @xmath250 terms in the sum ) , we obtain our main ansatz , @xmath251 for @xmath246 . we have only considered terms with @xmath246 in the sum ( [ vkansatzcorr ] ) ; terms with repeated sub - indexes can be expressed as lower order ones . for example , if the index @xmath252 is present @xmath32 times , and the others are all different , we find : @xmath253+\sigma_{j_1,\dots , j_{m - k+1}}[(1-\langle s_{j_1}\rangle)^k-(-\langle s_{j_1}\rangle)^k]\nonumber\end{aligned}\ ] ] as can be see expanding @xmath254 and keeping in mind that @xmath69 . the number of such terms in the sum ( [ vkansatzcorr ] ) is @xmath255 , so they give smaller contribution that terms with all sub - indexes different . proceeding order by order from @xmath256 , we see that our main ansatz ( [ ansatz ] ) follows from ( [ vkansatzcorr ] ) . we point out that in systems of heterogeneous particles we do not have a closed description for the global , extensive , variable @xmath22 so van kampen s expansion can not be used . instead we derive the implications of van kampen s ansatz over the correlations of the microscopic variables . ( [ ansatz ] ) is a simple and convenient expression that in general allows to close the equation for the moments ( [ momentsgeneral],[correlationsgeneral ] ) . often , however it is not necessary , and a weaker condition of the form ( [ vkansatzcorr ] ) , that directly follows from van kampen s ansatz without further assumptions , is sufficient . van kampen s ansatz is generally valid when the macroscopic equations have a single attracting fixed point , when the system displays small fluctuations around the macroscopic state . the general method explained here is expected to be valid under similar conditions . an interesting topic for future research will be whether a system that has a single attracting fixed point in the absence of diversity always maintains this globally stable state when diversity is present , and whether a system that does not posses this globally stable fixed point can acquire it when diversity is added . in the kirman model with distributed influence , the averages and correlations obey : @xmath257\label{kirmancorrelations}\end{aligned}\ ] ] with @xmath258 . note that , due to the particular form of the rates , these equations are indeed closed . the first equation leads to a steady state value @xmath259 , which implies @xmath260 ( a property that comes from the symmetry @xmath95 ) . ( [ kirmancorrelations ] ) is a linear system of equations for the correlations . the steady state correlations can always be obtained by inverting the matrix that gives the couplings . obtaining a closed expression for @xmath51 $ ] in terms of the moments of @xmath100 is , however , not straightforward . from ( [ kirmancorrelations ] ) , we see that in the steady state : @xmath261}{2(2\epsilon+\overline{\lambda})}\label{sigmakir},\ ] ] from where we obtain @xmath262=\sum_{i , j}\sigma_{i , j}=\frac{n(\epsilon+\overline{\lambda}/2)+2c(1 - 1/n)}{2(2\epsilon+\overline{\lambda})}\label{sigmakir2},\ ] ] with @xmath263 . multiplying ( [ sigmakir ] ) by @xmath87 and summing over @xmath88 , one obtains : @xmath264 where @xmath265 . this is obtained again multiplying ( [ sigmakir ] ) by @xmath266 and summing over @xmath267 : @xmath268 where @xmath269 . using the ansatz @xmath270 we see that the last term of ( [ d ] ) is @xmath224 ( while the other are of @xmath30 ) , so to the first order we obtain : @xmath96=\frac{n}{4}\left[1+\frac{\overline{\lambda}}{2\epsilon}+\frac{\sigma^2[\lambda]}{2\epsilon\left(4\epsilon+\overline{\lambda}\right)}\right]+o(n^0),\label{kirman1st}\ ] ] with @xmath99=\overline { \lambda^2}-\overline\lambda^2 $ ] . we have seen how the application of the ansatz ( [ ansatz ] ) allows one to obtain closed expression for the global average and variance . interestingly , in this particular example , it is possible to include all higher order terms to obtain an exact expression for @xmath271 ( which gives the exact expression for @xmath51 $ ] trough ( [ c],[sigmakir2 ] ) ) , details are given in the appendix : @xmath272 the second equality holds as long as @xmath273 ( note that @xmath274 ) . a sufficient condition for this is @xmath275 . when the @xmath1 s are i.i.d . random variables , the probability that this condition is satisfied approaches one as @xmath0 grows . this condition is actually necessary and sufficient for the first equality in ( [ dexact ] ) to hold ( see appendix).we finally obtain the following exact expression for the variance : @xmath96=\frac{n}{4}\left[1 + \frac{2\overline{\lambda}(1 - 1/n)}{4\epsilon+\overline{\lambda } } + ( n-3 + 2/n)\frac{\overline{\frac{\lambda^2}{n(4\epsilon+\overline{\lambda})+2\lambda}}}{2\epsilon+\overline{\frac{\lambda^2}{n(4\epsilon+\overline{\lambda})+2\lambda}}}\right]\label{kirmanexact}\ ] ] expressions ( [ kirman1st ] , [ kirmanexact ] ) refer to the variance of @xmath22 in a population with given values for the parameters of each agent , @xmath1 , so the averages are population averages i.e. @xmath276 . in the case that the parameters of the agents are random variables , the population averages themselves , @xmath277 , become random variables . to compute the expected ( average ) value of ( [ kirman1st ] , [ kirmanexact ] ) , @xmath278}$ ] , one has to average over the distribution of @xmath277 , which depends on the distribution @xmath49 of the @xmath279 ( we are assuming @xmath279 i.i.d . random variables ) . this averages were obtained numerically , by evaluating expressions ( [ kirman1st ] , [ kirmanexact ] ) over the same realizations of the @xmath1 s that were used in the numerical simulations . one can use the approximation @xmath280 , that works better the larger the @xmath0 and the lower the variance @xmath281 , and that , due to the law of large numbers , is valid in the limit @xmath282 . in fig.2 of the main text we compare the average of the analytical expression ( [ kirmanexact ] ) with results coming from numerical simulations . we find perfect agreement and see that at first order the dependence of @xmath51 $ ] with @xmath283 is linear and independent of the form of the distribution , as indicated by ( [ kirman1st ] ) . higher order corrections are noticeable for higher levels of diversity . in the case of heterogeneity in the preference of the agents for the states , as indicated in the main text , the variance is given by : @xmath61_{st}&=&\frac{n}{4(\epsilon+\frac{2\lambda}{n})}\left[\overline{\epsilon^+}(1+\frac{\lambda}{\epsilon})-\frac{\overline{\epsilon^+}^2}{\epsilon}\left(\frac{\lambda}{2\epsilon}+1\right)-2\frac{\sigma^{2}[\epsilon]}{2\epsilon+\lambda}\right]\label{varkirmanprefer},\end{aligned}\ ] ] in this case , the average of ( [ varkirmanprefer ] ) over the distribution of parameters can be easily computed , giving : @xmath284}_{st}&=&\frac{n}{4(\epsilon+\frac{2\lambda}{n})}\bigg[\widehat{\epsilon^+}\left(2+\frac{\lambda}{\epsilon}\right)-\widehat{\epsilon^+}^2\left(\frac{\lambda}{2\epsilon^2}+\frac{1}{\epsilon}\right)\nonumber\\ & & \hspace{1.5cm}-\sigma^2_{\epsilon^+}\left(\frac{2\epsilon+\lambda / n}{\epsilon(2\epsilon+\lambda)}+\frac{\lambda}{2\epsilon^2n}\right)\bigg]\label{varkirmanprefermed},\end{aligned}\ ] ] the correlation function can be derived as follows ( we exemplify the derivation in the case of distributed influence , for other types of heterogeneity , the derivation is similar):([kirmanaverage ] ) is an equation for the conditional averages @xmath285 if we set @xmath286 as initial conditions . it implies : @xmath287 with @xmath288 . noticing that ( [ kirmanaverage ] ) is equal to @xmath289 , we obtain : @xmath290 using now @xmath291(t)=\langle \langle n(t_0+t)|n(t_0)\rangle n(t_0)\rangle_{st}-\langle n\rangle_{st}^2=\sum_{i , j}\langle\langle s_i(t_0+t)|\{s_k(t_0)\}\rangle s_j(t_0)\rangle-\frac{n^2}{4}$ ] ( remember @xmath292 ) , and after some straightforward algebra , we obtain : @xmath293(t)=(\sigma_{st}^2-c/\overline{\lambda})e^{-(2\epsilon+\overline{\lambda})t}+c/\overline{\lambda}e^{-2\epsilon t}.\label{kst}\ ] ] from ( [ sigmakir ] ) we get @xmath294 , showing that ( [ kst ] ) is equal to the expression displayed in the main text . we start with equation ( [ sigmakir ] ) : @xmath261}{2(2\epsilon+\overline{\lambda})}.\ ] ] using the rescaled variables @xmath295 , and defining @xmath296 , we obtain : @xmath297 defining now @xmath298 , we arrive to : @xmath299,\\ t_{m+1}-g_1&=&t_m+\frac{n}{2}\sum_{n=1}^m\left[\left(\frac{2}{n\overline{\tilde{\lambda}}-1}\right)^n\left(\frac{2\overline{\tilde{\lambda}^{n+1}}}{n\overline{\tilde{\lambda}}-1}+g_1\overline{\tilde{\lambda}^{n}}\right)\right].\end{aligned}\ ] ] if @xmath300 , we see that : @xmath301 going back to the original variables , we finally obtain , with the notation of the main text : @xmath302 which can be rewritten in the form ( [ dexact ] ) , completing the proof . the condition of convergence is : @xmath303 a necessary and sufficient condition for this is @xmath304 . when the parameters @xmath1 are i.i.d . r. v. the probability of this typically approaches @xmath7 as @xmath0 grows .

we establish the well - posedness of compressible vortex sheets and entropy waves in two - dimensional steady supersonic euler flows over lipschitz walls under a boundary perturbation . in particular , when the total variation of the incoming flow perturbation around the background strong vortex sheet / entropy wave is small , we prove that the two - dimensional steady supersonic euler flows containing a strong vortex sheet / entropy wave past a lipschitz wall arestable . both the lipschitz wall ( whose boundary slope function has small total variation ) and incoming flow perturb the background strong vortex sheet / entropy wave . the weak waves are reflected after nonlinear waves interact with the strong vortex sheet / entropy wave and the wall boundary . using the wave - front tracking method , the existence of solutions in over lipschitz walls is first shown , when the incoming flow perturbation of the background strong vortex sheet / entropy wave has small total variation . then we establish thecontraction of the solutions with respect to the incoming flows . to achieve this , a lyapunov functional , equivalent to thedistance between two solutions containing strong vortex sheets / entropy waves , is carefully constructed to include the nonlinear waves generated both by the wall boundary and from the incoming flow . this functional is then shown to decrease in the flow direction , leading to thestability , as well as the uniqueness , of the solutions . furthermore , the uniqueness of solutions extends to a larger class of viscosity solutions .

1205.4429

we study the well - posedness of two - dimensional steady supersonic euler flows past a curved lipschitz wall containing strong vortex sheets / entropy waves in the @xmath2norm . the inviscid compressible flows are governed by the two - dimensional steady euler system : @xmath3 with @xmath4 , @xmath5 , @xmath6 , and @xmath7 representing the fluid velocity , scalar pressure , density , and total energy , respectively . furthermore , the total energy @xmath7 is explicitly given by @xmath8 where the internal energy @xmath9 can be written as a function of @xmath10 defined through the thermodynamical relations . the temperature @xmath11 and entropy @xmath12 are the other two thermodynamic variables . in the case of an ideal gas , the pressure _ p _ and internal energy _ e _ can be expressed as @xmath13 with the adiabatic index @xmath14 given by @xmath15 in particular , in terms of the density @xmath6 and entropy _ s _ , we have @xmath16 the constants @xmath17 , and @xmath18 in the above relations are all greater than zero . when the entropy @xmath12 = constant , the flow is called _ isentropic_. in this case , the pressure @xmath5 can be written as a function of the density @xmath6 , @xmath19 , and the flow is governed by the isentropic euler equations : @xmath20 then , by scaling , the pressure - density relation is @xmath21 the adiabatic exponent @xmath22 corresponds to the isentropic polytropic gas . the limiting case @xmath23 corresponds to the isothermal flow . define @xmath24 as the sonic speed . for polytropic gases , the sonic speed is @xmath25 . the flow type is classified by the _ mach number _ @xmath26 . when @xmath27 , system or governs a _ supersonic _ flow ( i.e. , @xmath28 ) , which has all real eigenvalues and is hyperbolic . for @xmath29 , system or governs a _ subsonic _ flow ( i.e. , @xmath30 ) , which has complex eigenvalues and is elliptic - hyperbolic mixed and composite . when @xmath31 , the flow is called sonic_. we are interested in whether compressible vortex sheets / entropy waves in supersonic flow over the lipschitz wall are always stable under the @xmath0 perturbation of the incoming flow . multidimensional steady supersonic euler flows are important in many physical applications ( cf . courant - friedrichs courant - friedrichs-1948 ) . in particular , when the upstream flow is a uniform steady flow above the plane wall in @xmath32 all the time , the flow downstream above a lipschitz wall in @xmath33 is governed by a steady euler flow after a sufficiently long time . moreover , compressible vortex sheets and entropy waves occur ubiquitously in nature and are fundamental waves . furthermore , since steady euler flows are asymptotic states and may be global attractors of the corresponding unsteady euler flows , it is important to establish the existence of steady euler flows and understand their qualitative behavior to shed light on the long - time asymptotic behavior of the unsteady compressible euler flows , one of the most fundamental problems in mathematical fluid dynamics which is still wide open . we observe that the stability of contact discontinuities for the cauchy problem for strictly hyperbolic systems in one space dimension under a @xmath0 perturbation has been studied by sabl@xmath34-tougeron sable-1993 and corli sabl@xmath34-tougeron corli - sable-1997 . in particular , the reflection coefficients , such as @xmath35 here , are required to be less than one , which is the stability condition for the mixed problem in the strip @xmath36 in the earlier works ; see , e.g. , sabl@xmath34-tougeron sable-1993 . working with the non - isentropic euler system and a uniform upstream flow , chen - zhang - zhu chen - zhang - zhu-2007 first proved the global existence in @xmath0 of supersonic euler flows containing a strong vortex sheet / entropy wave under the @xmath0 perturbation of the lipschitz wall by using the glimm scheme . the essential difference between system as analyzed in chen - zhang - zhu-2007 ( and in sections 27 here ) and strictly hyperbolic systems as considered in corli - sable-1997 , sable-1993 is that two of the four characteristic eigenvalues coincide and have two linearly independent eigenvectors which determine precisely the compressible vortex sheets and entropy waves so that two independent parameters are required to describe them , respectively . in this paper , for completeness , we first show , via the wave - front tracking method , the existence of solutions to the problem when a small @xmath0 perturbation is added to the uniform incoming flow . then the @xmath2stability of entropy solutions containing strong vortex sheets / entropy waves is established . as corollaries of these results , the estimates on the uniformly lipschitz semigroup @xmath37 of entropy solutions generated by the wave - front tracking approximations are obtained , and the uniqueness of weak solutions containing strong vortex sheets / entropy waves is established in a larger set of solutions , namely the class of viscosity solutions . in the following , we focus mainly on the problem in the region @xmath38 over the lipschitz wall for the supersonic euler flows @xmath39 governed by system , given that the corresponding problem for the isentropic system is simpler to analyze . the subsequent figure provides the schematic diagram for the problem we study : the boundary and initial data in the problem are as follows : * there is a lipschitz function @xmath40 such that @xmath41 and @xmath42 denote @xmath43 , @xmath44 , and @xmath45 = @xmath46 as the outer normal vectors to @xmath47 at the respective points @xmath48 ( cf . [ figure1 ] ) . * the incoming flow @xmath49 at @xmath50 is composed of two parts : 1 . the upstream flow @xmath51 consists of one straight vortex sheet / entropy wave @xmath52 and two constant vectors @xmath53 , when @xmath54 , and @xmath55 , when @xmath56 , satisfying @xmath57 where @xmath58 is the sonic speed of state @xmath59 . 2 . the @xmath0 perturbation @xmath60 at @xmath61 so that @xmath62 . then we consider the following initial - boundary value problem for system : @xmath63 * definition 1.1 * ( _ admissible entropy solutions _ ) . a @xmath0 function @xmath64 is said to be an entropy solution of the initial - boundary value problem and if and only if the following conditions hold : * @xmath65 is a weak solution of ( 2.1 ) and satisfies @xmath66 * @xmath65 satisfies the _ steady entropy clausius inequality _ : @xmath67 in the distributional sense in @xmath38 including the lipschitz wall boundary . one of the essential developments within this paper is to develop suitable methods to deal with the challenges caused by the nonstrictly hyperbolicity of the system and the lipschitz wall boundary , in comparison with the previous progress with the strictly hyperbolic systems of conservation laws , particularly to the analysis of the cauchy problem . for supersonic euler flow with a strong shock - front emanating from the wedge vertex , chen - li chen - li-2008 worked out the issue for a lipschitz wedge boundary . we now discuss some main differences in our work here from the cauchy problem and the resulting key difficulties . we remark that , in the case of the cauchy problem concerning only @xmath68 waves , the decrease of the lyapunov functional and the @xmath2stability of the solutions were obtained through the cancellation of distances on both sides of waves . in the presence of a strong shock , for the @xmath2stability of solutions of the cauchy problem for strictly hyperbolic systems of conservation laws , the lyapunov functional was found to decrease by employing the strength of the strong shock to control the strengths of weak waves of the other families ( e.g. , see lewicka - trivisa lewicka - trivisa-2002 ) . in contrast with our lipschitz wall problem , which is a problem of initial - boundary value type , there is no such cancellation by the boundary as only one - side is possible near it . furthermore , no strong vortex sheets / entropy waves ( characteristic discontinuities ) nor strong shocks are present to handle the strength of the weak waves of the other families , and the terms in the estimates for the first and fourth family carry different signs . as such , it is difficult to say whether the functional can be made to decrease for our case of strong vortex sheets and entropy waves with multiplicity of eigenvalues . one of the key steps to resolve this is to use the physical feature of the boundary condition that the flow of two solutions near the boundary must run in parallel ( also see chen - li-2008 ) . this observation helps us to obtain additional quantitative relations near the boundary . then , applying suitable weights and adjustments in the coefficients of the lyapunov functional and using the cancellation between the different families , the functional is found to decrease in the flow direction . the rest of the paper is organized as follows . in section 2 , we recall some fundamental properties of the two - dimensional steady euler system and discuss related nonlinear waves and wave interaction estimates . in section 3 , the wave - front tracking algorithm is discussed , working in the presence of strong vortex sheets / entropy waves , the suitable interaction potential @xmath69 is constructed , including the effect of the lipschitz wall , and the existence of entropy solutions in @xmath0 is established for the initial - boundary value problem . in section 4 , we construct the lyapunov functional @xmath70 ( equivalent to the @xmath2distance between two entropy solutions @xmath65 and @xmath71 ) to include the nonlinear waves produced by the wall boundary vertices . then , in section 5 , the monotone decrease of the functional @xmath70 is established in the flow direction , leading to the @xmath2stability of the solutions containing strong vortex sheets / entropy waves . using the estimates established in sections 35 , in section 6 , we obtain the existence of a lipschitz semigroup of solutions generated by a wave - front tracking approximation , as well as some estimates on the uniformly lipschitz semigroup @xmath37 produced by the limit of wave - front tracking approximations . moreover , the uniqueness of solutions with strong vortex sheets / entropy waves is obtained in the larger class of viscosity solutions . in this section , we first present some basic properties of the steady euler system . then related nonlinear waves and interaction estimates are discussed , which will be employed in the later sections . consider the following vector functions of the solution @xmath65 : @xmath72 @xmath73 where @xmath74 . then the steady euler equations in can be expressed in the following conservative form : @xmath75 when @xmath76 is a smooth solution , system is equivalent to @xmath77 then the roots of the fourth degree polynomial @xmath78 are the eigenvalues of ; that is , the solutions of the equation @xmath79 where @xmath80 is the sonic speed . for supersonic flows ( i.e. @xmath28 ) , system is hyperbolic . specifically , when @xmath81 , system has four real eigenvalues in the @xmath82-direction : @xmath83 with the four corresponding linearly independent eigenvectors given by @xmath84 where @xmath85 the re - normalization factors such that @xmath86 , given that the @xmath87th - characteristic fields , @xmath88 , are genuinely nonlinear . the second and third linearly degenerate characteristic fields satisfy @xmath89 , @xmath90 , which correspond to vortex sheets and entropy waves , respectively . the wave curves in the phase space are now described . the rankine - hugoniot jump conditions for are @xmath91 = \left[h(u)\right],\ ] ] and the discontinuity propagates with the speed @xmath92 . there are two different waves associated with the fields @xmath93 , with the corresponding linearly independent right eigenvectors @xmath94 , in : _ vortex sheets _ : @xmath95 _ entropy waves _ : @xmath96 albeit the two contact discontinuities , the vortex sheet and the entropy wave , above match as a single discontinuity in the physical @xmath97plane , two independent parameters are needed to describe them in the phase space @xmath98 since there are two linearly independent eigenvectors corresponding to the repeated eigenvalues @xmath99 of the linearly degenerate characteristics fields . the nonlinear waves associated with @xmath100 , are shock waves and rarefaction waves . the shock waves have their speeds of propagation given by @xmath101 where @xmath102 and @xmath103 . substituting @xmath104 @xmath105 , into , the @xmath87-hugoniot curve @xmath106 through the state @xmath107 is @xmath108 = \frac{c^{2}_{0}}{b_{0}}[\rho ] , \hspace{3 mm } [ u ] = -\sigma_{d}[v ] , \hspace{3 mm } \rho_{0}(\sigma_{d}u_{0 } - v_{0})[v ] = [ p ] , \qquad d = 1,4.\ ] ] written as @xmath109 , @xmath110 , the half curves of @xmath106 for @xmath111 in the phase space are said to be the shock curves on which any state forms a shock with the below state @xmath107 in the @xmath97plane respecting the entropy condition . furthermore , for each @xmath112 or @xmath113 , the curves @xmath114 and @xmath115 at the state @xmath107 have the same curvature . if @xmath65 is a piecewise smooth solution ( see also chen - zhang - zhu-2007 ) , then any of the following conditions below is equivalent to the entropy inequality in definition 1.1 for a shock wave : * _ the physical entropy condition _ : the density increases across the shock in the flow direction , @xmath116 * _ the lax entropy condition _ : on the @xmath87th - shock , the shock speed @xmath117 satisfies @xmath118 the rarefaction wave curves @xmath119 through the state @xmath107 in the state space are given by @xmath120 we next discuss several essential properties of the nonlinear waves and related wave interaction estimates in lemmas 2.12.7 below . these facts will be used in the later sections . we also refer the reader to chen - zhang - zhu chen - zhang - zhu-2007 for further details . riemann problems and riemann solutions _ we focus on the related riemann problems and their solutions in this section , which serve as the building blocks for the front tracking algorithm for the initial - boundary value problem and . _ riemann problem of lateral - type . _ we note that the _ straight - sided wall problem _ is the case when problem and is considered with the boundary @xmath121 . it can be seen that , when the angle between the straight - sided wall and the flow direction of the incoming flow is zero , problem and has an entropy solution made up of two constants states @xmath122 and @xmath123 , satisfying @xmath124 in the subdomains @xmath125 and @xmath126 of @xmath38 separated by a straight vortex sheet / entropy wave . these are precisely the states @xmath127 and @xmath128 below and above the large vortex sheet / entropy wave . the principal aim of this paper is to establish the @xmath2well - posedness for problem and for the solutions near the background solution containing a strong vortex sheet / entropy wave @xmath129 with @xmath121 . it has been observed in courant - friedrichs-1948 that , if the angle between the flow direction of the front state and the wall at a boundary vertex is smaller than @xmath130 and larger than the extreme angle determined by the incoming flow state and @xmath131 , then a unique @xmath132-shock is generated , separating the front - state from the supersonic back - state . if the angle between the flow direction of the front - state and the wall at a boundary vertex is larger than @xmath130 and less than the extreme angle , then a @xmath132-rarefaction wave is produced , emanating from the vertex . these waves are easily seen through the shock polar analysis ( cf . chen - zhang - zhu-2007 , courant - friedrichs-1948 ) . this signifies that , when the angle between the flow direction of the front - state and the wall at a boundary vertex is close to @xmath130 , the lateral riemann problem can be uniquely solved . for further details , see lemma 2.3 and chen - zhang - zhu-2007 . for an indepth discussion , we also refer to courant friedrichs courant - friedrichs-1948 . _ riemann problem involving only weak waves . _ consider the subsequent initial value problem with piecewise constant initial data : @xmath133 u|_{x = x_{0 } } = \underline{u } = \begin{cases } u_{a } , \hspace{2 mm } y > y_{0 } , \\ u_{b } , \hspace{2 mm } y < y_{0 } , \end{cases } \end{cases}\ ] ] with the constant states @xmath134 and @xmath135 denoting the _ above _ state and _ below _ state with respect to the line @xmath136 , respectively . then there is @xmath137 so that , for any states @xmath138 in the neighborhood @xmath139 of @xmath128 , or @xmath138 in the neighborhood @xmath140 of @xmath127 , the initial value problem has a unique admissible solution consisting of four waves , consisting of shocks , rarefaction waves , vortex sheets and/or entropy waves . _ riemann problem involving the strong vortex sheets / entropy waves . _ from now on , the notation @xmath141 will be used to write @xmath142 as the solution of the riemann problem , where @xmath143 , and @xmath144 is the strength of the @xmath145-wave . for any wave with @xmath146 and @xmath147 , we also use @xmath148 to denote the strong vortex sheet / entropy wave that connects @xmath135 and @xmath134 with strength @xmath149 . that is , @xmath150 particularly , we observe that @xmath151 we write @xmath152 for any @xmath146 . then we have @xmath153 _ the vector function g(@xmath154 ) satisfies _ @xmath155 _ and _ @xmath156 _ furthermore , for the plane vortex sheet and entropy wave with the lower state @xmath157 , upper state @xmath158 , and strength @xmath159 _ , @xmath160 these can be easily obtained from direct calculations and are thus omitted . the properties in above play a fundamental role in achieving the necessary estimates on the strengths of reflected weak waves in the interaction between the strong vortex sheet / entropy wave and weak waves ( see the proofs for lemmas 2.42.7 ) . _ 2.2 . wave interactions and reflection estimates . _ in the following , several essential estimates are provided on wave interactions and reflections . for their proofs and all the related details , we refer to chen - zhang - zhu-2007 . for the weak wave interaction away from both the strong vortex sheet / entropy wave and the wall boundary in the regions @xmath125 or @xmath126 , we have the following estimate : @xmath161 _ assume that @xmath162 are three states with @xmath163 , @xmath164 . then @xmath165 with _ @xmath166 _ where @xmath167 and _ @xmath168 we write @xmath169 for the points @xmath170 in the @xmath97plane with @xmath171 . define @xmath172 \omega_{l+1 } = \lbrace ( x , y ) : x \in \left[a_{l } , a_{l+1}\right ] , y > b_{l } + ( x - a_{l})\text{tan}(\theta_{{l } , { l+1}})\rbrace , \\[2 mm ] \gamma_{l+1 } = \lbrace ( x , y ) : x \in \left(a_{l } , a_{l+1}\right ) , y = b_{l } + ( x - a_{l})\text{tan}(\theta_{{l } , { l+1}})\rbrace , \end{array } \right.\end{aligned}\ ] ] and the outer normal vector to @xmath173 : @xmath174 with the constant state @xmath175 , consider the following initial - boundary value problem : @xmath176 @xmath177 _ suppose @xmath178 and @xmath179 with _ @xmath180 _ then there exists a unique solution @xmath181 of problem such that @xmath182 and @xmath183 . moreover , _ @xmath184 _ where @xmath185 , @xmath186 , @xmath187 , and @xmath188 are @xmath189-functions of @xmath190 , @xmath191 , @xmath192 , @xmath193 , @xmath194 , and @xmath134 satisfying _ @xmath195 _ and @xmath188 is bounded . in particular , @xmath196 at the origin . _ this lemma has two purposes . the first is to estimate the weak waves generated by the vertices on the lipschitz wall boundary . this boundedness will be used to control the boundary perturbation ; see in the construction of the wave interaction potential @xmath197 . the second is to estimate the strength of the reflected wave @xmath198 with respect to the incident wave @xmath199 . property of the coefficients will play an important role to control the reflected waves . _ estimates on the interaction between the strong vortex sheet / entropy wave and weak waves from below . _ estimate below plays a key role in ensuring the @xmath1stability of entropy solutions , especially for the existence of the constants @xmath200 and @xmath201 in lemma 5.1 ( see below ) . this estimate also ensures the existence of @xmath202 ( @xmath35 , 1 ) in the construction of the wave interaction potential @xmath197 in . @xmath203 _ let @xmath204 and @xmath147 with _ @xmath205 _ then there exists a unique @xmath206 such that the riemann problem admits an admissible solution that consists of a weak @xmath207wave of strength @xmath208 , a strong vortex sheet of strength @xmath209 , a strong entropy wave of strength @xmath210 , and a weak @xmath211wave of strength @xmath198 : _ @xmath212 _ moreover , _ @xmath213 _ where @xmath214 , and @xmath215 is bounded . _ @xmath216 _ the coefficient @xmath217 in the strength @xmath218 of a weak 4-wave in lemma 2.4 remains bounded away from zero . _ @xmath219 by lemma 2.4 , we can find a unique solution @xmath220 as a @xmath221-function of @xmath222 @xmath223 , and @xmath224 to @xmath225 that is , @xmath226 where we have omitted @xmath224 for simplicity . from chen - zhang - zhu-2007 , we know that @xmath227 differentiate with respect to @xmath228 , and let @xmath229 , and @xmath230 . we obtain @xmath231 by lemma 2.1 , we have + @xmath232 @xmath233 this completes the proof . _ estimates on the interaction between the strong vortex sheet / entropy wave and weak waves from above . _ we have @xmath234 _ let @xmath146 and @xmath235 with _ @xmath236 _ then there exists a unique @xmath206 such that the riemann problem admits an admissible solution that consists of a weak @xmath207wave of strength @xmath208 , a strong vortex sheet of strength @xmath209 , a strong entropy wave of strength @xmath210 , and a weak @xmath211wave of strength @xmath198 : _ @xmath237 _ moreover , _ @xmath238 @xmath239 _ where @xmath240 is bounded and @xmath241 _ the constant @xmath242 here is used in the definition of weighted strength @xmath243 of weak waves in . @xmath244 _ the coefficient @xmath245 in the strength @xmath246 of a weak 1-wave in lemma 2.6 remains bounded away from zero , while the reflection coefficient @xmath247 . _ @xmath248 by lemma 2.6 , we can find a unique solution @xmath220 as a @xmath221-function of @xmath249 , @xmath250 , @xmath251 , @xmath252 , and @xmath224 to @xmath253 that is , @xmath254 where we have omitted @xmath224 for simplicity . from chen - zhang - zhu-2007 , we know that @xmath255 differentiate with respect to @xmath250 and let @xmath256 , and @xmath230 . we obtain @xmath257 by lemma 2.1 , we have @xmath258 however , for the reflection coefficient @xmath259 , we have @xmath260 where @xmath259 is not necessarily bounded away from zero , but is less than one . we start off here with a brief description of the wave - front tracking method to be employed throughout in sections 47 and then establish the existence of entropy solutions when the perturbation of the incoming flow has small total variation at @xmath61 . the main scheme in the method of wave - front tracking is to construct approximate solutions within a class of piecewise constant functions . at the beginning , we approximate the initial data by a piecewise constant function . then we solve the resulting riemann problems exactly with the exception of the rarefaction waves , which are replaced by rarefaction fans with many small wave - fronts of equal strengths . the outgoing fronts are continued up to the first time when two waves collide and a new riemann problem is solved . in this process , one has to modify the algorithm and introduce a simplified riemann solver in order to keep the number of wave - fronts finite for all @xmath261 in the flow direction . we refer the reader to bressan @xcite and baiti - jenssen @xcite for related references . the riemann solvers_. as seen in section 2 , the solution to the riemann problem @xmath262 is a self - similar solution given by at most five states separated by shocks , vortex sheets , entropy waves , or rarefaction waves . to connect the state @xmath263 to @xmath224 , there exist @xmath221curves @xmath264 with arc length parametrization such that @xmath265 for some @xmath266 , and @xmath267 , @xmath268 . next , we discuss the construction of front tracking approximations for our initial - boundary value problem . let @xmath269 denote the initial approximation parameter . for given initial data @xmath270 and with @xmath271 , consider @xmath272 a sequence of piecewise constant functions approximating @xmath270 in the @xmath1norm , and the wall boundary is also approximated as described in section 2 . set @xmath273 to be the total number of jumps in the initial data @xmath272 and the tangential angle function of the wall boundary . let @xmath274 be a parameter so that a rarafaction wave is replaced by a step function whose `` steps '' are no further apart than @xmath275 . the discontinuity between two steps is set to propagate with a speed equal to the rankine - hugoniot speed of the jump connecting the states corresponding to the two steps . at any time , the simplified riemann solver ( defined below ) is employed , the constant @xmath276 denotes the speed of the generated non - physical wave , which is strictly greater than all the wave speeds of system ( 2.1 ) . note that the strength of the non - physical wave is the error generated when the simplified riemann solver is applied . @xmath277 the accurate riemann solver ( * ars * ) is the exact solution to the riemann problem , with the condition that every rarefaction wave @xmath278 , @xmath105 , is divided into equal parts and replaced by a piecewise constant rarefaction fan of several new wave - fronts of equal strength . @xmath279 when only weak waves are involved , the simplified riemann solver ( @xmath280 ) here is the same as the one described in baiti - jenssen-1998 , bressan-2000 . that is , all new waves are put together in a single @xmath281 , travelling faster than all characteristic speeds . in the case of a weak wave interacting with the strong vortex sheet / entropy wave , the purpose of ( @xmath280 ) is to ignore the strength of the weak wave , while preserving the strength of the strong vortex sheet / entropy wave , and to place the error in the non - physical wave in the following manner : . the riemann problem @xmath282 is solved as follows : @xmath283 with @xmath284 as the speed of the strong vortex sheet / entropy wave , and the state @xmath285 is solved in a way that @xmath286 is the strong vortex sheet / entropy wave starting from @xmath287 and @xmath288 . hence , we find that ( * srs * ) keeps the same strength of the strong vortex sheet / entropy wave , and the error appears in the non - physical fronts . . the riemann problem @xmath282 is solved as follows : @xmath289 with @xmath290 denoting the speed of the strong vortex sheet / entropy wave . construction of wave front tracking approximations _ given @xmath269 , the corresponding front tracking approximate solution @xmath291 is built up as follows . at @xmath61 , all the riemann problems in @xmath272 are solved by using the accurate riemann solver . furthermore , one can change the speed of one of the incoming fronts so that , at any time @xmath33 , there is at most one collision involving only two incoming fronts . of course , this adjustment of speed can be chosen arbitrarily small . let @xmath292 be a fixed small parameter with @xmath293 , as @xmath294 , which will be determined later . for convenience , the index @xmath145 in @xmath144 will be dropped henceforward , and we will write @xmath144 as @xmath295 when there is no ambiguity involved ; the same applies for @xmath296 ; and we will moreover employ the same notation @xmath295 as a wave and its strength as before . _ case 1 : two weak waves with strengths @xmath295 and @xmath296 interact at some @xmath33_. the riemann problem produced by this collision is solved in the following way : * if @xmath297 and the two waves are physical , then the accurate riemann solver is employed . * if @xmath298 and the two waves are physical , or there is a non - physical wave , then the simplified riemann solver is employed . _ case 2 : a weak wave @xmath295 interacts with the strong vortex sheet / entropy wave and one weak wave at some @xmath33_. the riemann problem produced by this collision is solved in the following way : * if @xmath299 and the weak wave is physical , then the accurate riemann solver is applied . * if @xmath300 and the weak wave is physical , or this wave is non - physical , then the simplified riemann solver is applied . _ case 3 : the flow perturbation due to the lipschitz wall boundary_. * when the change of the angle of the boundary is larger than @xmath292 and the weak wave is physical , then the accurate riemann solver is employed to solve the lateral riemann problem . * if the change of the angle of the boundary is less than @xmath292 , then this perturbation is ignored . _ case 4 : the physical wave collides with the boundary_. the accurate riemann solver is employed to solve the lateral riemann problem . _ case 5 : the non - physical wave collides with the boundary_. we can allow these waves to cross the boundary . glimm s functional and wave interaction potential _ the goal in this subsection is to construct the suitable glimm - type functional and the associated wave interaction potential @xmath69 for our initial - boundary value problem . this involves a careful incorporation of the additional nonlinear waves generated from the wall boundary vertices . @xmath301 ( _ approaching waves _ ) . ( i ) two weak fronts @xmath295 and @xmath296 , located at points @xmath302 and of the characteristic families @xmath303 , @xmath304 @xmath305 @xmath306 , respectively , are said to be approaching each other if the following two conditions are concurrently satisfied : * @xmath307 and @xmath308 are both in one of the two intervals into which @xmath309 is partitioned by the location of the strong vortex sheet / entropy wave . that is , both waves are either in @xmath310 or @xmath311 ; * either @xmath312 or else @xmath303 = @xmath304 and at least one of them is a genuinely nonlinear shock . in this case , we write @xmath313 @xmath305 @xmath314 . \(ii ) we say that a weak wave @xmath295 of the characteristic family @xmath303 is approaching the strong vortex sheet / entropy wave if either @xmath315 and @xmath316 , or @xmath317 and @xmath318 . we then write @xmath319 . \(iii ) we say that a weak wave @xmath295 of the characteristic family @xmath303 is approaching the boundary if @xmath315 and @xmath318 . we then write @xmath320 . define the total ( weighted ) strength of weak waves in @xmath321 as @xmath322 here , for a weak wave @xmath295 of the @xmath145-family , its weighted strength is defined as @xmath323 where @xmath324 and the coefficient @xmath242 as given in lemma 2.6 . next , the wave interaction potential @xmath197 is defined as @xmath325 here the constants @xmath202 ( @xmath35 , 1 ) and @xmath326 , while @xmath327 is a constant to be specified later . to control the total variation of the new waves produced by the boundary vertices , @xmath328 in our wave interaction potential @xmath197 is an added term , compared to that for the cauchy problem . _ the glimm - type functional @xmath329 is defined as follows _ @xmath330 _ where the states @xmath331 and @xmath332 are the below state and the above state of the strong vortex sheet / entropy wave respectively at time " x , @xmath333 and @xmath334 are the below and above state of the strong vortex sheet / entropy wave respectively at @xmath61 , and @xmath18 is a large positive constant to be determined later . _ we remark that @xmath335 , @xmath69 , and @xmath329 remain unchanged between any pair of subsequent interaction times . however , we will demonstrate that , across an interaction time " @xmath82 , both @xmath69 and @xmath329 decrease . @xmath336 _ assume that @xmath337 is sufficiently small . then v@xmath338 will remain sufficiently small for all @xmath33 . moreover , the quantity @xmath339 has a uniform bound for any @xmath271_. @xmath219 with the glimm - type functional @xmath329 , consider @xmath340 where @xmath341 and @xmath342 denote the times " before and after the interaction time " @xmath33 , respectively . _ case 1 : two weak waves @xmath295 and @xmath296 collide_. then the states @xmath343 and @xmath344 do not alter across this interaction time " @xmath33 . hence , we have @xmath345 where @xmath346 and @xmath347 are constants independent of @xmath269 . _ case 2 : a weak wave @xmath295 of the 1-family interacts with the boundary_. @xmath348 where @xmath349 . _ case 3 : a new 4-wave @xmath295 produced by the lipschitz wall boundary_. @xmath350 where @xmath351 is large . in the following two cases , the states @xmath331 and @xmath332 change across this interaction time " @xmath33 . _ case 4 : a weak wave @xmath295 of the 4-family collides with the strong vortex sheet / entropy wave from below_. @xmath352 _ case 5 : a weak wave @xmath295 of the 1-family collides with the strong vortex sheet / entropy wave from above_. @xmath353 in the cases above , @xmath354 , @xmath355 in connection with the weight @xmath356 , and the constant @xmath357 is large . next , we establish that the _ total ( weighted ) strength of waves _ in @xmath321 remain sufficiently small for all @xmath33 if it is sufficiently small at @xmath50 . more precisely , @xmath358 this can be proved as follows : _ ( i ) time " @xmath359 is the first interaction_. given that @xmath360 and @xmath361 in cases 15 above , we conclude that , for @xmath18 sufficiently large and @xmath292 small enough , @xmath362 therefore , @xmath363 _ ( ii ) @xmath364 and @xmath365 for any @xmath366_. then , for the next interaction time " @xmath367 , similar to case 1 , we also conclude @xmath368 therefore , all together , we obtain @xmath369 this implies that @xmath370 for all @xmath33 , since @xmath371 is independent of @xmath82 . furthermore , the total variation of @xmath321 is uniformly bounded . more precisely , we conclude that @xmath372 this completes the proof . in order to have a front tracking approximate solution @xmath373 defined for any time @xmath33 , along with a uniform bound on the total variation , we also need to have that the number of wave - fronts in @xmath373 is finite . this is given by the subsequent lemma . @xmath374 _ for any fixed @xmath271 small enough , the number of wave - fronts in @xmath375 is finite and the approximate solutions @xmath375 are defined for all @xmath33 . moreover , for any @xmath33 , the total strength of the all non - physical waves is of order @xmath376_. @xmath219 we first note the total interaction potential @xmath197 remains unchanged when there is no interaction and decreases across an interaction time " @xmath33 as discussed in cases 15 in proposition 3.1 . furthermore , from cases 15 and the subsequent analysis above , we have concluded that @xmath370 . hence , one can fix some number @xmath377 such that @xmath378 now , following an argument similar to the one given in baiti - jenssen-1998 , we reach the following conclusions . note that initially @xmath379 is bounded and @xmath380 decreases thereafter for each case . moreover , in the case where the interaction potential between the incoming waves or the change of the angle of the boundary is larger than @xmath292 , @xmath380 decreases by at least @xmath381 in these interactions , as implied by the bounds given in . following the wave - front tracking method in our problem , new physical waves can be only produced by such interactions . furthermore , when the weak wave @xmath295 of 1-family collides with the wall boundary , we solve the lateral riemann problem and have shown earlier that , after this interaction , there is only a reflected wave of 4-family with the reflection coefficient @xmath382 . hence , before and after this interaction , the number of the waves keeps the same , and this implies that the number of the waves is finite . finally , because non - physical waves are generated only when physical waves collide , we can also conclude that the number of non - physical wave fronts are finite ; and , provided that two waves can only collide once , the number of interactions is also finite . consequently , it follows that the approximate solutions @xmath321 are defined for all times @xmath33 . the similar argument allows us to conclude that the total strength of all non - physical wave fronts at any @xmath82 is of order @xmath383 . this completes the proof lemma 3.2 . following the line of arguments given in baiti - jenssen-1998,bressan-1992 for the wave - front tracking algorithm and lemma 3.1 above , we finish this section with the following theorem for the global existence of entropy solutions to the initial - boundary value problem and . @xmath384 _ suppose that @xmath385 is small enough . then , for the initial - boundary value problem and , there exists a global weak solution in bv satisfying the _ steady clausius entropy _ inequality . _ [ [ section ] ] to show that the front tracking approximations , constructed for the existence analysis in section 3 , converge to a unique limit , we estimate the distance between any two @xmath269-approximate @xmath65 and @xmath71 of problem and of initial - boundary value type . to this end , we develop the lyapunov functional @xmath386 , equivalent to the @xmath1distance : @xmath387 and prove that the functional @xmath388 is _ almost decreasing _ along pairs of solutions , @xmath389 for some constant @xmath390 . here @xmath65 and @xmath71 are two approximate solutions constructed via the wave - front tracking method , and the small approximation parameter @xmath269 is responsible for controlling the subsequent errors : errors in the approximation of the initial data and the boundary . errors in the speeds of shock , vortex sheet , entropy wave , and rarefaction fronts . the total strength of all non - physical fronts . the maximum strength of rarefaction fronts . along the line of arguments presented in bressan - liu - yang-1999 , lewicka - trivisa-2002 , liu - yang-1999 , with `` time '' @xmath82 fixed , at each @xmath391 , one connects the state @xmath392 with @xmath393 in the state space by going along the hugoniot curves @xmath394 , and @xmath395 . depending on the location of the strong vortex sheet / entropy wave in @xmath392 and @xmath393 , the distance between @xmath392 and @xmath393 is estimated along discontinuity waves in possibly different `` directions '' , determining the strength of the @xmath145-hugoniot wave @xmath396 in the following way : suppose that @xmath392 and @xmath393 are both in @xmath310 and @xmath311 . then one begins at the state @xmath392 and moves along the hugoniot curves to reach the state @xmath393 . suppose that @xmath392 is in @xmath310 and @xmath393 is in @xmath311 . then one begins at the state @xmath392 and moves along the hugoniot curves to reach the state @xmath393 . suppose that @xmath393 is in @xmath310 and @xmath392 is in @xmath311 . then one begins at the state @xmath393 and moves along the hugoniot curves to reach the state @xmath392 . define the @xmath1weighted strengths of the waves in the solution of the riemann problem @xmath397 or @xmath398 as follows : @xmath399 with the constants @xmath400 , @xmath401 , and @xmath402 above to be specified later on , based on the estimates of wave interactions and reflections in lemmas 2.22.7 . we define the following lyapunov functional , @xmath403 where the weights are given by @xmath404 the constants @xmath405 and @xmath406 are to be determined later . here @xmath69 denotes the total wave interaction potential incorporating the boundary effect as defined in , and @xmath407 denotes the total strength of waves in @xmath65 and @xmath71 , which approach the @xmath145-wave @xmath408 , defined in the following manner ( for @xmath391 where there is no jump in @xmath65 or @xmath71 ) : @xmath409 next , we define the following global weights @xmath410 : [ cols="<,<,<,<",options="header " , ] under the assumption that @xmath411 is small enough with @xmath412 , @xmath413 @xmath414 , one concludes @xmath415 where the constant @xmath416 is independent of @xmath269 and `` time '' @xmath82 . here _ we define the strength of any large wave of the @xmath417- or @xmath418-characteristic family to equal to some fixed number b ( bigger than all strengths of small waves ) , _ and the concepts small " and large " mean the waves that connect the states in the same or in the distinct domains @xmath419 and @xmath420 , respectively . the summands in are defined as follows , @xmath421 where , at each @xmath82 , @xmath295 stands for the ( non - weighted ) strength of the wave @xmath422 , located at the point @xmath307 and belonging to the characteristic family @xmath423 ; @xmath424 , @xmath425 is the set of all waves ( in @xmath65 and @xmath71 ) and the set of all large ( strong ) characteristic discontinuities ( in @xmath65 and @xmath71 ) respectively . consequently , there holds @xmath426 for any @xmath261 with the constant @xmath390 depending only on the quantities independent of @xmath82 : the strength of the strong vortex sheet / entropy wave and @xmath411 . we now analyze the evolution of the lyapunov functional @xmath70 in the flow direction @xmath33 . for @xmath427 , we call @xmath428 the speed of the @xmath145-wave @xmath408 ( along the hugoniot curve in the phase space ) . then , at a time " @xmath33 which is not the interaction time of the waves either in @xmath429 or @xmath430 , an explicit computation gives @xmath431 where @xmath432 denotes the speed of the hugoniot wave @xmath422 , @xmath433 stands for the points close to the boundary , and @xmath434 is the slope of the boundary . we present the notation @xmath435 e_{b , j } & = { \lvertq_{j}(b)\rvert}w_{j}(b)\left(\dot{y}_{b } + \lambda_{j}(b)\right ) , \label{4.8}\end{aligned}\ ] ] where @xmath436 = @xmath437 , @xmath438 , and @xmath439 = @xmath440 . then can be written as @xmath441 our central aim is to prove the bounds : @xmath442 where the quantities denoted by the landau symbol @xmath443(1 ) are independent of the constants @xmath444 and @xmath445 . from together with the uniform bound on the total strengths of waves , we obtain @xmath446 integration of over the interval @xmath447 $ ] yields @xmath448 we remark that , at each interaction time " @xmath82 when two fronts of @xmath65 or two fronts of @xmath71 interact , by the glimm interaction estimates , all the weight functions @xmath449 decrease , if the constant @xmath406 in the lyapunov functional is taken to be sufficiently large . furthermore , due to the self - similar property of the riemann solutions , @xmath70 decreases at this time " . in the next section , we establish the bounds , particularly and , when @xmath295 is a strong vortex sheet / entropy wave in @xmath450 and near the lipschitz wall boundary , respectively . for the case of the non - physical waves in @xmath450 , as well as the case that the weak wave @xmath451 , which appears when @xmath65 and @xmath71 are both in @xmath310 or @xmath311 , estimates and are shown similarly based on the arguments in bressan - liu - yang bressan - liu - yang-1999 , provided that @xmath452 is sufficiently small and @xmath405 is sufficiently large . in what follows , we focus only on the other two cases , namely and . : _ the first strong vortex sheet / entropy wave @xmath295 in @xmath65 or @xmath71 is crossed_. then , by lemma 2.4 , we have the estimates : @xmath453 moreover , the essential estimate @xmath454 given in lemma 2.4 ensures the existence of desired weights @xmath200 and @xmath201 in the following way . @xmath455 _ there exist @xmath200 , @xmath201 , and @xmath456 satisfying _ @xmath457 with lemma 5.1 , we estimate @xmath458 for @xmath459 , starting with @xmath460 : by and , @xmath461 where @xmath462 , and @xmath463 is a positive constant . the term @xmath464 here is the total strength of all the weak waves in @xmath65 and @xmath71 which approach the @xmath382-wave @xmath465 , and the term @xmath466 is from the weight @xmath467 . for @xmath468 , since @xmath469 , reduces to @xmath470 where @xmath471 . for @xmath145 = 4 , @xmath472 where @xmath473 , and @xmath463 is a positive constant . the term @xmath474 is the total strength of all the weak waves in @xmath65 and @xmath71 which approach the @xmath132-wave @xmath475 , and the term @xmath466 is from the weight @xmath476 . for the weighted @xmath1strength @xmath408 in , we choose @xmath200 small enough relatively to @xmath477 and @xmath201 large enough relatively to @xmath478 , choose @xmath405 large enough and the total variation of @xmath65 and @xmath71 so small , and use to obtain @xmath479 * case 2 : * _ the weak wave @xmath295 between the two strong vortex sheets / entropy waves in @xmath65 and @xmath71 is crossed_. for @xmath480 , we have @xmath481 for the cases when @xmath468 , we have @xmath482 for @xmath145 = 4 , we have @xmath483 then we have @xmath484 since @xmath485 when @xmath486 , we obtain @xmath487 if all the weights @xmath401 are small enough and @xmath405 is sufficiently large . notice that the choice of the upper or lower superscripts depends on the family number @xmath423 . : _ the second strong vortex sheet / entropy wave @xmath295 in @xmath65 or @xmath71 is crossed_. for this case , by lemma 2.6 , we have @xmath488 moreover , the essential estimate @xmath489 in lemma 2.6 ensures the existence of desired weights @xmath490 and @xmath491 in the following manner . @xmath492 _ there exist @xmath490 , @xmath491 , and @xmath493 satisfying _ @xmath494 with lemma 5.2 , we estimate @xmath458 for @xmath495 as follows : by , @xmath496 where @xmath497 , and @xmath498 is a positive constant . the term @xmath464 here is the total strength of all the weak waves in @xmath65 and @xmath71 which approach the @xmath382-wave @xmath499 , and the term @xmath500 is from the weight @xmath501 . for @xmath502 , since @xmath469 , reduces to @xmath503 where @xmath471 . by and , @xmath504 where @xmath505 , and @xmath463 is a positive constant . the term @xmath474 here is the total strength of all the weak waves in @xmath65 and @xmath71 which approach the @xmath132-wave @xmath506 , and the term @xmath466 is from the weight @xmath507 . for the weighted @xmath1strength @xmath408 in ( 4.1 ) , when @xmath491 is small enough relatively to @xmath478 and @xmath490 is large enough relatively to @xmath477 , @xmath405 is large enough , applying and , the total variation of @xmath508 and @xmath509 is so small that @xmath510 which yields . : _ close to the lipschitz wall boundary . _ this case differs from the cauchy problem . here we will use the particular property of the boundary condition : the flows of @xmath65 and @xmath71 are tangent to the lipschitz wall , implying that they must be parallel with each other along the boundary . then a piecewise constant weak solution is constructed only along the hugoniot curves determined by the riemann data @xmath511 and @xmath512 , the states of solutions @xmath65 and @xmath71 , respectively , close to the boundary . * proposition 5.2 . * _ suppose that u@xmath513 = @xmath514 and v@xmath513 = @xmath515 are states in a small neighborhood @xmath516 of @xmath127 satisfying @xmath517 , and @xmath518 . denote by @xmath519 the strength of the @xmath520 shock in the riemann problem determined by u@xmath513 and v@xmath513 , and by @xmath521 the corresponding @xmath520-characteristic speed . then _ @xmath522 & & |h_{4}(b)| \leq |h_{1}(b)| + \mathcal{o}(1)|h_{2}(b)||\lambda_{2 } - \dot{z}_{b}| + |h_{1}(b)|\mathcal{o}(1)|\dot{z}_{b}| , \\[2 mm ] & & |h_{1}(b)| = \bar{\mathcal{o}}(1)|h_{4}(b)| , \hspace{5 mm } \frac{1}{2 } < \tilde{\mathcal{o}}(1 ) < \frac{3}{2},\end{aligned}\ ] ] _ where @xmath523 is the slope of the lipschitz wall . _ * we do the proof by analyzing the following two cases . _ case 1 : @xmath524 and @xmath525 that corresponds to the case @xmath526_. starting at the state @xmath135 , we move along the hugoniot curves of the second and third families to reach @xmath527 . note that these two families are the contact hugoniot curves , and so @xmath528 and @xmath529 are constant along the hugoniot curves . given that @xmath530 , @xmath531 , and @xmath532 , the quantity @xmath533 remains unchanged as the initial value @xmath534 , i.e. , @xmath523 in this process by the boundary condition . therefore , we conclude that @xmath535 , equivalently , @xmath536 _ case 2 : @xmath537 that corresponds to @xmath538_. starting at the state @xmath511 , we move along the 1-hugoniot curve to reach @xmath539 , then possibly move along the 2-contact hugoniot curve to reach @xmath540 , the 3-hugoniot curve to reach @xmath541 , and the 4-hugoniot curve to reach @xmath512 . to make clear some essential relations among the strengths @xmath542 , we project @xmath543 onto the @xmath544plane . let @xmath545 be the projection of @xmath546 onto the @xmath508-axis , @xmath547 be the projection of @xmath548 onto the @xmath4plane ; and so on . at the background state @xmath127 , there holds @xmath549 @xmath550 we first note that @xmath551 . given that @xmath552 along with finite characteristic speeds @xmath553 and @xmath554 , there always holds @xmath555 near the state @xmath127 . so we can conclude that , in the @xmath4plane , the derivative @xmath556 along the 1-curve is always larger than @xmath523 . this implies that @xmath557 . meanwhile , we have @xmath558 and @xmath559 . hence , @xmath560 thus , we conclude that there is some distance along the 4-hugoniot curve to reach @xmath527 . therefore , @xmath561 . next , we present an essential estimate to bound @xmath562 more precisely in terms of @xmath562 . to that end , define the signed length of @xmath563 and @xmath564 by @xmath565 and @xmath566 on the @xmath4plane : @xmath567 -\|(u_{1 } - u_{b})|_{(u , v)}\|&\mbox { if $ h_{1 } < 0 $ , } \nonumber \\ \end{array } \right.\end{aligned}\ ] ] and @xmath568 -\|(v_{b } - u_{3})|_{(u , v)}\|&\mbox { if $ h_{4 } < 0$. } \nonumber \\ \end{array } \right.\end{aligned}\ ] ] secondly , we note that @xmath569 moreover , since @xmath570 , we can similarly conclude @xmath571 using the following projections on the @xmath4plane , @xmath572 @xmath573 thirdly , we note that @xmath574 where @xmath575 , @xmath576 denotes the angle between @xmath577 and @xmath578 , @xmath579 denotes the angle between @xmath580 and @xmath577 , @xmath581 for @xmath582 , and @xmath583 hence , there holds @xmath584 at @xmath127 , there also holds @xmath585 = -@xmath586 and @xmath587 , which implies @xmath588 hence we note the following key estimate : @xmath589 estimate now implies @xmath590 yielding @xmath591 given that @xmath592 is always small enough , and this is guaranteed by the sufficiently small total variation of the initial perturbation @xmath593 and the perturbation of the boundary . we note that the requirement @xmath595 in proposition 5.2 is just the boundary condition because @xmath523 here is the slope of the lipschtiz wall . applying proposition 5.2 now yields @xmath596 e_{b , j } & = & |q_{j}(b)|w_{j}(b)(\dot{z}_{b } + \lambda_{j } ) = |q_{j}(b)|w_{j}(b)(\dot{z}_{b } - \lambda_{j } ) + 2 \lambda_j |q_{j}(b)|w_{j}(b)\nonumber\\ & = & \mathcal{o}(1 ) w^{b}_{j}|h_{j}(b)|(\dot{z}_{b } - \lambda_{j } ) + \mathcal{o}(1)\lambda_{j } w^{b}_{j}|h_{j}(b)| \\ & = & \mathcal{o}(1)h_{1}(b ) = \mathcal{o}(1)h_{4}(b ) , \hspace{6 mm } j = 2 , 3 , \nonumber \\[2 mm ] e_{b,4 } & = & |q_{4}(b)|w_{4}(b)(\dot{z}_{b } + \lambda_{4})\nonumber\\ & = & 4b\kappa_1 w^{b}_{4}|h_{4}(b)||\lambda_{1}| + \mathcal{o}(1)|h_{4}(b)| \nonumber \\ & \leq & 4b\kappa_{1}|\lambda_{1}|w^{b}_{4}\big(|h_{1}(b)| + \mathcal{o}(1)|h_{2}(b)||\lambda_{2 } - \dot{z}_{b}| + \mathcal{o}(1)|h_{1}(b)||\dot{z}_{b}|\big ) + \mathcal{o}(1)|h_{4}(b)| . \nonumber\end{aligned}\ ] ] using lemma 5.1 , we can choose @xmath200 and @xmath201 such that @xmath597 then , with the total variation of the incoming flow perturbation and the boundary perturbation small enough and @xmath405 large enough , one has @xmath598 provided that @xmath592 is sufficiently small . this is guaranteed since the total variation of the incoming flow perturbation and the boundary perturbation are sufficiently small . as a corollary of the essential estimates in sections 35 , we can now establish the existence of the semigroup @xmath37 generated by the wave - front tracking method , as well as the lipschitz continuity of @xmath37 . * proposition 6.1 . * _ suppose that @xmath599 is small enough . then the map @xmath600 produced by the wave - front tracking algorithm is a uniformly lipschitz continuous semigroup satisfying the properties _ : * @xmath601 , for all @xmath602 ; * @xmath603 , for all @xmath604 . * proof . * since @xmath605 is generated by the wave - front tracking algorithm , property ( i ) is immediate . next , property ( ii ) is proved as follows . take a pair of front tracking @xmath269-approximate solutions @xmath606 and @xmath607 of and with @xmath608 and @xmath609 as the initial data , respectively . using and , at any @xmath261 , we have @xmath610 hence , the @xmath269-semigroup is lipschitz continuous . * definition 6.1 . * for a given @xmath611 , we define the domain : @xmath612 * remark 6.1 . * given a solution @xmath76 to the initial - boundary value problem of and , we note that , if @xmath613 at any fixed @xmath261 , then @xmath614 at @xmath61 and @xmath615 for @xmath33 as a strong vortex sheet / entropy wave is present . the semigroup @xmath37 generated by the wave - front tracking algorithm is provided by the next theorem . * theorem 6.1 . * _ suppose that @xmath616 is small enough . then , in the @xmath2norm , @xmath605 produced by the wave - front tracking algorithm is a cauchy sequence . denote this unique limit by @xmath37 such that , for any @xmath33 , @xmath617 . then the map @xmath618 is a uniformly lipschtiz semigroup in @xmath2 . _ _ in particular , the entropy solution to the initial - boundary problem and constructed by the wave - front tracking algorithm is unique and @xmath2 stable_. based on the essential estimates in sections 35 , the proof of theorem 6.1 can be shown in the same way as the argument given in bressan - colombo-1995 . also see chen - li chen - li-2008 . in this section , as an immediate consequence of the estimates obtained in sections 46 , we find that the semigroup @xmath37 produced by the wave - front tracking method is the only _ standard riemann semigroup _ ( srs ) in the sense of definition 7.1 given below . in other words , the semigroup defined by the wave - front tracking method is the canonical trajectory of the standard riemann semigroup ( srs ) . this yields the uniqueness of entropy solutions in a broader class of viscosity solutions as introduced by bressan in bressan-1995 . furthermore , it coincides with the semigroup trajectory generated by the wave - front tracking method . * definition 7.1 . * problem and is said to have a standard riemann semigroup ( srs ) if , for some small @xmath619 , there exist a continuous mapping @xmath620 and a constant @xmath621 satisfying the following properties : * ( _ semigroup property _ ) : @xmath622 ; * ( _ lipschitz continuity _ ) : @xmath623 ; * ( _ consistency with the riemann solver ) : _ given piecewise constant initial data @xmath624 , then , for all @xmath625 $ ] , the function @xmath626 coincides with the solution of and obtained by piecing together the standard riemann solutions and the lateral riemann solutions . _ _ * theorem 7.1 . * _ suppose that problem and has a standard riemann semigroup @xmath627 . consider the semigroup @xmath37 produced by the wave - front tracking method , that is , @xmath628 . assume @xmath624 . then , for all @xmath33 , @xmath629_. _ furthermore , a continuous map @xmath630 \mapsto \mathcal{d}$ ] is a viscosity solution of problem and defined in bressan-1995 if and only if @xmath631.\ ] ] in particular , a continuous map @xmath630 \mapsto \mathcal{d}$ ] is a viscosity solution if and only if @xmath632.\ ] ] _ the proof here follows a similar argument to the one presented in bressan-1995 . the only difference is that there is a strong vortex sheets / entropy waves in our problem . nonetheless , one can proceed with the proof by considering the convergence of the wave - front tracking method which is shown in section 3 . * remark 7.1 . * in the simpler cases of the isentropic or isothermal euler flow , as well as the potential flow , as far as the @xmath2stability problem is of concern , we realize the same results as for the full euler system . * acknowledgements : * the research of gui - qiang chen was supported in part by the national science foundation under grants dms-0935967 and dms-0807551 , the uk epsrc science and innovation award to the oxford centre for nonlinear pde ( ep / e035027/1 ) , the nsfc under a joint project grant 10728101 , and the royal society wolfson research merit award ( uk ) . the research of vaibhav kukreja was supported in part by the national science foundation under grants dms-0935967 and dms-0807551 , the uk epsrc science and innovation award to the oxford centre for nonlinear pde ( ep / e035027/1 ) . lewicka - trivisa-2002 m. lewicka and k. trivisa , on the @xmath1 well posedness of systems of conservation laws near solutions containing two large shocks , j. differential equations , 179 ( 2002 ) , 133177 .

the near - infrared is an important part of the spectrum in astronomy , especially in cosmology because the light from objects in the early universe is redshifted to these wavelengths . however , deep near - infrared observations are extremely difficult to make from ground - based telescopes due to the bright background from the atmosphere . nearly all of this background comes from the bright and narrow emission lines of atmospheric hydroxyl ( oh ) molecules . the atmospheric background can not be easily removed from data because the brightness fluctuates unpredictably on short timescales . the sensitivity of ground - based optical astronomy far exceeds that of near - infrared astronomy because of this long - standing problem . gnosis is a prototype astrophotonic instrument that utilizes oh suppression fibers " consisting of fiber bragg gratings and photonic lanterns to suppress the 103 brightest atmospheric emission doublets between 1.47 and 1.7 . gnosis was commissioned at the 3.9 m anglo - australian telescope with the iris2 spectrograph to demonstrate the potential of oh suppression fibers , but may be potentially used with any telescope and spectrograph combination . unlike previous atmospheric suppression techniques gnosis suppresses the lines before dispersion and in a manner that depends purely on wavelength . we present the instrument design and report the results of laboratory and on - sky tests from commissioning . while these tests demonstrated high throughput ( 60% ) and excellent suppression of the skylines by the oh suppression fibers , surprisingly gnosis produced no significant reduction in the interline background and the sensitivity of gnosis+iris2 is about the same as iris2 . it is unclear whether the lack of reduction in the interline background is due to physical sources or systematic errors as the observations are detector noise dominated . oh suppression fibers could potentially impact ground - based astronomy at the level of adaptive optics or greater . however , until a clear reduction in the interline background and the corresponding increasing in sensitivity is demonstrated optimized oh suppression fibers paired with a fiber - fed spectrograph will at least provide a real benefit at low resolving powers .

1212.1201

near - infrared ( nir ) observations ( 0.92.5@xmath0 ) are important in practically all areas of astronomy and astrophysics . for example , low - mass stars and brown dwarfs emit a substantial fraction of their light at these wavelengths and nir spectroscopy is the most efficient way to study these objects . nir spectroscopy is also one of the best ways to study the early universe because optical and ultraviolet emission lines from distant galaxies are redshifted to nir wavelengths due to hubble expansion . unfortunately , deep nir observations from the ground are extremely difficult to make due to the presence of a bright atmospheric background . the background in the range 0.91.8@xmath0 is predominantly from the de - excitation of atmospheric hydroxyl ( oh ) molecules at an altitude of @xmath1 90 km @xcite . the nir background is @xmath1 1000 times brighter than the optical background and can not be simply subtracted from astronomical observations @xcite because its brightness fluctuates on short timescales @xcite . solving the nir sky background problem is an important challenge in observational astronomy . previous attempts at a ground - based solution have not been able to suppress oh emission lines over a broad wavelength range while maintaining high throughput between the lines , which is critical for a wide range of science cases . for example , observations may be made in a very narrow wavelength range between oh emission lines using ultra - narrow band filters @xcite . however , this requires a specific unambiguously identifiable feature within the narrow wavelength range from the object . as a result , the number and nature of objects that may be observed by this technique is severely limited . more sophisticated approaches attempt to remove oh emission lines by dispersing the light with a diffraction grating at high resolution , selectively masking out the oh lines , and then recombining the light @xcite . unfortunately , the diffraction grating and the system optics ( which the light must pass through twice ) unavoidably scatters the bright oh light and the scattered portions can not be effectively removed . thus , oh emission lines are better dealt with before the light reaches any dispersing element . there have been attempts to use holographic filters for this purpose . @xcite demonstrated a device with 10 notches 10db deep and 0.1 nm wide with 85% throughput between the notches . using several of these holographic filters in series would be sufficient to suppress on the order of 100 oh doublets pre - dispersion , but the internotch throughput would be very low in this configuration . see @xcite for an in - depth comparison of these oh suppression techniques . oh suppression using aperiodic fiber bragg gratings @xcite overcomes many of the shortcomings of these previous approaches and is the best available solution to the nir sky background problem from the ground @xcite . fiber bragg gratings ( fbgs ) are common photonic devices widely used in telecommunications . basic fbgs are single - mode fibers ( smfs ) with a periodic refractive index modulation written into the fiber core by exposing it to ultraviolet light . the periodic refractive index modulation induces strong reflections at the bragg wavelength , @xmath2 , where @xmath3 is the spatial period of the refractive index modulation , and @xmath4 is the effective index of the core . reflectivities close to 100% and narrow bandwidths of 0.1 nm are possible with simple periodic fbgs @xcite . however , basic periodic fbgs have limited use for oh suppression . each periodic index modulation can be thought of as producing a single notch . a very large number of periodic index modulations would be required to suppress the dense forest of oh lines in the @xmath5 and @xmath6 bands . writing multiple periodic index modulations in succession in the same fiber results in high loss in the spectral regions between the notches because of the excessive exposure to ultraviolet light @xcite . aperiodic fbgs are designed with all notches treated simultaneously resulting in a single complex refractive index modulation that encodes @xmath150 irregularly spaced notches over a span of @xmath1100 nm with high internotch throughput . figure [ fig : afbgresponse ] shows the wavelength response of a single aperiodic fbg with 50 notches between 1.47 and 1.58@xmath0 . aperiodic fbgs are an attractive means of filtering atmospheric oh lines . however , when used in astronomy , where the wavefront exiting the telescope is distorted by atmospheric turbulence , coupling light into small core diameter smfs is challenging . as a result , large core diameter multi - mode fibers ( mmfs ) are more commonly used in astronomy . unfortunately , mmfs smear out the narrow notches of fbgs into broad , shallow notches because the bragg condition is different for each fiber mode . oh suppression requires a fiber with the light collecting ability of an mmf and the suppression characteristics of an smf fbg . the solution to this problem is a device called a photonic lantern @xcite . the device consists of a multi - mode ( mm ) port connected to an array of smfs by a taper transition . the photonic lantern converts the modes of the mm port into the supermodes of the smf array and vice versa . by splicing photonic lanterns to an array of fbgs , we have an `` oh suppression fiber '' that is easy to couple light into and exhibits the exact same transmission characteristics as an fbg in an smf . the gnosis grating unit is the first oh suppression unit to utilize these oh suppression fibers . it is independent of telescope and spectrograph but we commissioned the unit at the 3.9 m anglo - australian telescope ( aat ) at siding spring observatory with the existing iris2 infrared imaging spectrograph @xcite to demonstrate the potential of oh suppression fibers . in addition to the grating unit , gnosis consists of a fore - optics unit and an iris2 interface unit , which connects the grating unit to the telescope and spectrograph , respectively . the optical light path is shown in figure [ fig : opticalpath ] . light exiting the back of the aat is collected by a 7 element integral field unit ( ifu ) spanning 1.2@xmath7 mounted inside the fore - optics unit . the light is transported by a fiber bundle to the grating unit , where oh suppression occurs . the oh suppressed light is transported by another fiber bundle to the iris2 spectrograph for measurement . in this paper , we present the design and performance of each gnosis subsystem and as a whole . we begin by discussing the two components of our oh suppression fibers , the fbgs and the photonic lanterns in sections [ section : fbg ] and [ section : photoniclantern ] . the oh suppression fibers are housed in the gnosis grating unit , which is described in section [ section : gratingunit ] . in sections [ section : foreoptics ] and [ section : iris2interface ] we discuss the gnosis fore - optics unit and iris2 interface unit , which connects the grating unit to the telescope and iris2 , respectively . section [ section : onsky ] summarizes the on - sky performance of gnosis and section [ section : discussion ] contains a discussion on the future of oh suppression fibers . with the best - fitting butterworth profile ( red line ) . the notch is deep , narrow , and square , which is ideal for oh suppression.[fig : h1notch],scaledwidth=45.0% ] the gnosis oh suppression fibers utilize two complex , multi - notch aperiodic fbgs @xcite in series to suppress oh lines over two - thirds of the @xmath6 band ( 1.471.7@xmath0 ) . the fbgs were manufactured by redfern optical components based on our design to suppress the 103 brightest oh doublets using the line positions and strengths from @xcite . the h1 fbg has notches in the first portion of the @xmath6 band ( 1.471.58@xmath0 ) and the h2 fbg has notches in the second portion ( 1.581.7@xmath0 ) . the fbgs are printed in a custom photosensitive smf ( nufern cms8 ) . fbgs are highly sensitive to both strain and temperature variations , which induce wavelength shifts . the h1 and h2 fbgs are packaged in 304 stainless steel tubes designed to induce a strain in the fiber that varies with temperature in such a way that it cancels the intrinsic thermal wavelength shift of the fbgs , allowing the notches to remain aligned with the oh lines over the temperature range between -10 and + 25@xmath8c . , scaledwidth=45.0% ] cccccccc no . & @xmath9 & @xmath10 & @xmath11 & @xmath12 & @xmath13 & @xmath14 & @xmath15 + & ( nm ) & ( nm ) & ( db ) & ( db ) & ( nm ) & ( nm ) & + 1 & 1466.505 & 0.007 & -30.6 & 4.4 & 0.202 & 0.010 & 8 + 2 & 1469.842 & 0.007 & -32.7 & 4.7 & 0.246 & 0.011 & 8 + 3 & 1470.220 & 0.007 & -18.2 & 4.1 & 0.186 & 0.008 & 8 + 4 & 1471.329 & 0.007 & -13.6 & 3.5 & 0.161 & 0.009 & 8 + 5 & 1474.001 & 0.007 & -25.8 & 3.8 & 0.211 & 0.007 & 8 + 6 & 1475.555 & 0.005 & -23.5 & 3.2 & 0.167 & 0.012 & 8 + 7 & 1477.235 & 0.004 & -28.4 & 3.1 & 0.171 & 0.014 & 8 + 8 & 1478.371 & 0.004 & -31.6 & 3.6 & 0.190 & 0.015 & 9 + 9 & 1479.982 & 0.005 & -31.7 & 3.9 & 0.320 & 0.018 & 9 + 10 & 1480.575 & 0.005 & -28.2 & 3.0 & 0.168 & 0.014 & 7 + 11 & 1483.307 & 0.004 & -33.3 & 4.0 & 0.188 & 0.016 & 7 + 12 & 1486.438 & 0.004 & -29.7 & 2.7 & 0.143 & 0.013 & 7 + 13 & 1488.694 & 0.004 & -28.7 & 3.6 & 0.442 & 0.021 & 7 + 14 & 1490.818 & 0.005 & -21.9 & 3.6 & 0.139 & 0.004 & 6 + 15 & 1490.984 & 0.005 & -22.4 & 1.4 & 0.146 & 0.015 & 6 + 16 & 1493.187 & 0.004 & -30.0 & 2.9 & 0.163 & 0.014 & 9 + 17 & 1500.675 & 0.006 & -11.5 & 1.6 & 0.203 & 0.011 & 7 + 18 & 1502.516 & 0.005 & -14.4 & 3.0 & 0.118 & 0.006 & 5 + 19 & 1502.696 & 0.006 & -14.6 & 1.2 & 0.125 & 0.013 & 5 + 20 & 1505.281 & 0.007 & -28.4 & 2.6 & 0.171 & 0.014 & 7 + 21 & 1505.553 & 0.007 & -34.4 & 4.6 & 0.171 & 0.013 & 9 + 22 & 1506.397 & 0.007 & -22.8 & 2.6 & 0.165 & 0.013 & 7 + 23 & 1506.895 & 0.007 & -32.2 & 3.2 & 0.171 & 0.013 & 8 + 24 & 1508.222 & 0.008 & -17.7 & 2.4 & 0.163 & 0.013 & 6 + 25 & 1508.826 & 0.008 & -28.4 & 2.5 & 0.168 & 0.013 & 8 + 26 & 1510.733 & 0.008 & -9.9 & 1.5 & 0.155 & 0.013 & 5 + 27 & 1511.369 & 0.008 & -21.9 & 2.4 & 0.163 & 0.012 & 8 + 28 & 1514.545 & 0.008 & -13.7 & 1.9 & 0.160 & 0.013 & 6 + 29 & 1518.711 & 0.007 & -32.3 & 3.3 & 0.187 & 0.015 & 8 + 30 & 1524.092 & 0.007 & -36.2 & 5.6 & 0.202 & 0.016 & 9 + 31 & 1528.781 & 0.004 & -34.9 & 4.8 & 0.171 & 0.013 & 9 + 32 & 1533.241 & 0.005 & -38.1 & 7.1 & 0.247 & 0.014 & 11 + 33 & 1539.536 & 0.004 & -34.2 & 4.5 & 0.173 & 0.013 & 9 + 34 & 1543.203 & 0.005 & -38.6 & 8.8 & 0.327 & 0.038 & 8 + 35 & 1543.861 & 0.009 & -8.5 & 1.3 & 0.263 & 0.015 & 7 + 36 & 1546.214 & 0.005 & -21.8 & 2.2 & 0.177 & 0.012 & 8 + 37 & 1547.423 & 0.005 & -16.8 & 1.8 & 0.164 & 0.012 & 8 + 38 & 1550.088 & 0.004 & -30.3 & 2.8 & 0.184 & 0.014 & 8 + 39 & 1550.979 & 0.004 & -32.0 & 3.5 & 0.182 & 0.014 & 9 + 40 & 1551.788 & 0.004 & -25.9 & 2.2 & 0.169 & 0.013 & 8 + 41 & 1554.035 & 0.005 & -34.7 & 4.9 & 0.306 & 0.016 & 11 + 42 & 1554.615 & 0.004 & -33.8 & 3.9 & 0.201 & 0.014 & 9 + 43 & 1557.018 & 0.004 & -30.7 & 2.7 & 0.174 & 0.014 & 8 + 44 & 1559.764 & 0.004 & -34.9 & 4.9 & 0.185 & 0.013 & 9 + 45 & 1563.134 & 0.004 & -35.0 & 4.5 & 0.193 & 0.015 & 7 + 46 & 1565.625 & 0.008 & -32.8 & 3.8 & 0.517 & 0.016 & 12 + 47 & 1570.254 & 0.004 & -29.9 & 2.3 & 0.171 & 0.013 & 8 + 48 & 1576.030 & 0.006 & -15.6 & 1.2 & 0.220 & 0.011 & 9 + 49 & 1578.114 & 0.006 & -17.2 & 2.5 & 0.132 & 0.007 & 7 + 50 & 1578.306 & 0.008 & -18.2 & 1.5 & 0.135 & 0.013 & 6 + cccccccc & @xmath9 & @xmath10 & @xmath11 & @xmath12 & @xmath13 & @xmath14 & @xmath15 + & ( nm ) & ( nm ) & ( db ) & ( db ) & ( nm ) & ( nm ) & + 1 & 1583.028 & 0.008 & -28.8 & 1.3 & 0.171 & 0.015 & 7 + 2 & 1583.330 & 0.009 & -29.2 & 1.2 & 0.183 & 0.021 & 9 + 3 & 1584.252 & 0.007 & -25.5 & 1.9 & 0.157 & 0.014 & 8 + 4 & 1584.807 & 0.008 & -29.1 & 1.2 & 0.178 & 0.017 & 8 + 5 & 1586.253 & 0.008 & -19.8 & 2.2 & 0.142 & 0.012 & 8 + 6 & 1586.932 & 0.009 & -28.2 & 1.2 & 0.180 & 0.016 & 8 + 7 & 1589.006 & 0.008 & -12.4 & 1.7 & 0.114 & 0.012 & 8 + 8 & 1589.732 & 0.009 & -23.3 & 2.0 & 0.205 & 0.025 & 7 + 9 & 1591.590 & 0.011 & -12.2 & 1.6 & 0.473 & 0.018 & 8 + 10 & 1592.503 & 0.008 & -2.8 & 0.6 & 0.102 & 0.012 & 5 + 11 & 1593.224 & 0.007 & -15.5 & 2.1 & 0.163 & 0.012 & 8 + 12 & 1597.263 & 0.009 & -29.0 & 1.1 & 0.195 & 0.017 & 8 + 13 & 1597.415 & 0.016 & -4.3 & 1.6 & 0.164 & 0.022 & 8 + 14 & 1603.086 & 0.007 & -29.8 & 0.8 & 0.201 & 0.016 & 9 + 15 & 1607.975 & 0.007 & -28.5 & 1.3 & 0.161 & 0.015 & 8 + 16 & 1612.866 & 0.007 & -30.4 & 0.9 & 0.232 & 0.017 & 10 + 17 & 1619.460 & 0.008 & -28.8 & 1.4 & 0.166 & 0.014 & 8 + 18 & 1623.539 & 0.007 & -30.9 & 0.9 & 0.270 & 0.017 & 11 + 19 & 1627.035 & 0.007 & -11.3 & 1.7 & 0.173 & 0.010 & 8 + 20 & 1627.971 & 0.006 & -7.4 & 1.2 & 0.162 & 0.010 & 8 + 21 & 1630.229 & 0.007 & -20.9 & 2.5 & 0.196 & 0.013 & 8 + 22 & 1631.553 & 0.005 & -15.6 & 4.4 & 0.182 & 0.015 & 9 + 23 & 1631.721 & 0.013 & -29.1 & 1.8 & 0.200 & 0.027 & 9 + 24 & 1634.178 & 0.008 & -26.5 & 2.2 & 0.211 & 0.014 & 9 + 25 & 1635.136 & 0.008 & -30.0 & 1.4 & 0.315 & 0.020 & 10 + 26 & 1636.040 & 0.007 & -22.6 & 2.1 & 0.171 & 0.012 & 9 + 27 & 1638.854 & 0.007 & -29.1 & 1.6 & 0.208 & 0.013 & 9 + 28 & 1641.471 & 0.007 & -25.6 & 2.0 & 0.156 & 0.012 & 9 + 29 & 1644.219 & 0.006 & -29.3 & 1.6 & 0.192 & 0.013 & 9 + 30 & 1644.765 & 0.007 & -24.3 & 2.5 & 0.222 & 0.013 & 9 + 31 & 1647.563 & 0.008 & -25.8 & 2.2 & 0.136 & 0.012 & 7 + 32 & 1647.737 & 0.008 & -25.9 & 2.4 & 0.137 & 0.014 & 7 + 33 & 1647.907 & 0.007 & -26.6 & 1.9 & 0.144 & 0.015 & 7 + 34 & 1650.239 & 0.007 & -29.2 & 1.4 & 0.175 & 0.016 & 8 + 35 & 1655.385 & 0.007 & -26.3 & 1.9 & 0.161 & 0.011 & 8 + 36 & 1658.635 & 0.007 & -17.5 & 2.2 & 0.237 & 0.011 & 10 + 37 & 1661.000 & 0.006 & -19.5 & 2.1 & 0.127 & 0.013 & 7 + 38 & 1661.208 & 0.008 & -19.2 & 2.3 & 0.124 & 0.012 & 7 + 39 & 1668.914 & 0.007 & -24.7 & 2.3 & 0.137 & 0.014 & 8 + 40 & 1669.238 & 0.007 & -30.1 & 1.7 & 0.149 & 0.014 & 7 + 41 & 1670.267 & 0.007 & -19.5 & 2.2 & 0.132 & 0.009 & 8 + 42 & 1670.882 & 0.007 & -28.2 & 1.7 & 0.149 & 0.015 & 8 + 43 & 1672.479 & 0.006 & -15.9 & 1.9 & 0.125 & 0.009 & 7 + 44 & 1673.251 & 0.007 & -25.9 & 2.0 & 0.167 & 0.019 & 8 + 45 & 1673.391 & 0.006 & -8.3 & 1.9 & 0.192 & 0.011 & 8 + 46 & 1675.378 & 0.008 & -10.2 & 1.7 & 0.112 & 0.009 & 8 + 47 & 1675.530 & 0.005 & -9.1 & 1.7 & 0.116 & 0.009 & 8 + 48 & 1675.623 & 0.007 & -11.7 & 2.0 & 0.119 & 0.002 & 6 + 49 & 1676.356 & 0.007 & -20.4 & 2.3 & 0.165 & 0.012 & 7 + 50 & 1679.400 & 0.007 & -2.5 & 0.6 & 0.090 & 0.008 & 5 + 51 & 1680.237 & 0.007 & -13.7 & 2.1 & 0.161 & 0.011 & 8 + 52 & 1684.044 & 0.007 & -27.4 & 1.9 & 0.177 & 0.014 & 8 + 53 & 1684.901 & 0.008 & -14.4 & 1.9 & 0.184 & 0.012 & 8 + 54 & 1690.362 & 0.008 & -29.6 & 1.7 & 0.191 & 0.017 & 8 + 55 & 1695.502 & 0.009 & -27.6 & 2.2 & 0.155 & 0.012 & 8 + we measured the position , depth , and width of each notch from the wavelength response of the h1 and h2 fbgs ( @xmath1610,000 ) . the fbgs were not designed with any particular functional form for the notch profile in mind , but we found that a butterworth profile provided a good fit to the empirical data . thus , each notch in the wavelength response was fitted with a butterworth profile , @xmath17^{2n}},\ ] ] where @xmath9 is the profile center , @xmath18 is the profile value at the center , @xmath13 is the profile width , @xmath19 is the profile value far away from the profile center , and @xmath15 is the profile index . the value of @xmath15 affects the steepness of the profile . from the butterworth parameters for each notch , we take @xmath9 to be the notch position , @xmath11 to be the notch depth , and @xmath13 to be the notch width . figure [ fig : h1notch ] shows one of the notches in the h1 fbg and the best - fitting butterworth profile ( red line ) . the h1 fbg notch parameters are listed in table [ table : h1notchtable ] . the h1 fbg notch parameters are averages measured from the wavelength response of eight separate devices obtained by scanning a narrow tunable laser source across the device bandwidth and comparing with a reference fiber . a total of 50 notches were measured in the range 14651580 nm in each of the eight h1 devices . the h2 fbg notch parameters are listed in table [ table : h2notchtable ] . the h2 fbg notch parameters are averages measured from the wavelength response of 106 separate devices obtained using a broadband source and an optical spectrum analyzer . a total of 55 notches were measured in the range 15801700 nm in each of the 106 h2 devices . tables [ table : h1notchtable ] and [ table : h2notchtable ] contain more notches than the original design of 103 . the additional notches may be printing errors from the manufacturing process . for example , notch 9 in table [ table : h2notchtable ] is not associated with any oh line from @xcite , but we include it here because it is a significant spectral feature of the fbgs . also , some notches are deliberately designed to be wider in order to suppress closely spaced oh lines with a single notch . , scaledwidth=45.0% ] the throughput of the fbgs between notches is of interest because this is the region where scientific observations are made . the throughput at several internotch wavelengths was measured using a cutback technique with a photonetics tunics tunable external cavity laser ( 15201640 nm ) and a thorlabs pm100d power meter with an s122c sensor head . the laser source was calibrated to @xmath200.2 nm before the measurements were carried out . the internotch measurement wavelengths were chosen to avoid any notches . based on the values listed in tables [ table : h1notchtable ] and [ table : h2notchtable ] the closest any internotch measurement value comes to the center of a notch is @xmath21 nm . given that the average notch width is @xmath22 nm , the internotch measurements should have safely avoided all fbg notches . the internotch throughput of the h1 and h2 fbgs in the range 15201640 nm are shown in figure [ fig : internotch_throughput ] . the average throughput is 0.954 and 0.896 for the h1 and h2 fbgs , respectively . a total of nine h1 fbgs were randomly selected and measured . the points with error bars show the average internotch throughput of all nine h1 fbgs and the 1@xmath23 variation . the points without error bars are the measured values from one of the nine h1 fbgs . a total of five h2 fbgs were randomly selected from an initial delivery of 19 devices and measured . again , the points with error bars show the average and 1@xmath23 variation of all five h2 fbgs and points without error bars are measurements from one of the 5 devices . the throughput in both devices decreases at longer wavelengths , but this is due mainly to the decrease in the transmission of the cms8 fiber at longer wavelengths ( see figure [ fig : cms8_loss ] ) . the performance of the h1 and h2 fbgs are highly satisfactory for oh suppression . on average the notches are deep ( 24db ) , narrow ( 0.19 nm ) , and square and the average throughput between the notches is high ( 0.92 ) . the gnosis oh suppression fibers use two 1@xmath2419 photonic lanterns ( one for input and one for output ) to provide the fibers with the light collecting ability of an mmf . the photonic lanterns were manufactured by nkt photonics @xcite and consist of a 50@xmath0 core diameter mm port connected to an array of 19 smfs by a taper transition with a taper ratio of 11 enclosed in a protective metal case . a 5 m long fc / pc - connectorized delivery fiber is fusion spliced to the mm port to facilitate connecting the photonic lanterns with other components . the smfs are approximately 3 m in length and not connectorized . the photonic lanterns are designed to efficiently convert the modes of the mm port into the supermodes of the smf array and vice versa . the smf array will only support @xmath25 supermodes independent of wavelength , where @xmath25 is equal to the number of smfs . thus , for maximum efficiency the mm port should be designed to support @xmath26 modes . however , the number of modes supported by the mm port is wavelength dependent ( @xmath27 ) . for gnosis the mm port is designed to support @xmath28 modes at @xmath29 with a @xmath30 core diameter , but there will be more ( fewer ) modes at shorter ( longer ) wavelengths . the mm port parameters determine the numerical aperture ( na ) by @xcite @xmath31 equation ( [ equation : lanternna ] ) gives an na of 0.086 ( @xmath32/5.8 ) for the gnosis photonic lanterns . we can not measure this value directly because of the delivery fiber fusion spliced to the mm port . instead , we measured the focal ratio degradation ( frd ) in the delivery fiber and computed the mm port na from the measured output na of the delivery fiber . we fusion spliced an amplified spontaneous emission ( ase ) source centered at 1.53@xmath0 to a randomly chosen smf . the na of the light exiting the delivery fiber was taken to be the value that enclosed 95% of the total energy in the far - field image . the measurement was repeated 10 times for each photonic lantern . the na of the light exiting the delivery fiber is [email protected] on average for the 14 photonic lanterns . the laboratory measurement of the delivery fiber frd ratio gave @xmath33/@xmath34 . thus , the average value of the mm port na is [email protected] , which is in agreement with the estimate from equation [ equation : lanternna ] . the measured mm port na values for each photonic lantern is listed in table [ table : plmeasurements ] . for minimum loss in the mm to sm conversion , the mm port must be fed at @xmath35 @xmath32/5.8 in order to avoid overfilling . to account for frd in the delivery fiber the oh suppression fibers should be fed at @xmath35 @xmath32/7 to ensure the light arrives at the mm port at @xmath35 @xmath32/5.8 . at the output end of the oh suppression fibers , it is important to know the f - ratio of the beam exiting our fibers to properly interface them with the subsequent components . based on our measurements of the output photonic lanterns the beam exiting our oh suppression fibers is @xmath1@xmath32/4.9 . at the design wavelength , the number of modes supported by the mm port exactly matches the number of supermodes of the smf array resulting in minimum loss in the taper transition . we determined the sm to mm throughput by splicing the 1.53@xmath0 ase source to a randomly chosen smf and measured the output power of the delivery fiber . the measurement was repeated 10 times for each photonic lantern . the average sm to mm throughput of the seven output photonic lanterns is [email protected] with the best throughput at 0.989 . we measured the mm to sm throughput by injecting a beam with an na of 0.086 ( @xmath32/5.8 ) from an ase source centered at 1.53@xmath0 into the delivery fiber and collecting the power from all 19 smfs . the average mm to sm throughput of the five measured input photonic lanterns is [email protected] . recall previously that we found that the delivery fiber must be fed at @xmath35 @xmath32/7 to compensate for frd . hence , in our measurement scheme we were overfilling the mm port , causing the throughput to be significantly lower than the sm to mm throughput . however , the gnosis fore - optics unit feeds the oh suppression fibers at @xmath1@xmath32/5 and the measured mm to sm throughput gives a sense of the performance under operating conditions . for reference , the throughput of the photonic lanterns near the design wavelength is listed in table [ table : plmeasurements ] . cccc input & input & output & output + mm to sm & na & sm to mm & na + throughput & & throughput & + 0.857 & 0.081 & 0.996 & 0.084 + 0.849 & 0.082 & 0.975 & 0.089 + 0.838 & 0.078 & 0.962 & 0.088 + not measured & 0.089 & 0.973 & 0.088 + 0.863 & 0.088 & 0.984 & 0.084 + 0.849 & 0.088 & 0.973 & 0.088 + not measured & 0.091 & 0.989 & 0.079 + we have only measured the throughput of the photonic lanterns near the design wavelength . in theory , the throughput of the photonic lantern depends on wavelength because the number of modes supported by the mm port depends on wavelength . in the suppression range of gnosis ( 1.471.7@xmath0 ) , the number of modes varies from 21 to 16 , ignoring dispersion . in the mm to sm conversion , the throughput is lower at the blue end because there are more modes in the mm port than smfs . in the sm to mm conversion , the throughput is lower at the red end because there are more smfs or supermodes than can be supported by the mm port . thus , when two identical photonic lanterns are used back - to - back , the loss at the blue end during the mm to sm conversion and the loss at the red end in the sm to mm conversion should produce a wavelength response that is peaked at the design wavelength . although this behavior has not been demonstrated for a symmetric photonic lantern system alone , the effect is present in an integrated oh suppression fiber . the wavelength response of an oh suppression fiber is discussed in section 4.3 and the throughput appears to be peaked at @xmath11.53@xmath0 . the gnosis grating unit contains seven independent oh suppression fibers . each oh suppression fiber consists of a 1@xmath2419 input photonic lantern , 19 pairs of h1+h2 fbgs , and an output 19@xmath241 photonic lantern all fusion spliced together and arranged on a plastic tray as shown in figure [ fig : gratingunit01 ] . the seven trays are mounted within an aluminum enclosure and padded with foam sheets . the grating unit enclosure is mounted in a standard equipment rack within the cassegrain cage during use . in assembling the oh suppression fibers , all splices were carried out using a fitel s175 v.2000 fusion splicer . cms8 to cms8 fusion splices were carried out using the standard identical smf prescription , which results in an average splice loss of @xmath10.05db . cms8 to smf-28 fusion splices were carried out using a custom prescription , which results in an average splice loss of @xmath10.15db . all splices are protected by splice sleeves and excess fiber is spooled into circular drums in each tray . cms8 fiber is much lossier than smf-28 so we minimized the length of cms8 fiber as much as possible during the assembly of the oh suppression fibers . we measured the throughputs of all seven oh suppression fibers before installation during the first commissioning run by simulating illumination by the telescope with a thorlabs s5fc1550s(p)-a2 sld source centered at 1.55@xmath0 . the fibers were fed with a @xmath1@xmath32/5 beam by butt - coupling a mmf to the fiber using a fc - to - fc connector and the output power was measured using a thorlabs pmd100d power meter with an s122c sensor head . the measured values are listed in table [ table : labthroughputmay2011 ] . the average throughput of the seven oh suppression fibers is @xmath36 with a maximum and minimum of 0.61 and 0.55 , respectively . we may compare this measurement to an estimate based on the measurements of individual components . from the cms8 to smf-28 splice loss ( 0.15db ) , cms8 to cms8 splice loss ( 0.05db ) , the average mm to sm throughput ( 0.851 ) of the input photonic lanterns at f/5.8 , the average sm to mm throughput ( 0.974 ) of the output photonic lanterns , and the average internotch throughput of the h1 ( 0.954 ) and h2 ( 0.896 ) fbgs in the range 15201640 nm , the estimated throughput of an oh suppression fiber is @xmath10.61 including fresnel reflections from the fc - to - fc connector . additionally , we must adjust the mm to sm throughput to an @xmath32/5 input beam . based on laboratory measurements of a single photonic lantern , the mm to sm throughput for an @xmath32/5 beam is @xmath10.25db lower than an @xmath32/5.8 beam , i.e. , 0.803 . with this adjustment , the estimated throughput is 0.575 , which is in agreement with the measured value for the integrated oh suppression fiber . figure [ fig : gratingthroughputvswavelength ] shows the wavelength response of oh suppression fiber 5 . the wavelength responses of the oh suppression fibers were measured after installation and alignment of the fore - optics unit ( see section [ section : foreoptics ] ) and iris2 interface unit ( see section [ section : iris2interface ] ) . iris2 was used to obtain a spectrum of the dome flat lamp through the entire system with and without an oh suppression fiber . the ratio of the two spectra gives the wavelength response of the oh suppression fiber . the notches of the fbgs appear as significant dips in the throughput , but they are not as deep or narrow as indicated in tables [ table : h1notchtable ] and [ table : h2notchtable ] because the resolution of iris2 is not high enough to resolve these notches . the internotch throughput is relatively high near the design wavelength of the photonic lanterns and there is some dependence on wavelength due to the linear loss of the cms8 and the wavelength - dependent loss from a symmetric photonic lantern system . the grating unit is interfaced with the aat using a fore - optics unit that is mounted at the aat cassegrain focus . the main function of the fore - optics unit is to re - image the central region of the aat focal plane onto an ifu and feed this light to the grating unit . the first optical element encountered by the @xmath32/8 beam from the aat is an acquisition mirror . the acquisition mirror is just a circular mirror with a central aperture . the central portion of the beam passes through the aperture to the ifu while the rest of the beam is diverted to an acquisition camera . the portion of the beam that passes through the acquisition mirror is then magnified to @xmath32/265 by an optical relay consisting of a magnifying achromatic doublet lens and a doublet field - flattening lens with a pupil stop positioned in the telescope pupil plane between the two lenses . each lens has an ar coating with a reflectance @xmath37% over the waveband from 1.0 to 1.7@xmath0 . the beam is then captured by an ifu consisting of an array of seven hexagonal lenslets made of fused silica arranged as shown in figure [ fig : ifuslit ] . the front face of each lenslet is polished and ar - coated . the lenslets are glued onto a fused silica substrate , which is fixed in a mount attached to the optical relay assembly . at the ifu surface the plate scale is 0.2@xmath7mm@xmath38 due to the magnification of the optical relay . thus , each 2 mm flat - to - flat lenslet spans 0.4@xmath7 on the sky and has a field of view ( fov ) of 0.14arcsec@xmath39 . the total ifu spans 1.2@xmath7 on - sky , which is about the median seeing in the @xmath6 band at siding spring observatory , and the total ifu fov is 0.97 arcsec@xmath39 . mm@xmath38 at the ifu surface . each 2 mm flat - to - flat hexagonal lenslet spans 0.4@xmath7 and the total ifu spans 1.2@xmath7 . the numbering indicates the iris2 interface unit slit element connected to each ifu element.[fig : ifuslit],scaledwidth=40.0% ] each lenslet feeds an @xmath32/6.5 beam into a 50@xmath0 core diameter mmf positioned at the lenslet focus at the back of the ifu substrate . the mmfs are approximately 5 m in length and feed a @xmath1@xmath32/5 beam to the oh suppression fibers in the grating unit , based on an frd estimate from @xcite and laboratory measurements . the fibers are jacketed , reinforced , and bundled together in a limited - bend armoured conduit . at the ifu end , the fibers are terminated with a glass ferrule and were attached to the substrate using a uv curing adhesive ( norland optical adhesive 61 ) prior to potting with a polyeurethane encapsulant ( opti - tec 4200 ) . during the attachment process each fiber was individually aligned with the corresponding lenslet by maximizing the power coupled into the fiber from a test source simulating the illumination of the fore - optics by the telescope . at the grating unit end , the fibers are terminated in fc / pc connectors the throughput of the fore - optics unit was measured during assembly and installation by illuminating the entrance aperture with an @xmath32/8 beam from a thorlabs s5fc1550s(p)-a2 sld source to simulate illumination by the aat . before the ifu was installed , the power at the ifu position was measured using a thorlabs pm100d power meter with an s122c sensor head and compared to the power at the entrance aperture yielding a throughput of @xmath40 for the fore - optics relay . after the ifu was installed , a xenics xeva-1.7 - 640 camera was used to image the front and back surfaces of the ifu . the fraction of incident power captured by each ifu element was estimated from these images , correcting for aperture losses at the array and non - uniformity in the illumination from the sld source . comparing the input power at each ifu element , which is the fraction of incident power captured by each ifu element times the output power of the relay , to the power at the end of each fiber gives the ifu ( plus fiber bundle ) throughput of each element . the average ifu throughput is @xmath41 . thus , the average total fore - optics unit throughput ( relay+ifu+fiber bundle ) is @xmath10.72 . the individual throughput of each fore - optics unit element is listed in table [ table : labthroughputmay2011 ] . the @xmath1@xmath32/4.9 beams from the gnosis oh suppression fibers are fed to the iris2 infrared imaging spectrograph by 50@xmath0 core diameter mmfs 12 m in length . for these observations , iris2 was positioned below the horseshoe of the telescope mount on the dome floor . the fibers are jacketed and enclosed together in a steel coil limited bend armored conduit . at the grating unit end , the fibers are fc / pc - connectorized and are butt - coupled to the oh suppression fibers using fc - to - fc connectors . at the iris2 end , the fibers are terminated in a linear slit block consisting of a 7-element fused silica v - groove array from ocean optics . the grooves are designed to hold the fibers , which have a 125@xmath0 diameter cladding , with a center - to - center spacing of 250@xmath0 . the fibers are glued to the grooves and they are encased in epoxy with a protective casing at the back end for strain relief . the slit block is mounted to the iris2 interface unit assembly , which sits over the iris2 dewar window . the interface unit assembly rests on the structure usually used to attach iris2 to the telescope via kinematic mounts and includes various adjustment mechanisms to align the slit block with the slit masks in the iris2 slit wheel . the interface unit includes an optical relay consisting of two achromatic doublet lenses with a magnification of 3 that images the fibers onto the iris2 slit plane , which is inside the instrument dewar . estimating the frd in the fiber bundle using @xcite , we expect an @xmath1@xmath32/4 beam from the fiber in the slit block which is magnified to @xmath1@xmath32/12 and the fibers are 150@xmath0 in diameter at the slit plane . the magnification yields a resolving power of @xmath422350 for iris2 and satisfies the maximum acceptance cone of iris2 ( @xmath32/8 ) . no modifications were made to iris2 except the addition of custom cold stops and slit masks . two cold stops ( @xmath32/10 and @xmath32/12 ) are available for use with gnosis . three slit masks were made by laser drilling with a slit of seven linear 250/200/180@xmath0 diameter holes spaced 750@xmath0 apart in the same position in the spectral axis as for the standard @xmath6-offset slit wheel . the throughput of the iris2 interface unit was measured before installation by simulating illumination by the aat . the entire gnosis system was connected and the entrance aperture of the fore - optics unit was illuminated with an @xmath32/8 beam from a thorlabs s5fc1550s(p)-a2 sld source . we measured the output power of each oh suppression fiber in the grating unit and the power in the slit image when the iris2 interface unit is connected to the grating unit . the average throughput of the iris2 interface unit is [email protected] and the measured value for each element is listed in table [ table : labthroughputmay2011 ] . there are additional losses from the alignment of the slit block with the slit mask . even with perfect alignment , low - level aberrations will result in a loss of a few percent depending on the size of the slit mask holes and the focal ratio out of the fibers in the slit block . first , focus was adjusted with the dome flat lamp until the central fiber spot image had an fwhm of @xmath43pixels . next , tip - tilt was adjusted followed by an alignment of the central fiber to within 0.1pixels of the central slit mask hole . then , in - plane rotation was adjusted until the position of the outer fiber images are within 0.1pixels of each other . the central fiber was again aligned to the central slit mask hole . lastly , small translational adjustments were then made to maximize the throughput . the throughput numbers vary from alignment to alignment , but it was typically above 90% and around 97% for the central fiber with the 250@xmath0 slit . there was a noticeable drop by a few percent in the throughput away from the central fiber , indicating that the magnification of the relay is slightly too large . this was confirmed by comparing the difference between the spot positions and the slit mask hole positions for all seven fibers . in calculations we assume an average throughput of 0.95 for slit alignment . ccccc sky pos & fore - optics & grating unit & iris2 interface & overall + n & 0.790 & 0.612 & 0.91 & 0.38 + wnw & 0.727 & 0.579 & 0.93 & 0.35 + wsw & 0.849 & 0.563 & 0.93 & 0.37 + c & 0.752 & 0.596 & 0.93 & 0.42 + ene & 0.752 & 0.583 & 0.93 & 0.43 + ese & 0.548 & 0.551 & 0.92 & 0.42 + s & 0.605 & 0.551 & 0.93 & 0.31 + average & 0.718 & 0.576 & 0.93 & 0.38 + we summarize the on - sky performance of gnosis below . an in - depth analysis of the on - sky performance of gnosis may be found in @xcite . gnosis was commissioned at the aat over five separate observing runs spread over the months of march , may , july , september , and november of 2011 . initially , the fore - optics unit was connected to the grating unit by butt - coupling with fc - to - fc connectors . from the may commissioning run onward , the fore - optics unit was fusion spliced directly to the oh suppression fibers within the grating unit . the system was configured to maximize the throughput of central sky fiber . with the remaining fibers we maximized the average throughput of the system by connecting the highest throughput fibers with each other . however , this results in greater fiber - to - fiber throughput variation . in most observations one fiber was configured to bypass the grating unit to serve as a control fiber without oh suppression . the data reduction procedure is similar to the procedure described in @xcite . spectroscopic observations were made with the @xmath6 broadband filter , @xmath32/12 coldstop , and the slit mask with 180 or 250@xmath0 diameter holes in multiple - read mode ( mrm ) . in mrm , the detector read noise is minimized ( @xmath18@xmath44 ) because the 1024@xmath241024 rockwell hawaii-1 detector is non - destructively read out during the exposure and the final image is a linear least - squares fit through all the reads . the detector dark current is @xmath10.0015@xmath44 s@xmath38 . each image was corrected for detector non - linearities and the spectrum in each fiber was extracted using `` gaussian summation extraction by least squares '' from @xcite . each spectrum was corrected for fiber - to - fiber throughput variations measured from observations of the dome flat lamp . the dome flat lamp spectrum in each fiber was extracted and integrated . the values were normalized to the mean value of all seven fibers to give the fiber - to - fiber throughput variation . the measured fiber - to - fiber variation is consistent with the laboratory values listed in table [ table : labthroughputmay2011 ] . the wavelength calibration for each fiber was determined from a xenon arc lamp observation . the arc lamp spectrum in each fiber was extracted and the pixel position of each xenon line was fit by a cubic polynomial . the spectra were also corrected for inter - quadrant cross - talk @xcite because the spectra span two quadrants of the iris2 detector . in each spectrum , the counts at wavelengths below the @xmath6 filter cutoff ( 1.5@xmath0 ) come from the detector . after correcting for detector dark current , the median count rate at @xmath45 was subtracted from the spectrum . the final correction we applied to all observations was an instrument response correction , which corrects for the variation in throughput with wavelength . the instrument response was measured from a0v standard star observations . the spectrum in each fiber was extracted and reduced as discussed above . then the instrument response for each fiber was taken to be the sky - subtracted spectrum divided by a model spectrum of vega @xcite normalized to the median value in the range 1.51.69@xmath0 . the average throughput of gnosis ( fore - optics + grating unit + iris2 relay ) was measured to be @xmath10.38 in a laboratory setting at 1.55@xmath0 ( see table [ table : labthroughputmay2011 ] ) . on - sky , we must include the additional losses from the telescope , the slit block alignment , the iris2 spectrograph , and aperture losses for a point source . the throughput of the aat is @xmath10.88 including reflections from the two mirrors and extra loss for the accumulation of dirt on the mirrors . the throughput of iris2 is @xmath10.12 @xcite . for the median seeing of 1.2@xmath7 , the aperture loss is approximately 0.30.5 for a point - source offset between 00.6@xmath7 from the center of the hexagonal array @xcite . thus , the end - to - end throughput of our system was expected to be @xmath10.04 for a diffuse source and @xmath10.018 for a point source with typical aperture losses . we measured the end - to - end throughput of the system from observations of a0v stars taken during the september commissioning run , which had the best observing conditions . the spectra were combined by summing each spectral pixel , neglecting the difference in the wavelength calibration of each fiber ( less than 1.5pixels over most of the detector ) . the combined spectrum of the a0v star was divided by a model vega spectrum scaled to the appropriate brightness and assuming a value for the aperture losses . the @xmath46-band seeing was measured to be @xmath11.5@xmath7 . although we are only able to estimate the aperture losses for this seeing , the a0v star observations were consistent with a throughput of @xmath10.020.04 in agreement with the expected value from laboratory measurements . , scaledwidth=45.0% ] the oh suppression performance of gnosis was measured from observations of blank sky . a total of 45 blank sky observations were taken during the september run at various locations on the sky . the exposure time of these observations was either 15 or 30 minutes . the instrument thermal background is significant in gnosis observations and a separate thermal background subtraction must be applied to blank sky observations . cold frames , observations where the fore - optics unit was removed and pointed at a container of liquid nitrogen , were obtained with the same exposure time as the blank sky observations . figure [ fig : cold ] shows the smoothed cold frame spectrum from the central gnosis fiber . alternatively , we may fit a thermal blackbody spectrum to the continuum points in between the oh lines @xmath47 and subtract the best - fitting model from our spectrum . figure [ fig : skyspec ] shows the cold - subtracted blank sky spectrum from six suppressed fibers and the control fiber from 8.75hr of observations from 2011 september 1 to 4 . these 21 blank sky observations were made after moonset and are all @xmath35 60@xmath8 from the moon . the spectra were flux - calibrated assuming an efficiency of 3.3% . the strong suppression of the oh lines in the range 1.51.7@xmath0 is evident . the suppression factor was measured for approximately half the lines and 78% meet or exceed the target specifications . the other lines that do not meet the target specifications mostly correspond to fbg notches that are too narrow to suppress the entire doublet . the fbgs are designed based on the doublet separations found in the oh line model of @xcite . the notches that do not meet the target specification have a much wider doublet separation according to @xcite . the doublet mean is the same , but the fbg notch is not wide enough to suppress the doublet according to @xcite . an example is shown in figure [ fig : abrams ] for the 31 q1(4.5 ) transition , which shows the @xcite values in red and the @xcite values in blue . , scaledwidth=45.0% ] nevertheless , this is a clear demonstration that fbgs can cleanly remove oh lines while maintaining relatively high throughput between the lines . the integrated background in the range 1.51.7@xmath0 was reduced by a factor of @xmath18 , but there was no significant reduction in the interline background as predicted by @xcite . this is unexpected given that @xcite recently found that scattered oh dominates the interline background in their @xmath6-band spectrum . the blank sky observations made with gnosis thus far are detector noise - dominated in the interline regions . the faintness of the interline background and low system throughput of our current configuration requires 30 minute exposures for an interline background of @xmath110adu per pixel , which corresponds to @xmath145@xmath44 per pixel . of this , approximately 27@xmath44 per pixel are from the detector dark current and the other 18@xmath44 per pixel are from the sky . thus , the detector s 8@xmath44 per pixel read noise gives @xmath117% uncertainty in the interline background . thus , it is possible that our oh suppression fibers are reducing the interline background , but we were unable to observe the reduction of the low interline signal among the detector noise . alternatively , there may be no reduction because of physical sources , but it is impossible to tell without observations that are not detector noise - dominated . the sensitivity of gnosis was measured from an observation of a low surface brightness galaxy hizoa j0836 - 43 ( @xmath6-band surface brightness of 17.3magarcsec@xmath39 ) . the galaxy is larger than the gnosis fov and the unknown aperture losses were not an issue . the sky - subtracted spectrum was divided by the square root of the non - subtracted spectrum to obtain the signal to noise per pixel . the median signal to noise per pixel was @xmath110 for a 30 minute exposure . the same analysis on the spectrum through the control fiber yields the same result , surprisingly indicating that there was no improvement in signal to noise per pixel when using gnosis oh suppression . this is because oh suppression mainly improves the signal to noise near the oh lines . in between the lines ( the region that dominates the sensitivity calculation ) the reduction in the background was much smaller than expected in this paper , we have given a thorough description of each subsystem of gnosis , the first instrument to use fiber bragg grating oh suppression fibers . these fibers were designed to demonstrate a reduction of the interline background increasing sensitivity by a factor of 40 or more when suppressing the brightest oh doublets in the @xmath6-band . the line spread function of iris2 shows that the spectrograph s diffraction grating scatters light many pixels from the line center and simulations show that the cumulative contribution from multiple lines leads to a continuum of scattered light that dominates the interline background @xcite . if the emission lines are suppressed before the light has the opportunity to be scattered by the spectrograph , the continuum of scattered light and therefore the interline background will be reduced . with the oh lines corresponding to @xmath48 transitions from @xcite shown by the dashed blue lines . these @xmath48 lines were not suppressed by design based on their relative strengths in @xcite but they match up well with the residual features in the spectrum.[fig : abramsr],scaledwidth=45.0% ] to demonstrate this effect , we carried out observations of blank sky with gnosis at the 3.9 m aat with the iris2 spectrograph . although the oh suppression fibers successfully suppressed most of the brightest oh doublets in the range 1.51.7@xmath0 at the target level or greater and reduced the integrated background by a factor of @xmath110 , the spectra showed no significant reduction in the interline background and thus the same median signal to noise per pixel when compared to a control fiber without oh suppression . a reduction in the interline background and the associated increase in sensitivity would led to oh suppression fibers having an impact on astronomy at the level of adaptive optics or greater . therefore , it is critically important to identify why no reduction was observed in these experiments and determine if improvements can be made to realize the full benefit of oh suppression . only observations that are detector noise - dominated have been made with oh suppression fibers . if no interline background reduction is seen in oh - suppressed observations that are not detector noise - dominated it would indicate either oh suppression fibers do not suppress scattered light or the atmosphere is more complex than modeled by @xcite . specifically , the oh spectrum may be different than the model of @xcite used to design the fbgs , which would result in poor suppression . we have presented evidence of this above and poorly suppressed lines such as these would scatter more light into the interline regions . alternatively , there may be unaccounted for atmospheric continuum sources from other molecular species that creates a continuum floor . it may also be possible that the unsuppressed oh lines are not as weak as indicated by @xcite and they scatter more light into the interline regions . as evidence , we found that the @xmath48 transitions found in @xcite match up well with residual features in our oh - suppressed spectrum of the sky as seen in figure [ fig : abramsr ] . these doublets according to @xcite are stronger than indicated by @xcite and would contribute more scattered light to the interline region . the evidence that inaccuracies in the fbg design because of inaccuracies in the oh line model of @xcite , which may result in more scattered light in the interline region than expected , is certainly suggestive . however , this is based on the assumption that oh suppression fibers suppress scattered light , which has yet to be demonstrated . thus , it is critically important that the next step be a clear demonstration of scattered light suppression with observations that are not detector noise - dominated . such observations require that the interline signal is greater and/or the detector noise is lower . in general , this may be accomplished by increasing the brightness of the source , increasing the throughput of the system and/or utilizing a low - noise detector . the system throughput may be increased mostly significantly by optimizing the oh suppression fibers and/or the spectrograph . presently , without first demonstrating that oh suppression fibers suppress scattered light , fabricating optimized fibers would be difficult to justify . thus , @xcite suggested that pairing gnosis with a fiber - fed spectrograph with a high efficiency volume phase holographic ( vph ) grating is the best way to go about improving system throughput . if a high - performance detector , a 1.7@xmath0 cutoff hawaii-2rg for example , were included the spectrograph would simultaneously increase throughput and lower detector noise , which would be ideal for oh - suppressed observations of blank sky . the australian astronomical observatory has plans to build such a spectrograph called praxis for further testing of oh suppression fibers on - sky by early 2014 @xcite . however , for the purpose of demonstrating scattered light suppression , observations of blank sky are not ideal because the sky spectrum may be considerably more complex than the spectrum the fbgs are designed to suppress . thus , the ideal source would be an oh line source in the laboratory , such as that created by @xcite . this would remove any complications from additional atmospheric sources , but the weaker suppressed oh lines are still present . alternatively , scattered light suppression may also be demonstrated with a single emission line from a bright arc lamp ( xenon has at least one line at 1.6733@xmath0 that coincides with one of the fbg notches ) . these observations only require gnosis and the iris2 spectrograph without any telescope , which is advantageous because they do not require building a new spectrograph . regardless of whether or not oh suppression fibers can be used to reduce the interline background and increase sensitivity , they provide real benefits for observations at low resolving powers ( 500@xmath493000 ) , which was shown by @xcite using observations of [ ] emission lines in seyfert galaxies and ch@xmath50 absorption in brown dwarfs . previously , observations at these resolving powers have been too low to resolve out the oh lines , but that is no longer necessary with oh suppression fibers . the performance of the oh suppression fibers is very good , but there is room for improvement especially the throughput . the biggest single loss in the oh suppression fibers comes from mm to sm conversion of the input photonic lantern . the beam feeding the delivery fiber of the input photonic lanterns is @xmath1@xmath32/5 and we argued in section [ section : photoniclantern ] that the oh suppression fibers must be fed at @xmath35 @xmath32/7 to avoid overfilling the mm port ( @xmath32/5.8 ) of the 1@xmath2419 photonic lanterns . thus , the mm to sm conversion throughput will increase if we slow the beam feeding the oh suppression fiber from @xmath32/5 to @xmath32/7 . for a back - to - back system with 1@xmath2419 photonic lanterns , the total throughput increases by @xmath10.5db ( 11% ) when slowing the input beam from @xmath32/5 to @xmath32/7 @xcite . however , doing so will decrease the fiber s fov . by conservation of tendue , the product of the fiber fov on - sky and the telescope diameter is proportional to the product of the fiber input na and the fiber core diameter . therefore , it is not possible to underfill the photonic lantern mm port by reducing the core diameter or input na without sacrificing the fiber s fov . as an alternative to slowing the input beam from @xmath32/5 to @xmath32/7 to avoid overfilling the mm port , we may increase the mm port na to match the input @xmath32/5 beam . if we keep the core diameter fixed , this corresponds to increasing @xmath25 , the number of smfs in the photonic lantern . based on equation ( [ equation : lanternna ] ) an mm port na of 0.12 ( the @xmath32/5 beam arrives at the mm port at @xmath32/4.2 due to frd ) with @xmath30 corresponds to @xmath51 . thus , a back - to - back system with 1@xmath2437 photonic lanterns fed at f/5 would be equivalent to a system with 1@xmath2419 photonic lanterns fed at @xmath32/7 and should yield a similar increase in throughput over the current configuration . the wavelength - dependent loss of a symmetric photonic lantern system may be addressed by designing the mm port of the input and output photonic lanterns to have different nas . if @xmath25 is the number of smfs then the mm port of the input photonic lantern should be designed to support @xmath52 modes at the blue end of the suppression range . as @xmath53 there will be @xmath54 modes at the red end of the suppression range and no penalty will be incurred in the mm to sm conversion at any wavelength . the mm port of the output photonic lantern should be designed to support @xmath55 modes at the red end of the suppression range , which would correspond to @xmath56 modes at the blue end of the suppression range . thus , in the sm to mm conversion , none of the modes will be lost at any wavelength . photonic lanterns with large @xmath25 are very cumbersome to handle and the gnosis grating unit is rather bulky and heavy ( 100 kg ) . a significant reduction in size and weight may be possible with oh suppression fibers consisting of fbgs printed in multi - core fibers ( mcfs ) with each end tapered down into an mmf . @xcite demonstrated such a device with an mcf containing @xmath1120 cores within a 230@xmath0 cladding and a single notch . mcfs with 37 cores can be manufactured and it should be a simple matter to taper down each end of the mcf into a 50@xmath0 mmf following a similar process for the manufacturing of the photonic lanterns . in addition to the reduction in size and weight no splices would be required boosting the throughput by @xmath10.35db ( 8% ) . assuming complex refractive index modulations can be imprinted into each mcf core with the same level of performance as the current gnosis fbgs and incorporating all the suggestions above would result in a throughput of at least @xmath10.73 . these next - generation oh suppression fibers are currently under development @xcite . thus far we have discussed improvements to the oh suppression fibers themselves but the other subsystems also require optimization . retrofitting gnosis to the existing iris2 spectrograph was acceptable for an initial demonstration of oh suppression fibers , but a fiber - fed spectrograph like praxis would be ideal for future science observations . in the current configuration , the instrument thermal background is high ( see figure [ fig : cold ] ) and emanates almost entirely from the slit block of the iris2 interface unit reducing the sensitivity of observations . a spectrograph with a vacuum feed - through would make the iris2 interface unit unnecessary and significantly reduce the thermal background . also , it would reduce the number of optical surfaces and there would be no slit block alignment errors , which would increase the system throughput by @xmath115% in addition to the increase in throughput due to the high efficiency vph grating and the increase in sensitivity due to the high performance detector . oh suppression fibers have the potential to significantly expand our window to the universe from the ground if their suppression of scattered light can be confirmed and the atmosphere does not contain bright continuum sources that are un - filterable . nevertheless , these fibers provide real benefits for spectroscopic observations at low resolving powers and optimized systems utilizing these types of fibers are very feasible . the gnosis team acknowledges funding by arc lief grant le100100164 . we thank the referee for providing insightful comments and suggestions which have greatly improved this paper . c.q.t . gratefully acknowledges support by the national science foundation graduate research fellowship under grant no . thanks stuart gilchrist , stuart jackson , rene pelton , billy robbins , andrew sheinis , and tim white .

in this paper , we consider periodic soft inclusions with periodicity , where the solution , , satisfies semi - linear elliptic equations of non - divergence in with a neumann data on . the difficulty lies in the non - divergence structure of the operator where the standard energy method based on the divergence theorem can not be applied . the main object is developing a viscosity method to find the homogenized equation satisfied by the limit of , called as , as approaches to zero . we introduce the concept of a compatibility condition between the equation and the neumann condition on the boundary for the existence of uniformly bounded periodic first correctors . the concept of second corrector has been developed to show the limit , , is the viscosity solution of a homogenized equation .

1111.2398

[ [ section ] ] let @xmath6 be a bounded and connected domain in @xmath7 with a smooth boundary . we are going to define a perforated domain @xmath8 by removing a @xmath1-periodic balls out of @xmath6 . for each @xmath9 , let @xmath10 be a ball with center @xmath11 and radius @xmath12 . let @xmath13 and @xmath14 the homogenization of partial differential equations in a perforated domain with dirichlet or neumann boundary value has been studied by many authors . please refer @xcite and @xcite for details . in this paper , we will consider the generalization of the following soft inclusions where the diffusion coefficients are zero on the holes : @xmath15 in @xcite , they show that @xmath16 converges to @xmath17 weakly in @xmath18(strongly in @xmath19 ) , and that the limit @xmath17 satisfies @xmath20 for some constant matrix @xmath21 where @xmath22^n } \chi_{t^{\mathfrak a}}$ ] . their method relies on the energy estimates and compensated compactness to pass the limit in the weak formulation . such energy method can not be applicable to nonlinear equations of non - divergence type since the solutions may have different order of energies , @xcite , even though they satisfy equations in the same class . in this paper , we are going to develop a viscosity method to find the homogenization process of the following semi - linear equation of non - divergence type : @xmath23 where @xmath24 , @xmath25 and @xmath26 . and the equation satisfies the following conditions . * conditions i : * a. [ con - unif - elliptic ] @xmath27 is uniformly elliptic : there are positive constants @xmath28 such that @xmath29 for all @xmath30 and for all @xmath31 . b. [ con - oblique ] @xmath32 satisfies the uniform oblique condition : there is a uniform constant @xmath33 satisfying @xmath34 . and , for the convenience , we also assume @xmath35 . c. [ con - y - periodicity ] @xmath36 , @xmath32 , @xmath37 and @xmath38 are periodic in @xmath39-variable : for every @xmath40 , we have @xmath41 d. [ con - c - decr ] @xmath42 and @xmath37 is non - increasing with @xmath43 variables . e. [ con - alpha ] @xmath44 for every @xmath45 . f. [ con - conti ] @xmath46 and @xmath47 are uniformly continuous with respect to @xmath39 variable . that is , for any given @xmath48 and @xmath49 , @xmath50 throughout this paper , we always assume the conditions i above . our first theorem concerns about the existence of compatibility constant for neumann problem . [ thm - main - c ] consider the equation defined as follow : @xmath51 assume that @xmath52 and @xmath53 is bounded . then , for any given @xmath30 , there is a unique constant @xmath54 that makes the equation has a soution @xmath55 . a. we are going to call @xmath56 a compatibility constant of the equation . b. now suppose that @xmath57 . if @xmath58 for all @xmath30 and the size of halls @xmath59 , then we call @xmath60 and @xmath61 ( or the equation ) satisfies the compatibility condition . we remark that laplace equation equipped the neumann boundary condition satisfies the compatibility condition . we will show it in chapter [ sec - com ] . now let us introduce our main theorem : [ thm - main-1 ] let @xmath2 be a viscosity solution of . suppose that our equation satisfies the conditions i and 1 . the equation satisfies the compatibility condition , 2 . @xmath16 is bounded uniformly on @xmath1 , and @xmath62 on the @xmath63 where @xmath64 and @xmath65 is same in definition [ def - lim - u ] . 3 . @xmath66 for uniform constant @xmath67 in theorem [ thm - cor2-elliptic ] . then , there exists an uniformly elliptic operator @xmath68 . and @xmath16 , solution of , converges to @xmath5 , solution of the equation , uniformly . @xmath69 we will use the condition ( 3 ) to prove the uniformly ellipticity of @xmath68 . hence it can be dropped if @xmath70(laplace case ) and @xmath71(neumann boundary case ) . see chapter [ sec - cor2 and ellipticity ] . and the condition ( 2 ) can be dropped if we can construct a barrier at any boundary points . we will show that such a barrier exists if @xmath6 satisfies exterior sphere condition in chapter [ sec - homogenization ] . [ cor - main-1 ] let @xmath16 be the solution of the equation . assume that satisfies all of the conditions ( 1 ) -(6 ) , condition ( 3 ) in theorem [ thm - main-1 ] and the compatibility condition . assume also that @xmath6 satisfies an exterior sphere condition . then @xmath16 converges uniformly to @xmath17 which is the solution of . [ cor - main-2 ] let @xmath16 be the solution of the equation with the condition . then @xmath16 converges uniformly to @xmath17 which is the solution of if the domain @xmath6 satisfies an exterior sphere condition . finally , we develop the following estimate . it tells us that @xmath16 is almost lipschitz continuous . [ [ section-1 ] ] we begin by recalling the definition of viscosity solutions . it is a definition of viscosity solution defined in @xcite . for a given function @xmath5 defined on @xmath72 , 1 . the superdifferential @xmath73 of order 2 at @xmath74 is defined by @xmath75 2 . the subdifferential @xmath76 of order 2 at @xmath74 is defined by @xmath77 3 . @xmath78 is the set of those points @xmath79 to which there corresponds a sequence @xmath80 such that @xmath81 for @xmath82 and such that @xmath83 , @xmath84 , @xmath85 , and @xmath86 as @xmath87 . @xmath88 is the set of those points @xmath79 to which there corresponds a sequence @xmath80 such that @xmath89 for @xmath82 and such that @xmath83 , @xmath84 , @xmath85 , and @xmath86 as @xmath87 . [ def - viscositysol ] for a given nonlinear equation that is defined in bounded domain @xmath6 , @xmath90 1 . an upper semi - continuous @xmath5 is a viscosity sub - solution of if @xmath91 2 . an lower semi - continuous @xmath5 is a viscosity super - solution of if @xmath92 3 . @xmath5 is called a viscosity solution of if @xmath5 is called a viscosity super- and sub - solution of . we employ the comparison principle given at @xcite . let @xmath93 and @xmath94 be viscosity super- and sub - solutions of respectively for given @xmath95 . then @xmath96 in @xmath6 . the following existence theorem for the viscosity solution can be found at @xcite . there is a unique viscosity solution @xmath2 of . we finish this section by introducing more intuitive concept of viscosity solution that is equivalent the definition above if our solution @xmath5 is in @xmath97 near the boundary : let @xmath98 for some bounded @xmath6 . then a continuous function @xmath5 is the viscosity super(sub)-solution of the equation at @xmath99 if @xmath100 touches @xmath5 by below at @xmath99 , then @xmath101 suppose that @xmath5 is a continuous viscosity super - solution(sub - solution ) of at all interior point of @xmath6 . suppose also that @xmath5 is @xmath97 near the @xmath63 . then @xmath5 is a viscosity super - solution(sub - solution ) of if @xmath102 in this section , we are going to define the compatibility condition and investigate their properties . before introducing the compatibility condition , we are going to find the ( periodic ) viscosity solution of the following equation defined on @xmath103 : @xmath104 we assume that all the functions @xmath105 , @xmath46 , @xmath61 and @xmath106 are periodic in @xmath39 variable . assume also that @xmath105 is uniformly elliptic with elliptic constant @xmath107 and @xmath108 and @xmath109 for @xmath110 in condition . then , we will prove the comparison for the viscosity solution of . [ lem - precom - cp ] let @xmath111 and @xmath112 be continuous , bounded and periodic viscosity super and sub - solution of the equation respectively . then we have @xmath113 first , assume that @xmath111 and @xmath112 are in @xmath114 . if the conclusion is not true , there exists @xmath115 such that @xmath116 . now we add a positive constant @xmath117 so that @xmath118 and then decrease @xmath119 until @xmath120 touches @xmath121 . set @xmath122 . then , from the assumption , @xmath123 and hence we can find @xmath124 such that @xmath125 touches @xmath121 at @xmath126 from above . in other words , @xmath127 first let us consider the case when @xmath128 . then , from @xmath129 and @xmath130 , we have @xmath131 which is a contradiction . hence @xmath126 is not on the boundary . therefore @xmath126 is supposed to be an interior point of @xmath132 . hence , @xmath133 the touching point , should be in the interior of @xmath103 . but , it also impossible because @xmath134 also be a viscosity super solution of and the super solution @xmath134 can not touch the sub - solution @xmath112 by above at any interior point . so , @xmath135 on @xmath103 . the case @xmath111 is lower semicontinuous and @xmath112 is upper - semicontinuous can be proved by the usual viscosity argument . see chapter 3 in @xcite . from the comparison , we directly prove the existence of the solution of equation . [ lem - precom - ex ] suppose that there exist a periodic bounded continuous ( viscosity ) super - solution @xmath136 and a sub - solution @xmath137 of the equation of equation . then , there exists the unique periodic viscosity solution @xmath138 of located between @xmath136 and @xmath137 . we first define @xmath139 . then , @xmath55 is well defined because of @xmath136 and @xmath137 . and , we also prove that @xmath55 is a viscosity solution by applying the argument in @xcite . finally , from the definition , @xmath140 for all @xmath40 . [ lem - precom - bdd ] for each @xmath141 , there exists the solution @xmath138 of the equation satisfying @xmath142 for some constant @xmath143 . for given @xmath46 and @xmath106 , let @xmath144 be the solution of the equation when @xmath145 and @xmath146 be the solution of when @xmath147 . then , if we have @xmath148 and @xmath149 , the conclusion comes from the linearity of the equation . so , we consider the case @xmath145 first . we may assume that @xmath150 without losing generality . select a ball @xmath151 which is a component of @xmath152 and a unit cube @xmath153 of @xmath154 with center 0 . set @xmath155 by the direct calculation , we can obtain @xmath156 where @xmath157 and @xmath158 are eigenvalues of @xmath159 . select a large @xmath160 so that @xmath161 . and define @xmath162 where @xmath163 . then , we have @xmath164 from @xmath165 on the boundary @xmath166 , we get @xmath167 now we define @xmath168 @xmath169 then , because of the shape of @xmath170 , @xmath171 in the viscosity sense for all interior points of @xmath172 and , on the boundary , @xmath173 on @xmath4 . therefore @xmath174 is a periodic viscosity super - solution , and we can observe that @xmath175 be a viscosity sub - solution . hence , by the lemma [ lem - precom - ex ] , there is a periodic viscosity solution @xmath176 of the equation between a sub - solution @xmath175 and a super - solution @xmath177 . in addition , we have @xmath178 and @xmath179 for general @xmath46 , consider the function @xmath180 and apply the above estimate , we can get the conclusion . and , if @xmath147 , then @xmath181 become a super and sub - solutions so , we can deduce that @xmath182 . [ lem - precom - osc ] let @xmath176 be the solution of and @xmath183 then we have @xmath184 for a uniform constant @xmath185 . for small @xmath186 , @xmath187 be a set of balls with radius @xmath188 instead of balls with radius @xmath59 . let @xmath189 , @xmath190 , @xmath191 , @xmath192 and @xmath193 . then @xmath194 satisfies @xmath195 let @xmath153 be a unit cell of @xmath103 which is punctured by a ball @xmath196 . we may assume that @xmath197 . since @xmath194 is nonnegative and @xmath198 is contained in @xmath103 , we can apply the harnack estimate ( in @xcite ) on @xmath199 in @xmath200 , and hence we have @xmath201 for some @xmath202 which depend only on @xmath203 , @xmath107 , @xmath108 , @xmath59 and @xmath204 . and hence , from the periodicity , we have @xmath205 let @xmath206 for some @xmath207 . then we have @xmath208 and @xmath209 on @xmath210 . then , if we select @xmath211 , @xmath111 is a super solution in @xmath212 . so , a comparison principle tells us @xmath213 in @xmath214 and hence @xmath215 so , by choosing @xmath204 properly between @xmath216 and @xmath217 , we have @xmath218 for some constant @xmath219 . similarly , we can obtain @xmath220 . combine these three results . then we can conclude @xmath221 where @xmath222 depends only on @xmath223 and @xmath59 . finally , applying lemma [ lem - precom - bdd ] to get @xmath224 , we get the conclusion . now , we are discussing the regularity of @xmath138 . the regularity of viscosity solution of bounded domain has been developed by many authors . especially , we will use the results in @xcite , @xcite and @xcite to get the regularity of @xmath138 . let us assume that @xmath225 and @xmath53 is bounded . let @xmath153 be a cell of @xmath154 . we may assume that the center of @xmath153 and @xmath226 is @xmath216 . by applying the interior estimate in @xcite , @xmath199 is @xmath227 at every interior points and hence for some open set @xmath228 which is contained in @xmath229 and containing @xmath230 , and @xmath231 where @xmath222 is a constant depending only on @xmath203 , @xmath107 , @xmath108 and @xmath59 . let @xmath232 be a function which has same value with @xmath199 in @xmath228 . then , @xmath199 satisfies @xmath233 so , from the @xcite , @xmath199 is @xmath234 in @xmath235 ( hence in @xmath236 ) and , from the a priori estimate in @xcite , we have the @xmath234 estimate @xmath237 in summary , we have [ lem - precom - c2 ] let @xmath55 is the ( viscosity ) solution of the equation with @xmath238 and @xmath239 . then @xmath55 is in @xmath240 for some @xmath241 and we have @xmath242 in this section , we are going to prove the theorem [ thm - main-1 ] by applying previous subsection . first , for fixed @xmath30 , consider the following approximated equation @xmath244 for each @xmath1 , we have the periodic viscosity solution @xmath245 of by lemma [ lem - precom - bdd ] . [ lem - precor1-ex ] for each @xmath141 , there exists @xmath138 satisfying and we have @xmath246 actually , we just need the result when @xmath57 . but , in this section , we can consider more general case(@xmath46and @xmath106 are not identically @xmath216 ) because it does not effect on the result . the following is about the oscillation of @xmath199 . [ lem - precor1-osc ] let @xmath176 be the solution of the equation and @xmath183 then we have @xmath247 let @xmath248 . then @xmath194 satisfies @xmath249 now apply lemma [ lem - precom - osc ] and then we can get the conclusion . we can also obtain the estimate of @xmath199 by lemma [ lem - precor1-osc ] and lemma [ lem - precom - c2 ] . [ lem - com - e - c2 ] suppose that @xmath105 , @xmath46 , @xmath61 and @xmath106 satisfies the condition in lemma [ lem - precom - c2 ] . then , we have @xmath250 where @xmath222 is depending only on @xmath203 , @xmath107 , @xmath108 and @xmath59 . @xmath251 from the condition in chapter 1 and lemma [ lem - precor1-osc ] , @xmath252 is bounded . hence we can apply lemma [ lem - precom - c2 ] to @xmath199 of and we can get the estimate . lemma [ lem - com - e - c2 ] and [ lem - precor1-ex ] tells us that @xmath253 is bounded uniformly on @xmath1 . so , from arzela - ascoli theorem , we can deduce that there is a @xmath254 , @xmath255 , and a subsequence @xmath256 where @xmath257 converges @xmath55 in @xmath258 and @xmath259 uniformly . and if we take @xmath260 , then @xmath261 and @xmath241 satisfy the equation . [ prop - exist - alpha ] if @xmath105 , @xmath46 , @xmath61 and @xmath106 satisfies the condition in lemma [ lem - precom - c2 ] then , we always find @xmath262 and @xmath263 which satisfy the equation . [ lem - alpha - uniq ] let @xmath243 be given as [ prop - exist - alpha ] . then such @xmath243 is unique . let @xmath264 and @xmath265 be two solutions of the equation with corresponding to constants @xmath266 and @xmath267 respectively . and , to obtain a contradiction , assume that @xmath266 and @xmath267 are not same . we may assume that @xmath268 without losing generality . since @xmath144 and @xmath146 are bounded , we can find a constant @xmath47 such that @xmath269 touches @xmath146 by above at @xmath270 . suppose that @xmath133 is a interior point , then @xmath271 has a local minimum at @xmath133 . but since @xmath271 is a solution of @xmath272 , @xmath271 can not have its minimum at interior point because of the strong maximum principle . so @xmath133 can not be in the interior of @xmath103 . suppose that @xmath273 . then , @xmath274 but , @xmath275 so we get a contradiction and hence @xmath266 should be the same with @xmath267 . from proposition [ prop - exist - alpha ] , there exist @xmath243 that makes the equation [ eqn - main - com ] has a solution @xmath55 . and by lemma [ lem - alpha - uniq ] , such a @xmath243 is unique . we can define a compatibility constant even the operator and boundary condition are nonlinear . more precisely , for given operator @xmath276 and boundary condition @xmath277 , and a vector @xmath30 , there is a constant @xmath241 and a periodic function @xmath278 satisfying the equation @xmath279 if the operator satisfies the conditions in @xcite . the proof is quite similar . as we told in the introduction , the laplace equation and the neumann boundary condition satisfies the compatibility condition . let @xmath153 be a one cell of @xmath103 having center @xmath216 and @xmath55 is a solution satisfying the following equation : @xmath280 then , by using divergence theorem , we have @xmath281 hence , @xmath243 should be @xmath216 . moreover , the operator satisfying the symmetric condition @xmath282 and @xmath283 then , we can show that @xmath60 and @xmath61 satisfies the compatibility condition . let @xmath138 be the solution of equation . assume that @xmath105 and @xmath61 satisfies @xmath282 and @xmath283 and the condition in lemma [ lem - precom - c2 ] . then , @xmath284 and hence @xmath105 and @xmath61 satisfies the compatibility condition . let @xmath285 . then @xmath286 and @xmath287 . apply it to the equation . then , @xmath288 it tells us that @xmath289 is also a solution of equation and hence @xmath290 by comparison ( lemma [ lem - precom - cp ] . from above , @xmath138 can not be nonnegative or nonpositive unless @xmath291 identically . so , we can conclude @xmath292 because of lemma [ lem - precor1-osc ] and hence @xmath293 converges to 0 . in this section , we are going to define the first corrector from the heuristic calculation and investigate their existence and regularity by using results in previous section . let us consider the asymptotic expansion of @xmath16 at @xmath294 . in other words , suppose that @xmath16 has the following asymptotic expansion . @xmath295 if @xmath17 is regular , then it is quite similar to the second polynomial @xmath296 near @xmath294 . so , we will identify @xmath17 with @xmath297 . finally , define @xmath298 to simplify the notation . then , by the calculation , we have first and second derivatives : @xmath299 from , we have @xmath300 we can observe that , at a first line of the equation , there is one @xmath301 order term . so , if @xmath55 and @xmath302 exists and regular enough , then @xmath55 should satisfy @xmath303 in the interior of @xmath304 . and , on the boundary , there are three @xmath305-order terms @xmath306 , @xmath307 , and @xmath308 . hence we could find a equation for @xmath55 : @xmath309 as we discuss before , there is a periodic solution @xmath55 of the equation above if @xmath60 and @xmath61(or , our main equation ) satisfies the compatibility condition . since the equation is linear , @xmath310 is linear with respect to @xmath311 . that is , if @xmath312 is the solution of the equation with @xmath313 , then @xmath314 . we are going to deduce the properties of @xmath55 from @xmath312 . we note that the solution of is not unique since @xmath315 is a solution of @xmath316 is a solution . so , we assume that @xmath312 is the solution of the equation when @xmath317 satisfying @xmath312 is nonnegative and @xmath318 . from lemma [ lem - precom - osc ] and lemma [ lem - precom - c2 ] , the @xmath114 norm of @xmath312 is bounded by constant which is depend on the size of holls @xmath59 . the following lemmas concerns about the relation between that constant @xmath222 and @xmath59 . [ lem - cor1-osc - uni ] let @xmath55 be the periodic solution of the equation with @xmath318 and assume that @xmath59 is small enough . then we have @xmath319 where @xmath320 . we are going to assume @xmath321 without losing generality . let @xmath153 be a unit cell of @xmath154 whose center is @xmath216 and @xmath322 . without losing any loss of generality , we may assume that the center of @xmath153 and @xmath226 is 0 . then , since @xmath144 satisfies @xmath323 the maximum and minimum should be achieved at a boundary point from the maximum principle ( in @xcite ) . so , @xmath324 by the definition of @xmath144 , @xmath144 is nonnegative and @xmath325 . let @xmath326 , @xmath327 and @xmath328 then , @xmath136 satisfies @xmath329 and then we have @xmath330 from the comparison . similarly , we can show @xmath331 let @xmath332 . then @xmath333 is nonnegative and satisfies @xmath334 whenever @xmath335 . by applying the harnack estimate ( in @xcite ) on @xmath333 in @xmath336 , we have @xmath337 which implies @xmath338 . now combining these three results , we have @xmath339 [ lem - cor1-dv1 ] let @xmath340 be the solution of the equation with @xmath317 . and suppose that the coefficient functions @xmath60 and @xmath61 satisfies @xmath238 . then , we have the following estimate @xmath341 where @xmath342 is the same as in [ lem - cor1-osc - uni ] and @xmath343 is a distance between @xmath39 and @xmath152 . in particular , @xmath344 if @xmath345 . it follows from lemma [ lem - cor1-osc - uni ] and the standard interior @xmath234 estimate(see chapter 6 of @xcite ) . [ lem - cor1-dv2 ] let @xmath340 be the solution of the equation with @xmath313 . and suppose that the coefficient functions @xmath60 and @xmath61 satisfies @xmath238 . then , we have @xmath346 where @xmath347 . let @xmath153 be a unit cell of @xmath154 whose center is @xmath216 and @xmath348 . we will show @xmath349 is bounded . from lemma [ lem - cor1-dv1 ] , @xmath350 in bounded in @xmath351 , so we just need to show that the gradient is bounded in @xmath352 . let us define the scaled function @xmath353 . then @xmath333 satisfies the following equation @xmath354 where @xmath355 . since @xmath356_{c^\alpha(b_2 \setminus b_1 ) } = \mathfrak{a}^\alpha [ a_{ij}(\cdot)]_{c^\alpha(b_{2\mathfrak{a } } \setminus b_\mathfrak{a})}$ ] , @xmath356_{c^\alpha(b_2 \setminus b_1 ) } + [ b^i(\mathfrak{a}~\cdot)]_{c^{1,\alpha}(b_2 \setminus b_1 ) } \le \lambda$ ] . additionally , since @xmath357_{c^{2,\alpha}(\partial b_2)}$ ] is bounded independently on @xmath59 from lemma [ lem - cor1-dv1 ] , we have the following estimate @xmath358_{c^{2,\alpha}(b_2 \setminus b_1 ) } \le c(n,\lambda,\lambda,\mu ) \left ( [ b^i(\mathfrak{a}~\cdot)]_{c^{1,\alpha}(\partial b_2 ) } + [ \phi]_{c^{2,\alpha}(\partial b_2 ) } \right ) \le c_1(n,\lambda,\lambda,\mu)\ ] ] by using the estimate in @xcite . especially , @xmath359 is bounded by @xmath202 whenever @xmath360 . in this section , we define the effective equation @xmath68 by finding the second corrector . and , we prove two important properties of @xmath68 : the uniform ellipticity and continuity of the effective equation @xmath68 . throughout this section , we assume that satisfies the compatibility condition and condition i in chapter 1 hold . let us define @xmath361 as a @xmath154-valued function whose components are @xmath340 and @xmath362 for a given vector @xmath363 and a symmetric matrix @xmath159 . additionally , let us define a matrix @xmath364 as @xmath365 from lemma [ lem - cor1-dv2 ] , we can deduce the following lemma . [ lem - cor2-bddz ] for any given a symmetric matrix @xmath159 and a point @xmath39 in @xmath103 , the following estimate holds : @xmath366 now , let us apply @xmath367 to the equation . then we have the following : @xmath368 we note that our second corrector @xmath302 should satisfy the above equation . we add the auxiliary term @xmath369 to the interior equation and @xmath370 to the boundary equation to guarantee the existence of second corrector . then , we have the following equation about @xmath39 variable for fixed @xmath371 and @xmath372 . @xmath373 from lemma [ lem - cor2-bddz ] @xmath364 is bounded . so , the equation is well defined . and , by adding the auxiliary term , we can find a bounded viscosity solution for each @xmath1 and we also can prove the comparison principle like lemma [ lem - precom - cp ] . since the proof of comparison principle is similar to that of lemma [ lem - precom - cp ] , we just state it without proof . ( comparison ) [ lem - cor2-com ] suppose that @xmath374 is a super - solution of and @xmath375 is a sub - solution of for fixed @xmath159 , @xmath376 , @xmath377 , @xmath378 and @xmath1 . then we have @xmath379 [ lem - cor2-exist ] for each @xmath159 , @xmath59 , @xmath377 , @xmath378 and @xmath1 , there is a periodic solution @xmath380 of the equation satisfying @xmath381 where @xmath222 is a constant depending only on @xmath203 , @xmath107 , @xmath108 , @xmath33 and @xmath59 . for fixed @xmath59 , let us define @xmath382 then , from lemma [ lem - cor2-bddz ] and [ lem - cor1-osc - uni ] , we have @xmath383 and , from the definition of @xmath384 , @xmath385 and @xmath386 are super and sub - solution of respectively . so , from the similar reason in lemma [ lem - precom - cp ] , we can find the solution @xmath302 of which satisfies @xmath387 [ lem - cor2-osc ] the solution @xmath302 of the equation satisfies @xmath388 where @xmath222 depends only on @xmath203 , @xmath107 , @xmath108 and @xmath33 . let @xmath389 and @xmath390 . then @xmath391 satisfies the following equation : @xmath392 from lemma [ lem - cor2-bddz ] , lemma [ lem - cor2-exist ] and lemma [ lem - cor1-osc - uni ] , we have @xmath393 we note that @xmath394 and @xmath395 are bounded uniformly on @xmath396 . therefore , the oscillation of @xmath302 is bounded because of lemma [ lem - precom - osc ] . let @xmath397 and @xmath398 . then , @xmath398 satisfies @xmath399 where @xmath400 + and @xmath401 . from proposition [ prop - exist - alpha ] , @xmath402 is in @xmath403 and hence @xmath364 is in @xmath404 for fixed @xmath159 , @xmath377 , and @xmath378 . and then from the condition in chapter 1 , @xmath405 is bounded uniformly on @xmath1 . so , we have @xmath234 estimate for @xmath406 . let @xmath407 be the solution of the equation of which satisfies the condition ( in condition i ) . then , @xmath408 is bounded uniformly on @xmath1 for given any @xmath159 , @xmath43 and @xmath378 . the proof of corollary above is almost same as that of lemma [ lem - precom - c2 ] . so we omit the proof . there is a unique limit of @xmath370 as @xmath409 . from lemma [ lem - cor2-exist ] and lemma [ lem - cor2-osc ] , there exists a subsequence @xmath410 such that @xmath411 converges to a constant @xmath412 . and , from similar argument in lemma [ lem - alpha - uniq ] , @xmath412 should be same even though we change the subsequence because of the uniform @xmath234 estimate of @xmath406 . that implies @xmath413 converges to @xmath412 . @xmath414 is the limit of @xmath415 for fixed @xmath159 , @xmath377 and @xmath378 . we prove later that the limit equation of @xmath16 satisfies the equation @xmath416 in chapter [ sec - homogenization ] . usually , it is called as an _ effective equation_. we will end this section by proving two important properties of @xmath68 , _ uniformly ellipticity _ and _ continuity_. [ thm - cor2-elliptic ] assume the conditions in condition i hold and the equation satisfies the compatibility condition . then , there is a positive real number @xmath67 depending only on @xmath203 , @xmath107 , @xmath108 and @xmath33 such that if the size of hole @xmath59 is less than or equal to @xmath67 , then @xmath68 is uniformly elliptic . in other words , there is a positive constant @xmath417 satisfying @xmath418 for any symmetric matrix @xmath159 and positive matrix @xmath419 . we will show @xmath420 for any given @xmath159 , @xmath419 , @xmath377 , and @xmath378 . actually , it is equivalent to prove @xmath421 . so , we first define @xmath422 . then , @xmath423 satisfies the following equation . @xmath424 we will construct a ( viscosity ) sub - solution @xmath425 such that @xmath426 converges to a positive constant . we consider the case @xmath427 because the general result can be obtained by scaling . to construct a sub - solution , we need to estimate the righthand side of the equation . first , from lemma [ lem - cor2-bddz ] , @xmath428 . hence we have @xmath429 for all @xmath430 . so , for small @xmath1 , we have @xmath431 from lemma [ lem - cor1-dv2 ] , @xmath432 is bounded uniformly on @xmath376 . more precisely , @xmath433 where @xmath202 is a constant in lemma [ lem - cor1-dv2 ] that depends only on @xmath203 , @xmath107 , @xmath108 , and @xmath33 . secondly , from lemma [ lem - cor1-dv1 ] , @xmath434 is small if @xmath39 is far from the boundary . more precisely , for any given @xmath435 , if the size of halls @xmath59 is less than or equal to @xmath436 , then we have @xmath437 where @xmath438 is a distant between @xmath39 and @xmath152 and the constant @xmath439 is same in lemma [ lem - cor1-dv1 ] . finally , from lemma [ lem - cor1-osc - uni ] , we have @xmath440 where @xmath441 is a constant which is independent of @xmath1 and @xmath59 . now we are ready to define the barrier . let us define the function @xmath442 as follow . @xmath443 we will select @xmath384 , @xmath444 , and @xmath435 later . then , the function @xmath442 is continuous and twice differentiable except the points on @xmath445 . and from the calculation , we have @xmath446 and , on the boundary @xmath210 , we have @xmath447 if we choose @xmath384 bigger than @xmath444 , then our function @xmath448 has sharp edge on @xmath445 , and hence there are no second order polynomials touching @xmath449 by above at any points on @xmath445 . select @xmath450 , @xmath451 . then , @xmath452 and , @xmath453 and , on the boundary , we have @xmath454 finally , select @xmath435 satisfying @xmath455 and @xmath456 . then , @xmath457 if @xmath458 and hence @xmath459 is a sub - solution of equation for small @xmath460 from and . and , from lemma [ lem - cor2-com ] , we have @xmath461 that is @xmath462 if @xmath463(laplace equation ) , @xmath464 and @xmath71(neumann boundary condition ) in equation , then we can prove the uniform ellipticity even the size of hall @xmath59 is large because we can use the divergence theorem . let @xmath153 be a one cell whose center is 0 and punctured by a ball @xmath465 . we identify @xmath466 with @xmath414 because the error between them is of order @xmath467 . then , we have the followng by using the divergence theorem : @xmath468 where @xmath469 and @xmath470 . so , we have the explicit formula of @xmath414 : @xmath471 and the uniform ellipticity comes automatically from above formula . [ prop - cor2-conti ] assume the conditions in condition i and the equation satisfies the compatibility condition . then , 1 . @xmath472 is continuous with respect to @xmath43 and @xmath99 variables . @xmath472 is non - incresing with @xmath43 variable . 1 . we will show that @xmath68 is continuous with @xmath99 variable for fixed @xmath159 , @xmath43 . and we omit the proof of the continuity with @xmath43 because that is quite similar to the proof of continuity with @xmath99 . now , suppose that @xmath159 , @xmath43 are fixed . and let @xmath473 then , the equation for second corrector can be modified to @xmath474 + for the simplicity of notation , we define @xmath475 and @xmath476 for some @xmath477 , @xmath478 . then , @xmath479 satisfies the following equation : @xmath480 + since @xmath46 and @xmath47 are continuous uniformly on @xmath39 , @xmath481 where @xmath482 is a nondecreasing function with @xmath483 . + hence , @xmath484 because of lemma [ lem - cor2-exist ] . and the conclusion comes by taking limit on both side . it can be shown by using similar argument above and the comparison principle . in this section , we are going to prove the limit of solutions satisfies the homogenized equation . first , assume that @xmath16 is bounded uniformly on @xmath1 . then we can define the limit of @xmath16 in the following way . [ def - lim - u ] define @xmath64 and @xmath65 as follow : @xmath485 we will prove that @xmath65 is a super - solution of the equation of the effective equation(equation [ eqn - rst-1 ] ) . and the lower semi - continuity of @xmath65(upper semi - continuity of @xmath64 ) comes from a similar argument as in @xcite . now we are going to prove our main theorem . suppose that @xmath65 is not a viscosity super - solution . then , there is a second polynomial @xmath297 touches @xmath486 from below at @xmath48 such that there exists @xmath487 satisfying @xmath488 in @xmath489 and @xmath490 and @xmath491 . for the simplicity , suppose that @xmath492 and @xmath493 . set @xmath494 . then , since @xmath495 is compact for any given @xmath496 , we can find @xmath497 which satisfies @xmath498 from the definition of @xmath65 , there is a subsequence @xmath499 \times \overline{\omega}$ ] which converges to @xmath500 satisfying @xmath501 and @xmath502 set @xmath503 . since @xmath504 and @xmath505 is compact , we can find a subsequence of @xmath506 which converges to some @xmath507 as @xmath508 . and hence we assume that @xmath509 and @xmath510 . since @xmath511 from the definition of @xmath512 , by taking limit infimun on both side , we have @xmath513 but , by the definition of @xmath514 , we have @xmath515 therefore @xmath516 and hence @xmath517 . that implies the sequence @xmath518 converges to 0 ( not as a subsequence ) . and , from above inequality , we also conclude that @xmath519 let @xmath520 . since the sequence @xmath521 related with @xmath522 converges to 0 as @xmath87 , we can find @xmath523 and @xmath524 satisfying @xmath525 and @xmath526 . after setting @xmath527 , we also find @xmath528 and @xmath529 satisfying @xmath530 and @xmath531 . in this way , we can obtain a sequence @xmath532 satisfying 1 . @xmath533 is a subsequence of @xmath534 , 2 . @xmath535 , @xmath536 , and @xmath537 as @xmath538 , and 3 . @xmath539 in @xmath505 where @xmath540 . let @xmath541 for a given @xmath542 . then , from @xmath543 , @xmath544 satisfies , @xmath545 and @xmath546 on @xmath547 . let @xmath548 , @xmath549 and @xmath550 . we note that @xmath551 . now , let us define the first and second corrector as follow . @xmath552 and , define @xmath553 where @xmath554 . we will show that @xmath555 is a sub - solution in a ball @xmath505 if we choose @xmath556 and @xmath384 properly . first , let us check the boundary condition on @xmath557 . @xmath558 from the continuity of @xmath68 , @xmath559 converges to @xmath560 since @xmath561 and @xmath562 . and @xmath560 is positive if we choose @xmath384 and @xmath556 small enough since @xmath563 is positive . hence we have @xmath564 for sufficiently large @xmath565 . now we are going to apply @xmath566 to our main equation . from the calculation , we have @xmath567 apply it to the equation . then we have @xmath568 from the definition of @xmath569 , @xmath570 whenever @xmath571 . hence we can make @xmath572 small by choosing @xmath573 small enough . and , from lemma [ lem - cor1-osc - uni ] and [ lem - cor2-osc ] , @xmath574 for some constant @xmath222 which is uniform on @xmath1 and @xmath565 . hence it is smaller than @xmath575 if @xmath565 is large enough . from those two calculations , @xmath576 satisfies the following by small @xmath556 and large @xmath565 . @xmath577 similarly , because of the continuity of @xmath47 and @xmath46 with @xmath99 variable , @xmath578 hence , from the continuity of @xmath68 , we have @xmath579 for large @xmath565 . in summary , @xmath555 is a sub - solution of equation in @xmath580 for large @xmath565 and @xmath581 on @xmath582 . hence , from the comparison , we have @xmath583 in @xmath584 . substitute @xmath585 instead of @xmath99 to above equation and take limit on both side . then , we have @xmath586 that is a contradiction . so , @xmath587 is nonpositive and hence @xmath65 is a super - solution at any point in @xmath6 . by using similar argument , we can show @xmath64 is a sub - solution of the equation in @xmath6 . from the second assumption , @xmath588 on @xmath63 . so , @xmath589 from the comparison principle . finally , since @xmath590 because of the definition of @xmath65 and @xmath64 , we conclude @xmath591 in this section , we are going to construct a barrier to show @xmath588 on the @xmath63 . at first , we are going to prove that the condition ( 2 ) in [ thm - main-1 ] holds if @xmath6 is convex . [ lem - bar ] let @xmath16 be the solution of the equation . assume that the equation satisfies the condition i and compatibility condition and the size of halls @xmath376 is less than or equal to @xmath592 where @xmath67 is same with the constant in theorem [ thm - cor2-elliptic ] . assume also that @xmath593 is convex . then @xmath16 is bounded uniformly on @xmath1 . moreover , for any given @xmath594 , we can find a barrier functions @xmath595 and @xmath596 which is bounded uniformly on @xmath1 and satisfying @xmath597 for the simplicity we assume that @xmath492 and @xmath598 and @xmath6 is contained in a half space @xmath599 . we also assume that @xmath600 and @xmath601 since the general case can be obtained by scaling . let @xmath312 be the solution of equation when @xmath313 with @xmath602 . let @xmath603 be a vector whose i - th component is @xmath340 . since @xmath601 , @xmath604 . choose @xmath605 and define @xmath606 for some @xmath384 . define also @xmath607 , and @xmath608 where @xmath609 is the n - th stanadard unit vector defined in @xmath154 . we finally define @xmath610 where @xmath611 is a constant satisfying @xmath612 and @xmath613 is a matrix with defined by @xmath614 . by adding @xmath615 , @xmath616 in @xmath6 and @xmath617 on @xmath63 . then , from the calculation , we have @xmath618 now , apply it to our main equation , then we have @xmath619 by the theorem [ thm - cor2-elliptic ] , @xmath620 if @xmath1 is small enough . from the definition , we can easily deduce that @xmath621 . and hence if we choose @xmath622 then , @xmath623 for every @xmath624 . and , at the boundary , @xmath625 finally , since @xmath626 on @xmath63 , @xmath627 is a super - solution of and hence @xmath628 on @xmath304 . and the uniform boundedness comes from the fact that @xmath629 , and @xmath202 and @xmath406 is bounded uniformly on @xmath1 . let @xmath630 . then , @xmath631 in @xmath6 and @xmath632 . by using similar argument , we can construct @xmath596 having properties @xmath633 in @xmath6 and @xmath634 . and such @xmath595 and @xmath596 satisfy . let @xmath16 be the solution of equation and assume all the conditions in lemma [ lem - bar ] . then , for any given boundary point @xmath378 , @xmath635 from lemma [ lem - bar ] , there is a function @xmath595 satisfying @xmath636 in @xmath6 for some constant @xmath222 which is uniform on @xmath1 . and from the definition of @xmath595 in lemma [ lem - bar ] , @xmath595 converges to @xmath297 uniformly . so , @xmath637 and hence @xmath638 . similarly , we can show @xmath639 . finally , since @xmath640 , @xmath641 . in this section , we construct a barrier to show @xmath642 on @xmath63 for non - convex domain @xmath6 . throughout this section , we do nt assume that @xmath6 is convex . instead , we assume that @xmath6 has a exterior sphere condition . in the other words , for given any @xmath378 on @xmath63 , there is a ball @xmath643 satisfying @xmath644 . let us also define the set of functions : @xmath645 for some nondecreasing function @xmath412 satisfying @xmath646 . we note that @xmath46 satisfying ( condition i ) is in @xmath647 . so , for given any @xmath38 in @xmath647 , we may find @xmath648 . we define @xmath649 . in other words , @xmath650 is the limit of @xmath651 and @xmath302 is the solution of the following equation for given @xmath38 : @xmath652 suppose that @xmath653 identically . then , @xmath654 is a positive constant which depends only on @xmath203 , @xmath107 , @xmath108 , and the size of holls @xmath376 . [ fig - q][ht ] it is enough to prove @xmath655 is a negative constant . let @xmath153 be a subset of @xmath103 having the shape in the picture and @xmath656 be a point in @xmath153 . we can choose @xmath153 that has a @xmath114 boundary . let @xmath657 be the solution of the following equation : @xmath658 then , from the regularity theory of viscosity solution in @xcite and hopf s lemma(lemma 3.4 in @xcite ) , @xmath657 satisfies the following : @xmath659 where @xmath660 and @xmath661 are positive constants depending only on @xmath203 , @xmath107 , @xmath108 and @xmath33 . now extend @xmath662 outside of @xmath153 and @xmath663 in @xmath664 . hence , if we choose @xmath665 , then @xmath657 satisfies @xmath666 ( in the viscosity sense ) since the sharp edge on @xmath667 does not allow any second polynomial to touch @xmath657 from above on @xmath667 . so , the function @xmath668 is a sub - solution of equation for small @xmath1 . hence the solution @xmath302 of with @xmath669 should be larger than @xmath670 by lemma . now we get the conclusion since @xmath671 . [ lem - bar - inv ] for given any @xmath672 and @xmath673 , there is a function @xmath674 satisfying @xmath675 by above lemma , there is a positive constant @xmath676 such that @xmath677 . from the definition of @xmath654 , if @xmath46 is independent on @xmath39 , then @xmath678 for each fixed @xmath74 . from this and the linearity of the equation , @xmath679 for all @xmath680 . now let us define @xmath681 as follow : @xmath682 then , @xmath681 is continuous since @xmath650 and @xmath683 is continuous and , @xmath684 let @xmath378 be a point on @xmath63 . we assume that @xmath6 has exterior sphere condition . so , there exists a ball @xmath685 such that @xmath686 . we may assume that @xmath687 without any loss of generality . since @xmath6 is bounded , we can find a large ball @xmath688 containing @xmath6 and @xmath689 . extend @xmath37 and @xmath38 when @xmath690 and hence @xmath691 also defined on @xmath688 . [ lem - bar - h ] assume that @xmath692 is uniformly elliptic . then , there is a function @xmath693 satisfying the following : @xmath694 we may assume that @xmath695 is defined in @xmath688 since it can be easily extended . let @xmath696 outside the ball @xmath689 . then , by defining @xmath697 properly inside the ball @xmath698 , we may assume that @xmath442 is in @xmath114 in @xmath688 . and , from the definition , the second and third statement are true for all @xmath699 and @xmath207 . so , we get the conclusion by choosing @xmath241 and @xmath384 large enough . let us define the function @xmath700 on @xmath688 . then , from proposition [ prop - cor2-conti ] and lemma [ lem - bar - h ] , @xmath395 is continuous and , by lemma [ lem - bar - inv ] , there is a continuous function @xmath681 satisfying the following : @xmath701 for any given @xmath673 . [ lem - bar - g ] let @xmath106 be the function defined as above for given @xmath673 . then , @xmath702 . from the definition and the comparison , @xmath703 . and from the construction of @xmath442 , @xmath704 by combining those two inequalities , we have @xmath705 . now consider the equation @xmath706 where @xmath442 is same in lemma [ lem - bar - h ] and @xmath106 is same in lemma [ lem - bar - g ] . since @xmath688 is convex , if we assume all the conditions in the theorem [ thm - main-1 ] except ( 2 ) , then @xmath138 converges to @xmath333 uniformly and @xmath333 satisfies the equation @xmath707 and hence , @xmath708 because of the comparison . let us define @xmath709 . then , from the uniform convergence , @xmath710 as @xmath409 . and , from the definition of @xmath442 , @xmath711 since @xmath138 is a super - solution in @xmath6 , we have the following by comparison : @xmath712 by taking @xmath713 on both side , we can conclude the following @xmath714 similarly , we can show @xmath715 . hence we get the following : let @xmath16 is the solution of equation . assume that the equation satisfies the condition i and the compatibility condition . suppose also that @xmath692 is uniformly elliptic . then , @xmath716 on @xmath63 if @xmath6 satisfies an exterior sphere condition . it follows immediately from theorem [ thm - main-1 ] and lemma above . [ [ section-2 ] ] in this section , we develop the following uniform estimates . those two estimate tell us about the shape of @xmath16 . it turns out that @xmath16 is almost lipschitz continuous with an error of @xmath1 order . let us consider the following equation : @xmath717 we note that the function @xmath47 is independent on @xmath99 variable . it is only difference between above equation and the main equation . throughout this section , we assume that the function @xmath718 and @xmath38 is differentiable with respect to @xmath43 and @xmath99 variable respectively . additionally , we assume that @xmath6 is convex and all the assumptions in chapter 1 . [ lem - disc - grad ] let @xmath16 be the solution of above equation and , the size of holes @xmath376 is smaller than @xmath67 where @xmath67 is same in theorem [ thm - cor2-elliptic ] . then , for given direction @xmath719 , @xmath720 is bounded uniformly on @xmath1 . that is , there exist @xmath721 and @xmath722 . since @xmath731 , @xmath732 and @xmath733 , we have @xmath734 because @xmath47 is differentiable in @xmath43 variable and @xmath46 is differentiable in @xmath99 variables , by the mean value theorem , we can find @xmath735 and @xmath736 which satisfies @xmath737 where @xmath738 . and from the boundary condition of @xmath728 and @xmath739 , @xmath740 on @xmath741 . now we are going to prove the boundedness of @xmath742 on @xmath743 . let @xmath99 be a point in @xmath744 . since @xmath745 , @xmath746 or @xmath747 . we assume @xmath746 since the other case is similar . let @xmath595 and @xmath596 be functions satisfying . then , we have @xmath748 it can be shown from the comparison if there is a super - solution which is bounded uniformly on @xmath1 , then we are done . let @xmath753 where @xmath754 is same in lemma [ lem - bar ] . then , we can show that @xmath136 is a super solution for large @xmath384 . hece @xmath742 is bounded uniformly on @xmath726 since @xmath136 is bounded uniformly on @xmath1 . [ lem - disc - flat ] let @xmath755 be the viscosity solution of . and suppose all the conditins in lemma [ lem - disc - grad ] are satisfied . then , there is a constant @xmath756 which is independent of @xmath1 satisfying @xmath757 for every @xmath477 , @xmath758 contained in a same @xmath1-cell of @xmath759 . . then @xmath761 satisfies @xmath762 & \text { in } \frac{1}{{\varepsilon } } \omega \setminus t^{\mathfrak a } \\ b^i(y ) d_i \widetilde{u}_{\varepsilon}(y ) = 0 & \text { on } \frac{1}{{\varepsilon}}\omega \bigcap \partial t^{\mathfrak a } \\ \widetilde{u}_{\varepsilon}= 0 & \text { on } \partial \frac{1}{{\varepsilon}}\omega \setminus t^{\mathfrak a}. \end{cases}\ ] ] let @xmath153 be an unit cell with center 0 in @xmath154 and let @xmath763 be the intersection between @xmath153 and @xmath152 . if @xmath764 is nonempty , then we can prove the lemma by the barriers @xmath765 in lemma [ lem - disc - grad ] . so , we will assume that @xmath766 and @xmath767 . 99 caffarelli , luis a. ; cabr , xavier _ fully nonlinear elliptic equations . _ american mathematical society colloquium publications , 43 . american mathematical society , providence , ri , 1995 . vi+104 pp . isbn : 0 - 8218 - 0437 - 5 crandall , michael g. ; ishii , hitoshi ; lions , pierre - louis _ user s guide to viscosity solutions of second order partial differential equations . * 27 * ( 1992 ) , no . 1 , 167 luis caffarelli ; ki - ahm lee , _ viscosity method for homogenization of highly oscillating obstacles _ indiana univ . math . j. 57 ( 2008 ) , 1715 - 1742 . caffarelli , luis a. ; souganidis , panagiotis e. ; wang , l. _ homogenization of fully nonlinear , uniformly elliptic and parabolic partial differential equations in stationary ergodic media . pure appl . math . * 58 * ( 2005 ) evans , l. c. periodic _ homogenisation of certain fully nonlinear partial differential equations . edinburgh sect . a * 120 * ( 1992 ) , no . 3 - 4 , 245265 . d. gilbarg , n. s. trudinger _ elliptic partial dierential equations of second order . _ jikov , v. v. ; kozlov , s. m. ; oleinik , o. a. _ homogenization of differential operators and integral functionals . _ translated from the russian by g. a. yosifian [ g. a. iosif yan ] . springer - verlag , berlin , 1994 . xii+570 pp . isbn : 3 - 540 - 54809 - 2 lions , p .- l . ; souganidis , p. e. _ correctors for the homogenization of hamilton - jacobi equations in the stationary ergodic setting . pure appl . math . * 56 * ( 2003 ) , no . 10 , 15011524 . lieberman , gary m. ; trudinger , neil s. _ nonlinear oblique boundary value problems for nonlinear elliptic equations . _ trans . * 295 * ( 1986 ) , no . 2 , 509546 . nguetseng , gabriel _ a general convergence result for a functional related to the theory of homogenization . _ siam j. math . 20 * ( 1989 ) , no .

we investigate the initial - boundary value problem for linearized gravitational theory in harmonic coordinates . rigorous techniques for hyperbolic systems are applied to establish well - posedness for various reductions of the system into a set of six wave equations . the results are used to formulate computational algorithms for cauchy evolution in a 3-dimensional bounded domain . numerical codes based upon these algorithms are shown to satisfy tests of robust stability for random constraint violating initial data and random boundary data ; and shown to give excellent performance for the evolution of typical physical data . the results are obtained for plane boundaries as well as piecewise cubic spherical boundaries cut out of a cartesian grid .

gr-qc0106026

grid boundaries pose major difficulties in current computational efforts to simulate 3-dimensional black holes by conventional cauchy evolution schemes . the initial - boundary value problem for einstein s equations consists of the evolution of initial cauchy data on a spacelike hypersurface and boundary data on a timelike hypersurface . this problem has only recently received mathematical attention . friedrich and nagy @xcite have given a full solution for a hyperbolic formulation of the einstein equations based upon a frame decomposition in which the connection and curvature enter as evolution variables . because this formulation was chosen to handle mathematical issues rather than for ease of numerical implementation , it is not clear how the results translate into practical input for formulations on which most computational algorithms are based . the proper implementation of boundary conditions depends on the particular reduction of einstein s equations into an evolution system and the choice of gauge conditions . the purpose of this paper is to elucidate , in a simple context , such elementary issues as : ( i ) which variables can be freely specified on the boundary ; ( ii ) how should the remaining variables be updated in a computational scheme ; ( iii ) how can the analytic results be implemented as a computational algorithm . for this purpose , we consider the evolution of the linearized einstein s equations in harmonic coordinates @xcite and demonstrate how a robustly stable and highly accurate computational evolution can be based upon a proper mathematical formulation of the initial - boundary value problem . harmonic coordinates were used to obtain the first hyperbolic formulation of einstein s equations @xcite . @xcite for a full account of hyperbolic formulations of general relativity . while harmonic coordinates have also been widely applied in carrying out analytic perturbation expansions , they have had little application in numerical relativity , presumably because of the restriction in gauge freedom @xcite . however , there has been no ultimate verdict on the suitability of harmonic coordinates for computation . in particular , their generalization to include gauge source functions @xcite appears to offer flexibility comparable to the explicit choice of lapse and shift in conventional numerical approaches to general relativity . there is no question that harmonic coordinates offer greater computational efficiency than any other hyperbolic formulation . @xcite for a recent application to the study of singularities in a space - time without boundary . here we use a reduced version of the harmonic formulation of the field equations . this allows us retain a symmetric hyperbolic system , and apply standard boundary algorithms , in a way that is consistent with the propagation of the constraints . we show , on an analytic level , that this leads to a well posed initial - boundary problem for linearized gravitational theory . our computational results are formulated in terms of a cartesian grid based upon background minkowskian coordinates . robustly stable evolution algorithms are obtained for plane boundaries aligned with the cartesian coordinates , which is the standard setup for three dimensional evolution codes . similar computational results @xcite , based upon long evolutions with random initial and boundary data , were previously found for the linearized version of the arnowitt - deser - misner formulation ( adm ) @xcite of the field equations , where the lack of a hyperbolic formulation required a less systematic approach which had no obvious generalization to other boundary shapes . in this paper , we also attain robust stability for spherical boundaries which are cut out of the cartesian grid in an irregular piecewise cubic fashion . this success gives optimism that the methods can applied to such problems as black hole excision and cauchy - characteristic matching , where spherical boundaries enter in a natural way . conventions : we use greek letters for space - time indices and latin letters for spatial indices , e.g. @xmath0 for standard minkowski coordinates . linear perturbations of a curved space metric @xmath1 about the minkowski metric @xmath2 are described by @xmath3 with similar notation for the corresponding curvature quantities , e.g. the linearized riemann tensor @xmath4 and linearized einstein tensor @xmath5 . indices are raised and lowered with the minkowski metric , with the result that @xmath6 . boundaries in the background geometry are described in the form @xmath7 with the spatial coordinates decomposed in the form @xmath8 , where the @xmath9 directions span the space tangent to the boundary . the linearized einstein tensor has the form @xmath10 where @xmath11 @xmath12 @xmath13 and @xmath14 we analyze these equations in standard background minkowskian coordinates @xmath15 . to linearized accuracy , @xmath16 in terms of the curved space connection @xmath17 associated with @xmath1 and the condition @xmath18 defines a linearized harmonic gauge . the diffeomorphisms of the curved spacetime induce equivalent metric perturbations according to the harmonic subclass of gauge transformations @xmath19 where the vector field @xmath20 satisfies @xmath21 . the linearized curvature tensor @xmath22 as well as the linearized einstein equations , are gauge invariant . we introduce a cauchy foliation @xmath23 in the perturbed spacetime such that it reduces to an inertial time slicing in the background minkowski spacetime . the unit normal to the cauchy hypersurfaces is given , to linearized accuracy , by @xmath24 the choice of an evolution direction @xmath25 , with unit lapse and shift in the background minkowski spacetime , defines a perturbative lapse @xmath26 and perturbative shift @xmath27 . harmonic evolution consists of solving the einstein s equations @xmath28 , subject to the harmonic conditions @xmath18 . this formulation led to the first existence and uniqueness theorems for solutions to the nonlinear einstein equations by considering them as a set of 10 nonlinear wave equations @xcite . in the linearized case , einstein s equations in harmonic coordinates reduce to ten flat space wave equations so that their mathematical analysis is simple . the cauchy data @xmath29 and @xmath30 at @xmath31 determine ten unique solutions @xmath32 of the wave equation @xmath33 , in the appropriate domain of dependence . these solutions satisfy @xmath34 so that they satisfy einstein s equations provided @xmath35 and @xmath36 , which can be arranged by choosing initial cauchy data satisfying constraints . for a detailed discussion , @xcite . although this standard harmonic evolution scheme led to the first existence and uniqueness theorem for einstein s equations , it is not straightforward to apply to the initial - boundary value problem . the ten wave equations for @xmath37 require ten individual pieces of boundary data in order to determine a unique solution . given initial data such that @xmath38 and @xmath39 at @xmath31 , as described above , the resulting solution satisfies the linearized einstein equations only in the domain of dependence of the cauchy data . in order that the solution of einstein s equations extend to the boundary , it is necessary that @xmath38 at the boundary . unfortunately , there is no version of boundary data for @xmath37 , e.g. dirichlet , neumann or sommerfeld data for the ten individual components , from which @xmath40 can be calculated at the boundary . here we consider reduced versions of the harmonic evolution scheme in which only six wave equations are solved and this problem does not arise . these reduced harmonic formulations are presented below . a linearized evolution scheme for the harmonic einstein system , can be based upon the six wave equations @xmath41 along with the four harmonic conditions @xmath42 because of the harmonic conditions , this system satisfies the spatial components @xmath43 of the linearized einstein s equations . as a result of the linearized bianchi identities @xmath44 , or @xmath45 the linearized hamiltonian constraint @xmath46 and linearized momentum constraints @xmath47 are also satisfied , throughout the domain of dependence of the cauchy data , provided that they are satisfied at the initial time @xmath31 . this constrains the initial values of @xmath48 according to @xmath49 and @xmath50 where @xmath51 . then , if these constraints are initially satisfied , the reduced harmonic einstein system determines a solution of the linearized einstein s equations . the well - posedness of the system follows directly from the well - posedness of the wave equations for @xmath52 . the auxiliary variables @xmath53 satisfy the ordinary differential equations @xmath54 where @xmath52 enters only in the role of source terms . these differential equations do not affect the well - posedness of the system and have unique integrals determined by the initial values of @xmath55 . the reduced harmonic ricci system consists of the six wave equations @xmath56 along with the four harmonic conditions ( [ eq : auxeqs ] ) , which can be re - expressed in the form @xmath57 where we have set @xmath58 . together these equations imply that the spatial components of the perturbed ricci tensor vanish , @xmath59 . in addition , the bianchi identities imply that the remaining components satisfy @xmath60 where , in terms of the ricci tensor , @xmath61 and @xmath62 . together with the evolution equations @xmath59 , the bianchi identities imply that the hamiltonian constraint satisfies the wave equation @xmath63 . if the hamiltonian and momentum constraints are satisfied at the initial time , then @xmath64 also vanishes at the initial time so that the uniqueness of the solution of the wave equation ensures the propagation of the hamiltonian constraint . in turn , eq . ( [ eq : momprop ] ) then ensures that the momentum constraint is propagated . thus the reduced harmonic ricci system of six wave equations and four harmonic equations leads to a solution of the linearized einstein s equations for initial cauchy data satisfying the constraints . the harmonic ricci system takes symmetric hyperbolic form when the wave equations are recast in first differential order form . thus the system is well posed . the formulation of the harmonic ricci system as a symmetric hyperbolic system and the description of its characteristics is given in appendix [ app : ricci ] . although well - posedness of the analytic problem does not the guarantee the stability of a numerical implementation it can simplify its attainment . the harmonic einstein and ricci systems are special cases of a one parameter class of reduced harmonic systems for the variable @xmath65 which satisfying the wave equations @xmath66 and the harmonic conditions @xmath38 , where @xmath67 . this system is symmetric hyperbolic when the auxiliary system @xmath38 is symmetric hyperbolic . this can be analyzed by setting @xmath68 in the auxiliary system , which then takes the form @xmath69 and implies @xmath70 the auxiliary system is symmetric hyperbolic when eq . ( [ eq : subwave ] ) is a wave equation whose wave speed @xmath71 is positive . this is satisfied for @xmath72 and @xmath73 . in the range @xmath74 , the wave speed is faster than the speed of light . only for the case @xmath75 , the harmonic ricci system , is the wave speed of the auxiliary system equal to the speed of light . the auxiliary system for the reduced harmonic einstein case has a well posed initial - boundary value problem but represents a borderline case . this could adversely affect the development of a stable code based upon a nonlinear version of the reduced harmonic einstein system . consider the initial - boundary problem consisting of evolving cauchy data prescribed in a set @xmath76 at time @xmath31 lying in the half - space @xmath77 with data prescribed on a set @xmath78 on the boundary @xmath79 . harmonic evolution takes its simplest form when einstein equations are expressed as a second differential order system . however , in order to apply standard methods it is necessary to recast the problem as a first order symmetric hyperbolic system . then the theory determines conditions on the boundary data for a well posed problem in the future domain of dependence @xmath80 . here @xmath80 is the maximal set of points whose past directed inextendible characteristic curves all intersect the union of @xmath76 and @xmath78 before leaving @xmath80 . appendix [ app : ricci ] describes the symmetric hyperbolic description of boundary conditions for the 3-dimensional wave equation . the basic ideas and their application to the initial - boundary value problem are simplest to explain in terms of the 1-dimensional wave equation , as follows . see appendix [ app : ricci ] for the analogous treatment of the 3-dimensional case . our presentation is based upon the formulation of maximally dissipative boundary conditions @xcite , the approach used by friedrich and nagy @xcite in the nonlinear case . an alternative description of boundary conditions for symmetric hyperbolic systems is given in ref . @xcite and for linearized gravity in ref . @xcite . the one - dimensional wave equation @xmath81 can be recast as the first order system of evolution equations @xmath82 where we have introduced the auxiliary variables @xmath83 and @xmath84 . given initial data @xmath85 at time @xmath31 subject to the constraint , @xmath86 , these equations determine a solution @xmath87 of the wave equation . note that the constraint is propagated by the evolution equations . in coordinates @xmath88 the system ( [ eq : swe.first])-([eq : swe.last ] ) has the symmetric hyperbolic form @xmath89 where the solution @xmath90 consists of the column matrix @xmath91 and @xmath92 in our case the source term @xmath93 but otherwise it plays no essential role in the analysis of the system . the contraction of eq . ( [ shscalar ] ) with the transpose @xmath94 give the flux equation @xmath95 this can be used to provide an estimate on the norm @xmath96 for establishing a well posed problem , in the half - space @xmath77 with boundary at @xmath79 , provided the flux arising with the normal component of @xmath97 satisfies the inequality @xmath98 this inequality determines the allowed boundary data for a well posed initial - boundary value problem . as expected from knowledge of the characteristics of the wave equation , the normal matrix @xmath99 has eigenvalues @xmath100 , with the corresponding eigenvectors @xmath101 these eigenvectors are associated with the variables @xmath102 re - expressing the solution vector in terms of the eigenvectors , @xmath103 the inequality ( [ eq : inequality ] implies that homogeneous boundary data must take the form @xmath104 where the parameter @xmath105 satisfies @xmath106 . the component @xmath107 , corresponding to the kernel of @xmath99 , propagates directly up the boundary and can not be prescribed as boundary data . non - homogeneous boundary data can be given in the form @xmath108 where @xmath109 is an arbitrary function representing the free boundary data at @xmath79 . ( [ eq : bdrydata ] ) shows how the scalar wave equation accepts a continuous range of boundary conditions . the well - known cases of sommerfeld , dirichlet or neumann boundary data are recovered by setting @xmath110 , or , @xmath111 , respectively . note that there are consistency conditions at the edge @xmath112 . for instance , dirichlet data corresponds to specifying @xmath113 on the boundary and this must be consistent with the initial data for @xmath83 . the harmonic einstein system consists of the six wave equations ( [ eq : gammawave ] ) and the four harmonic conditions ( [ eq : auxeqs ] ) . since the wave equations for @xmath52 are independent of the auxiliary variables @xmath48 , the well - posedness of the initial - boundary value problem for @xmath52 follows immediately . furthermore , since the harmonic conditions propagate the auxiliary variables up the boundary by ordinary differential equations , the harmonic einstein system has a well posed initial - boundary value problem . a unique solution in the appropriate domain of dependence is determined by the initial cauchy data @xmath52 and @xmath114 at @xmath31 , the initial data @xmath48 at the edge @xmath115 and the boundary data @xmath52 at @xmath79 given in any of the forms described in appendix [ app : ricci ] ( e.g. dirichlet , neumann , sommerfeld ) . a solution of the linearized harmonic einstein system satisfies @xmath116 = 0 . as a result the bianchi identities imply @xmath117 and @xmath118 so that the constraints are satisfied provided the constraint eq s . ( [ eq : auxil1 ] ) and ( [ eq : auxil2 ] ) are satisfied at @xmath31 . the free boundary data for this system consists of six functions . however , as shown in ref . @xcite , the vacuum bianchi identities satisfied by the weyl tensor imply that only two independent pieces of weyl data can be freely specified at the boundary . we give the corresponding analysis for the linearized einstein system in appendix [ app : weyl ] . this makes it clear that only two of the six pieces of metric boundary data are gauge invariant . this is in accord with the four degrees of gauge freedom consisting of the choice of linearized lapse @xmath26 ( one free function ) and linearized shift @xmath119 ( three free functions ) . a linearized evolution requires a unique lapse and shift , whose values can be specified explicitly as space - time functions or specified implicitly in terms of initial and boundary data subject to dynamical equations . in the case of a harmonic gauge , in order to assess whether explicit space - time specification of the lapse and shift is advantageous for the purpose of numerical evolution , it is instructive to see how it affects the initial - boundary value problem for the reduced harmonic einstein system . in the linearized harmonic formulation without boundary , a gauge transformation ( [ eq : gauge ] ) to a shift - free gauge @xmath120 is always possible within the domain of dependence of the initial cauchy data @xcite . in the presence of a boundary , consider a gauge transformation with @xmath121 , so that @xmath122 for harmonic evolution of constrained initial data , both @xmath123 and @xmath124 satisfy the wave equation . at @xmath31 , choose cauchy data for @xmath123 satisfying @xmath125 and @xmath126 on the boundary @xmath79 , require that @xmath123 satisfy @xmath127 then @xmath128 has vanishing cauchy data at @xmath31 , vanishing dirichlet boundary data at @xmath79 and satisfies the wave equation , so that @xmath129 . thus , for evolution of constrained data with a boundary , a shift - free harmonic gauge is possible . however , in this gauge , boundary data for all 6 components of @xmath52 can no longer be freely specified since the harmonic condition implies @xmath130 this relates neumann data for @xmath131 to dirichlet data for @xmath132 and would complicate any shift - free numerical evolution scheme . as an example , one could freely specify dirichlet boundary data for the 3 components @xmath132 and ( i ) obtain neumann boundary data @xmath133 from the @xmath134-components of eq . ( [ eq : noshift ] ) , ( ii ) evolve @xmath135 in terms of initial cauchy data and ( iii ) obtain neumann boundary data @xmath136 from the @xmath137-component of eq . ( [ eq : noshift ] ) . note the nonlocality of step ( ii ) , which would have to be carried out `` on the fly '' during a numerical evolution . the requirement that the shift vanish reduces the free boundary data to three components . it is also possible to eliminate an additional free piece of data by choosing a unit lapse , i.e. setting @xmath138 . suppose the shift has been set to zero , so that the harmonic condition implies @xmath139 . consider a gauge transformation @xmath140 , where @xmath141 satisfies @xmath142 so that the shift remains zero . for harmonic evolution of constrained data , both @xmath143 and @xmath144 satisfy the wave equation . at @xmath31 , choose cauchy data for @xmath143 satisfying @xmath145 and @xmath146 ( where we assume the cauchy data is given on a non - compact set so that there are no global obstructions to a solution @xmath143 ) . on the boundary @xmath79 , require @xmath143 satisfy @xmath147 then @xmath148 because it is a solution of the wave equation with vanishing cauchy data and dirichlet boundary data . ( alternatively , the lapse can be gauged to unity by a transformation satisfying @xmath149 , so that the harmonic source function @xmath150 still drops out of eq . ( [ eq : einstein ] ) for the einstein tensor . ) a unit lapse and zero shift implies that @xmath151 so that @xmath152 can not be freely specified at the boundary . coupled with our previous results , imposition of a unit lapse and zero - shift reduces the free boundary data to the two trace - free transverse components @xmath153 , in accord with the two degrees of gauge - free radiative freedom associated with the weyl tensor . ( in addition , the initial cauchy data must satisfy eq . ( [ eq : noshift ] ) , @xmath151 and @xmath154 ) . a similar result arises in the study of the unit - lapse zero - shift initial - boundary value problem for the linearized adm equations @xcite . however , in the case of harmonic evolution , it is clear that explicit specification of the lapse and shift leads to a more complicated initial - boundary value problem . it is more natural to retain the freedom of specifying 6 pieces of boundary data , which then determine the lapse and shift implicitly during the course of the evolution . the reduction of the free boundary data can be accomplished by other gauge conditions on the boundary , which are not directly based upon the lapse and shift . an example , which plays a central role in the friedrich - nagy formulation , is the specification of the mean curvature of the boundary . to linearized accuracy , the unit outward normal to the boundary at @xmath79 is @xmath155 the associated mean extrinsic curvature @xmath156 is given to linear order by @xmath157 although the extrinsic curvature of a planar boundary vanishes in the background minkowski space , the linear perturbation of the background induces a non - vanishing linearized extrinsic curvature tensor . under a gauge transformation induced by the @xmath20 , the mean curvature transforms according to @xmath158 a gauge deformation of the boundary in the embedding space - time makes it possible to obtain any mean curvature by solving a wave equation intrinsic to the boundary . in this respect , the mean extrinsic curvature of the boundary is pure `` boundary gauge '' and can be specified to eliminate one degree of gauge freedom . when the harmonic conditions are satisfied , the mean curvature of the boundary reduces to @xmath159 . this discussion shows that there are various ways that the six free pieces of boundary data can be restricted by gauge conditions . such restrictions can be important for an analytic understanding of the initial - boundary problem but their usefulness for numerical simulation is a separate issue , especially in applications where the boundary does not align with the numerical grid as discussed in sec . [ sec : implem ] . the underlying equations of the reduced harmonic ricci system are the six wave equations ( [ eq : hwave ] ) and the four harmonic conditions ( [ eq : gauge_t ] ) and ( [ eq : gauge_i ] ) . the symmetric hyperbolic formulation of this system and the analysis of its characteristics is given in appendix [ app : ricci ] . the variables consist of @xmath160 where @xmath161 , @xmath162 , @xmath163 and @xmath164 . in a well posed initial - boundary value problem , there are seven free functions that may be specified at the boundary . for example , in the analogue of the dirichlet case , the free boundary data consists of @xmath165 and @xmath166 . in the context of the second order differential form given by eq s . ( [ eq : hwave ] ) , with the harmonic conditions ( [ eq : gauge_t ] ) and ( [ eq : gauge_i ] ) , the boundary data remains the same , e.g. dirichlet data @xmath167 and @xmath166 . this determines the evolution of @xmath168 by the wave equation , which then provides source terms for the symmetric hyperbolic subsystem ( [ eq : gauge_t ] ) and ( [ eq : gauge_i ] ) . this subsystem implies that @xmath169 , so that the evolution of @xmath166 is also governed by the wave equation , with initial cauchy data @xmath166 and @xmath170 where @xmath170 is provided by the initial cauchy data @xmath171 via eq . ( [ eq : gauge_i ] ) ) . the evolution of @xmath171 is then obtained by integration of eq . ( [ eq : gauge_i ] ) . unlike the reduced harmonic einstein system where the constraints propagate up the boundary by ordinary differential equations , the initial - boundary value problem for the reduced harmonic ricci system does not necessarily satisfy einstein s equations even if the constraints are initially satisfied . the bianchi identities ( [ eq : momprop ] ) imply ( [ eq : hamprop ] ) so that both @xmath172 and @xmath173 would initially vanish but , since @xmath172 satisfies the wave equation , @xmath172 would vanish throughout the evolution domain only if it vanished on the boundary . in that case , eq . ( [ eq : momprop ] ) would imply that the momentum constraints were also satisfied throughout the evolution domain . thus evolution of constrained initial data for the harmonic ricci system yields a solution of the einstein equations if and only if the hamiltonian constraint is satisfied on the boundary . this is equivalent to requiring that @xmath169 on the boundary . if the evolution equations are satisfied , we can express this in the form @xmath174 where @xmath175 and @xmath176 . this allows formulation of the following well posed initial - boundary value problem for a solution satisfying the constraints . we prescribe initial cauchy data that satisfies the constraints for @xmath177 , @xmath105 , @xmath166 and @xmath171 , and free boundary data for @xmath177 and @xmath166 . the system of wave equations then determines @xmath177 . this allows integration of eq . ( [ eq : boxphi ] ) on the boundary to obtain dirichlet boundary values for determination of @xmath105 as a solution of the wave equation . the remaining fields @xmath166 and @xmath171 are then determined as a symmetric hyperbolic subsystem . note that the boundary constraint ( [ eq : boxphi ] ) reduces the free boundary data from seven independent functions ( for unconstrained solutions ) to six , in agreement with the free boundary data for solutions of the reduced harmonic einstein system . numerical error is an essential new factor in the computational implementation of the preceding analytic results . the initial cauchy data can not be expected to obey the constraints exactly . in particular , machine roundoff error always produces an essentially random component to the data which , in a linear system , evolves independently of the intended physical simulation . it is of practical importance that a numerical evolution handle such random data without producing exponential growth ( and without an inordinate amount of numerical damping ) . we designate as _ robustly stable _ an evolution code for the linearized initial - boundary problem which does not exhibit exponential growth for random ( constraint violating ) initial data and random boundary data . this is the criterion previously used to establish robust stability for adm evolution with specified lapse and shift @xcite . we test for robust stability using the 3-stage methodology proposed for evolution - boundary codes in ref . the tests check that the @xmath178 norm of the hamiltonian constraint @xmath179 does not exhibit exponential growth under the following conditions . * _ stage i : _ evolution on a 3-torus with random initial cauchy data . * _ stage ii : _ evolution on a 2-torus with plane boundaries , i.e. @xmath180 $ ] , with random initial cauchy data and random boundary data . * _ stage iii : _ evolution with a cubic boundary with random initial cauchy data and random boundary data . stage i tests robust stability of the evolution code in the absence of a boundary . stage ii tests robust stability of the boundary - evolution code for a smooth boundary ( topology @xmath181 ) . stage iii tests robust stability of the boundary - evolution code for a cubic boundary with faces , edges and corners , as standard practice for computational grids based upon cartesian coordinates . we have established robust stability for the evolution - boundary codes , described in sec s . [ sec : hec ] and [ sec : hrc ] , the reduced harmonic einstein and ricci systems . the tests are performed according to the procedures outlined in ref . @xcite for an evolution of 2000 crossing times ( a time of @xmath182 , where @xmath183 is the linear size of the computational domain ) on a uniform @xmath184 spatial grid with a time step @xmath185 ( which is slightly less than half the courant - friedrichs - lewy limit and typical of values used in numerical relativity ) . for purposes such as singularity excision or cauchy - characteristic matching , there are computational strategies based upon spherical boundaries . the extension of a robustly stable adm boundary - evolution algorithm with given lapse and shift from a cubic grid boundary to a spherical boundary is problematic @xcite . robustly stable evolution - boundary algorithms for the reduced harmonic ricci and einstein systems with spherical boundaries are presented in sec . [ sec : sphb ] . numerical evolution of the fields @xmath168 ( or @xmath52 ) and @xmath144 ( or @xmath186 ) is implemented on a uniform spatial grid @xmath187 } , y_{[j ] } , z_{[k ] } ) = ( i \ , \delta x , \ , j \ , \delta x , \ , k \ , \delta x)$ ] , with time levels @xmath188 } = n \ , \delta t$ ] . thus a field component @xmath84 is represented by its values @xmath189}_{[i , j , k ] } = f(t^{[n ] } , x_{[i ] } , y_{[j ] } , z_{[k]})$ ] . in order to obtain compact finite difference stencils for imposing boundary conditions , the fields @xmath171 ( or @xmath190 ) are represented on staggered grids at staggered time - levels , where @xmath191}_{[i+1/2 , j+1/2 , k+1/2 ] } = f(t^{[n+1/2 ] } , x_{[i+1/2 ] } , y_{[j+1/2 ] } , z_{[k+1/2]})$ ] . the extensive tests of stability reported here were performed using a leapfrog evolution algorithm in order not to bias the tests by introducing excessive dissipation . we expect that the tests would also be satisfied by more dissipative algorithms . this was borne out by a limited number of evolutions using an iterative crank - nicholson algorithm ( implemented as in ref . @xcite ) . * stage i * evolution of @xmath52 is done by a standard three - level leapfrog scheme , with the wave equation in second - differential - order form , i.e. @xmath192}_{[i , j , k ] } - 2 \ , \gamma^{ij[n ] } _ { [ i , j , k ] } + \gamma^{ij[n-1]}_{[i , j , k ] } } { ( \delta t)^2 } & = & \gamma^{ij[n]}_{,xx[i , j , k ] } + \gamma^{ij[n]}_{,yy[i , j , k ] } + \gamma^{ij[n]}_{,zz[i , j , k ] } \ , .\end{aligned}\ ] ] second spatial derivatives are computed via @xmath193 } = \frac { f_{[i+1,j , k ] } - 2\ , f_{[i , j , k ] } + f_{[i-1,j , k]}}{(\delta x)^2 } \ , .\ ] ] on a grid staggered in space and at staggered time levels , evolution of @xmath190 is carried out by the finite difference version of eq . ( [ eq : einstaux1 ] ) , i.e. @xmath194}_{[i+1/2,j+1/2,k+1/2 ] } - \gamma^{it[n-1/2]}_{[i+1/2,j+1/2,k+1/2]}}{\delta t } & = & \\ - \gamma^{ix[n]}_{,x[i+1/2,j+1/2,k+1/2 ] } - \gamma^{iy[n]}_{,y[i+1/2,j+1/2,k+1/2 ] } & -&\gamma^{iz[n]}_{,z[i+1/2,j+1/2,k+1/2]}. \nonumber\end{aligned}\ ] ] here first spatial derivatives are evaluated at the center of the integer grid cells , i.e. @xmath195 } = \frac{1}{4 } \big [ \frac { f_{[i+1,j , k ] } - f_{[i , j , k ] } } { \delta x } & + & \frac { f_{[i+1,j+1,k ] } - f_{[i , j+1,k ] } } { \delta x } \\ + \frac { f_{[i+1,j , k+1 ] } - f_{[i , j , k+1]}}{\delta x } & + & \frac { f_{[i+1,j+1,k+1 ] } - f_{[i , j+1,k+1]}}{\delta x } \big ] \ , . \nonumber\end{aligned}\ ] ] since @xmath186 is represented on the integer grid which is staggered in space and time with respect to the grid representation of @xmath190 , the finite difference equation used for updating @xmath186 is similar to eq . ( [ eq : fde.gamma^it ] ) used to update @xmath190 . * stage ii * the hierarchy of ten linearized equations for the reduced harmonic einstein system makes the initial - boundary value problem particularly simple to implement as a finite difference algorithm . let the plane boundary be given by the @xmath196-th grid point , with the stage i evolution algorithm applied to all points with @xmath197 . update of @xmath198}$ ] requires @xmath199}$ ] as boundary data . the same boundary information allows update of @xmath200}$ ] , which , in turn , allows update of @xmath201}$ ] with no additional boundary data . however , it is interesting to note that specification of free boundary data for @xmath202}$ ] or @xmath203}$ ] would not affect stability simply because no evolution equation uses that data . * stage iii * let the evolution domain be the cube @xmath204 with the grid structured so that the boundary lies on ( non - staggered ) grid points , i.e. , @xmath205 } = \pm l$ ] , etc . the boundary data consist of @xmath52 on the faces , edges and corners of the cube . the field @xmath190 can then be updated at all staggered grid points inside the cube , including those neighboring the boundary . for instance , update of @xmath206}$ ] involves use of @xmath52 at the points @xmath207}$ ] , all of which are on or inside the boundary . similarly , evolution of @xmath186 can be carried out at all interior grid points without further boundary data . robust stability of the evolution - boundary algorithm is demonstrated by the graph of the hamiltonian constraint in fig . [ fig : stage.iii ] . the linear growth results from momentum constraint violation in the initial data . * stage i * evolution of @xmath168 is carried out identically as the evolution of @xmath52 in the einstein system ( see eq . ( [ eq : fde.gamma^ij ] ) ) . the fields @xmath168 and @xmath166 are represented on the integer grid while @xmath171 is represented on a half - integer grid staggered in space and in time . thus the evolution equations ( [ eq : gauge_t ] ) and ( [ eq : gauge_i ] ) for @xmath166 and @xmath171 have finite difference form @xmath208}_{[i , j , k ] } - \phi^{[n]}_{[i , j , k]}}{\delta t } + \left ( \partial_i h^{it } \right)^{[n+1/2]}_{[i , j , k ] } + \frac{1}{2 } \delta_{ij } \frac{h^{ij[n+1]}_{[i , j , k ] } - h^{ij[n]}_{[i , j , k]}}{\delta t } & = & 0 \\ \label{eq : fde.h^it } \frac { h^{it[n+1/2]}_{[i+1/2,j+1/2,k+1/2 ] } - h^{it[n-1/2]}_{[i+1/2,j+1/2,k+1/2 ] } } { \delta t } + \left ( \partial_i h^{it } - \frac{1}{2 } \delta_{jk } \partial^i h^{jk } + \partial_j h^{ij } \right)^{[n]}_{[i+1/2,j+1/2,k+1/2 ] } & = & 0 , \end{aligned}\ ] ] where the spatial derivative terms are computed according to eq . ( [ eq : fde.f_x ] ) . * stage ii * _ unconstrained plane boundary_. as shown in sec . [ sec : hrs ] , seven free functions can be prescribed as unconstrained boundary data for @xmath209 and @xmath166 in the reduced harmonic ricci system . again let the boundary be defined by the @xmath196-th grid point , with @xmath197 for interior points . then the boundary data @xmath210}$ ] and @xmath211}$ ] allows the evolution algorithm to be applied to update @xmath168 and @xmath171 at all interior points , e.g. to update @xmath212}$ ] and @xmath213}$ ] , which in turn allows update of @xmath166 at all interior points . _ constrained plane boundary_. conservation of the constraints in the reduced harmonic ricci system requires that the hamiltonian constraint be enforced at the boundary . in order to obtain a finite difference approximation to a solution of einstein s equations , the unconstrained evolution - boundary algorithm must be modified to enforce the hamiltonian constraint on the boundary . on a @xmath214const boundary , we accomplish this , in accord with the discussion in sec . [ sec : hrs ] , by prescribing freely the function @xmath166 and the five components of the traceless symmetric tensor @xmath215 . the missing ingredient , @xmath216 is updated at the boundary according to eq . ( [ eq : boxphi ] ) . in the finite difference algorithm , in order to be able to apply eq . ( [ eq : boxphi ] ) via a centered three - point stencil , we introduce a guard point at @xmath217}$ ] , where the boundary data @xmath215 is provided . ( see fig . [ fig : stencil.ii ] . ) assuming that the fields @xmath218 and @xmath166 are known for @xmath219}$ ] , @xmath220 and that the boundary data @xmath221}$ ] is known for @xmath219}$ ] , we use the following evolution - boundary algorithm to compute these fields at @xmath222}$ ] : * \(i ) we update the fields @xmath168 at time - level @xmath222}$ ] at all grid points within the numerical domain of dependence of the data known at @xmath188}$ ] , i.e. , at all points which require no boundary data . * \(ii ) at the guard point @xmath223 we assign dirichlet boundary data to the five independent components of the symmetric traceless @xmath224}_{[k_0 + 1]}$ ] by prescribing boundary values for @xmath225 and setting @xmath226 . * \(iii ) at the guard point @xmath223 we update the fields @xmath227}_{[k_0 + 1 ] } = h^{ij[n]}_{[k_0 + 1 ] } + \left(\delta t\right ) \ , \left(\partial_t h^{ij}\right)^{[n+1/2]}_{[k_0 + 1]}$ ] . * \(iv ) at the boundary point @xmath196 we update @xmath227}_{[k_0]}$ ] using the field - equation @xmath228 , written in the finite - difference form @xmath229}_{[k_0 ] } - 2 \ , h^{ij[n]}_{[k_0 ] } + h^{ij[n-1]}_{[k_0 ] } } { \delta t^2 } = \left(\nabla^2 h^{ij}\right)^{[n]}_{[k_0 ] } .\ ] ] * \(v ) at the boundary point @xmath196 we compute the boundary values @xmath230}_{[k_0]}$ ] using the finite - difference version of eq . ( [ eq : boxphi ] ) , i.e. @xmath231}_{[k_0 ] } - 2 \ , h^{[n]}_{[k_0 ] } + h^{[n-1]}_{[k_0 ] } } { \delta t^2 } -\left(\partial_i \partial_j h^{ij}\right)^{[n]}_{[k_0]}=0 .\ ] ] in eq s . ( [ eq : fde.box_hij ] ) - ( [ eq : fde.box_phi ] ) all derivatives are computed using centered three - point expressions . * \(vi ) from knowledge of @xmath227}_{[k_0]}$ ] and @xmath230}_{[k_0]}$ ] we construct @xmath232}_{[k_0]}$ ] . * \(vii ) we assign boundary data for @xmath233}_{[k_0]}$ ] and update @xmath233}$ ] and @xmath234}$ ] according to eq s . ( [ eq : fde.phi ] ) - ( [ eq : fde.h^it ] ) . * stage iii * _ unconstrained cubic boundary_. the unconstrained cubic boundary is essentially the same as the unconstrained plane boundary , i.e. , the functions @xmath209 and @xmath166 are provided at the boundary . then the evolution equations are applied to update @xmath168 , @xmath171 and @xmath166 at all interior grid points . recall that @xmath171 is represented on a staggered grid with some interior points located a half grid - step away from the boundary . nevertheless , @xmath171 can be updated at such points using eq . ( [ eq : fde.h^it ] ) . _ constrained cubic boundary_. the algorithm for enforcing the hamiltonian constraint on a cubic boundary is an extension of the algorithm for a plane boundary . the boundary functions @xmath215 are provided at a set of guard points that are @xmath235 outside the boundary of the cube , such that at grid - points on the boundary of the cube we can use the field equations @xmath236 as well as the boundary constraint eq . ( [ eq : boxphi ] ) to update @xmath177 and @xmath105 at the boundary points . boundary data for @xmath166 is provided at grid points on the boundary of the cube , in a similar fashion to the constrained plane boundary . robust stability of the evolution - boundary algorithm for the constrained cubic boundary is demonstrated by the graph of the hamiltonian constraint in fig . [ fig : stage.iii ] . in addition to the robust stability test results , in fig . [ fig : stage.iii ] we have also included results from two physical runs , based on the plane - wave solution eq s . ( [ eq : ctest.plwave - first ] ) - ( [ eq : ctest.plwave - last ] ) . the longer physical run ( up to @xmath237 ) was performed with a grid - size of @xmath238 , while the shorter run ( up to @xmath239 ) was performed with a grid - size of @xmath240 . comparison of the physical runs with different gridsizes indicates that the long - term polynomial growth in the hamiltonian constraint violation can be controlled by grid - resolution . the implementation of both the reduced harmonic einstein and ricci systems display robust stability in stage i , ii and iii tests . we now extend these results to a a spherical boundary cut out of a cartesian grid , with the evolution domain defined by the interior of a sphere of radius @xmath241 . in order to update fields at these interior grid points , values are needed at a set of `` guard '' points consisting of grid points on or outside the boundary . some of these field values constitute free boundary data . we extend our test of robust stability to a fourth stage which checks that the @xmath178 norm of the hamiltonian constraint @xmath172 does not exhibit exponential growth under the following condition . * _ stage iv : _ evolution with a spherical boundary with random initial cauchy data and random free boundary data at all guard points . as before , we perform the tests for an evolution of 2000 crossing times ( @xmath242 , where @xmath243 is the diameter of the computational domain ) on a uniform @xmath184 spatial grid with a time step slightly less than half the courant - friedrichs - lewy limit . we have established stage iv robust stability for the following implementations of reduced harmonic einstein and ricci revolution . let the evolution domain be defined by the interior of a sphere of radius @xmath241 . the update of @xmath186 at an interior point @xmath244 $ ] requires the values of @xmath190 at the points @xmath245}$ ] , some of which might be `` staggered - boundary '' points outside the evolution domain . these staggered - boundary points are inside the spherical shell @xmath246 , if @xmath247 . at these points we update @xmath190 by the same finite - difference equation ( [ eq : fde.gamma^it ] ) as for the interior points . this defines a set of non - staggered boundary points for the field @xmath52 determined by two conditions : ( i ) that the @xmath52 can be updated at all interior points using eq . ( [ eq : fde.gamma^ij ] ) and ( ii ) that @xmath190 can be updated at all interior and staggered - boundary points . both conditions are satisfied if we define the set of non - staggered boundary points by @xmath248}^2+y_{[j]}^2+z_{[k]}^2 } < r + \delta r_0 $ ] , where @xmath247 . at this set of points we obtain values for @xmath52 via eq . ( [ eq : fde.gamma^ij ] ) , with @xmath249 provided in a `` guard - shell '' @xmath250 , where @xmath251 . the radius @xmath241 of the spherical boundary of the reduced harmonic einstein system is related to the linear size @xmath252 of the computational grid by @xmath253 this ensures that all guard points fall inside the domain @xmath254 ^ 3 $ ] . _ unconstrained spherical boundary_. the stage iv unconstrained evolution - boundary algorithm for the reduced harmonic ricci system is similar to that of the reduced harmonic einstein system . we use two spherical shells , of thickness @xmath255 and @xmath256 , with @xmath257 and @xmath258 . the algorithm is the following : * \(i ) we update all fields @xmath168 at time - level @xmath222}$ ] at all non - staggered grid points within the sphere of radius @xmath259 . * \(ii ) we provide boundary data @xmath260}$ ] at all non - staggered grid points within the spherical shell @xmath250 . we update @xmath232 } = h^{ij[n ] } + \delta t \ , \left(\partial_t h\right)^{ij[n+1/2]}$ ] within the same spherical shell . * \(iii ) we provide boundary data @xmath233}$ ] within the spherical shell @xmath261 . * \(iv ) we update the field @xmath233}$ ] inside the sphere of radius @xmath241 according to eq . ( [ eq : gauge_t ] ) , and update the fields @xmath234}$ ] inside the sphere of radius @xmath262 according to eq . ( [ eq : gauge_i ] ) . _ constrained spherical boundary_. the evolution - boundary algorithm for the reduced harmonic ricci system with constrained spherical boundary is an extension of the algorithm with unconstrained boundary . in addition to the spherical shells defined in the case of the unconstrained boundary , we define an additional set of guard points @xmath263}^2+y_{[j]}^2+z_{[k]}^2 } < r + \delta r_0 + \delta r_1 + \delta r_2 $ ] , where the boundary data @xmath215 is provided . the quantity @xmath264 is defined by two conditions : ( i ) the fields @xmath177 can be updated according to eq . ( [ eq : fde.box_hij ] ) at all non - staggered grid points @xmath265}^2+y_{[j]}^2+z_{[k]}^2 } < r + \delta r_0 + \delta r_1 $ ] , and ( ii ) the hamiltonian constraint eq . ( [ eq : boxphi ] ) can be enforced at the same set of grid points . both of these conditions are satisfied if @xmath266 . the evolution - boundary algorithm for the reduced harmonic ricci system with a constrained spherical boundary is the following : * \(i ) we update all fields @xmath168 at time - level @xmath222}$ ] at all non - staggered grid points within the sphere or radius @xmath259 . * \(ii ) we update the fields @xmath227}$ ] at the set of boundary points @xmath267}^2+y_{[j]}^2+z_{[k]}^2 } < r + \delta r_0 + \delta r_1 $ ] via eq . ( [ eq : fde.box_hij ] ) . then we update the field @xmath230}$ ] at these points via eq . ( [ eq : fde.box_phi ] ) . from knowledge of @xmath227}$ ] and @xmath230}$ ] we construct @xmath232}$ ] at the same set of grid points . * \(iii ) we provide boundary data @xmath224}$ ] at the set of guard points @xmath263}^2+y_{[j]}^2+z_{[k]}^2 } < r + \delta r_0 + \delta r_1 + \delta r_2 $ ] , then we update @xmath227 } = h^{ij[n ] } + \delta t \ , \left(\partial_t h^{ij } \right)^{[n+1/2]}$ ] at the same set of grid points . * \(iv ) we assign boundary data for @xmath233}$ ] at the set of boundary points @xmath248}^2+y_{[j]}^2+z_{[k]}^2 } < r + \delta r_0 $ ] . * \(v ) we update @xmath233}$ ] and @xmath234}$ ] according to eq s . ( [ eq : gauge_t ] ) - ( [ eq : gauge_i ] ) . the radius @xmath241 of the constrained spherical boundary for the reduced harmonic ricci system and the linear size @xmath268 of the computational grid are related by @xmath269 this ensures that all boundary points and all guard points fall inside the domain @xmath254 ^ 3 $ ] . the graphs of the hamiltonian constraint in fig . [ fig : stage.iv ] illustrate robust stability for a spherical boundary for the reduced harmonic einstein system and for the reduced harmonic ricci system with constrained boundary data . comparison of figs . [ fig : stage.iii ] - [ fig : stage.iv ] shows that there is no significant difference between the stage iii and stage iv performances in terms of numerical stability . in order to calibrate the performance of the algorithms we carried out convergence tests based upon analytic solutions constructed from a superpotential @xmath270 symmetric in @xmath271 and @xmath272 and antisymmetric in @xmath273 $ ] , @xmath274 $ ] , such that @xmath275 . as a result of these symmetry properties , the tensor @xmath276 is symmetric and satisfies the linearized harmonic einstein equations @xmath277 and @xmath278 . in our first testbed we choose @xmath270 as a superposition of two solutions , @xmath279 with the remaining independent components of @xmath270 set to zero . the solution @xmath280 is defined by @xmath281}{\omega_a^2 } , \quad i \neq j\end{aligned}\ ] ] and @xmath282 is defined by @xmath283 } { \omega_b^2 } , \quad b^{ij } = b^{tt } = 0 . \label{eq : ctest.plwave - last}\end{aligned}\ ] ] here @xmath284 is a plane wave propagating with frequency @xmath285 along the diagonal of the@xmath286 plane , so that a wave crest leaving @xmath287 travels a distance @xmath288 before arriving at @xmath289 . since the topology of stages i and ii imply periodicity in the @xmath290 direction , we set @xmath291 similarly , the frequency @xmath292 of the functions @xmath293 is set to @xmath294 in stage iii we use the same choices , while in the stage iv tests we set @xmath295 the amplitudes @xmath296 were chosen to be @xmath297 convergence runs used the plane wave solution . in stages i - iii we used the grid sizes @xmath298 while in stage iv we used @xmath299 with the additional gridsize @xmath300 in the case of the the reduced harmonic ricci system with constrained spherical boundary . the time - step was set to @xmath301 . in the stage iv test of the reduced harmonic einstein system the widths of the boundary shells were chosen to be @xmath302 and @xmath303 . the same parameters were used when testing the algorithm for the reduced harmonic ricci system with unconstrained spherical boundary . the evolution - boundary algorithm for the reduced harmonic ricci system with constrained spherical boundary was tested using the parameters @xmath304 , and @xmath305 . the code was used to evolve the solutions from @xmath31 to @xmath306 ( @xmath307 in stage iv ) , at which time convergence was tested by measuring the @xmath178 and the @xmath308 norms of @xmath309 for the einstein system and of @xmath310 for the ricci system , which test convergence of the hamiltonian constraint . the norms were evaluated in the entire evolution domain . in addition , we also checked convergence of the metric components to their analytic values . in addition to plane wave tests , we tested the qualitative performance in stage iv using an offset spherical wave based upon the superpotential ( with shifted origin ) @xmath311 where @xmath312 and @xmath313 .\ ] ] the parameters @xmath314 , and @xmath315 are set to @xmath316 evolution requires cauchy data at @xmath31 and boundary data at the guard points . the cauchy data @xmath317 was provided by giving @xmath318 } , \gamma^{ij[-1 ] } , \gamma^{tt[0]}$ ] and @xmath319}$ ] at all interior and guard points . in addition , we provided boundary data at each time - step by giving @xmath320}$ ] at all guard points . the metric and hamiltonian constraint were found to be 2nd order convergent for stages i - iv . in particular , in stage iv , the norm of @xmath309 vanished to @xmath321 . in the case of the ricci system the cauchy data @xmath322 was provided by giving @xmath323 } , h^{ij[-1 ] } , \phi^{[0]}$ ] and @xmath324}$ ] at all interior and boundary points . in addition , when the hamiltonian constraint was numerically imposed at the boundary , we also provided @xmath325 by giving @xmath326}$ ] . we first tested the code without numerically imposing the hamiltonian constraint . in this case we provided boundary data at each time - step by giving @xmath327}$ ] and @xmath328}$ ] at all guard points . next we tested the code with the hamiltonian constraint numerically imposed at the boundary . thus we first prescribed the traceless @xmath329}$ ] and @xmath328}$ ] at all guard points , then computed @xmath105 at each time - step via the boundary constraint eq . ( [ eq : boxphi ] ) . in all cases we found the numerically evolved metric functions converge to their analytic values to @xmath330 . in stage i , @xmath310 vanished to roundoff accuracy , while in stages ii - iii it vanished to second order accuracy . in particular , for the stage iii algorithm with constrained boundary , @xmath331 converged to zero as @xmath321 . in stage iv , with constrained boundary , we found that the @xmath308 norm of @xmath332 vanishes to first order accuracy . however , the @xmath178 norm decreases linearly with grid size only for @xmath333 and @xmath334 but fails to show further decrease for @xmath335 and @xmath336 . this anomalous behavior of the @xmath178 norm stems from the random way in which guard points are required at different sites near the boundary . this introduces an unavoidable nonsmoothness to the second order error in the metric components , which in turn leads to @xmath337 error in the second spatial derivatives occurring in @xmath310 or in the hamiltonian constraint . unlike the einstein system in which the constraints propagate tangent to the boundary , this error in the ricci system propagates along the light cone into the interior however , since its origin is a thin boundary shell whose width is @xmath338 , the @xmath308 norm of @xmath310 remains convergent to first order . we expect that the convergence of the hamiltonian constraint for a spherical boundary would be improved by matching the interior solution on the cartesian grid to an exterior solution on a spherical grid aligned with the boundary , as is standard practice in treating irregular shaped boundaries . we also tested the code s ability to evolve an outgoing spherical wave traveling off center with respect to a spherical boundary of radius @xmath241 . [ fig:2dplots ] illustrates a simulation performed using the stage iv algorithm for the reduced harmonic ricci system , with the hamiltonian constraint numerically enforced at the boundary . the metric fields were evolved from @xmath31 to @xmath339 , using a grid of @xmath340 . after the analytic wave has propagated out of the computational domain , the remnant error is two orders of magnitude smaller than the initial signal . this shows that artificial reflection off the boundary is well controlled even in the computationally challenging case of a piecewise cubic spherical boundary . we thank helmut friedrich for numerous discussions of the initial - boundary value problem . our numerical code was based upon an earlier collaboration with roberto gmez . this work has been partially supported by nsf grant phy 9988663 to the university of pittsburgh . computer time was provided by the pittsburgh supercomputing center and by npaci . in order to study the evolution of the system consisting of eq s . ( [ eq : hwave ] ) , ( [ eq : gauge_t ] ) , and ( [ eq : gauge_i ] ) in the half - space @xmath77 with boundary at @xmath79 , we employ the auxiliary variables @xmath341 . in terms of the variables @xmath342 the system takes the form @xmath343 next we define the 34-dimensional vector @xmath90 by @xmath344 the system of equations ( [ eq : ev_system_first ] ) - ( [ eq : ev_system_last ] ) then has the form @xmath345 where @xmath346 is the identity matrix , @xmath347 and @xmath348 where @xmath349 the matrix @xmath350 has the eigenvalue @xmath351 , with multiplicity @xmath352 and eigenvectors @xmath353 the eigenvalue @xmath111 , with multiplicity @xmath352 and eigenvectors @xmath354 and the kernel of the matrix has dimension @xmath355 , with a basis @xmath356 in the eigen - basis defined by @xmath350 the vector @xmath90 defined in eq . ( [ eq : udef1 ] ) takes the form @xmath357 with @xmath358 non - homogeneous boundary data can be given in terms of a free column vector field @xmath109 in the form @xmath359 where @xmath105 can be any @xmath360 matrix satisfying @xmath361 the three simplest matrices that satisfy the condition ( [ eq : hcond ] ) are @xmath362 and @xmath363 . the first of these corresponds to specifying neumann data for @xmath168 and dirichlet data for @xmath364 . using the zero matrix as a candidate for @xmath105 corresponds to giving sommerfeld data for @xmath168 and specifying the quantity @xmath365 . last , picking @xmath105 to be the identity matrix corresponds to giving dirichlet data for @xmath167 as well as for @xmath166 . note that the evolution system ( [ eq : ev_system_first ] ) - ( [ eq : ev_system_last ] ) accepts a much richer class of boundary conditions than the three we just mentioned . one simply needs to pick a matrix @xmath105 that satisfies eq . ( [ eq : hcond ] ) and the choice of @xmath105 defines the seven free functions that are to be specified at the boundary . the curvature tensor , which provides gauge invariant fields , decomposes into the ricci curvature , which vanishes if the evolution and constraint equations are satisfied , and the weyl curvature @xmath366 . in order to analyze the boundary freedom , it is convenient make the following choice of a complete , independent set of 10 linearized weyl tensor components @xmath367 : @xmath368 , @xmath369 , @xmath370 , @xmath371 , @xmath372 , and @xmath373 . we use the linearized vacuum bianchi identities @xmath374\mu\nu}=0 $ ] and @xmath375 to show that the weyl data which can be freely specified on the boundary can be reduced to the 2 independent components @xmath376 . first , the identity @xmath377 implies ( after using the trace - free property of the weyl tensor ) @xmath378 which determines the boundary behavior of @xmath379 in terms of the remaining 9 weyl components . next , note that the identity @xmath380 implies @xmath381 or , taking a @xmath23-derivative and using eq . ( [ eq : trac ] ) , that @xmath382 this gives a propagation equation intrinsic to the boundary which determines the time dependence of @xmath383 in terms of the boundary data for @xmath376 . ( note that @xmath383 propagates up the boundary with velocity @xmath384 in one mode and in a cone with velocity @xmath385 in the other mode . ) next , the identity @xmath386 determines the time dependence of @xmath387 ; and the identity @xmath388 determines the time dependence of @xmath371 . this reduces the free weyl data on the boundary to the 4 independent components @xmath376 and @xmath389 . however , the specification of @xmath389 , in addition to @xmath376 , would lead to an inconsistent boundary value problem . this can be seen from the identity @xmath390 which determines neumann data for @xmath389 in terms of dirichlet data for @xmath376 and other known quantities . similarly , the identity @xmath391 determines neumann data for @xmath376 . thus , since the components of the weyl tensor satisfy the wave equation , the specification of both @xmath376 and @xmath389 as free dirichlet boundary data leads to an inconsistent boundary value problem . the determination of boundary boundary values for @xmath389 from boundary data for @xmath376 is a global problem which first requires solving the wave equation to determine @xmath376 from its boundary and initial data . then the time derivative of the trace - free part of eq . ( [ eq : bianch ] ) yields @xmath392 which propagates @xmath389 up the boundary in terms of initial data . defining @xmath393 and @xmath394 , with @xmath395 , this reduces to @xmath396 which has propagation velocity @xmath385 . ( note that this is but one of the variations consistent with the maximally dissipative condition used by friedrich and nagy @xcite . in the case of unit lapse and vanishing shift , assigning boundary data for @xmath172 is equivalent to assigning data for the trace - free part of the intrinsic 2-metric of the boundary foliation , consistent with results found in ref .

this paper investigates simultaneous wireless information and power transfer ( swipt ) in energy harvesting ( eh ) relay systems . unlike existing swipt schemes requiring the channel state information ( csi ) for _ coherent _ information delivery , we propose a _ noncoherent _ swipt framework for decode - and - forward ( df ) relay systems bypassing the need for csi and consequently saving energy in the network . the proposed swipt framework embraces _ power - splitting noncoherent df _ ( ps - ncdf ) and _ time - switching noncoherent df _ ( ts - ncdf ) in a unified form , and supports arbitrary-ary noncoherent frequency - shift keying ( fsk ) and differential phase - shift keying ( dpsk ) . exact ( noncoherent ) maximum - likelihood detectors ( mlds ) for ps - ncdf and ts - ncdf are developed in a unified form , which involves integral evaluations yet serves as the optimum performance benchmark for noncoherent swipt . to reduce the computational cost of the exact mlds , we also propose _ closed - form _ approximate mlds achieving near - optimum performance , thus serving as a _ practical _ solution for noncoherent swipt . numerical results demonstrate a performance tradeoff between the first and second hops through the adjustment of time switching or power splitting parameters , whose optimal values minimizing the symbol - error rate ( ser ) are strictly between 0 and 1 . we demonstrate that-fsk results in a significant energy saving over-dpsk for ; thus-fsk may be more suitable for eh relay systems . energy harvesting , maximum likelihood , noncoherent swipt , wireless information transfer ( wit ) , wireless power transfer ( wpt ) .

1505.07188

the wireless information transfer ( wit ) and wireless power transfer ( wpt ) technologies share the common origin of the electromagnetic waves , which is a direct consequence of the fact that radio frequency ( rf ) signals carry energy as well as information at the same time . nevertheless , prior research activities and industrial developments on wit have been carried out independently from those on wpt , due to the fact that wit and wpt operate with rather different receiver power sensitivities , i.e. , @xmath2 dbm for information receivers and @xmath3 dbm for energy receivers @xcite . as a result , wit and wpt have distinct receiver architectures and design objectives . in particular , wit makes use of the _ active _ mixer for maximal information delivery rate @xcite , whereas wpt utilizes the _ passive _ rectenna for maximal energy transmission efficiency @xcite . it is not until recently that the dual use of rf signals for _ simultaneous wireless information and power transfer _ ( swipt ) @xcite has been studied , giving rise to an appealing solution for energy - constrained wireless systems such as cooperative relay networks . recent advances on swipt have demonstrated conflicting objectives in designing wit for maximal information delivery rate and wpt for maximal energy transmission efficiency , as witnessed by the fundamental _ rate - energy tradeoff _ @xcite . varshney first introduced the notion of the rate - energy region for swipt in additive white gaussian noise ( awgn ) channels @xcite , which was then generalized to frequency - selective fading channels @xcite . these pioneering studies , however , were based on an ideal receiver circuit which simultaneously decodes information and harvests energy from the same signal without any loss @xcite . it has been recognized that this ideal receiver architecture @xcite may be difficult to implement given the current state - of - the - art of electronic circuits . a practical receiver architecture incorporating both energy harvesting ( eh ) and information decoding ( i d ) circuits into a single terminal was proposed in @xcite . based on this practical architecture , the wireless terminal may operate either in the time switching mode @xcite , in which the receiver perform eh and i d in a time - division fashion , or in the power splitting mode @xcite , in which the receiver splits the received signal into two streams for separate eh and i d at the same time . the corresponding rate - energy tradeoff was studied for single - input single - output ( siso ) @xcite and multiple - input multiple - output ( mimo ) channels @xcite . the power allocation , beamforming design , and optimization schemes were studied to achieve various tradeoffs between eh and i d . the throughput maximization for a multi - user system was addressed in @xcite . the optimum power allocation maximizing the information rate while guaranteeing that the harvested energy is above a threshold was studied for broadband systems where perfect instantaneous channel state information ( csi ) was available at the transmitter @xcite . moreover , beamforming scheme maximizing the harvested energy subject to information rate constraint was studied under imperfect csi in @xcite . similarly , beamforming design maximizing the information rate subject to eh constraint was addressed in @xcite . the precoder maximizing the energy efficiency in mimo wiretap channel subject to secrecy rate and power constraints was studied in @xcite . these works @xcite represent typical applications of swipt in point - to - point ( p2p ) communications . recently , the swipt has also found its application in energy - constrained wireless relay networks . in eh relay systems , the source and destination typically have dedicated power supplies from battery or power grid , while the relays are powered by the rf signals radiated by the source . the throughput performance of an amplify - and - forward ( af ) relay network employing eh relays was studied for both delay - limited and delay - tolerant applications in @xcite . the benefits of eh relays and the differences between eh relay networks and conventional relay networks were investigated in @xcite . in @xcite , a joint power allocation and relay selection scheme for af relay networks with eh ability was studied . a greedy switching strategy between eh and data relaying was investigated for a dual - hop af network , and the outage probability was analyzed in @xcite . besides af relay systems , swipt has also been studied for decode - and - forward ( df ) relay systems . specifically , a df network consisting of multiple source - destination pairs and one eh relay was studied in @xcite , where the outage probability was studied taking into account the spatial randomness of the source - destination pairs . for the same system , the diversity performance and power allocation were studied in @xcite . similarly , a df eh network consisting of one or multiple sources , multiple relays , and one destination was studied in @xcite , and a more general df system consisting of a multiple source - destination pairs and multiple eh relays in the presence of random interferences was studied in @xcite , where the small - scale fading and the large - scale path - loss were taken into account in the outage probability analysis . due to the far - field isotropic rf radiation , the energy transfer efficiency of wpt ( e.g. , microwave power transfer ) via rf signals is very low . beamforming techniques ( based on large antenna arrays ) can substantially boost the energy efficiency by forming focused energy beams using the instantaneous csi . the existing works @xcite on swipt have either explicitly or implicitly assumed _ coherent _ wit which requires the instantaneous csi for information decoding . for eh relay systems featuring coherent wit , the source needs to periodically send training symbols for channel estimation , which results in an increased network signaling overhead . moreover , sophisticated channel estimation algorithm needs to be implemented in a _ distributed _ manner among spatially separated wireless terminals , which inevitably increases the hardware complexity and results in extra energy consumption . the consumed energy due to channel estimation and training becomes a serious issue especially for eh relay systems , as the consumed power reduces the net harvested energy for future data relaying . therefore , _ coherent _ swipt requiring channel estimation might be a costly approach for energy - constrained applications . as a remedy , the _ noncoherent _ swipt which eliminates the need for training signalings and the channel estimation , and consequently , consumes much less energy , might be an energy - efficient and cost - effective solution for eh relay systems . to the best of our knowledge , however , the _ noncoherent _ swipt has not been studied in the literature , which motivated this work . the existing works on swipt have been devoted to the _ information - theoretic _ performance analysis in terms of the rate - energy tradeoff , ergodic capacity , outage probability , etc . however , insights into the swipt regarding the bit or symbol error rate performance associated with an actual detection scheme from the _ communication - theoretic _ perspective are not available . to the best of our knowledge , the maximum - likelihood detector ( mld ) achieving noncoherent swipt has not been studied yet . in this paper , we will tackle the noncoherent mlds which characterize the optimum performance benchmark for i d in eh relay systems . the main objective of this paper is to develop the _ noncoherent _ swipt framework and practical mlds for eh relay networks . we consider an eh relay system consisting of a source , a destination , and multiple eh relays , which can harvest energy directly from the rf signals radiated by the source and utilize the harvested energy to decode - and - forward the source information to the destination . the main contributions of the paper are summarized as follows : * a unified noncoherent swipt framework embracing both _ power - splitting noncoherent df _ ( ps - ncdf ) and _ time - switching noncoherent df _ ( ts - ncdf ) protocols is developed , which supports any @xmath0-dpsk or @xmath0-fsk signalings enabling the noncoherent swipt . * following the unified noncoherent swipt framework , we derive the exact mlds for ps - ncdf and ts - ncdf in a unified form , for both @xmath0-fsk and @xmath0-dpsk signalings . the exact mlds involve numerical integral evaluations ; yet are useful for characterizing the optimum performance benchmark for i d in noncoherent swipt . * to reduce the computational cost of the exact mlds , we develop _ closed - form _ approximate mlds with substantially lower complexity . numerical results demonstrate that the proposed approximate mlds achieve almost identical performance to the exact mlds , thus constituting a practical solution for _ noncoherent _ swipt . * useful insights into the noncoherent swipt are obtained from the communication - theoretic perspective in terms of the symbol - error rate ( ser ) . specifically , we demonstrate a performance tradeoff between the first and second hops through the adjustment of the time switching coefficient or power splitting factor . moreover , the optimal values of these system parameters corresponding to the minimum symbol - error rate ( ser ) are strictly between 0 and 1 . finally , we demonstrate that @xmath0-fsk is an energy - efficient solution for noncoherent swipt , especially when @xmath1 . * as a byproduct of this work , we obtain generic analytical results dealing with the distributions of circularly symmetric complex gaussian ( cscg ) random variables and their transformations . these generic results may be of general usefulness for other different applications . moreover , an _ exact _ expression of the transition ( error ) probability is developed for @xmath0-dpsk . in addition , a tight closed - form approximation of the transition probability is developed . to the best of our knowledge , these analytical results have not been reported in the literature . section [ sec : sysmodel ] introduces the system model of eh relay networks . section [ sec : ncdf_relaying ] develops the noncoherent swipt framework for eh relay networks . section [ sec : math ] presents the mathematical fundamentals developed in this paper . section [ sec : mld ] proposes the closed - form approximate as well as the exact mlds . section [ sec : numerical ] evaluates the performance of the proposed detectors and section [ sec : con ] concludes the paper . _ notation _ : we use @xmath4 to denote that @xmath5 is defined by @xmath6 . also , @xmath7 , @xmath8 , @xmath9 , @xmath10 , @xmath11 , and @xmath12 denote the conjugate , transpose , conjugate transpose , 2-norm , expectation , and natural logarithm , respectively . @xmath13 denotes an @xmath14 identity matrix , @xmath15 denotes an all - zero column vector , @xmath16 denotes a column vector with 1 at its @xmath17-th entry and 0 elsewhere , and @xmath18 denotes the phase of @xmath19 over a @xmath20 interval of interest . finally , @xmath21 indicates that @xmath22 is a cscg random vector with mean @xmath23 and covariance @xmath24 . consider an eh relay system comprising a source terminal @xmath25 , a destination terminal @xmath26 , and @xmath27 relays @xmath28 , @xmath29 . the source and destination are powered by dedicated energy sources such as battery or power grid , while the relays have no power supply and instead can harvest energy from the rf signals radiated by the source . using the harvested energy , the relays assist the source - destination communication through the noncoherent df relaying , enabling swipt in eh relay systems where the instantaneous csi is not needed . we consider a composite fading model where the wireless channels are subject to both small - scale fading and large - scale path - loss . note that the path - loss can lead to a serious deterioration in the power transfer efficiency and thus can not be neglected in eh systems @xcite . let @xmath30 denote the path - loss associated with the channel from the transmit terminal @xmath31 to the receive terminal @xmath32 , and @xmath33 denotes the corresponding small - scale rayleigh fading coefficient , for @xmath34 . note that this channel modeling implies that the actual variances of the fading coefficients @xmath35 are absorbed into the corresponding path - loss components @xmath30 , and thus , a general _ asymmetric _ fading model is essentially considered by choosing distinct path - loss components @xmath30 for different wireless links . we consider the _ noncoherent _ communication where the instantaneous csis , @xmath35 , @xmath34 , are unknown to all terminals in the network . the swipt is accomplished through the noncoherent signalings such as @xmath0-fsk and @xmath0-dpsk . specifically , the source broadcasts rf signals according to the noncoherent @xmath0-fsk or @xmath0-dpsk modulation . the relays utilize the harvested energy to _ noncoherently _ decode and forward the information to the destination , where the ultimate _ noncoherent _ detection is performed . thus , the end - to - end information transfer is accomplished through pure noncoherent signalings . due to the absence of the csi , sophisticated protocols such as distributed space - time coding , beamforming , and relay selection which require the csi can not be applied . instead , we adopt the time - division based protocol which coordinates multiple terminals transmissions in different time intervals . specifically , while the source transmits its signal , all relays stay silent and harvest energy from the source rf signals . after the source transmission , the relays decode the source information and forward it to the destination in different time intervals , allowing the destination to combine multiple received signals and thus enabling the diversity gain . note that the time - division relaying protocol allows the developed detectors to be readily integrated into existing time - division multiple access ( tdma ) systems , and it has been considered in numerous works including eh relay systems @xcite and self - powered relay networks @xcite . for ease of the exposition , we focus on the single - antenna transmission over frequency - flat fading channels in this paper . however , the extension to the multi - antenna and/or frequency - selective fading is possible . specifically , by applying the multi - carrier transmission such as orthogonal frequency division multiplexing ( ofdm ) , the frequency - selective fading is automatically turned into flat fading , thus enabling the application of our scheme . moreover , the extension to the multiple - input multiple - output ( mimo ) configuration over each hop is also possible . the remaining issue is to design the noncoherent / differential space - time codes specifically suitable for eh relay systems , which may be considered as a future work . for eh systems , the main motivation to use noncoherent modulation schemes such as noncoherent fsk and dpsk - qam modulation in this paper because it is typically used for coherent communication where the instantaneous csi is available . ] coupled with appropriate noncoherent detection schemes is that the periodic training symbols and sophisticated channel estimation algorithms are completely eliminated , which results in substantial reduction of the signaling overhead and processing burden , and consequently reduces the energy consumption in the network . in this section , by modifying the power splitting and time switching receiver structures for _ coherent _ swipt @xcite where the instantaneous csi is required , we develop a unified _ noncoherent _ swipt framework embracing both ps - ncdf and ts - ncdf requiring no instantaneous csi . in the ps - ncdf protocol , the wpt and wit circuits at each relay terminal function simultaneously while the power fed to each circuit is a portion split from the front - end rf received signal . let @xmath36 ( sec ) denote the total transmission time divided into @xmath37 blocks , each of which of duration @xmath38 ( sec ) is allocated to one of the transmitter including the source and @xmath27 relays , as illustrated in fig . [ fig : psncdf ] . each block consists of @xmath39 symbols of duration @xmath40 ( sec ) each . in the first block , the source broadcasts its signal with power @xmath41 ( watts ) . each relay splits its received rf signal power into two portions : the @xmath42 portion is utilized for wpt and the harvested energy at relay @xmath28 is @xmath43 , @xmath44 , where @xmath45 is the eh efficiency @xcite . the remaining @xmath46 portion is for wit . thus , the harvested power @xmath47 ( watts ) available for data relaying at @xmath28 is given by @xmath48 in the remaining @xmath27 blocks , the relays take turns to decode and forward the source information to the destination using full portion of the harvested power @xmath47 in . an extreme case of @xmath49 represents a pure wit system with no eh , while @xmath50 corresponds to a pure wpt system with no i d . in this paper , we consider a general swipt system featuring both wit and wpt ; thus , we have @xmath51 . for the @xmath0-dpsk signaling , the source message @xmath52 in the @xmath53-th symbol time , @xmath54 , is conveyed via two consecutive information - bearing symbols @xmath55 and @xmath56 according to the differential encoding @xmath57 , where @xmath58 is the initial reference signal . according to the power splitting , the received rf signal power at each relay is split in the proportion of @xmath42 and @xmath46 , where the fractions @xmath42 and @xmath46 of the power are utilized for eh and i d , respectively . thus , the received signal at @xmath28 , @xmath44 , intended for i d is @xmath59 where @xmath40 ( sec ) is the symbol time , @xmath60 is the awgn introduced at the receive antenna at @xmath28 , and @xmath61 is the awgn due to the i d circuit at @xmath28 ( which includes the rf - to - baseband conversion noise , the analog - to - digital conversion noise , circuit loss , etc ) @xcite . the destination does not have a power splitter and the received signal is solely used for i d . thus , the received sinal at the destination from the direct source - destination link is given by @xmath62 where @xmath63 and @xmath64 are the awgns due to the receive antenna and i d circuit at @xmath26 , respectively . each relay node assists the communication from the source to the destination via the df protocol , where @xmath28 detects the source message by ml detection and forwards the detected message @xmath65 with full harvested power @xmath47 in to the destination in a time - division manner via the @xmath0-dpsk transmission . specifically , the transmitted symbols @xmath66 , @xmath54 , at @xmath28 are successively related according to the differential encoding @xmath67 , where @xmath68 is the initial reference signal at @xmath28 , @xmath44 . the signal received at the destination from @xmath28 is given by @xmath69 where @xmath70 and @xmath71 are awgns due to the receive antenna and i d circuit , respectively . it is assumed that all awgns are mutually independent . for noncoherent @xmath0-fsk , the source message @xmath72 is transmitted over one of the @xmath0 orthogonal carriers . the received rf signal at each receive terminal is fed to @xmath0 orthogonal subband correlators , yielding an @xmath73 baseband equivalent signal vector @xmath74^t$ ] , @xmath75 . the baseband equivalent signal received at the destination from the direct link is @xmath76 where @xmath77 and @xmath78 are the awgns due to the receive antenna and the i d circuit at @xmath26 , respectively . due to power splitting at each relay , the received signal intended for i d at @xmath28 is scaled by the factor @xmath79 , and the baseband equivalent received signal at @xmath28 , @xmath44 , is given by @xmath80 where @xmath81 and @xmath82 are awgns due to the receive antenna and the i d circuit at @xmath28 , respectively . each relay decodes the source message and forwards the decoded message @xmath83 with full power @xmath47 in to the destination in a time - division fashion . the baseband equivalent signal received at the destination from @xmath28 is given by @xmath84 where @xmath85 and @xmath86 are awgns due to receive antenna and i d circuit at @xmath26 , respectively . the awgns are assumed mutually independent . in the ts - ncdf protocol , the wpt and wit functionalities take place in a time - division fashion . specifically , the total transmission time @xmath36 ( sec ) is divided into two phases : the wpt phase of duration @xmath87 and the wit phase of duration @xmath88 , as illustrated in fig . [ fig : tsncdf ] . in the wpt phase , the source radiates rf energy flow with power @xmath41 to the relays , where only the eh circuits are activated . thus , the harvested energy at @xmath28 is @xmath89 , @xmath44 , where @xmath45 is the energy conversion efficiency @xcite . in the wit phase , only the i d circuit is activated at each relay . to coordinate multiple terminals transmissions , the wit phase is further divided into @xmath37 blocks of duration @xmath90 each . in the first block , the source broadcasts information with power @xmath41 to all relays . the relay @xmath28 utilizes its harvested energy @xmath91 to decode and forward the source information to the destination in the @xmath92-th block , @xmath44 . thus , the harvested power @xmath47 available for data relaying at @xmath28 is given by @xmath93 an extreme case of @xmath94 represents a pure wit system with no eh , and @xmath95 corresponds to a pure wpt system with no i d . in this paper , we consider @xmath96 , which corresponds to the general swipt relay system featuring both wit and wpt . in the ts - ncdf protocol , the signal received at each terminal during the wit phase is solely utilized for i d , i.e. , no power splitting is involved . consequently , the signal model for ts - ncdf employing the @xmath0-dpsk is depicted as follows : [ eq : yij_dpsk_ts ] @xmath97 @xmath44 , where @xmath98 and @xmath99 are awgns due to the receive antenna and i d circuit , respectively , @xmath34 . the source symbol @xmath55 is related to the message @xmath100 through @xmath57 . similarly , @xmath67 . the signal model for ts - ncdf employing noncoherent @xmath0-fsk is [ eq : by0d_ts ] @xmath101 for @xmath44 , where @xmath102 and @xmath103 , @xmath34 . since we consider two protocols , namely ps - ncdf and ts - ncdf , employing two different noncoherent modulations based on @xmath0-dpsk or @xmath0-fsk , there are in total four possible schemes considering different combinations , which make the derivations of the detectors rather tedious . for ease of development , we now unify the system models of ps - ncdf and ts - ncdf such that the detectors for ps - ncdf and ts - ncdf are derived in a general unified form . the key to the unifying process is a set of reparameterizations of the noise variances and the signal - to - noise ratios ( snrs ) as follows : [ eq : parameters ] @xmath104 where @xmath105 and @xmath106 . in the @xmath0-dpsk transmission , each message is conveyed via two consecutive symbols . thus , it is convenient to represent two consecutive received symbols as a vector . to that end , we introduce the following vectors : @xmath107 , ~~ \bm{s}{\stackrel{\delta}{=}}\left[\begin{array}{c } s(l-1 ) \\ s(l ) \end{array } \right ] , ~~ \bm{s}_r{\stackrel{\delta}{=}}\left[\begin{array}{c } s_r(l-1 ) \\ s_r(l ) \end{array } \right].\label{eq : dpsk_vector}\end{aligned}\ ] ] using definitions in , we can unify ps - ncdf and ts - ncdf for @xmath0-dpsk as follows : [ eq : dpsk_unified ] @xmath108 where @xmath109 for @xmath34 . by appropriately choosing the parameters according to , the unified model can represent either ps - ncdf or ts - ncdf . based on the parameters in , the system models of ps - ncdf and ts - ncdf associated with noncoherent @xmath0-fsk can be expressed in a unified form as follows : [ eq : fsk_unified ] @xmath110 where @xmath111 for @xmath112 and @xmath44 . for ease of readability , a list of symbols are summarized in table [ table : symbols ] . under the noncoherent swipt framework , the instantaneous csis @xmath35 , @xmath113 , are not available to the whole network , which makes the noncoherent detection necessary . to ensure the end - to - end optimality of the noncoherent communication , it is required that the destination as well as the relays perform the ml detection . specifically , each relay performs its own ( noncoherent ) ml detection individually and forwards its estimated message to the destination . due to potential decoding errors at the relays , the ultimate ( noncoherent ) ml detection at the destination must take into account the transition decoding error probabilities in all the relays , which will be treated in more detail in what follows . to facilitate the development of the noncoherent detection schemes , it is necessary to study the distributions involving combinations of cscg and rayleigh random variables / vectors , as indicated by the unified swipt frameworks in and . we carry out a generic probability analysis involving independent random varaibles / vectors as follows : @xmath114 we are interested in the following random vectors : @xmath115 where @xmath116^t$ ] , @xmath117 is any complex constant , and @xmath118 is any integer number . note that and are generic forms of @xmath119 in for @xmath0-dpsk and @xmath120 in for @xmath0-fsk , respectively . [ lemma : pdf ] given the random variables / vectors in , the pdfs of @xmath121 in and @xmath122 in are , respectively , given by @xmath123 where @xmath124^t$ ] , @xmath125^t$ ] , and @xmath126 is defined as @xmath127 see appendix a. [ co : pdf ] let @xmath128 and @xmath129 . then the pdf of @xmath130 , @xmath131 , and the pdf of @xmath132 , @xmath133 , are given by @xmath134 and @xmath135 . due to rotational invariance of the cscg distribution @xcite , it can be shown that @xmath136 follows the same distribution as @xmath137 , which yields the above result . a comprehensive study of the distribution of the product of two independent cscg random variables , namely , @xmath138 " , was also presented in @xcite . however , the result in @xcite was given in terms of modified bessel function . if one utilizes the result of @xcite , then the pdf of @xmath139 " may be derived as an _ infinite _ convolution integral of the modified bessel function with the gaussian pdf , which is prohibitively more complex than the result obtained here . note that the _ finite _ integral of has a well - behaved integrand involving only _ elementary _ functions , and it has a fast convergence speed which allows for fast evaluations at very high accuracy , as will be demonstrated later . in order to develop the mld at the destination , the transition ( error ) probability of the dpsk detector at the relay needs to be determined . to that end , it is useful to study the distribution of the phase difference of two consecutively received dpsk signals . [ lemma : phase_dist ] given the random variables / vectors in , we consider the @xmath140 random vector @xmath141^t= x_1\bm{c } + \bm{x}_0 $ ] , where @xmath142^t$ ] . then the pdf of @xmath143 is @xmath144,\label{eq : phase_dist } \end{aligned}\ ] ] where @xmath145 and @xmath146 is defined in _ any _ @xmath20 interval of interest . in is a periodic function with period @xmath20 . thus , one may consider the domain of @xmath146 as any @xmath20 interval of interest such as @xmath147 or @xmath148 . ] see appendix b. the phase difference distribution of psk / dpsk signals was studied for a simplistic awgn channel with no fading @xcite , where the pdf was given in an integral form . in contrast , in addition to the awgn @xmath149 , we impose rayleigh fading where the fading coefficient is modeled by @xmath150 . for the fading scenario , the obtained pdf is given by a truly _ closed - form _ expression , which is more suitable for further analysis . the results in lemma [ lemma : pdf ] and corollary [ co : pdf ] provide a useful characterization of the joint magnitude and phase distribution of sophisticated gaussian transformations , while the result in lemma [ lemma : phase_dist ] characterizes the distribution of the input - output phase difference involving gaussian random variables / vectors . these generic analytical results which have not been reported in the literature may be useful in a different context involving different applications . to facilitate the evaluation of @xmath151 in , we now derive some alternative expressions of @xmath151 , which are suitable for numerical computations . [ lemma : equivalent_forms ] @xmath151 in can be equivalently expressed as @xmath152 , & \beta\epsilon<1,\\ \beta { \mathrm{e } } ^\beta k_0(2\beta ) , & \beta\epsilon=1 , \\ \frac { { \mathrm{e } } ^{\frac{1}{\epsilon}}}{\epsilon}\bigg[k_0\bigg(2\sqrt{\frac{\beta}{\epsilon}}\bigg)+ \displaystyle\int_{\frac{1}{\epsilon}}^{\sqrt{\frac{\beta}{\epsilon}}}\frac{1}{x } { \mathrm{e } } ^{-\big(x+\frac{\beta}{\epsilon x}\big ) } { \mathrm{d } } x \bigg ] , & \beta\epsilon>1 , \end{array } \right.\label{eq : alternative2 } \end{aligned}\ ] ] where @xmath153 is the zero - order modified bessel function of the second kind ( * ? ? ? * eq . ( 8.432.1 ) ) . for @xmath154 , can be rewritten as @xmath155 $ ] . similarly , @xmath156 $ ] for @xmath157 . for @xmath158 , we have @xmath159 . these expressions have a common integral - form @xmath160 with @xmath161 . by change of variable , it can be shown that @xmath162 . thus , we have @xmath163 , followed by ( * ? ? ? * eq . ( 3.471.9 ) ) . finally , substituting @xmath164 for @xmath157 or @xmath154 and @xmath165 for @xmath166 yields . the alternative form in indicates that the original integral of can be seen as the expectation of the random transformation @xmath167 , where @xmath168 is exponentially distributed with unit mean . the alternative form in has a closed - form solution only for @xmath166 . for @xmath169 , we are able to boil down the original integral into two terms , where the first term is given in an exact closed - form while the second term is given by a sub - integral . the benefit of using this alternative form is that the sub - integral is taken over a smaller integral interval and thus can be more easily computed as compared to the original integral in . we now exploit these alternative forms to derive a very useful piecewise approximation for the original integral with very high accuracy . [ lemma : numerical_computation ] for @xmath169 , @xmath151 in can be approximated by the @xmath170-th order expression , @xmath171 , @xmath172 , where @xmath173 is the @xmath170-th remainder and @xmath174 is given by @xmath175 , & \beta\epsilon<1 , \frac { { \mathrm{e } } ^{\frac{1}{\epsilon}}}{\epsilon}>1,\vspace{2 mm } \\ \frac { { \mathrm{e } } ^{\frac{1}{\epsilon}}}{\epsilon}\bigg[k_0\bigg(2\sqrt{\frac{\beta}{\epsilon}}\bigg)-\psi_n\bigg(\beta,\sqrt{\frac{\beta}{\epsilon}}\bigg ) \bigg ] , & \beta\epsilon<1 , \frac { { \mathrm{e } } ^{\frac{1}{\epsilon}}}{\epsilon}\leq 1 , \vspace{3 mm } \\ \frac { { \mathrm{e } } ^{\frac{1}{\epsilon}}}{\epsilon}\bigg[k_0\bigg(2\sqrt{\frac{\beta}{\epsilon}}\bigg)+\psi_n\bigg(\frac{1}{\epsilon},\sqrt{\frac{\beta}{\epsilon}}\bigg ) \bigg ] , & \beta\epsilon>1 , \end{array } \right.\label{eq : n_approx}\end{aligned}\ ] ] where @xmath176 , @xmath177 , @xmath178 , and @xmath179 . also , @xmath180 is given by @xmath181 where @xmath182 and @xmath183 , @xmath184 , are , respectively , the weights and zeros of the legendre polynomial of order @xmath170 ( * ? ? ? * table 25.4 ) , and @xmath185 . see appendix c. the first piece of is a direct consequence of followed by the taylor series expansion of @xmath167 around the mean @xmath186=1 $ ] . the second and third pieces of are direct consequences of the gaussian - legendre quadrature carried out for the sub - integrals in . note that the closed - form piecewise approximation of @xmath151 by means of @xmath174 in is valid for any natural number @xmath172 , and an improved accuracy is expected by increasing the order @xmath170 . however , from the computational cost point - of - view , it is desirable to choose a fairly small order @xmath170 which attains the required accuracy . to that end , we set @xmath187 to guarantee extremely low computational complexity , and the expression of reduces to [ eq:2_approx ] i_2(,)= ^-(1+^2 ^ 2 - 2 ^ 2 ^ 3-^2 ^ 4 ^ 2 ) , & @xmath188,[eq:2_approx_a ] + , & @xmath189 , [ eq:2_approx_b ] + , & @xmath157,[eq:2_approx_c ] where we have @xmath190 , @xmath191 , and @xmath192 for @xmath187 . through extensive evaluations in section [ sec : numerical ] , we will demonstrate that the piecewise approximation @xmath193 in is indeed very accurate for a wide range of @xmath194 and @xmath195 values . considering its very low computational cost , we will utilize @xmath193 in the next section to facilitate the implementation of the approximate mlds . following the unified noncoherent swipt framework in section [ sec : ncdf_relaying ] , we will develop unified noncoherent detectors in this section . specifically , we begin by finding the exact mlds at each relay , followed by the transition probability analysis . finally , the exact mlds as well as the closed - form approximate mlds at the destination are derived for both @xmath0-dpsk and @xmath0-fsk . from the unified first - hop model in and , we see that the input - output relation is equivalent to the conventional p2p communication in rayleigh fading . thus , the mld at the relay can be derived similar to that for p2p communication . specifically , the mlds of @xmath0-dpsk at @xmath28 , @xmath44 , for ps - ncdf and ts - ncdf are expressed in a unified form ( * ? ? ? 5.2.8 ) @xmath196 similarly , the mlds of @xmath0-fsk at @xmath28 for ps - ncdf and ts - ncdf are given by ( * ? ? ? 5.4.2 ) @xmath197 due to potential detection errors at the relay , the detected message @xmath198 at @xmath28 may be different from the actual source transmitted message @xmath100 . in fact , a potential decision error at @xmath28 may transform the correct message @xmath72 to any incorrect message @xmath65 . thus , the _ transition probability _ , @xmath199 , namely the probability that the message @xmath198 is detected while @xmath100 is actually transmitted serves as useful side information for the destination detection . with orthogonal signalings such as @xmath0-fsk , the transition from the correct message @xmath100 to any one of the @xmath200 incorrect message is equiprobable @xcite . thus , the transition probability for @xmath0-fsk can be expressed in terms of the ser , @xmath201 , as follows : @xmath202 where @xmath203 for non - orthogonal signalings such as @xmath0-dpsk , however , the transition errors are not equiprobable because each constellation point is not equidistant from all other constellation points @xcite . thus , the simple relationship in for @xmath0-fsk does not hold for @xmath0-dpsk . with non - trivial derivations , we derive the _ exact _ transition probability for @xmath0-dpsk as follows . [ lemma : transition_prob ] for the mld of @xmath0-dpsk in at relay @xmath28 , @xmath44 , the exact transition probability in rayleigh fading is given by @xmath204 , where @xmath205 { \mathrm{d } } \theta,\label{eq : transition_prob } \end{aligned}\ ] ] @xmath206 . see appendix d. the exact transition probability of @xmath0-dpsk depends only on the transition phase indexed by @xmath207 . it is not hard to mathematically prove @xmath208 for @xmath209 . this property can be attributed to the symmetry of the psk constellations . it suggests that in order to characterize @xmath0 transition probabilities for @xmath210 , one only needs to calculate @xmath211 transition probabilities @xmath212 for @xmath213 via , and the remaining @xmath211 transition probabilities , @xmath212 , @xmath214 , follow immediately by the symmetry @xmath208 . [ lemma : transition_prob_approx ] the transition probability of @xmath0-dpsk in rayleigh fading can be approximated as @xmath215 where @xmath216 , & m\geq 4 . \end{array } \right.\end{aligned}\ ] ] see appendix e. the approximate transition probability in exploits the fact that the error transmissions from the transmitted psk phase to its two closest neighbouring phases are most likely and euqi - probable . moreover , error transitions to any farther constellations points other than the two closest constellation points are negligible . in this subsection , we develop the mlds at the destination for both ps - ncdf and ts - ncdf protocols in a unified form , which exploits explicitly the transition probabilities at the relays to ameliorate the error propagations from the relays . [ thm : exact_mld ] the exact mlds of @xmath0-dpsk at the destination for both ps - ncdf and ts - ncdf are given in a unified form as follows : @xmath217\!\bigg\},\label{eq : mld_dpsk}\end{aligned}\ ] ] where @xmath218 , @xmath219 , @xmath220 , and @xmath221 is the transition probability given in . similarly , the exact mlds of @xmath0-fsk at the destination for ps - ncdf and ts - ncdf are given in a unified form as @xmath222\bigg\}. \label{eq : mld_fsk}\end{aligned}\ ] ] see appendix f. the exact ( optimum ) mlds in and require the evaluations of the integral @xmath151 in . in addition , the transition probability involved in the exact mld of is given by a finite integral - form in , which makes the exact mlds complex to implement in practice . nevertheless , it is very meaningful to study the exact mlds as they serve as the optimum performance benchmark for swipt in noncoherent df systems . with the exact mlds , one may use them as a theoretical performance upper bound to evaluate the performance of suboptimum detectors , if developed . in the following , to make the mlds suitable for practical systems , we develop _ closed - form _ approximations for the mlds . following the exact mlds in theorem [ thm : exact_mld ] , applying the approximation of @xmath225 for @xmath0-fsk and @xmath0-dpsk and the transition probability approximation in lemma [ lemma : transition_prob_approx ] for @xmath0-dpsk immediately gives the approximate mlds . the ml decision metrics for the exact mlds in theorem [ thm : exact_mld ] are given by the weighted sum of the decision statistics associated with the direct link and the relay branches . for each relay branch , the weight function is proportional to the transition probability and the decision statistic is involved in the integral function @xmath151 . for the approximate mlds in theorem [ th : approx_mld ] , the exact integral function @xmath151 is replaced by its tight closed - form approximation @xmath193 . in addition , for @xmath0-dpsk , the approximate transition probability is used in place of the exact transition probability , which reduces the sum of @xmath200 transition terms to 2 terms , irrespective of @xmath0 . in this section , we carry out monte carlo simulations to gain useful design insights into the proposed ps - ncdf and ts - ncdf protocols . the eh efficiency at the relays is set to @xmath226 as in @xcite . we adopt the _ bounded path - loss model _ as in @xcite in which the path - loss is strictly larger than one . specifically , the path - loss from transmit terminal @xmath31 to receive terminal @xmath32 is modeled by @xmath227 , where @xmath228 is the path - loss exponent and @xmath229 is the transmitter - receiver distance for @xmath34 . the path - loss exponent is set to @xmath230 which corresponds to the urban cellular communication environment @xcite . the source - destination distance is fixed to @xmath231 ( m ) , while the source - relay distance @xmath232 and relay - destination distance @xmath233 can vary arbitrarily between 0 and 3 subject to @xmath234 , for @xmath44 . for ease of simulations , the noise variances are set to be @xmath235 , such that each receive terminal has a total noise variance @xmath236 , i.e. , @xmath237 , where @xmath238 and @xmath239 , @xmath34 , are the variances of the receive antenna noise and the i d circuit noise , respectively . the error performance is parameterized by the _ transmitter snr _ , defined as @xmath240 . we will evaluate the error performance of ps - ncdf and ts - ncdf at the same information rate @xmath241 ( bps ) , where @xmath39 is the total number of information - bearing symbols transmitted over @xmath36 ( sec ) . note that @xmath39 new symbols are transmitted in the first block , which are relayed in the following @xmath27 blocks by the relays . the symbol time @xmath40 is related to @xmath242 as @xmath243 for simplicity , we will fix @xmath244 such that @xmath245 , which depends only on @xmath0 . vs the exact numerical integration @xmath151 for various @xmath194 and @xmath195 values . ] we first check the accuracy of the proposed piecewise approximation @xmath193 in . as a benchmark , we use the matlab numerical integration function integral " to directly compute @xmath151 in at a relative error tolerance of @xmath246 , which is referred to as the exact " result . we evaluate @xmath193 for @xmath247 $ ] db under various @xmath194 values of interest ranging from @xmath3 to @xmath248 db in fig . [ fig : integral_approx_vs_beta ] . since @xmath195 spans over a very wide range of values for each @xmath194 , the three subcases involved in are all considered in fig . [ fig : integral_approx_vs_beta ] for each @xmath194 value . in this case , the piecewise approximation of is applied according to the conditions for the subcases . it can be observed that @xmath193 and @xmath151 overlap for all possible values of @xmath194 and @xmath195 , which confirms that @xmath193 in is indeed a very good approximation . we validate the proposed closed - form approximate mlds in and through comparisons to the exact ( integral - form ) mlds in and , for binary dpsk ( bdpsk ) and binary fsk ( bfsk ) signalings with fixed symbol rate @xmath249 ( bps ) . specifically , the ts - ncdf protocol with time switching coefficient @xmath250 in a single - relay ( @xmath251 ) eh system with @xmath252 ( m ) is considered in fig . [ fig : mlds_ts_single_relay ] , and the ps - ncdf protocol with power splitting factor @xmath253 for a two - relay ( @xmath254 ) eh system with @xmath255 ( m ) and @xmath256 ( m ) is considered in fig . [ fig : mlds_ps_two_relay ] . as demonstrated in these figures , the sers of the approximate mlds are in excellent agreement with those of the exact mlds over the whole snr range , indicating that the proposed closed - form approximate mlds are useful practical solutions for noncoherent swipt . . [ fig : mlds_comparison_benchamrk ] we now compare ps - ncdf and ts - ncdf in terms of the ser at the same information rate @xmath242 . [ fig : ps_ts_fixed_ns ] shows the ser versus the time switching coefficient @xmath257 for ts - ncdf or power splitting factor @xmath42 for ps - ncdf in a single - relay ( @xmath251 ) network with @xmath252 ( m ) , where binary signalings ( @xmath258 ) with @xmath249 ( bps ) and @xmath259 db ( see fig . [ fig : ps_ts_vary_coeff_30db_m2_fixed_ns ] ) and octonary signalings ( @xmath260 ) with @xmath261 ( bps ) and @xmath262 db ( see fig . [ fig : ps_ts_vary_coeff_40db_m8_fixed_ns_snr ] ) are considered . as shown in fig . [ fig : ps_ts_fixed_ns ] , the sers exhibit some global minima around @xmath263 for ts - ncdf and around @xmath264 for ps - ncdf . this is because there exists a performance tradeoff between the first and second hops through the adjustment of @xmath257 ( for ts - ncdf ) or @xmath42 ( for ps - ncdf ) . specifically , when @xmath257 ( or @xmath42 ) increases , more power ( proportional to @xmath257 or @xmath42 ) can be harvested and utilized by the second - hop data relaying , resulting in an increased snr over the second - hop . on the other hand , for ts - ncdf , as @xmath257 increases , less time ( proportional to @xmath265 ) is available for the first - hop information transmission . thus , the symbol time @xmath40 for ts - ncdf must decrease so as to maintain the same target information rate @xmath242 , which results in smaller received symbol energy and consequently decreased snr over the first - hop . similarly , for ps - ncdf , as @xmath42 increases , less power ( proportional to @xmath46 ) is split into the wit circuit for information decoding ; thus , the first - hop snr decreases . therefore , adjusting the values of @xmath257 ( or @xmath42 ) results in different impacts on the snrs over the two hops , and consequently , a tradeoff exists between the first- and second - hop snrs through the adjustment of @xmath257 ( or @xmath42 ) . this can be also explained mathematically . by substituting into , we see that @xmath266 is monotonically decreasing in @xmath257 ( or @xmath42 ) and @xmath267 is monotonically increasing in @xmath257 ( or @xmath42 ) . due to the opposite behaviors of the first- and second - hop snrs as a function of @xmath257 ( or @xmath42 ) , the optimal value of @xmath257 ( or @xmath42 ) minimizing the overall error probability is strictly between 0 and 1 . + in this subsection , we are interested in evaluating the performance of a variety of noncoherent signalings for @xmath268 . to that end , we focus on a single - relay network with @xmath252 ( m ) employing the near - optimum eh parameters , i.e. , @xmath264 for ps - ncdf and @xmath263 for ts - ncdf . note that @xmath264 and @xmath263 minimize , respectively , the sers of ps - ncdf and ts - ncdf , as shown in fig . [ fig : ps_ts_fixed_ns ] . however , it is difficult to _ analytically _ determine the optimum eh parameters . an interesting phenomenon observed is that while @xmath0-dpsk outperforms @xmath0-fsk for @xmath269 , the performance gain diminishes as @xmath0 increases , as shown in figs . [ fig : ps_vs_ts_m2_fixed_ns_snr ] and [ fig : ps_vs_ts_m4_fixed_ns_snr ] . in particular , for @xmath270 shown in fig . [ fig : ps_vs_ts_m4_fixed_ns_snr ] , the performance gain of dpsk over fsk is almost negligible . as @xmath0 increases to @xmath1 , we see that @xmath0-fsk achieves a significant gain over @xmath0-dpsk . in particular , for a target ser of @xmath271 , the performance gain of @xmath0-fsk over @xmath0-dpsk is approximately 4 db and 8 db for @xmath260 and @xmath272 , respectively , as shown in figs . [ fig : ps_vs_ts_m8_fixed_ns_snr ] and [ fig : ps_vs_ts_m16_fixed_ns_snr ] . this implies that @xmath0-fsk achieves a promising energy gain over @xmath0-dpsk for large @xmath0 . however , it is important to note that this energy gain comes at the cost of higher required bandwidth . in particular , the minimum double - sideband ( dsb ) bandwidth at an intermediate frequency ( if ) assuming ideal rectangular nyquist filtering @xcite is @xmath273 for @xmath0-dpsk and @xmath274 for @xmath0-fsk ( @xmath0 times that of @xmath0-dpsk ) . nevertheless , considering the significant energy gain , @xmath0-fsk may be a more suitable solution , especially when @xmath1 , for eh relay systems due to the limited energy . we have proposed a noncoherent swipt framework embracing both ps - ncdf and ts - ncdf in a unified form , supporting @xmath0-fsk and @xmath0-dpsk . following this framework , we developed exact noncoherent mlds for ps - ncdf and ts - ncdf in a unified form , which involves integral evaluations . nevertheless , the exact mlds are useful for characterizing the optimum performance benchmark for noncoherent swipt . furthermore , to make the mlds suitable for practical implementation , we developed _ closed - form _ approximate mlds achieving near - optimum performance . numerical results demonstrated a performance tradeoff between the first and second hops through the adjustment of the time switching coefficient or power splitting factors . moreover , the optimal values of these system parameters corresponding to the minimum ser are strictly between 0 and 1 . finally , we demonstrated that @xmath0-fsk is an energy - efficient solution for noncoherent swipt , especially when @xmath1 . the conditional distribution of @xmath121 given @xmath275 is @xmath276 , where @xmath277.\ ] ] it can be shown that @xmath278 and @xmath279.\ ] ] after some manipulations , the conditional pdf is written as @xmath280 the unconditional pdf is @xmath281 , which is evaluated to @xmath282 applying change of variable @xmath283 in and after some manipulations , we have @xmath284 which is the same as upon some appropriate parameterizations . the derivation of the pdf of @xmath122 follows similar lines as the derivation of @xmath285 , which is briefly summarized as follows . the conditional pdf of @xmath122 given @xmath286 is given by @xmath287 similarly , the unconditional pdf is obtained as @xmath288 using similar change of variable in , we can obtain the final expression of . it follows that @xmath289 , where @xmath290,\end{aligned}\ ] ] where @xmath291 . applying the property of the multivariate gaussian distribution , the conditional distribution of @xmath292 given @xmath293 follows a gaussian distribution , i.e. , @xmath294 , where @xmath295 and @xmath296 . let @xmath297 . it follows that @xmath298 . thus , the unconditional pdf of @xmath299 is @xmath300\bigg\ } { \mathrm{d } } x. \label{eq : app_z_pdf1}\end{aligned}\ ] ] according to ( * ? ? ? * eq . ( 3.471.9 ) ) , is evaluated to @xmath301 where @xmath153 is the zero - order modified bessel function of the second kind ( * ? ? ? * eq . ( 8.432.1 ) ) . using the cartesian - to - polar transformation , @xmath302 , the joint pdf of @xmath303 and @xmath304 is @xmath305k_0\bigg(\frac{2r_0}{\sigma_0}\sqrt{\frac{1}{\omega_0+\omega_1}+ \frac{|\vartheta_0|^2}{\sigma_0 ^ 2}}\bigg ) . \!\!\end{aligned}\ ] ] since @xmath306 , the pdf of @xmath307 is @xmath308 , which evaluates to @xmath309 it is verified by mathematica@xmath310 that the following equation holds @xmath311 for @xmath312 and @xmath313 . letting @xmath314 and @xmath315 in , we solve the integral of in closed - form as @xmath316}\bigg[1+\frac{\cos\theta}{\sqrt{\sin^2\theta+ \frac{\sigma_0 ^ 2}{|\vartheta_0|^2(\omega_0+\omega_1)}}}\arccos\bigg(-\frac{\cos\theta}{\sqrt{\sin^2\theta+ \frac{\sigma_0 ^ 2}{|\vartheta_0|^2(\omega_0+\omega_1)}}}\bigg)\bigg].\ ] ] finally , substituting @xmath317 and @xmath236 into the above expression yields the final expression in . from , we can rewrite @xmath151 for @xmath157 as @xmath318 over the integration interval @xmath319 in , we have @xmath320 , @xmath321 , and @xmath322 , meaning that the overall integrand is a finite bounded function . moreover , the integrand is always between 0 and 1 , which ensures fast convergence when approximated by the gauss - legendre quadrature ( glq ) @xcite . note that the glq is one of the most efficient and accurate non - adaptive quadrature rules for a given number of arithmetic operations @xcite . applying the @xmath170-th order glq rule of ( * eq . ( 25.4.29 ) ) in yields @xmath323 , where @xmath324 $ ] and @xmath180 is given in . this gives the third case of . for @xmath154 in , it is easy to verify that the overall integrand is not necessarily smaller than one in this case . thus , simply applying the glq rule for @xmath154 in might give an approximation with slower convergence speed compared to the glq approximation for @xmath157 . however , if we further enforce @xmath325 , then the pre - bracket coefficient @xmath326 in can scale down the approximation error for the integral , which can significantly accelerate the convergence speed for the approximation . thus , for @xmath154 and @xmath325 , the @xmath170-th order glq rule of ( * ? ? ? * eq . ( 25.4.29 ) ) readily applies , which yields @xmath327 $ ] . this is the same as the second case of . for the remaining case with @xmath154 and @xmath328 , we start with the alternative form of @xmath151 in , which can be written as @xmath329 $ ] where the expectation is taken with respect to @xmath330 and @xmath330 is an exponential random variable with unit mean . thus , the integral @xmath151 can be interpreted as the statistical mean of the random variable @xmath331 . using the @xmath170-th order estimate of the mean of random variables , we can approximate @xmath151 by @xmath174 as follows ( * ? ? ? * eq . ( 5.85 ) ) : @xmath332=1},\label{eq : app_talor1}\end{aligned}\ ] ] where @xmath333|^n]$ ] , @xmath334 . using the binomial theorem and the exponential distribution of @xmath330 , it is not hard to show that @xmath335 . the function @xmath336 can be represented as a composite function with @xmath337 $ ] , where @xmath338 and @xmath339 . thus , fa@xmath340 di bruno s formula @xcite is applied as follows : @xmath341 } { { \mathrm{d } } x}\bigg|_{x=1 } = \sum\limits_{\substack{\sum_{i=1}^n im_i = n\\ \forall m_i\in \mathbb{z}^{+}}}\frac{n!}{\prod\limits_{j=1}^n m_j!}f_0^{(\varsigma)}(t)\bigg|_{t=\tau}\prod_{k=1}^n\bigg(\frac{g_0^{(k)}(x)}{k!}\bigg)^{m_k}\bigg |_{x=1},\label{eq : app_taylor_der}\end{aligned}\ ] ] where @xmath342 , @xmath343 , and @xmath179 . it can be shown that @xmath344 and @xmath345 . applying these relationships in , substituting the expression into , and after tedious manipulations , we obtain the first case of . according to the mld rule in , the phase @xmath346 is a sufficient statistic for the mld . let @xmath347 denote the decision region for the message @xmath100 , where @xmath348 , @xmath349 . then the ml decision rule is in favor of the message @xmath198 if @xmath350 , @xmath351 . thus , the transition probability is written as @xmath352\pi}{m}\leq \theta_r-\theta_m \leq \frac{\big[2(m_r - m)+1\big]\pi}{m}\bigg|m\bigg).\label{eq : app_tp2}\end{aligned}\ ] ] the unified system model for @xmath28 in can be rewritten as @xmath353^t+\bm{n}^\prime_{0r}$ ] , where @xmath354 and @xmath355 . it follows that @xmath356 and @xmath357 . applying the result in lemma [ lemma : phase_dist ] , the pdf of @xmath358 is given by @xmath359 . \label{eq : app_phase_dist } \end{aligned}\ ] ] thus , the transition probability in can be written as [ eq : app_symmetry_tp ] @xmath360 where is due to the change of variable @xmath361 and @xmath362 . note that the symmetry @xmath363 can be also attributed to the rotational invariance property of the psk signal constellation . according to , the transition probability can be calculated as @xmath364 finally , substituting into yields the final expression in . noticing that the transmitted psk phase will most likely be mistaken for its two closest neighboring phases with equal probability @xcite , we can approximate the transition probability as follows : for the two closest error transitions with @xmath365 or @xmath200 , we have @xmath366 ; for the correct symbol transition with @xmath367 , we have @xmath368 . for all other error transitions , we simply ignore the probabilities . this approximation immediately gives . it remains to derive the ser for @xmath0-dpsk . for @xmath258 , a closed - from exact expression of the ser is available , namely , @xmath369 . for @xmath210 , however , no closed - form exact ser expression is reported in the literature for fading channels . to proceed , we resort to an upper bound on the ser for awgn channels in ( * ? ? ? * eq . ( 8.95 ) ) , which is the best known bound for awgn channels . taking the statistical average of this bound over the rayleigh distribution by means of ( * ? ? ? * eq . ( 6.283.2 ) ) , we obtain a novel upper bound on the ( average ) ser of @xmath0-dpsk as @xmath370 $ ] . overall , the ser of @xmath0-dpsk is given by . applying total probability theorem and lemma [ lemma : transition_prob ] , the likelihood function ( lf ) conditioned on message @xmath100 for the unified @xmath0-dpsk model in can be written as @xmath371 from , we see that @xmath372 , where @xmath373.\end{aligned}\ ] ] after some manipulations , we have @xmath374 from , we see that applying lemma [ lemma : pdf ] immediately yields the lf @xmath375 as follows : @xmath376 substituting and into and using the definitions of @xmath377 , @xmath378 , and @xmath379 given in theorem [ thm : exact_mld ] , we obtain the mld in for @xmath0-dpsk . for @xmath0-fsk in , following similar procedure as in for @xmath0-dpsk , the lf can be written as @xmath380,\label{eq : app_fsk_lf2}\end{aligned}\ ] ] furthermore , the pdf of @xmath381 in , conditioned on the knowledge of @xmath100 , is @xmath382 applying lemma [ lemma : pdf ] , the pdf of @xmath120 in given @xmath198 is given by @xmath383 finally , substituting and into yields the mld in for @xmath0-fsk . a. a. nasir , x. zhou , s. durrani , and r. a. kennedy , `` relaying protocols for wireless energy harvesting and information processing , '' _ ieee trans . wireless commun . _ , vol . 12 , pp . 36223636 , july 2013 . z. ding , i. krikidis , b. sharif , and h. v. poor , `` wireless information and power transfer in cooperative networks with spatially random relays , '' _ ieee trans . wireless commun . _ , vol . 13 , pp . 44404453 , aug . 2014 . c. xing , s. ma , z. fei , y .- c . wu , and h. poor , `` a general robust linear transceiver design for multi - hop amplify - and - forward mimo relaying systems , '' _ ieee trans . signal process . _ , vol . 61 , pp . 11961209 , mar . 2013 . c. xing , m. xia , f. gao , and y .- c . wu , `` robust transceiver with tomlinson - harashima precoding for amplify - and - forward mimo relaying systems , '' _ ieee j. sel . areas commun . _ , vol . 30 , pp . 13701382 , sept . p. liu , s. gazor , i .- kim , and d. i. kim , `` noncoherent amplify - and - forward cooperative networks : robust detection and performance analysis , '' _ ieee trans . _ , vol . 61 , pp . 36443659 , sept .

two - neutron correlation in the channel in quasi two - dimensional ( 2d ) neutron systems at zero temperature is studied by means of the bcs theory with finite - range effective nuclear forces . the dineutron correlation in low density neutron systems confined in an infinite slab is investigated in a simplified model that neutron motion of one direction is frozen . when the slab is thin enough , two neutrons form a tightly bound dineutron with a small size in the quasi-2d system , and a bose dineutron gas is found in low density limit . with increase of fermi momentum , the neutron system changes from the bose - gas phase to the superfluid cooper pair phase . the density dependence of the neutron pairing shows the bcs - bec crossover phenomena at finite low - density region . in the transition region , the size shrinking of neutron pair and enhancement of pairing gap are found . the relation to dineutron correlation at surface of neutron - rich nuclei is also discussed .

0902.3717

two - neutron correlations in neutron - rich nuclei is presently one of the fore - front subjects in the physics of unstable nuclei . in two - neutron halo nuclei @xcite such as @xmath1li , the dineutron correlation was theoretically predicted in many studies @xcite and is supported by experiments @xcite . dineutron correlations are discussed also in light neutron - rich nuclei such as in @xmath2he @xcite and also in medium - heavy neutron - rich nuclei @xcite as well as asymmetric nuclear matter ( for example , refs . @xcite and references therein ) . dineutron correlations are characterized by strong spatial correlations of two neutrons in the @xmath0 channel . although the neutron - neutron interaction is too weak to form a two - body bound state , the attraction in the @xmath0 channel is rather strong as is known from the large scattering length @xmath3 fm @xcite . originally the possible existence of a dineutron near the surface of nuclei was predicted by migdal in 1972 @xcite . in his work , it was shown that two interacting particles in a potential well can form a bound state if there is a single - particle level with energy close to zero . this idea describes the binding mechanism of a borromean system and is extended to the dineutron cluster picture and the dineutron correlation in two - neutron halo nuclei . recently , dineutron correlations have also been discussed from the point of view of bose - einstein condensation ( bec ) , which is considered along similar lines as deuteron and @xmath4-particle condensation suggested in dilute nuclear matter @xcite as well as @xmath4 condensate states in excited states of @xmath5 even nuclei @xcite . this @xmath6-bec feature stems from the surprisingly small size of cooper pairs recently revealed in the surface of nuclei @xcite . in nuclear systems , the @xmath7 @xmath8 pair correlation in the bardeen - cooper - schrieffer ( bcs ) phase has been known to be very important @xcite . from the analysis of the spatial structure of the cooper pair in neutron matter , a size shrinking is found to occur at finite low density @xcite . it indicates that the spatially correlated neutron pair may appear in the transition region from weak coupling bcs regime to strong coupling bec regime . similar bcs - bec crossover features of the spatially correlated neutron pair were predicted also at the surface of medium - heavy nuclei @xcite . since nuclear pairing increases as the size of nuclei shrinks ( @xmath9 mev ) , one may expect that the bec features are enhanced with respect to the infinite matter case . thus , the spatial structure of neutron pairs at the surface of neutron - rich nuclei attracts presently great interest . in particular , bcs - bec cross over features are expected to appear in the environment of dilute neutron matter , such as realized in neutron skins @xcite . this has been put into evidence in recent studies @xcite where it was found that the extension of a @xmath6-cooper pair strongly decreases when going from inside to the surface before expanding again when leaving the nucleus . the minimum is very pronounced and the cooper pair attains a minimum size of about 2 - 3fm . this is very small size reaching the dimension of the deuteron , that is of a bound state . this is also highlighted by the fact that in @xmath1li even the single cooper pair is very small in the surface @xcite . the fact that @xmath6-cooper pairs resemble a bound state in the nuclear surface and even become appreciably smaller than in low density neutron matter with maximum gap value , is intriguing . in this work we want to explore the rather extreme hypothesis that the binding mechanism may be influenced by some quasi two - dimensional features . by quasi two - dimensional ( 2d) , we mean that the bound cooper pairs are confined within a surface layer of about @xmath10 fm and that the degree of freedom in radial direction is approximately frozen . such a scenario could for example be realized by the fact that the density distribution of valence neutrons in very neutron rich and heavy nuclei are in radial , say @xmath11-direction , concentrated in a surface layer and that the density distribution in @xmath11-direction can be approximated by a frozen gaussian packet , whereas the motion of the neutrons within the layer is free . pursuing the picture to its extreme , one could imagine a slab of low density neutrons where in the transverse ( @xmath11 ) direction only a single @xmath12 level below the fermi energy is active . such situation also could eventually be relevant in the lasagne phase in neutron star crusts , where a low - density neutron gas may appear as a layer on the surface of the lasagne structure @xcite . in any case , one could imagine that in the outer layer of a neutron - rich nucleus , the pairing properties of the valence neutrons are more or less decoupled from the inert core part . for example , very neutron - rich sn isotopes with @xmath13 could represent such a scenario . a quasi-2d slab of low density neutron matter with the transverse degrees of freedom approximately frozen could then , as already mentioned , mimic the situation . we here insist on the 2d character because it is well known that the dimensionality has great influence on bound state formation and , thus , also on the formation of tight cooper pairs . such quasi-2d paired systems with low density also have been investigated in cold atom physics e.g. hydrogen atoms on a liquid @xmath14he surface and laser cooled atoms trapped in an planar potential @xcite . at any rate , and for the different reasons mentioned above , it seemed to us interesting to study an idealized situation of quasi-2d low density neutron matter realized within an infinitely extended slab . the study may turn out academic in the end but if future more detailed and not completely trivial investigations will exhibit a link to the surprisingly strong bound state features of cooper pairs in the nuclear surface , mentioned above , our effort will have been worth while . in this respect it is of interest to mention that , indeed , our study will show that in 2d matter a cooper pair attains a minimal size of about 2 fm whereas in 3d this value is over twice as large . in the present work , we investigate properties of neutron pairs in quasi-2d neutron systems by using finite - range effective nuclear forces . concerning the degree of freedom in the transverse @xmath11 direction perpendicular to the 2d plane , we for simplicity assume a gaussian packet as mentioned above . we study the two - neutron wave function in this model of quasi-2d system and show that two neutrons form a tightly bound state with a small size when the width @xmath15 of the gaussian in @xmath11-direction is small enough . we also investigate pairing properties of neutron matter based on the bcs theory . pairing gap and size of the cooper pair in quasi-2d neutron systems are analyzed , and bcs - bec crossover phenomena are discussed . the behaviour of @xmath8 pairing in infinite three - dimensional ( 3d ) matter has been extensively investigated based on the bcs theory @xcite for symmetric matter and neutron matter ( for example , refs . @xcite and references therein ) . as for a slab of nuclear matter , properties of @xmath8 pairing in symmetric nuclear matter have been studied in refs . @xcite and it was shown that the pairing gap increases at the surface of the slab . as for @xmath16 pairing , the gradual transition from bcs to bec involving the deuteron formation was predicted in dilute symmetric matter @xcite . competition of alpha and deuteron condensation was also discussed @xcite . there are many varieties of bcs calculations of nuclear matter concerning , e.g. , the treatment of single - particle spectra and the pairing interaction . for convenience , effective nuclear forces have been used in some calculations on pairing in nuclear matter instead of dealing with realistic bare nuclear forces . it was shown that the gogny force @xcite , a finite - range effective nuclear force , gives similar behaviour of nuclear matter pairing to that calculated from realistic nuclear forces with the brueckner - hf ( bhf ) approximation , as well as reasonable pairing properties in finite nuclei @xcite . therefore , we adopt the gogny d1s force in the present bcs calculations . we also use the minnesota force @xcite which reproduces features of @xmath17-wave @xmath18-@xmath18 scattering in @xmath16 and @xmath8 channels , and nucleus - nucleus scattering in the light nuclear region such as the @xmath4-@xmath4 system , as well as properties of light nuclei . the minnesota force is often used in structure studies of light neutron - rich nuclei . this paper is organized as follows . in the next section , we explain our simplified model of quasi-2d systems and formulation of the present work . the effective nuclear forces are described in section [ sec : interaction ] . in section [ sec : results ] , the results obtained for a two - neutron system and neutron matter are shown , and in section [ sec : discussion ] discussions are outlined . finally , we give a summary and outlook in section [ sec : summary ] . here we explain our framework of the simplified model of quasi-2d neutron system confined in a slab with a certain thickness . we first describe the wave function and the hamiltonian for a quasi-2d neutron matter system . we then explain the isolated two - neutron system and give the formalism for the quasi-2d system within the bcs theory . according to the outline of the introduction , we propose a model for a system of neutrons confined in a slab with a certain thickness . we assume that the neutrons in transverse ( @xmath11 ) direction are represented by a simple gaussian packet of size @xmath15 . as already mentioned , this size may mimic the concentration of the amplitudes of single particle wave functions in a surface layer , where @xmath11 corresponds to the radial direction in finite neutron - rich nuclei . therefore , the wave function for a single neutron is written as @xmath19 @xmath20,\ ] ] where @xmath21 indicates the coordinates in the slab , @xmath22 , and @xmath23 is the intrinsic spin . the thickness , i.e. , the rms width , of the slab is @xmath24 . in a similar way , a @xmath18-neutron wave function is given as @xmath25 here @xmath26 and @xmath27 are the antisymmetrized many - body wave functions . neutrons are interacting with each other via two - body nuclear forces . the interaction @xmath28 between two neutrons is the two - body effective nuclear force in the @xmath8 channel . here @xmath29 . because the @xmath17-wave neutron pair correlation is studied in the present work , only the central part of the effective nuclear force is considered . the gogny and minnesota forces are represented by a superposition of two gaussians , @xmath30.\label{eq : gogny}\ ] ] we freeze the the degree of freedom in transverse ( @xmath11 ) direction , as described in the introduction . by integrating over @xmath31 coordinates the three - dimensional schrdinger equation for @xmath26 is reduced to the two - dimensional equation for @xmath27 with respect to @xmath32 . then we get the following equation for quasi-2d neutron systems , @xmath33 @xmath34 where the 2d - kinetic term and the 2d - interaction term are @xmath35 @xmath36,\\ z_{b_m}(a)&\equiv&\left(\frac{b_m^2}{b_m^2 + 2a^2}\right)^{1/2 } \label{eq : strength2d}.\end{aligned}\ ] ] as seen from ( [ eq : v2d ] ) and ( [ eq : strength2d ] ) , the strength of the two - body interaction @xmath37 in the quasi-2d system depends on the thickness of the slab with the factor @xmath38 . the factor decreases with the increase of the width parameter @xmath15 . the thickness of the slab is only reflected in a weakening of the interaction for bigger slab sizes in our genuinely two - dimensional model . these are , of course , strongly simplified , may be oversimplified , assumptions to describe the detailed behaviour of the valence neutrons because the pauli effect from the core should be more complicated and a long tail of weakly bound neutrons may be not so simple . however , it is expected that such effects may effectively be taken into account by a modification of the width parameter @xmath15 . in the present work , we analyze properties of neutron pairs in quasi-2d neutron systems for various values of the slab size @xmath15 . for later comparison we consider the @xmath17-wave relative wave function @xmath39 for an isolated two - neutron system in the quasi-2d space . it is obtained by solving the schrdinger equation , @xmath40 @xmath41 is the spatial wave function normalized as @xmath42 , and @xmath43 is the intrinsic - spin wave function for the spin - zero channel . it is important to note that in two dimensions , a bound state can form at any arbitrarily small attraction @xcite . since the nuclear force in the @xmath0 channel is attractive at low energy , a two - neutron bound state in the quasi-2d space is obtained with an arbitrary width @xmath15 . we calculate the exact solution of the bound state with a gaussian expansion method . the size of the two - neutron bound state is given as @xmath44 please note that the size in 3d is given as @xmath45 . in this section , we investigate the behaviour of neutron pairs in quasi-2d infinite neutron matter in the slab by means of the bcs theory , which is equivalent to the hfb approximation in a homogeneous case @xcite . the application of bcs theory with the gogny force to a slab of nuclear matter was , e.g. , already performed in ref . @xcite . let us explain the equations for @xmath0 pairing in quasi-2d neutron matter . we apply the bcs theory to the 2d schrdinger equation ( [ eq : schrodinger2d ] ) . the @xmath17-wave pairing gap depends on @xmath46 and it is written as @xmath47 @xmath48 here @xmath49 and @xmath50 denote @xmath46 and @xmath51 , respectively , and @xmath52 is the pairing force , @xmath53.\ ] ] @xmath54 is the hartree - fock potential , @xmath55 where the exchange term is given as @xmath56.\ ] ] the occupation probability is @xmath57 and the chemical potential @xmath58 is determined so as to satisfy the number equation , @xmath59 where @xmath60 and @xmath61 are the density and the fermi momentum in 2d , respectively . in this paper , the @xmath61 dependence of pairing properties in neutron matter is discussed as a function of the ratio @xmath62 , where @xmath63 @xmath64 is the fermi momentum at the normal density of 3d symmetric nuclear matter . the energy per nucleon for the bcs state is @xmath652v^2_{k}+e_{pair},\ ] ] where the integrated pair energy @xmath66 is @xmath67 with @xmath68 the energy per nucleon for the hf state @xmath69 is determined from @xmath70 to analyze the spatial structure of a @xmath6-cooper pair it is useful to study the pair wave function in the coordinate space . it is defined by the fourier transform of the anomalous density @xmath71 , @xmath72 here @xmath73 , and @xmath74 is the normalization factor so that @xmath75 . the pair size ( coherence length ) @xmath76 is calculated from the root - mean - square distance of the pair wave function @xmath77 where @xmath78 in the low - density limit @xmath79 , the pair wave function @xmath80 and the energy @xmath81 approach @xmath82 and @xmath83 of the two - neutron bound state of section ii.b . as already mentioned , in this work we use the gogny d1s force @xcite and the minnesota force @xcite . in the @xmath8 channel , the gogny force is density independent and is represented by a superposition of two gaussians ( eq . ( [ eq : gogny ] ) ) . the parameters for the gogny d1s force are , @xmath84 the minnesota force is also given by a sum of finite - range gaussians , @xmath85 we adopt for the exchange - mixture parameter the value of @xmath86 . then the first ( second ) term gives the two - body force in the @xmath7 @xmath8 ( @xmath87 @xmath16 ) channel . in principle the gogny force represents an effective in medium force and the minnesota force a phenomenological nuclear force in free space . it has , however , been shown that the gogny d1s force gives similar behaviour for the @xmath0 pairing gap of symmetric nuclear matter as that calculated from realistic nuclear forces with the bhf+bcs approximation , and also describes pairing properties in finite nuclei @xcite . the minnesota force reasonably reproduces the @xmath17-wave @xmath18-@xmath18 scattering in the @xmath7 @xmath8 channel as well as the @xmath87 @xmath16 channel . it also well describes nucleus - nucleus scattering such as the @xmath4-@xmath4 system , as well as properties of light nuclei . because the minnesota force has no density - dependent term in the @xmath87 @xmath16 channel , it can not describe the saturation properties of symmetric nuclear matter . however , it gives a similar density dependence of neutron matter energy as the one obtained with the gogny d1s force , as shown later . the values of the @xmath88-@xmath88 scattering length @xmath89 in the @xmath0 channel obtained with the gogny d1s force and minnesota force are @xmath90 fm and @xmath91 fm , respectively . these values semiquantitatively agree with the experimental value @xmath3 fm @xcite . in fig . [ fig : phase ] , the calculated phase shifts of @xmath18-@xmath18 scattering in the @xmath0 channel are shown as well as those obtained with a realistic force . both the gogny d1s and minnesota forces show a reasonable reproduction of the @xmath0-wave scattering . figure [ fig : fig3d ] shows the properties of infinite 3d neutron matter calculated with the gogny d1s and minnesota forces by means of the bcs theory as well as from the hf calculations . figures [ fig : fig3d](a ) and [ fig : fig3d](b ) show the the energy ( @xmath92 ) and the integrated pair energy ( @xmath66 ) per nucleon , and the chemical potential ( @xmath58 ) of the bcs state are shown as well as the energy of the hf state ( @xmath93 ) for 3d neutron matter with the gogny d1s and minnesota forces , respectively . the density dependence of the neutron matter energy obtained with both forces is in reasonable agreement with that calculated with ab initio methods using realistic nuclear forces @xcite . in fig . [ fig : fig3d](c ) , the pairing gap of neutron matter is shown . the difference between the gogny d1s and minnesota forces is not large in the low - density region @xmath94 . in the @xmath95 region , the minnesota force gives a much smaller pairing gap than the gogny d1s force because it has a larger repulsive core with 108.15 mev hight at @xmath96 . let us make a comment on the pairing gap in 3d neutron matter investigated in different approaches . there are many calculations of the pairing gap in nuclear matter using realistic nuclear forces ( for reviews @xcite ) . in the standard bhf+bcs approximation , the pairing gap in 3d neutron matter is about 3 mev at the peak position @xmath97 ( figure 12 of ref . @xcite ) , which is consistent with the pairing gap calculated in bcs theory with the gogny force . in different approximations with medium polarization effects , a quenching of the pairing gap in neutron matter was suggested @xcite . the size @xmath98 of a cooper pair in 3d neutron matter is plotted as a function of @xmath62 in fig . [ fig : fig3d](d ) . as is already discussed in refs . @xcite , the pair size shrinking in 3d neutron matter is seen at finite low @xmath61 . similarly to the pairing gap , the difference of the pair size between the gogny and minnesota forces is small in the low - density region . in the @xmath99 region , the pair size increases more rapidly because of the weaker pairing with the minnesota force than is the case with the gogny d1s force . as we show later , also in quasi-2d neutron matter these two forces give qualitatively similar pairing behaviour for low @xmath61 . with both forces , the @xmath100 region ( @xmath101 is the average inter - neutron distance @xmath102 ) appears at finite low @xmath61 , i.e. , at low density . it shows the bcs - bec crossover feature as suggested by matsuo @xcite . = 7 cm = 6 cm in this section , two - neutron correlations in quasi-2d neutron systems are investigated by performing bcs calculations for 2d neutron matter with the gogny d1s and minnesota forces . the results for various values of @xmath15 of the confinement in the frozen direction ( @xmath11 ) are analyzed , and the behaviour of the cooper pair in quasi-2d neutron matter is discussed as a function of @xmath62 . by solving the schrdinger equation ( [ eq : schro - nn ] ) , the wave function @xmath41 of two isolated neutrons in a quasi-2d system is obtained . while a two - body bound state in 3d forms only at sufficiently strong attraction , a bound state in 2d can form , as already mentioned , at any arbitrarily small attraction . since the @xmath0 nuclear force is attractive at low energy , the two - neutron bound state is formed in quasi-2d systems with arbitrary width @xmath15 . the energy @xmath103 and the pair size @xmath104 of the bound state for @xmath105 , @xmath106 and @xmath107 fm are given in table [ tab : rmsr-2n ] . energy and size of the dineutron decrease as the width @xmath15 becomes small . in the quasi-2d system with a small width of @xmath105 fm , the dineutron size is as small as that of the deuteron in 3d . on the other hand , in the quasi-2d system with a width @xmath107 fm , the dineutron has a small binding energy , less than 0.1 mev , and a size larger than 20 fm . we , therefore , see that in our model the width of the slab has a strong influence on the two body correlations . it is very interesting that , although two neutrons are unbound in 3d , they form a tightly bound dineutron state with a size comparable to deuteron dimensions when they are confined in a sufficiently thin slab . to form the small - sized dineutron , the reduction of the dimension from three to two is crucial . in the present model , neutrons are confined in a gaussian packet with respect to the frozen @xmath11-direction . in slab direction , the potential part @xmath37 is given by almost the same form as the original effective nuclear force @xmath108 except for the factor @xmath38 in ( [ eq : v2d ] ) . this factor is smaller than one , depending on @xmath15 . for example , for the case of @xmath105 fm and the interaction range @xmath109 fm as with the long - range attractive term of the gogny d1s force , one gets @xmath110 , that is the strength of @xmath37 is smaller than that in the original interaction @xmath108 . let us remind the reader that the size of the dineutron state is determined by the balance of the kinetic energy and the potential energy . in a 2d system , the kinetic effect is weaker and a small - sized two - particle bound state can more easily be formed than in the 3d case . in fig . [ fig : wf-2n ] , the wave function @xmath41 and the probability density @xmath111 of two isolated neutrons in a quasi-2d system are shown as a function of the relative coordinate @xmath112 . in the quasi-2d system with @xmath105 and @xmath106 fm , the probability density has a peak around @xmath113 fm , which shows the significant spatial correlation between the neutrons . in a quasi-2d system with @xmath107 fm , the density distributes in a broad region , and the spatial correlation is much weaker . . [ tab : rmsr-2n ] energy @xmath83 and size @xmath114 of the two - neutron bound state @xmath41 in the quasi-2d system with the width parameter @xmath115 and 4 fm . the energy and the size @xmath116 for a deuteron in 3d space obtained with the minnesota force is also shown for comparison . [ cols="^,^,^,^,^ " , ] = 10 cm we now extend the bcs calculations to quasi-2d neutron matter with varying width parameters @xmath15 . = 6 cm = 10 cm the pairing gap at the fermi momentum @xmath117 as a function of @xmath62 is shown in fig . [ fig : gap - linear ] . the gogny d1s and minnesota forces give qualitatively similar results in the low @xmath61 region . at @xmath118 , the pairing gap is more suppressed in case of the minnesota force , because , as already mentioned , the short - range repulsion is larger than with the gogny force . for @xmath105 fm , the pairing gap obtained with the gogny d1s force has a peak about 4 mev high at @xmath119 , while that calculated with the minnesota force has a peak @xmath120 mev high at @xmath121 . with the increase of the width @xmath15 of the slab , the pairing gap @xmath122 is quenched because the attraction of the pairing force is weaker in a wider slab . in fig . [ fig : hfene ] , the energy ( @xmath92 ) and the integrated pair energy ( @xmath66 ) per nucleon , and the chemical potential ( @xmath58 ) of the bcs state are shown as well as the energy of the hf state ( @xmath93 ) calculated without pairing . compared with the energy of 3d neutron matter shown in fig . [ fig : fig3d ] , one of the striking features of quasi-2d neutron matter is that @xmath92 , @xmath66 , and the chemical potential have finite negative values even in @xmath123 limit . this is because , in the @xmath123 limit , the system goes to the strong coupling bec limit , where cooper pairs become identical with the two - particle bound state . in this limit , @xmath124 and @xmath125 become equal to the dineutron binding energy . as soon as the chemical potential turns negative , there is true binding between the two neutrons and the strong coupling bec feature is born out . therefore , in quasi-2d neutron matter one can , contrary to the 3d case , follow the transition from bec to the bcs regime . on the other hand , because of the same reason , the pairing gap @xmath122 in the quasi-2d neutron matter shows a linear behaviour at the low @xmath61 limit ( fig . [ fig : gap - linear ] ) , while the pairing gap disappears with a horizontal tangent at small @xmath61 in 3d neutron matter ( fig . [ fig : fig3d](c ) ) . = 10 cm = 6 cm the pair size @xmath76 in quasi-2d neutron matter is plotted as a function of @xmath62 in fig . [ fig : distance ] . in the @xmath123 limit , the pair size @xmath76 equals the size of the two - neutron bound state ( @xmath114 in ( [ eq : xinn ] ) ) . with the increase of @xmath61 , the pair size reduces first before it becomes large again with further increase of @xmath61 . thus the minimum pair size @xmath76 is found to be smaller than the size @xmath114 of the isolated two - neutron bound state in the quasi-2d system . this size reduction of the @xmath6-cooper pair indicates the enhancement of the dineutron correlation at finite low @xmath61 , i.e. due to the existence of a fermi sea . it should be pointed out that the pair size significantly depends also on the width @xmath15 of the slab . a striking difference of the pair size between the quasi-2d and 3d neutron matter is that the pair size at @xmath123 limit is finite in the quasi-2d neutron matter while it becomes infinite in the 3d neutron matter . in the 3d neutron matter , the cooper pair size @xmath98 has a minimum around @xmath126 , and it rapidly increases as @xmath61 goes 0 . interestingly , the size shrinking of the cooper pair in quasi-2d neutron matter has a close analogy with that of the deuteron in 3d symmetric matter predicted by lombardo and schuck @xcite . as mentioned above , the spatial correlation of cooper pairs enhances at finite low @xmath61 in quasi-2d neutron matter with a small width @xmath15 , where the bcs - bec crossover phenomena is expected . to discuss pairing properties from the point of view of the bcs - bec crossover , it is useful to compare the pair size @xmath76 with the average inter - neutron distance @xmath127 as discussed in the works for 3d neutron matter @xcite . the ratio @xmath128 in quasi-2d neutron matter is shown in fig . [ fig : rmsr - r ] as well as @xmath129 for the 3d neutron matter . strictly speaking , because the transition from the weak coupling bcs phase to strong coupling bec is known to be a crossover @xcite , there is no critical boundary of the transition . however , we consider that bcs - bec crossover features arise when the cooper pair size is smaller than the average inter - neutron distance by adopting the same criterion as in refs . @xcite . with this criterion , the bcs - bec crossover phenomena appears at @xmath130 in quasi-2d neutron matter with @xmath105 fm , and @xmath131 for @xmath106 fm , while in 3d neutron matter it is seen at @xmath132 . in the quasi-2d neutron matter , the ratio @xmath128 monotonically deceases as @xmath61 becomes small , and finally the system goes to the strong coupling bec limit . to reveal the features of bcs - bec crossover it is also useful to analyze the spatial structure of the @xmath6-cooper pair wave function as done in refs . figures [ fig : wf-2d - rpp ] and [ fig : wf - phi ] show the probability density @xmath133 and pair wave function @xmath134 of the cooper pair in the quasi-2d neutron matter . let us discuss the results for @xmath106 fm . at low @xmath61 such as @xmath135 and @xmath136 ( fig . [ fig : wf-2d - rpp](b4 ) and ( b3 ) ) , the spatial correlation of the cooper pair is strong and it indicates a bec - like dineutron pair . from @xmath135 to @xmath136 , the first peak becomes narrow due to the pauli principle , stemming from the other cooper pairs . consequently , the size shrinking of the dineutron occurs at a finite low @xmath61 , as mentioned . with further increase of @xmath61 , the nodal structure appears at @xmath137 which indicates the transition from the bcs - bec crossover to the bcs regime , and at @xmath138 the short - range correlation disappears as in the weak coupling bcs phase . the pair wave functions in quasi-2d neutron matter with @xmath106 fm are shown in fig . [ fig : wf - phi ] . we also show the two - neutron wave function @xmath41 for the dineutron bound state in the quasi-2d system and @xmath139 in 3d neutron matter . in the @xmath123 limit , the pair wave function @xmath134 becomes equal to the isolated two - neutron wave function @xmath41 of the dineutron state in case of quasi-2d neutron matter , while @xmath139 in 3d neutron matter becomes infinitely broad because there is no two - neutron bound state in a 3d system . = 16 cm = 12 cm in this section we tentatively associate the strong pairing correlations of valence neutrons in neutron - rich nuclei with the ones in a 2d system . it has recently become evident that in neutron - rich or neutron halo nuclei pairing correlations may show a strong maximum in the surface . this is for instance evidenced by the fact that the size of the cooper pairs has a sharp minimum in the surface . this feature has been shown to hold true in the case of a single cooper pair @xcite as well as in cases of many cooper pairs @xcite . though for the moment , it can not be excluded that the pronounced small cooper pair sizes in the nuclear surface is due to a somewhat trivial , i.e. geometrical effect@xcite , we here make the hypothesis that this is due to a real strong correlation effect , due to the presence of a surface and hence of a 2d influence . the valence nucleons may be confined in a surface shell for different reasons . for example , if valence neutrons are , in a heavy nucleus , distributed around a high lying principal shell , the corresponding density distribution , for instance if the shell does not contain an @xmath17 wave , may be concentrated in the surface region . even though low angular momentum orbits with radial nodes have some strength distribution in the inner region , their probability amplitudes may show peak structure in the surface , and therefore , the pairing is expected to be induced by the surface neutrons due to the possible mixing of many states having significant overlap of their radial part in the surface region . on the other hand it is also well known that finite size _ enhances _ pairing . this is already documented by the fact that the nuclear gap varies on average like @xmath140 mev , that is the presence of a surface certainly plays a role . this finite size enhancement is a quite subtle effect as theoretical investigations show @xcite and implies several competing effects . how much the scenario is influenced by quasi 2d features is still an open question but an interesting and suggestive hypothesis which we wanted to explore in this work . so one may consider that the strongly correlated cooper pairs with about a rms radius of 2 fm live in a surface sheet about @xmath10 fm across . in section [ sec : formulation ] we mimicked the transverse motion of the neutrons by a simple frozen gaussian @xmath141 . the motion perpendicular to @xmath11 is considered as planar . this is , of course , a highly idealized representation of the situation as mentioned several times . it may nevertheless capture some 2d features which possibly underly the strong pairing correlations in the neutron skin of neutron - rich nuclei . in the previous analysis of pair size in quasi 2d neutron matter in section [ sec : results ] , the bcs - bec crossover may occur only in dilute 2d matter for densities @xmath142 @xmath143 . let us estimate how many small sized cooper pairs can condensate in an area @xmath144 of the nuclear surface . for the dineutron condensation , more than one pair should exist at least . from the condition @xmath142 @xmath143 , two dineutrons can condensate in the surface area for @xmath145 fm , and four dineutrons for @xmath146 fm . recently , two - neutron correlations in neutron - skin nuclei have been investigated in light nuclei such as @xmath2he and @xmath147c @xcite as well as medium - heavy nuclei @xcite . @xmath2he is considered to be a neutron - skin nucleus having four valence neutrons surrounding an @xmath4 core , and therefore this nucleus is one of the simple examples to discuss neutron pair correlations in neutron skins . in refs . @xcite , it was argued that the @xmath2he ground state has spin - zero two - neutron correlations . from the above - mentioned discussion of quasi-2d neutron matter , the correlation in the ground state is expected to be of bcs - type rather than in the bcs - bec crossover region , because the valence neutron distribution concentrates in the surface with a distance @xmath148 fm , which is smaller than the bcs - bec crossover condition @xmath145 fm for the condensation of two dineutrons . instead of the ground state , the possible existence of an excited state @xmath2he(@xmath149 ) with a dilute gas - like state of dineutrons was suggested by one of the authors @xcite . according to the theoretical results of ref . @xcite , the dineutrons exist around the region with @xmath150 fm which satisfies the above - mentioned condition for the bcs - bec crossover . therefore , this state could have bec - like dineutrons at the surface . in medium - heavy nuclei , the spatial structure of spin - zero neutron pairing at the surface was investigated based on hfb calculations in refs . @xcite . to discuss two - neutron correlations in the neutron skin in connection with quasi-2d neutron matter , one should analyze the radial and angular behaviour separately and focus on the angular correlations ( opening angle ) of neutron pairing , as well as on the size of the cooper pairs . the transition from bcs phase to bcs - bec crossover is characterized by the change of the pair wave function from a weak - coupling nodal structure to an enhanced single bound state like peak . if such a structure change is found in the angular behaviour it may be an indication of the bcs - bec crossover , while a clear evidence may not be seen in the radial structure because it might be affected by pauli effect from constituent neutrons of the core . the spatial structure of spin - zero neutron pairing is displayed in ref . @xcite , which shows that the nodal structure along the angular direction is seen inside from the surface and it disappears on the outer side of the surface . more detailed analysis of the angular behaviour is requested . on the other hand , a good measure of the correlations also is the size of the cooper pairs . their rms radius seems systematically very small in the outer part of the nuclear surface @xcite , that is much smaller than the smallest value in homogeneous 3d matter . whether this also always goes along with a maximum in the pair correlations at the same place in a finite nucleus , remains to be seen . let us also comment on other effects omitted in the present model of infinite quasi-2d neutron matter . at the surface of a realistic finite nuclear system , valence neutrons feel the centrifugal force and the spin - orbit force which may work so as to break @xmath6-cooper pairs . periodicity condition on the spherical surface is another effect to be considered . two - neutron correlations have extensively been investigated in two - neutron halo nuclei by means of three - body models of two neutrons and a core @xcite . recently , it was suggested that the internal two - neutron wave function analyzed as a function of the distance @xmath151 from the nuclear center shows behaviour similar to the pair wave function of 3d neutron matter in the bcs - bec crossover region @xcite . in the analysis of ref . @xcite , the inter - neutron wave function has a nodal structure at small @xmath151 , and it shows enhancement of the spatial correlations with increasing @xmath151 . moreover , in a further analysis of the neutron pair in @xmath1li , size shrinking was found at the surface @xcite . it was found that there the pair is as small as 3 fm at @xmath152 fm @xcite . the formation of the small sized dineutron at the surface of @xmath1li may be understood with the picture of a quasi-2d neutron system . as shown in ref . @xcite with a three - body model , the spin - zero component of two neutrons shows a strong angular correlation and its distribution concentrates at the surface with @xmath153 fm . the situation corresponds to the quasi-2d neutron where the valence neutrons are confined at the surface region and the neutron pair is formed due to the angular correlation . it may be interpreted as the reduction of the dimension from three to two , being crucial for the small - sized neutron pair in two - neutron halo nuclei . at the small @xmath151 region , i.e. near the core , the size of the neutron pair increases and nodal structure appears in the relative wave function in the three - body model . this is because the pauli effect from the constituent neutrons of @xmath154li core raises the fermi momentum of the valence neutrons . this behaviour is similar to the weakening of the pair correlation in the bcs region at @xmath155 in the quasi-2d neutron matter . in spite of the analogous behaviour of the neutron pair in @xmath1li to that of the cooper pair in the quasi-2d neutron matter , one should keep in mind that two - neutron halo nuclei has only one pair of neutrons and the phenomena there is not regarded as the cooper pair condensation in the bcs picture . several investigations have , however , revealed that the features of a single cooper pair and the ones of a cooper pair embedded in several others are not drastically different . it is also interesting to consider two - proton correlation in proton - rich nuclei in analogy to the two - neutron correlation in neutron - rich nuclei . while the energy of infinite proton matter diverges without charge neutrality , proton - rich matter such as proton halo or proton skin structure can exist in finite nuclei . moreover , since two - proton decays are allowed from some proton - rich nuclei such as @xmath156be,@xmath2c,@xmath157o and @xmath158ne @xcite , more direct information of @xmath8 pair correlation could be experimentally obtained than for the case of neutron - rich nuclei . in the same way as neutron systems , we calculate a wave function @xmath159 for the two - proton bound state in quasi-2d systems . coulomb force ( @xmath160 ) is added to the two - body effective nuclear forces . in fig . [ fig : wf-2p ] , the obtained two - proton wave functions in the @xmath0-wave channel are shown in comparison with those of the dineutron state . because of the coulomb force , the spatial correlation of the @xmath8 pair is slightly more suppressed in the two - proton system than in the two - neutron system . however , the wave functions for @xmath105 fm and @xmath106 fm still have a pronounced peak . it suggests that the spatial correlation of the @xmath8 pair also is present in proton systems . singlet @xmath17 ( @xmath0 ) @xmath6 correlations in quasi-2d neutron systems were studied in a simplified model of neutrons confined in a slab with a certain thickness . the finite - range effective nuclear forces , the gogny d1s and minnesota forces , were adopted for the study . the strength of the two - body interaction in the quasi-2d systems depends on the width @xmath15 of the slab . namely , the interaction is stronger for thinner slabs than for wider ones . when the width of a slab is small enough , e.g. around @xmath105 fm , two neutrons form a tightly bound dineutron with a size as small as the deuteron size . the behaviour of the @xmath6-cooper pair in the quasi-2d neutron matter was investigated by means of the bcs theory . in quasi-2d neutron matter , the transition region of the bcs - bec crossover appears at @xmath130 and for @xmath161 fm . the size shrinking of the cooper pair occurs at finite low @xmath61 because of the pauli effect . it is also important to note that , as the width @xmath15 of the slab decreases , the pair correlation in 2d is enhanced and the cooper pair size becomes small because of the stronger attraction in the thinner slab . this is one of the characteristics different from 3d . in particular , the reduction of the dimension from 3d to 2d is crucial to form the strongly correlated neutron pair . the chemical potential becomes negative for @xmath162 and , therefore , there is true two - neutron binding in quasi-2d neutron matter and a genuine bec regime is born out . the relation of the quasi-2d neutron matter to finite neutron - rich nuclei was also discussed . several studies of the past have revealed that the size of cooper pairs changes very substantially from inside to outside of a nucleus @xcite . the subject has been taken up again more recently and the radius dependence of a neutron - neutron ( @xmath6 ) cooper pair has been investigated more systematically @xcite . it thereby turned out that the old believe that the size of a cooper pair is of the order of the diameter of a large nucleus must be revisited when considered locally , i.e. , as a function of the nuclear radius . namely , detailed bcs calculations with the d1s gogny force have shown that the rms radius of an @xmath6-cooper pair , though indeed large in the interior , becomes very small in the outer surface of a nucleus , where the density is low , before expanding again when leaving the nucleus . the minimum extension of a cooper pair attains a size as small as 2 fm , that is the same as the size of a deuteron , a bound state . although the variation of the pair radius as a function of the density qualitatively follows the one of infinite matter , a factor of two reduction of the pair diameter with respect to the homogeneous case is a quite surprisingly strong effect which calls for an explanation . we considered this pair size shrinking in the surface of neutron - rich nuclei with the present calculations of quasi-2d neutron systems , based on the hypothesis that the valence neutrons in a neutron - rich nucleus live in radial direction in a surface sheet about @xmath10 fm across because the wave functions of the active shell are essentially concentrated in the surface while the nucleons are free to move within the sheet . the idea behind this scenario is that the neutrons in the surface sheet are essentially constrained to a two - dimensional world which could give a certain clue to the smallness of the cooper pairs , since in two dimensions even an isolated @xmath6 pair forms a bound state . pursuing this idea , we further idealized the situation assuming a planar slab of neutrons where for simplicity the motion perpendicular to the slab is frozen into a gaussian wave packet in order to simulate the surface concentration of the valence neutrons in radial direction . freezing the perpendicular motion means that the neutrons are two - dimensional in the direction of the slab . the thickness of the slab only influences the strength of the pairing force , as already mentioned above . we , indeed , find that for reasonable slab thicknesses of @xmath163 fm , the pair radius is of the order of 2@xmath1643 fm in agreement with realistic 3d hfb calculations @xcite . we are , of course , aware of the fact that our model may be strongly oversimplified and further studies are required in order to ascertain whether a 2d aspect indeed helps to interpret the occurrence of astoundingly small cooper pairs in the nuclear surface . one also should keep in mind that a local picture may be a little too simple in the sense that the pair size certainly fluctuates a lot . nevertheless , a classical picture often grasps a good deal of reality , even in so tiny quantum object as nuclei . neutron pairs much smaller than the dimension of a nucleus may approximately be considered as bosons and a strong coupling bose - einstein condensation scenario of @xmath6 pairs may be realized in the surface of neutron - rich nuclei . bound state like cooper pairs in the nuclear surface may also allow to reinterpret neutron transfer reactions into superfluid nuclei which are known to have a strongly enhanced cross section @xcite . 2d aspects may also be relevant in the lasagne phase in neutron star crust . in practical application to such systems , one should extend the model by taking into account finite temperature and also excitation in the @xmath11-direction for higher neutron density . bcs and bec problems of two - dimensional fermion systems have been discussed also in the field of condensed matter physics in relation to electron systems and atomic systems @xcite . phenomena of bec and bcs - bec crossover in quasi-2d systems of trapped gases have been discussed in recent works @xcite , and it has been suggested that the condensation in a quasi-2d gas is sensitive to the frequency @xmath165 of the confinement in the `` frozen '' direction . this feature may have an analogy to the present results for the quasi-2d neutron matter . the authors would like to thank k. hagino , h. horiuchi , n. pillet , g. rpke , h. sagawa and n. sandulescu for valuable discussions on nuclear pairing and cooper pairs in the nuclear surface . fruitful discussions during the yipqs international molecule workshop held at yukawa institute for theoretical physics ( yitp ) in october 2008 were useful to complete this work . the computational calculations of this work were performed by supercomputers at yitp and also by supercomputer projects of high energy accelerator research organization ( kek ) . this work was supported by grant - in - aid for scientific research from japan society for the promotion of science ( jsps ) . it is also supported by the grant - in - aid for the global coe program `` the next generation of physics , spun from universality and emergence '' from the ministry of education , culture , sports , science and technology ( mext ) of japan . 0 i.tanihata et al . , * 55 * , 2676 ( 1985 ) . p. g. hansen and b. jonson , europhys . lett . * 4 * , 409 ( 1987 ) . g. f. bertsch and h. esbensen , annals phys . * 209 * , 327 ( 1991 ) . m. v. zhukov , b. v. danilin , d. v. fedorov , j. m. bang , i. j. thompson and j. s. vaagen , phys . * 231 * , 151 ( 1993 ) . h. esbensen , g. f. bertsch and k. hencken , phys . c * 56 * , 3054 ( 1997 ) . a538 * , 355c ( 1992 ) . s. aoyama , k. kat , and k. ikeda , prog . suppl . * 142 * , 35 ( 2001 ) ; t. myo , s. aoyama , k. kat , and k. ikeda , prog . . phys . * 108 * , 133 ( 2002 ) . t. myo , s. aoyama , k. kat and k. ikeda , prog . phys . * 108 * , 133 ( 2002 ) . t. myo , s. aoyama , k. kat and k. ikeda , phys . * b576 * , 281 ( 2003 ) . k. hagino and h. sagawa , phys . c * 72 * , 044321 ( 2005 ) . t. myo , k. kat , h. toki , k. ikeda , phys . c * 76 * , 024305 ( 2007 ) . et al . _ , lett . * 70 * , 730 ( 1993 ) . d. sackett _ et al . _ , c * 48 * , 118 ( 1993 ) . s. shimoura _ et al . _ , . lett . * b348 * , 29 ( 1995 ) . m. zinser _ _ , nucl . phys . * a619 * , 151 ( 1997 ) . t. nakamura , _ et al . lett . * 96 * , 252502 ( 2006 ) . y. kanada - enyo , phys . c * 76 * , 044323 ( 2007 ) . n. itagaki , m. ito , k. arai , s. aoyama and t. kokalova , phys . c * 78 * , 017306 ( 2008 ) . k. hagino , n. takahashi and h. sagawa , phys . c * 77 * , 054317 ( 2008 ) . m. matsuo , k. mizuyama and y. serizawa , phys . c * 71 * , 064326 ( 2005 ) . n. pillet , n. sandulescu and p. schuck , phys . c * 76 * , 024310 ( 2007 ) . m. baldo , j. cugmon , a. lejeune and u. lombardo , nucl . phys . * a515 * , 409 ( 1990 ) . t. takatsuka and r. tamagaki , prog . theor . phys . suppl . * 112 * , 27 ( 1993 ) . f. v. de blasio , m. hjorth - jensen , o. elgaroy , l. engvik , g. lazzari , m. baldo and h. j. schulze , phys . c * 56 * , 2332 ( 1997 ) . d. j. dean and m. hjorth - jensen , rev . * 75 * , 607 ( 2003 ) . m. baldo , e. e. sapershtein and s. v. tolokonnikov , nucl . * a749 * , 42 ( 2005 ) . m. matsuo , phys . c * 73 * , 044309 ( 2006 ) . j. margueron , h. sagawa and k. hagino , phys . c * 76 * , 064316 ( 2007 ) . a. a. isayev , phys . c * 78 * , 014306 ( 2008 ) . g. f. de tramond and b. gabioud , phys . c * 36 * , 691 ( 1987 ) . a. b. migdal , yad . fiz . * 16 * , 427 ( 1972 ) ; sov . * 16 * , 238 ( 1973 ) . h. bando and a. kuriyama , prog . extra number , 69 ( 1968 ) . m. baldo , u. lombardo and p. schuck , phys . rev . c * 52 * , 975 ( 1995 ) . g. ropke , a. schnell , p. schuck and p. nozieres , phys . lett . * 80 * , 3177 ( 1998 ) . u. lombardo , p. nozieres , p. schuck , h. j. schulze and a. sedrakian , phys . c * 64 * , 064314 ( 2001 ) . a. tohsaki , h. horiuchi , p. schuck and g. ropke , phys . lett . * 87 * , 192501 ( 2001 ) . t. yamada and p. schuck , phys . c * 69 * , 024309 ( 2004 ) . y. funaki , a. tohsaki , h. horiuchi , p. schuck and g. ropke , phys . c * 67 * , 051306(r ) ( 2003 ) . p. ring and p. schuck , _ the nuclear many - body problem _ ( springer - verlag , new york , 1980 ) . d.m . brink and r.a . broglia , _ nuclear superfluidity _ ( cambridge university press , new york , 2005 ) . i. tanihata , d. hirata , t. kobayashi , s. shimoura , k. sugimoto and h. toki , phys . lett . * b289 * , 261 ( 1992 ) . t. suzuki _ et al . _ , phys . lett . * 75 * , 3241 ( 1995 ) . a. ozawa , et al . , nucl.phys . * a691 * , 599(2001 ) . k. hagino , h. sagawa and p. schuck , arxiv:0812.0658 [ nucl - th ] . d. g. ravenhall , c. j. pethick and j. r. wilson , phys . rev . lett . * 50 * , 2066 ( 1983 ) . c. p. lorenz , d. g. ravenhall and c. j. pethick , phys . lett . * 70 * , 379 ( 1993 ) . k. oyamatsu , nucl . phys . * a561 * , 431 ( 1993 ) . g. watanabe , k. sato , k. yasuoka and t. ebisuzaki , phys . c * 66 * , 012801(r ) ( 2002 ) ; phys . c * 68 * , 035806 ( 2003 ) . a. i. safonov , s. a. vasilyev , i. s. yasnikov , i. i. lukashevich and s. jaakkola , phys . * 81 * , 4545 ( 1998 ) . d. s. petrov , m. holzmann , and g.v.shlyapnikov , phys . lett . * 84 * , 2551 ( 2000 ) . j. bardeen , l. n. cooper and j. r. schrieffer , j. bardeen , l. n. cooper and j. r. schrieffer , phys . * 108 * 1175 ( 1957 ) . m. baldo , u. lombardo , e. saperstein and m. zverev , phys . * b459 * , 437 ( 1999 ) . m. farine and p. schuck , * b459 * , 444 ( 1999 ) . m. baldo , m. farine , u. lombardo , e. e. saperstein , p. schuck and m. v. zverev , eur . j. a * 18 * , 17 ( 2003 ) . s. s. pankratov , m. baldo , u. lombardo , e. e. saperstein and m. v. zverev , nucl . * a811 * , 127 ( 2008 ) . u. lombardo and p. schuck , phys . rev . c * 63 * , 038201 ( 2001 ) . j. decharge and d. gogny , phys . c * 21 * , 1568 ( 1980 ) . j. f. berger , m. girod , and d. gogny , comput . . commun . * 63 * , 365 ( 1991 ) . h. kucharek , p. ring , p. schuck , r. bengtsson and m. girod , phys . * b216 * , 249 ( 1989 ) . e. garrido , p. sarriguren , e. moya de guerra and p. schuck , phys . c * 60 * , 064312 ( 1999 ) . d. r. thompson , m. lemere and y. c. tang nucl . a286 * , 53 ( 1977 ) . l. d. landau and e. m. lifshitz , _ quantum mechanics : non relativistic theory _ , course of theoretical physics vol . * 3 * ( pergamon , new york , 1989 ) . a. akmal , v. r. pandharipande and d. g. ravenhall , phys . c * 58 * , 1804 ( 1998 ) . j. carlson , j. . j. morales , v. r. pandharipande and d. g. ravenhall , phys . c * 68 * , 025802 ( 2003 ) . schulze , j. cugnon , a. lejeune , m. baldo , and u. lombardo , phys . b375 * , 1 ( 1996 ) . u. lombardo , p. schuck and w. zuo , phys . c * 64 * , 021301(r ) ( 2001 ) . c. shen , u. lombardo , p. schuck , w. zuo and n. sandulescu , phys . c * 67 * , 061302(r ) ( 2003 ) . u. lombardo , p . schuck and c. shen , nucl . a731 * , 392 ( 2004 ) . l. g. cao , u. lombardo and p. schuck , phys . c * 74 * , 064301 ( 2006 ) . v. g. j. stoks , r. a. m. klomp , c. p. f. terheggen and j. j. de swart , phys . c * 49 * , 2950 ( 1994 ) . m. randeria , j .- m . duan , and l .- y . shieh , phys . lett . * 62 * , 981 ( 1989 ) ; phys . b * 41 * , 327 ( 1990 ) . v. m. loktev , r. n. quick and s. g. sharapov , phys . rep . * 349 * , 1 ( 2001 ) . f. pistolesi and g. c. strinati , phys . rev . b * 53 * , 15168 ( 1996 ) . j. r. engelbrecht , m. randeria and c. a. r. sade melo , phys . rev . b * 55 * , 15153 ( 1997 ) . t. papenbrock and g. f. bertsch , phys . c * 59 * , 2052 ( 1999 ) . p. nozieres and s. schmitt - rink , j. low . temp . phys . * 59 * 195 ( 1985 ) . k. hagino , h. sagawa , j. carbonell and p. schuck , phys . lett . * 99 * , 022506 ( 2007 ) . n. sandulescu , private communication . m. farine , f. w. j. hekking , p. schuck and x. vias , phys . b * 68 * , 024507 ( 2003 ) . d. baye , nucl . * a627 * , 305 ( 1997 ) . k. arai , y. ogawa , y. suzuki and k. varga , prog . suppl . no.*142 * , 97 ( 2001 ) . p. descouvemont , e. m. tursunov and d. baye , nucl . * a765 * , 370 ( 2006 ) [ arxiv : nucl - th/0512016 ] . l. v. grigorenko , r. c. johnson , i. g. mukha , i. j. thompson and m. v. zhukov , phys . c * 64 * , 054002 ( 2001 ) . f. c. barker , phys . c * 68 * , 054602 ( 2003 ) . i. mukha _ et al . _ , phys . c * 77 * , 061303(r ) ( 2008 ) . w. von oertzen , a. vitturi , rep . . phys . * 64 * , 1427 ( 2001 ) .

measurement quantum mechanics , the theory of a quantum system which undergoes a measurement process , is introduced by a loop of mathematical equivalencies connecting previously proposed approaches . the unique phenomenological parameter of the theory is linked to the physical properties of an informational environment acting as a measurement apparatus which allows for an objective role of the observer . comparison with a recently reported experiment suggests how to investigate novel interesting regimes for the quantum zeno effect .

cond-mat9603182

the description of the measurement process has been a topic debated from the early developments of quantum mechanics @xcite . besides being a basic issue in the interpretation of the quantum formalism it has also a practical interest in predicting the results of experiments pointing out some paradoxical aspects of the quantum laws when compared to the classical ones . the development of devices whose noise figures are close to the quantum limit makes these experiments within the technological feasibility and demands for a systematization of the theory of quantum measurement with deeper understanding of its experimental implications @xcite . in ordinary quantum mechanics measurements are taken into account by postulating the wave function collapse @xcite . this approach has the unpleasant feature of introducing an extra assumption in the theory which , moreover , regards only instantaneous and perfect measurements . in the last twenty years relevant steps have been made in upgrading the von neumann postulate with more realistic and satisfactory approaches . these approaches essentially recognize that a measured system is not isolated but in interaction with a measurement apparatus . the way this basic fact is taken into account in the evolution law of the measured system has been developed according to different languages and points of view . a group algebra approach to the problem of an open quantum system was proposed by postulating ( completely positive ) semigroup properties for the dynamical law of a system in interaction with a markovian environment @xcite . in this case the open system is described by a semigroup master equation later successfully used in modeling quantum optics experiments @xcite . the semigroup master equation , still preserving positivity and trace of the density matrix operator , introduces decoherentization by dynamically quenching the off - diagonal elements of the density matrix , a key property used to explain the absence of superposition states in a measurement apparatus @xcite or in a general macroscopic system @xcite thought as systems interacting with an environment . open quantum systems were gathered to measured systems by obtaining an all alike semigroup master equation for a continuous measurement process @xcite modeled by repeated instantaneous effect - valued measurements @xcite , i.e. , partial localization or decoherentization kicks @xcite given to the density matrix at random values of the measured observable . in the model @xcite the semigroup master equation comes out only after averaging the instantaneous results of the measured observable with probability distribution in agreement with standard quantum mechanics @xcite . in absence of this average , i.e. , for a particular _ selection _ of the measurement outcomes , a nonlinear stochastic differential equation was proposed to describe the evolution of the density matrix during a continuous measurement process @xcite . since in this case during the measurement process the density matrix coincides with its square , a nonlinear stochastic differential equation could be derived for the corresponding wave function @xcite crowning previous attempts to include the lders postulate @xcite , a generalization of the von neumann one to _ selective _ measurements , into a stochastic schrdinger equation @xcite . this equation , obtained also in the framework of the quantum filtering theory @xcite , is a special case of the more general quantum state diffusion equation for open systems @xcite . its nonlinearity can be avoided only renouncing to make predictions on the outcome of the measured observable . indeed , a linear stochastic differential equation was proposed for the so called _ a posteriori _ states @xcite , i.e. , those unnormalized states which describe the quantum system when the measurement result is already known . as originally suggested by feynman @xcite the path integral approach to quantum mechanics is a quite appropriate framework for incorporating the effect of a measurement . the hint was picked up by modeling a continuous measurement of position having a certain result ( selective measurement ) by a restriction of the feynman path integral @xcite . the method was extended to measurements of a generic observable , function of momentum and position , by a restriction of the quantum propagator in the phase - space formulation @xcite . the restriction was originally proposed as a gaussian functional damping the paths proportionally to the time - averaged squared difference between the value taken by the observable on the paths and the measurement result , normalized to a given variance @xcite . however , the corresponding quantum propagator has desirable dynamical semigroup properties if the gaussian restriction is chosen linear in time @xcite which is equivalent to make the previously proposed variance scaling with the inverse of the measurement time @xcite . the modifications introduced in the quantum propagator by the restricted path integral approach can be incorporated into an effective hamiltonian depending on the selected measurement result . an effective wave equation was then derived which is linear in the wave function if the selected measurement result is considered known @xcite . on the other hand , if the selected measurement result is to be determined according to the evolution of the measured system the effective wave equation is read as a nonlinear stochastic differential equation . equivalence to the previously proposed stochastic equations occurs @xcite . the selective constraint imposed to the restricted path integral approach can be removed by summing over all possible measurement outcomes . in this way a nonselective process described by a density matrix was obtained @xcite . the effect of the interaction with the measurement apparatus modifies the corresponding quantum propagator through an influence functional @xcite which turns out to be equivalent to those obtained with other methods @xcite . from the above incomplete list of approaches , apart from an evident difference in the languages , a rather unified picture of the problem of quantum measurement emerges . it deserves uprising to the systematic theory of a quantum system evolving under the effect of a measurement process . such theory , named , for brevity , measurement quantum mechanics , is formally presented in section ii . a less formal introduction to measurement quantum mechanics is attainable through the analysis of a model of measurement device which is rather general . this is important not only as a justification of the formal approach to the theory but , above all , for clearly defining the meaning of a measurement in relation to the observer . as noted by cini @xcite paradoxical features arise from not considering the objective role of the observer in a measurement process . the evolution of a measured system in interaction with a measurement apparatus can not depend on the observer looking or not at the pointer of the apparatus . in order to take into account such objectivity the measurement apparatus must be classical with respect to the observer @xcite . in section iii we propose a simple model of measurement apparatus shaped as an environment of particles linearly interacting with the measured system and in contact with a heat reservoir at a fixed temperature @xcite . in the high temperature limit the particles behave as an informational environment which extracts information in objective way . noticeably , the formal structure of measurement quantum mechanics is obtained in the same limit . a further analysis of measurement quantum mechanics is given in section iv in connection to a recent optical experiment @xcite showing quantum zeno effect . comparison of the experimental results with the theoretical predictions indicate that the experiment @xcite was performed in a regime of very strong coupling of the measurement apparatus with the measured system . repetition of this experiment or similar ones in a weaker coupling regime is desirable for investigating interesting quantum features . some final remarks are given in section v. in this section we show the mathematical equivalence of the five approaches to the problem of measurement in quantum systems briefly described in the introduction . the demonstration is organized in steps relating neighboring pairs of approaches and giving rise to the equivalence loop sketched in fig . no one of these equivalence steps is novel . however , we reconsider all them in a unified framework and language with naturally emerging definitions for concepts as selective and nonselective measurements as well as _ a priori _ and _ a posteriori _ analysis of a measurement process . as a result we get a theory which takes into account the effect of a measurement process in ordinary quantum dynamics through a phenomenological parameter coupling the measured system to the measurement apparatus . we enter the loop of fig . 1 at the group algebra approach to the master equation and go on in the clockwise sense . ( 314,314)(-40,-260 ) = 0.75pt ( 97.0,-45.0)(120,90 ) ( -53.3,-153.5)(120,90 ) ( 4.7,-329.0)(120,90 ) ( 189.3,-329.0)(120,90 ) ( 246.3,-153.5)(120,90 ) ( 64.7,-30.0)(4,3)14.4 ( 64.7,-30.0)(-4,-3)14.4 ( 249.3,-30.0)(-4,3)14.4 ( 249.3,-30.0)(4,-3)14.4 ( 7.7,-205.5)(-1,3)5.7 ( 7.7,-205.5)(1,-3)5.7 ( 306.3,-205.5)(1,3)5.7 ( 306.3,-205.5)(-1,-3)5.7 ( 157.0,-314.0)(1,0)18.0 ( 157.0,-314.0)(-1,0)18.0 ( 157.0,-157.0)(0,0)*mqm * the dynamics of a system interacting with an environment is conveniently described in terms of a reduced density matrix operator @xmath0 obtained by tracing out the environment variables from the density matrix operator of the whole system + environment . the unitary evolution of @xmath0 for the isolated system is modified to an irreversible one by the interaction with the environment . in the limit of a markovian environment a dynamical law described by a completely positive semigroup has been postulated @xcite resulting in the following master equation for the reduced density matrix operator @xmath1 + { 1\over2 } \sum_\nu \left ( \left [ \hat l_\nu(t ) \hat \rho(t ) , \hat l_\nu(t)^{\dag } \right ] + \left [ \hat l_\nu(t ) , \hat \rho(t ) \hat l_\nu(t)^{\dag } \right ] \right ) \label{lindbeq}\ ] ] where @xmath2 is the hamiltonian operator for a general nonautonomous system and @xmath3 are the lindblad operators representing the influence of the environment on the system . the above equation is thought to describe the general case of an open quantum system . the evolution of a quantum system subjected to a measurement process is a particular case where the environment is the measurement apparatus and the lindblad operators are proportional to the measured quantities . if , for simplicity , we consider the measurement of a single observable represented by the hermitian operator @xmath4 , the corresponding lindblad operator can be chosen as @xmath5 . note that we allow for an explicit time dependence of @xmath6 in order to include the case of general observables such as the continuous quantum nondemolition ones @xcite . the function @xmath7 has dimensions @xmath8=[t^{-1 } a^{-2}]$ ] and represents , as we shall see in the next section , the coupling of the measured system to the measurement apparatus . the measurement process is continuous in time and the measurement coupling is time dependent as requested in the description of a general experimental situation . due to the presence of commutators in the right hand side of ( [ lindbeq ] ) the trace of the reduced density matrix operator is a conserved quantity which we assume to be unity @xmath9 we shall see later , when introducing selective measurements , that the reduced density matrix operator @xmath10 corresponds to an incoherent mixture of pure states associated to selective processes . in antithesis , the process described by @xmath10 is called nonselective and eq . ( [ trace ] ) can be interpreted as a normalization relation for the probability distribution over the selective processes . according to this interpretation , the result of a nonselective measurement , @xmath11 , and its associated variance , @xmath12 , can be evaluated by the trace rules @xmath13 \label{avens}\ ] ] and @xmath14 , \label{var}\ ] ] respectively . here overlining is used to denote the statistical average over the selective measurement results @xmath15 to be defined in the following together with the corresponding probability distribution . the equivalence of the master equation ( [ lindbeq ] ) to a density matrix propagator expressed in terms of an influence functional is our first step in the loop of fig . 1 . by moving to the coordinate representation of the reduced density matrix operator @xmath16 the corresponding evolution equation becomes the partial differential equation @xmath17 ^ 2 \bigg ] \rho(q_1,q_2,t ) \end{aligned}\ ] ] which can be transformed into the integral equation @xmath18 the two - point green function ( density matrix propagator ) @xmath19 has a phase - space path - integral representation which can be derived by standard methods @xcite . for a small time interval @xmath20 one has @xmath21 ) and inserting at the appropriate places the four identities @xmath22 where @xmath23 we get @xmath24 \delta t \right . \nonumber \\ & & ~-{1\over2 } \kappa(t '' - \delta t ) \left [ a \left(p_1^{(1)},q_1^{(1)},t '' - \delta t \right ) - a \left(p_2^{(1)},q_2^{(1)},t '' - \delta t \right ) \right]^2 \delta t \nonumber \\ & & ~\left . - { i\over \hbar } \left [ p_2^{(1 ) } { q_2 '' - q_2^{(1 ) } \over \delta t } -h \left(p_2^{(1)},q_2^{(1)},t '' - \delta t \right ) \right ] \delta t \right ) \rho(q_1^{(1)},q_2^{(1)},t '' - \delta t ) + { \cal o}(\delta t^2 ) . \end{aligned}\ ] ] by iterating this relation @xmath25 times with @xmath26 and then taking the limit @xmath27 , a functional measure arises @xmath28~ d[q]_{q',t'}^{q'',t '' } = \lim_{n \to \infty } \prod_{n=1}^{n } { dp^{(n ) } \over 2 \pi \hbar } \prod_{n=1}^{n-1}dq^{(n)}\ ] ] with boundary conditions imposed only to the @xmath29 paths . by comparison with eq . ( [ inteq ] ) we conclude @xmath30~ d[q_1]_{q_1',t'}^{q_1'',t '' } \int d[p_2]~ d[q_2]_{q_2',t'}^{q_2'',t''}~ \exp \left ( { i\over \hbar } s[p_1,q_1 ] - { i\over \hbar } s[p_2,q_2 ] ~- z [ p_1,q_1,p_2,q_2 ] \right ) \label{green2}\end{aligned}\ ] ] where @xmath31 = \int_{t'}^{t '' } dt~ [ p \dot q - h(p , q , t ) ] \label{s}\ ] ] @xmath32 = { 1 \over 2 } \int_{t'}^{t '' } dt~ \kappa(t ) \left [ a(p_1,q_1,t ) - a(p_2,q_2,t ) \right]^2 . \label{iphase}\ ] ] the effect of the measurement in the two - point green function is represented by the functional @xmath33 which reduces to the identity for @xmath34 , i.e. , in absence of a measurement process . the functional @xmath33 is the feynman - vernon influence functional @xcite evaluated by tracing out the degrees of freedom of the environment in the phase - space path - integral formulation of quantum mechanics . as we shall see in the next section , the influence functional approach allows us to give an enlightening interpretation of the nonselective measurement processes with an explicit expression of the parameter @xmath7 . the second step in the equivalence loop of fig . 1 is accomplished by a formal manipulation of the influence functional @xcite . by using the identity @xmath35 ^ 2 \right ) = \nonumber \\ & & ~ \int d[a]~ \exp \left ( - \int_{t'}^{t '' } dt~ \kappa(t ) [ a(p_1,q_1,t ) - a(t)]^2 - \int_{t'}^{t '' } dt~ \kappa(t ) [ a(q_2,p_2,t ) - a(t)]^2 \right ) \label{identity}\end{aligned}\ ] ] where the functional measure arises by slicing the interval @xmath36 $ ] into @xmath27 subintervals at times @xmath37 with @xmath26 , i.e. , @xmath38 = \lim_{n \to \infty } \prod_{n=1}^{n } da^{(n ) } \sqrt{2 \kappa(t^{(n ) } ) \delta t \over \pi } , \label{d[a]}\ ] ] the two - point green function @xmath19 can be decomposed into a couple of one - point green functions @xmath39}$ ] @xmath40~ ~g_{[a]}(q_1'',t'';q_1',t ' ) ~g_{[a]}(q_2'',t'';q_2',t')^ * \label{g2ing1a}\ ] ] where @xmath41}(q'',t'';q',t ' ) = \int d[p ] d[q]_{q',t'}^{q'',t '' } \exp \left ( { i \over \hbar } s[p , q ] - \int_{t'}^{t '' } dt~ \kappa(t ) [ a(p , q , t ) - a(t)]^2 \right ) . \label{g[a]}\ ] ] this one - point green function is the phase - space generalization of the restricted path - integral approach originally proposed on physical grounds as a damping of the feynman paths incompatible with the measurement result @xcite . here the measurement result is the function @xmath15 which , according to the interpretation of the functional measure @xmath42 $ ] , is , in general , continuous but not differentiable . if we assume that the measured system is in a pure state @xmath43 at the beginning of the measurement process , i.e. , @xmath44 eq . ( [ g2ing1a ] ) allows us to introduce a pure state @xmath45}(t ) \rangle$ ] which is the evolution at time @xmath46 of the initial one @xmath43 . indeed , the reduced density matrix can be thought as the functional integral @xmath47 ~\psi_{[a]}(q_1,t ) \psi_{[a]}(q_2,t)^ * \label{decoa}\ ] ] with the wave function of the pure state defined by the one - point green function @xmath39}$ ] @xmath48}(q'',t '' ) = \int dq'~ g_{[a]}(q'',t'';q',t ' ) ~ \psi(q',t ' ) . \label{psiag}\ ] ] the decomposition ( [ g2ing1a ] ) introduces the announced statistical description of a nonselective measurement process in terms of selective processes corresponding to different measurement results @xmath15 and characterized by a certain probability distribution . due to ( [ decoa ] ) the conservation of the trace of the reduced density matrix operator gives @xmath49~ \parallel \psi_{[a]}(t ) \parallel^2 . \label{proba}\ ] ] therefore , @xmath50}(t ) \parallel^2 $ ] is the probability distribution for the selective process with measurement result @xmath51 $ ] . once the probability distribution is known , the nonselective measurement result ( [ avens ] ) and variance ( [ var ] ) which , according to ( [ decoa ] ) , are written also as @xmath52 ~ \langle \psi_{[a]}(t ) | \hat a(t ) |\psi_{[a]}(t ) \rangle \label{avea1}\ ] ] and @xmath53 ~ \langle \psi_{[a]}(t ) | \left ( \hat a(t ) - \overline{a(t ) } \right)^2 |\psi_{[a]}(t ) \rangle , \label{vara1}\ ] ] can be expressed in terms of the selective results @xmath15 by @xmath52~ \parallel \psi_{[a]}(t ) \parallel^2 a(t ) \label{avea2}\ ] ] and @xmath54~ \parallel \psi_{[a]}(t ) \parallel^2 \left ( a(t ) - \overline{a(t ) } \right ) \left ( a(t-\tau ) - \overline{a(t ) } \right ) , \label{var2}\ ] ] respectively . the equivalence between ( [ avea1 ] ) and ( [ avea2 ] ) as well as that one between ( [ vara1 ] ) and ( [ var2 ] ) can be demonstrated explicitly by using the definition of the functional measure @xmath42 $ ] and the functional expression of the wave function @xmath55}(q , t)$ ] . the prescription @xmath56 in ( [ var2 ] ) is used to avoid the divergence which occurs by integrating the term @xmath57 . the third step of our equivalence loop , namely the existence of a differential equation for the wave function @xmath55}(q , t)$ ] , follows from the same standard method used for the reduced density matrix . the explicit representation of the functional integral giving the one - point green function @xmath39}$ ] allows us to write the difference of the wave function between two close times @xmath58 and @xmath59 up to terms @xmath60 and find the differential equation @xmath61}(q , t ) = \left [ -{i \over \hbar } h \left ( -i \hbar { \partial \over \partial q},q , t \right ) - \kappa(t ) \left [ a \left ( -i \hbar { \partial \over \partial q},q , t \right ) - a(t ) \right]^2 \right ] \psi_{[a]}(q , t ) . \label{psia}\ ] ] the mathematical properties of this differential equation crucially depend on two possible ways of analyzing a selective measurement process . a first possibility is to consider an analysis _ a posteriori_. the result of the measurement @xmath15 is already known and one wants to complete the quantum mechanical information on the measured system by evaluating the associated wave function @xmath55}(q , t)$ ] . in this case ( [ psia ] ) is an effective wave equation linear in the wave function . this equation is deterministic but not regular , in general , depending upon the nature of the function @xmath15 . the effect of the measurement appears as an anti - hermitian term added to the hamiltonian of the unmeasured system . due to this term the norm of the wave function @xmath55}(q , t)$ ] is not conserved in agreement with the probabilistic normalization ( [ proba ] ) . a second possibility is to consider an analysis _ a priori_. the result of the measurement @xmath15 is to be predicted not deterministically but randomly with probability distribution @xmath50}(t ) \parallel^2 $ ] . equation ( [ psia ] ) becomes a stochastic differential equation nonlinear in @xmath55}(q , t)$ ] . this can be seen more explicitly by introducing an appropriate noise @xmath62}(t)$ ] which relates the selective result @xmath15 to the expectation value of the operator @xmath6 in the state @xmath63}(t ) \rangle$ ] . comparison of ( [ avea1 ] ) and ( [ avea2 ] ) allows us to write @xmath64}(t ) | \hat a(t ) |\psi_{[a]}(t ) \rangle \over \langle \psi_{[a]}(t ) | \psi_{[a]}(t ) \rangle } + \eta_{[a]}(t ) \label{aintoa}\ ] ] where @xmath62}(t)$ ] is defined by @xmath65~ \parallel \psi_{[a]}(t ) \parallel^2 \eta_{[a]}(t ) = 0 .\ ] ] the nonlinearity of the stochastic equation for @xmath55}(q , t)$ ] is made clear by inserting ( [ aintoa ] ) into ( [ psia ] ) . this stochastic equation is also unconventional in the sense that it contains the square of the noise term @xmath62}(t)$ ] . it is evident that eq . ( [ psia ] ) is fully appropriate in the _ a posteriori _ analysis of a selective measurement process . however , in the _ a priori _ analysis one has to deal with an unconventional stochastic equation containing a complicated noise function defined in terms of a time - dependent probability distribution . it would be more desirable to have a standard stochastic equation with a white noise . this can be achieved , together with the fourth step in our equivalence loop , by a change of variable in the measurement result . analogously to eq . ( [ aintoa ] ) we write @xmath66}(t ) + { \xi(t ) \over 2 \sqrt{\kappa(t ) } } \label{aintoaxi}\ ] ] where @xmath67}(t)$ ] is considered known for the moment and @xmath68 is a noise whose characterization is obtained through the following considerations . the two - point green function already decomposed in ( [ g2ing1a ] ) into a couple of one - point green functions @xmath39}$ ] with functional measure @xmath42 $ ] can alternatively be decomposed into another couple of one - point green functions @xmath69}$ ] @xmath70 ~g_{[\xi]}(q_1'',t'';q_1',t ' ) ~g_{[\xi]}(q_2'',t'';q_2',t')^ * \label{g2ing1xi}\ ] ] with gaussian measure @xmath71 = \lim_{n \to \infty } \prod_{n=1}^{n } d\xi^{(n ) } \sqrt{\delta t \over 2 \pi } \exp \left ( -{1\over2 } { \xi^{(n)}}^2 \delta t \right ) \label{d[xi]}\ ] ] which makes @xmath69}$ ] exponentially linear in @xmath72 @xmath73}(q'',t'';q',t ' ) = \exp \left(\int_{t'}^{t '' } dt~ { 1\over4 } \xi(t)^2 \right ) g_{[a]}(q'',t'';q',t ' ) = \int d[p ] d[q]_{q',t'}^{q'',t '' } \nonumber \\ & & ~ \times \exp \left ( { i \over \hbar } s[p , q ] - \int_{t'}^{t '' } dt~ \kappa(t ) \left [ [ a(p , q , t)- { a}_{[\xi]}(t)]^2 - 2 [ a(p , q , t)- { a}_{[\xi]}(t ) ] { \xi(t ) \over 2 \sqrt{\kappa(t ) } } \right ] \right ) .\end{aligned}\ ] ] with respect to the gaussian measure @xmath74 $ ] , the noise @xmath68 has average @xmath75~\xi(t ) = 0 \label{ximedio}\ ] ] and covariance @xmath76~ \xi(t_1)~\xi(t_2 ) = \delta(t_1-t_2 ) , \ ] ] i.e. , it is a white noise . as in the previous case , the reduced density matrix can be written as a functional integral @xmath77 ~\psi_{[\xi]}(q_1,t ) \psi_{[\xi]}(q_2,t)^ * \label{decoxi}\ ] ] over pure states with wave functions @xmath78}(q , t)$ ] defined by @xmath79}(q'',t '' ) = \int dq'~g_{[\xi]}(q'',t'';q',t ' ) ~ \psi(q',t ' ) .\ ] ] according to the expression of the green function @xmath69}$ ] , the integral statement for @xmath78}(q , t)$ ] can be transformed into an operatorial expression by the usual slicing procedure @xmath80}(t ) \right]^2 \nonumber \\ & + & \sqrt{\kappa(t ) } \left [ \hat a(\hat p,\hat q , t ) - { a}_{[\xi]}(t ) \right ] \xi(t ) \bigg ] \bigg ) ~ |\psi_{[\xi]}(t ' ) \rangle \label{stocsol}\end{aligned}\ ] ] where @xmath81 means chronological ordering . by introducing the wiener process @xmath82 with ito algebra @xmath83 and @xmath84 ( @xmath85 ) , eq . ( [ stocsol ] ) is recognized as the formal solution of the following ito stochastic differential equation @xcite @xmath86}(t ) \rangle & = & \left [ - { i \over \hbar } \hat h(\hat p,\hat q , t ) - { 1 \over 2 } \kappa(t ) \left [ \hat a(\hat p,\hat q , t)- { a}_{[\xi]}(t ) \right]^2 \right]~ |\psi_{[\xi]}(t ) \rangle ~dt \nonumber \\ & & ~ + \sqrt{\kappa(t ) } \left [ \hat a(\hat p,\hat q , t ) - { a}_{[\xi]}(t ) \right ] ~|\psi_{[\xi]}(t ) \rangle~dw(t ) . \label{stoceq}\end{aligned}\ ] ] in ( [ stoceq ] ) we still have to specify the function @xmath67}(t)$ ] . according to the decomposition ( [ decoxi ] ) , the conservation of the trace of the reduced density matrix operator gives @xmath87 ~\langle \psi_{[\xi]}(t ) | \psi_{[\xi]}(t ) \rangle .\ ] ] since the gaussian measure @xmath74 $ ] is normalized , up to a zero - average fluctuation we have @xmath88}(t ) | \psi_{[\xi]}(t ) \rangle = 1 $ ] . the normalization of the state @xmath89}(t ) \rangle$ ] uniquely determines the expression of @xmath67}(t)$ ] . indeed , by evaluating the ito differential @xmath90}(t ) | \psi_{[\xi]}(t ) \rangle = 2 \sqrt{\kappa(t ) } ~ \langle \psi_{[\xi]}(t ) dw(t ) + { \cal o}(dt^{3/2})\ ] ] we see that the norm of the state @xmath89}(t ) \rangle$ ] is conserved only if we impose @xmath91}(t ) = \langle \psi_{[\xi]}(t ) | \hat a(t ) |\psi_{[\xi]}(t ) \rangle . \label{nlxi}\ ] ] in this case the stochastic differential equation ( [ stoceq ] ) becomes the nonlinear quantum state diffusion equation proposed in @xcite . unlike the effective wave equation ( [ psia ] ) , the quantum state diffusion equation ( [ stoceq ] ) is suitable for the _ a priori _ analysis of the selective processes but awkward when dealing with the _ a posteriori _ analysis . indeed , in the latter case one should guess the noise realization @xmath68 giving rise , according to ( [ aintoaxi ] ) and ( [ nlxi ] ) , to the known value of the measurement result @xmath15 . as in the decomposition of the reduced density matrix in terms of the pure states @xmath45}(t ) \rangle$ ] , the nonselective measurement result ( [ avens ] ) and variance ( [ var ] ) can be expressed in terms of the pure states @xmath89}(t ) \rangle$ ] by @xmath92 ~ \langle \psi_{[\xi]}(t ) | \hat a(t ) |\psi_{[\xi]}(t ) \rangle \label{avexi}\ ] ] and @xmath93 ~ \langle \psi_{[\xi]}(t ) | \left ( \hat a(t ) - \overline{a(t ) } \right)^2 |\psi_{[\xi]}(t ) \rangle , \label{varxi}\ ] ] respectively . as fifth and final step , the equivalence loop of fig 1 is closed by regaining the starting differential equation for the reduced density matrix operator latexmath:[\[\hat \rho(t ) = \overline { overlining means functional integration with measure @xmath74 $ ] . by using ito algebra we get @xmath95}(t ) \rangle ~ \langle \psi_{[\xi]}(t ) | + d | \psi_{[\xi]}(t ) \rangle ~ d \langle \psi_{[\xi]}(t ) | } \nonumber \\ & = & \left ( - { i \over \hbar } \left [ \hat h(t ) , \hat \rho(t ) \right ] - { 1 \over 2 } \kappa(t ) \left [ \hat a(t ) , \left [ \hat a(t ) , \hat \rho(t ) \right ] \right ] \right ) dt + { \cal o}(dt^{3/2})\end{aligned}\ ] ] which is eq . ( [ lindbeq ] ) with the previously discussed choice for the lindblad operators . deeper insight into the problem of measurement in quantum mechanics can be gained by completing the formal discussion of the previous section with general but explicit models for a measurement process . for this purpose the influence functional approach is a quite appropriate one and some attempts have already been done in this direction @xcite . here we discuss the measurement process in terms of an informational environment made of particles linearly interacting with the measured system and in contact with a heat reservoir at fixed temperature . each particle selectively measures the evolution of the system while the collection of them represents an apparatus performing a nonselective measurement . this model has two nice features . firstly , it is simple from a mathematical point of view and the corresponding influence functional can be evaluated exactly @xcite . secondly , very clear and controllable approximations can be made on the influence functional in order to reduce it to the formal expression given in the previous section . these approximations are physically related to the fact that the informational environment can be defined as a many - body system in which each particle , still interacting quantum mechanically with the measured system , behaves classically from the point of view of the observer . only in this case the decision of the observer to look or not to look at the pointer of the instrument does not influence the result of the measurement itself @xcite . we start by considering the hamiltonian of a measured system interacting with an environment @xmath96 where @xmath97 is the hamiltonian of the measured system with phase - space coordinates @xmath98 and @xmath99 and @xmath100 is the hamiltonian of the interacting environment . the environment is made by different sets of particles with mass @xmath101 and phase - space coordinates @xmath102 and @xmath103 , where @xmath104 labels the sets and @xmath105 the particles in each set . the particles of each set @xmath104 interact linearly with an observable @xmath106 of the system through a function @xmath107 which transduces a displacement of the observables @xmath108 into displacements of the coordinates @xmath102 @xmath109 ^ 2 \right ) . \label{henv}\ ] ] the interaction of the @xmath105th particle of the environment with the @xmath104th observable is characterized by a proper angular frequency @xmath110 . in terms of spectral response we can say that each particle measures the fourier component of @xmath111 at its proper angular frequency @xmath110 . we assume that at time @xmath112 the system is described by the density matrix @xmath113 and each particle of the environment is at thermal equilibrium with temperature @xmath114 . since the equilibrium value for the coordinates @xmath102 depends on the coordinates @xmath98 and @xmath99 , the assumption of thermal equilibrium at time @xmath112 implies a correlation between the environment and the system at the same instant of time @xcite . the density matrix for the total system at time @xmath112 can be written as @xmath115 where @xmath116 and @xmath117 .\ ] ] the initial displacements @xmath118 are uniquely expressed in terms of the values assumed by the measured observable at the initial points @xmath119 and @xmath120 by requiring translational and time reversal invariance @xcite . the steps required to get the expression of the influence functional are standard @xcite . at time @xmath121 the reduced density matrix of the system , obtained by tracing out the coordinates of the environment in the total density matrix @xmath122 can be obtained by propagating the initial density matrix @xmath113 @xmath123 by the two - point green function @xmath124 d[q_1]_{q_1',t'}^{q_1'',t '' } \int d[p_2 ] d[q_2]_{q_2',t'}^{q_2'',t '' } \exp \left ( \frac{i}{\hbar } s[p_1,q_1 ] - \frac{i}{\hbar } s[p_2,q_2 ] \right ) f[p_1,q_1,p_2,q_2]\end{aligned}\ ] ] where @xmath125 $ ] is given by ( [ s ] ) and the influence functional @xmath126 is @xmath127 = \int dq '' \int dq_1 ' dq_2 ' \int d[p_1 ] d[q_1]_{q_1',t'}^{q'',t '' } \int d[p_2 ] d[q_2]_{q_2',t'}^{q'',t '' } \nonumber \\&&~\times \exp \left ( \frac{i}{\hbar } s_{env}[p_1,q_1,a_1 ] - \frac{i}{\hbar } s_{env}[p_2,q_2,a_2 ] \right ) \rho_{env } ( \delta q_1',\delta q_2',t ' ) \label{inf}\end{aligned}\ ] ] with @xmath128 = \int_{t'}^{t '' } dt~ \left ( \sum_{\nu n } p_{\nu n } \dot q_{\nu n } - h_{env}(p , q-\lambda a , t ) \right ) .\ ] ] in the above formulas the notation @xmath129 , @xmath130 , @xmath131 is a shortening for @xmath132 , @xmath133 , @xmath134 and @xmath135 , @xmath136 for @xmath137 , @xmath138 . the integrations in ( [ inf ] ) are gaussian and can be performed giving @xmath139 = \prod_{\nu } \exp \bigg ( - \int_{t'}^{t '' } dt \int_{t'}^{t } ds ~ \lambda_\nu(t ) \left [ a_{\nu}(p_1,q_1,t ) - a_{\nu}(p_2,q_2,t ) \right ] \big ( \alpha_\nu(t - s ) \lambda_\nu(s ) \nonumber \\ & & \times \left [ a_{\nu}(p_1,q_1,s ) - a_{\nu}(p_2,q_2,s ) \right ] + i \beta_\nu(t - s ) { d \over ds } \left ( \lambda_\nu(s ) \left [ a_{\nu}(p_1,q_1,s ) + a_{\nu}(p_2,q_2,s ) \right ] \right ) \big ) \bigg ) \label{gfi}\end{aligned}\ ] ] where the two kernels @xmath140 \label{flu1}\ ] ] and @xmath141 \label{dis}\ ] ] describe fluctuation and dissipation phenomena , respectively @xcite . now we turn the attention to the requirement that the environment is informational , i.e. , classical with respect to the observer so that the readout of information through the coordinates @xmath130 has an objective value . since a classical system is one whose quantized structure can not be appreciated we must impose the thermal fluctuations to be large in comparison to the quanta of the environment @xmath142 in this case we have @xmath143 . \label{flu2}\ ] ] memory effects in the fluctuation and dissipation kernels are an inessential complication which can be avoided by assuming an ensemble of particles in the environment with a continuous spectrum of frequencies . if , for simplicity , we choose the same frequency density @xmath144 for each set @xmath104 we get @xmath145 \simeq \int_{0}^{\infty } d\omega { d n_\nu \over d \omega } \omega^2 \cos[\omega ( t - s ) ] = \omega \delta ( t - s ) .\ ] ] of course , the above assumption of continuous frequency spectrum implies that the condition ( [ classic ] ) can not be satisfied in the whole frequency range @xmath146 by a finite temperature . therefore , we should assume the condition ( [ classic ] ) to be valid only in a finite range @xmath147 $ ] which contains the most significant part of the fourier spectrum of @xmath148 and is relevant in monitoring the measured observables . in this case the condition ( [ classic ] ) allows us to neglect the dissipation term , proportional to ( [ dis ] ) , with respect to the fluctuation one , proportional to ( [ flu2 ] ) . in conclusion , the influence functional for an informational environment can be written as @xmath149 = \prod_{\nu } \exp \left ( - \frac12 \int_{t'}^{t '' } dt ~\kappa_\nu(t ) \left [ a_{\nu}(p_1,q_1,t ) - a_{\nu}(p_2,q_2,t ) \right]^2 \right ) \label{fvernon}\ ] ] with @xmath150 equations ( [ fvernon ] ) and ( [ kappone ] ) generalize to more than one monitored quantity the expression of the functional already introduced in ( [ iphase ] ) for nonselective measurements . note that @xmath151 can be written as @xmath152 where @xmath153 and @xmath154 . the two parameters @xmath155 and @xmath156 can be made indefinitely small by increasing the density of meters ( particles ) in the informational environment proportional to @xmath157 , or their temperature @xmath114 , respectively . these characteristic length and time play the role of similar parameters introduced _ ad hoc _ in the spontaneous localization @xcite and in the restricted path - integral @xcite approaches . we close this section by evaluating the measurement result arising as a readout from the informational environment and its corresponding variance . the observer reads the result of the nonselective measurement of the @xmath104th observable by looking at a pointer which responds to the coordinates of the measurement apparatus . this response is a weighted sum over the proper frequencies of the informational environment and is characterized by a normalized response function @xmath158 such that the pointer displacement and its variance are @xmath159 @xmath160 where @xmath161 \label{tr1}\ ] ] @xmath162 . \label{tr2}\ ] ] due to the condition ( [ classic ] ) the thermal relaxation time of the measurement apparatus turns out to be much smaller than the characteristic timescales of the measured system . this allows us to use an adiabatic approximation for the total density matrix operator and factorize it as @xmath163 where @xmath10 is the reduced density matrix operator of the system which takes into account the influence of the informational environment and @xmath164 is the density matrix operator of the informational environment at thermal equilibrium around the instantaneous value of the measured observables . in the representation of the environment coordinates @xmath164 has matrix elements @xmath165 which are operators with respect to the system coordinates through the displacements @xmath166 .\ ] ] in this case the traces in ( [ tr1 ] ) and ( [ tr2 ] ) contain gaussian integrals over the environment coordinates which can be performed giving @xmath167 = \lambda_\nu(t ) { \rm tr } \left [ \hat a_\nu(t ) \hat \rho(t ) \right]\ ] ] @xmath168 = { \hbar \over 2 m \omega_{\nu n } } \coth \left ( { \hbar \omega_{\nu n } \over 2 k_b t } \right ) + \lambda_\nu(t)^2 { \rm tr } \left [ \hat a_\nu(t)^2 \hat \rho(t ) \right ] .\ ] ] in the limit ( [ classic ] ) and for a continuous spectrum of frequencies with the previously chosen density we have @xmath169 @xmath170 where @xmath171 and @xmath172 are given by ( [ avens ] ) and ( [ var ] ) , respectively . up to the transduction factor @xmath173 and @xmath174 , respectively , eq.s ( [ p ] ) and ( [ dp ] ) are the result of the measured observable @xmath175 and its variance read from the pointer . the result of the measurement is the nonselective outcome of the observable @xmath175 obtained from measurement quantum mechanics . the measurement variance is the sum of a classical variance associated to the measurement apparatus and the quantum variance associated to the measured system . the nonselective result and variance ( [ p ] ) and ( [ dp ] ) can be expressed in terms of selective processes . the decomposition of @xmath171 and @xmath172 has already been discussed in the previous section . concerning the decomposition of the variance associated to the measurement apparatus we note that the nonselective measurement performed by the informational environment is achieved by summing over the frequencies of the particles in the environment . therefore , the contribution to the measurement variance given by the selective process at frequency @xmath176 is @xmath177 as expected from the equipartition theorem . equation ( [ dp ] ) sets the definition of classical and quantum measurements . the first ones are those characterized by a dominance of the variance associated to the measurement apparatus , i.e. , by a dominance of the thermal ( or brownian ) noise . in the second ones the quantum variance of the measured system is larger than the thermal noise . this second case promises a richer variety of experimental scenarios because the measurement variance depends on the preparation of the measured system and/or the strength of the coupling to the measurement apparatus . in many experimental situations one is dealing with an average of measurements on an individual quantum system each time prepared in the same initial state or a single measurement on an ensemble of independent identical quantum systems with the same initial conditions . in both cases or in a combination of them averaged measurement results , instead of individual measurement results , are actually registered as outcome of the experiment . according to the discussion of section ii we have two ways for theoretically reproducing the outcome of such an experiment . we can consider _ a priori _ selective measurements and obtain the experiment outcome by averaging the corresponding selective results in the sense of eq . ( [ avexi ] ) . we can also consider a nonselective measurement and evaluate the experiment outcome directly from eq . ( [ avens ] ) . the choice between the two methods may depend on the particular problem one is faced to . the method based on a nonselective process is more direct but it is based upon the solution of a master equation for the reduced density matrix operator which could be much more difficult than solving many times the the diffusion state equation ( [ stoceq ] ) and averaging the selective measurement results . in this section we will deal with an experimental situation which can be described by a simple 2 by 2 density matrix . the choice of the direct method based on nonselective measurements is , therefore , the natural one . let us consider a system with time - independent hamiltonian and discrete energy spectrum @xmath178 the evolution of the system subjected both to an external time - dependent perturbation @xmath179 and to a continuous nonselective measurement of the observable represented by the operator @xmath180 is given by @xmath181 - { 1 \over 2 } \kappa(t ) \left [ \hat a , \left [ \hat a , \hat \rho(t ) \right ] \right ] .\ ] ] an interesting situation is attained in the case the external perturbation @xmath182 stimulates the system to make transitions among the unperturbed levels @xmath105 and the occupancy of some of these levels is measured . at what extent does the measurement disturb the stimulated transitions ? eventually , inhibition of the stimulated transitions due to the occupancy measurement occurs and one has an example of what is called quantum zeno effect @xcite . from a theoretical point of view a two - level system is the simplest one for studying the interplay between the effects of the measurement process and of the external perturbation . this situation has been also experimentally investigated @xcite . an ensemble of about 5000 @xmath183 ions was stored in a penning trap . two hyperfine levels of the ground state of @xmath183 , created by a static magnetic field and hereafter called levels 1 and 2 , were driven by a radiofrequency , resonant between levels 1 and 2 , turned on for @xmath184 ms . the amplitude of the radiofrequency was adjusted to make the initially vanishing occupancy of level 2 to be unity at time @xmath114 in absence of other disturbances . during the radiofrequency pulse , @xmath105 optical pulses of length @xmath185 ms and frequency equal to the transition frequency between level 1 and a third level 3 were also applied . the number of photons emitted in the spontaneous transition @xmath186 , the transition @xmath187 being forbidden , was roughly proportional to the occupancy of level 1 . the optical pulses acted , therefore , as a measurement of the occupancy of level 1 . the occupancy at time @xmath114 of level 1 was observed to be frozen near its initial unity value , i.e. , the stimulated transition @xmath188 in the period @xmath114 to be inhibited , proportionally to the number @xmath105 of optical pulses . before trying a direct interpretation of the experiment @xcite , let us consider the case of a continuous measurement of the occupancy of level 1 in a two - level system subjected to stimulated transitions . in the representation of the unperturbed hamiltonian @xmath189 where @xmath190 we assume a perturbation @xmath182 with matrix elements @xmath191 and @xmath192 with @xmath193 real and @xmath194 . in the same representation the matrix elements of the measured occupancy of level 1 are @xmath195 and @xmath196 . the master equation for the reduced density matrix operator then gives @xmath197 \nonumber \\ \dot{\rho}_{22}(t)&= & - { i\over \hbar } \left [ v_{21 } \rho_{12}(t ) - \rho_{21}(t ) v_{12 } \right ] \\ \dot{\rho}_{12}(t)&= & \left [ -{i\over\hbar } ( e_1-e_2 ) - { \kappa \over 2 } \right ] \rho_{12}(t ) - { i\over \hbar } \left [ v_{12 } \rho_{22}(t ) - \rho_{11}(t ) v_{12 } \right ] \nonumber\end{aligned}\ ] ] and @xmath198 . the measurement coupling @xmath199 is assumed constant . by summing and subtracting the first two equations we find @xmath200 \rho_{12}(t ) - { i\over \hbar } v_{12 } \left ( \rho_{22}(t ) - \rho_{11}(t ) \right ) . \nonumber\end{aligned}\ ] ] the first equation is the conservation of the trace of the reduced density matrix operator in the case of a nonselective measurement , @xmath201 . by defining @xmath202 \qquad \rho_{22}(t ) - \rho_{11}(t ) = \gamma(t)\ ] ] with @xmath203 , @xmath204 and @xmath205 real , the other two equations give @xmath206 where we have introduced the rabi angular frequency @xmath207 the above system has a simple solution for @xmath208 ( resonance ) and in this case we get @xmath209 \label{rho11 } \\ \rho_{12}(t)&= & e^ { -\frac{i}{\hbar}(e_1-e_2)t -{1\over4 } \kappa t -i\omega t_0 } \bigg [ { \rm re } \left(\rho_{12}(0 ) e^{i \omega t_0}\right ) e^ { -{1\over4 } \kappa t } \nonumber \\ & & + i { \rm i m } \left ( \rho_{12}(0 ) e^{i \omega t_0}\right ) \left ( \cos(wt ) - \frac{\kappa}{4w } \sin(wt ) \right ) - i \left ( \rho_{22}(0)-\rho_{11}(0)\right ) \frac{\omega_r}{2w}\sin(wt ) \bigg ] \label{rho12}\end{aligned}\ ] ] where @xmath210 the angular frequency @xmath211 coincides with the rabi angular frequency @xmath212 when @xmath213 . in this case the effect of the measurement disappears and the system oscillates between levels 1 and 2 with angular frequency @xmath212 . in the opposite limit of strong measurement coupling the frequency @xmath211 is imaginary and an overdamped regime is achieved in which transitions are inhibited . the border between the two regimes is at @xmath214 corresponding to a critical measurement coupling @xmath215 the behavior of the measured occupancy of level 1 which , according to eq . ( [ avens ] ) , turns out to be @xmath216 is shown in fig . 2 for different values of the measurement coupling starting from initial conditions @xmath217 and @xmath218 . the transition between the rabi - like regime and the zeno - like regime is marked by the disappearing of the oscillatory behavior at @xmath219 . the quantum variance associated to the nonselective measurement of level 1 can be evaluated according to eq . ( [ var ] ) and is @xmath220 . it vanishes in the limit of strong measurement coupling and oscillates with angular frequency @xmath212 between 0 and @xmath221 in the opposite limit @xmath222 . at the critical measurement coupling @xmath219 , after a short transient of the order of @xmath223 , the quantum variance approaches the constant and maximum value @xmath224 . it is worth to note that an identical behavior of the reduced density matrix operator is obtained if we consider @xmath189 as the measured quantity . in this case eq.s ( [ rho11 ] ) and ( [ rho12 ] ) still hold with the substitution @xmath225 where @xmath226 is the measurement coupling associated to the measurement of @xmath189 . this observation allows us to compare fig . 2 with fig . 1 of ref . @xcite where the quantum zeno effect for the same two - level system investigated here was analyzed _ a posteriori _ in the case of a single selective measurement of energy . as expected , quantitative differences between nonselective measurements and single selective measurements are obtained but in both cases zeno inhibition occurs for strong measurement coupling . the behavior shown in fig . 2 agrees also with that found in @xcite where the quantum zeno effect is analyzed within the quantum trajectory approach @xcite . the discussion relative to fig . 2 can be made general . in the limit @xmath227 the influence of the measurement disappears and ordinary quantum mechanics is recovered . in the limit @xmath228 the influence of the measurement dominates and the average measurement result ( [ avens ] ) as well as the corresponding variance ( [ var ] ) are frozen into the values assumed at the beginning of the measurement . in the case of fig . 2 these values are @xmath217 and @xmath229 , respectively . moreover , in the limit @xmath228 the off - diagonal elements of the reduced density matrix operator vanish and a classical behavior is obtained . in order to recover the results of the experiment @xcite the previous analysis for a continuous nonselective measurement with @xmath199 constant must be generalized to a series of measurement pulses spaced by intervals of no measurement . according to the experimental procedure we consider @xmath105 nonselective measurements of the occupancy of level 1 with coupling @xmath199 during the intervals @xmath230 $ ] , @xmath231 , with @xmath185 ms and @xmath184 ms . during the remaining part of the interval @xmath232 $ ] the system is subjected only to the radiofrequency perturbation at @xmath233 mhz whose amplitude @xmath193 is fixed by the condition @xmath234 that gives , in absence of measurement , @xmath235 if @xmath236 . since the radiofrequency is resonant @xmath237 . the time evolution of the reduced density matrix operator corresponding to the described process is obtained by successive iterations of ( [ rho11 ] ) and ( [ rho12 ] ) with the same formulas evaluated for @xmath213 . the theoretical probability for the transition @xmath188 at the end of the interval @xmath114 @xmath238 can be compared with the corresponding experimental data @xmath239 available for @xmath240 . the uncertainty for the experimental transition probabilities is estimated to be @xmath241 @xcite . 3 we show the sum of the squared differences between the theoretical and experimental transition probabilities normalized to the experimental uncertainties , the @xmath242 , as a function the phenomenological parameter @xmath243 . in the same figure we show also the probability @xmath244 that the observed value for the chi - square should exceed the value @xmath245 by chance . in our case the number of degrees of freedom is @xmath246 . the two arrows indicate the chi - square value obtained by the same authors of @xcite using a theoretical model based on the instantaneous von neumann collapse without ( higher value ) or with ( lower value ) corrections for the finite duration of the measurement pulses , the effect of optical pumping from level 2 to level 1 and the measured value of @xmath212 . depending upon the statistical confidence level we adopt , from fig . 3 we see that values of @xmath247 are required to fit properly the experimental data . the minimum value of @xmath242 obtained from measurement quantum mechanics slightly differs from the @xmath242 obtained on the basis of the von neumann postulate including experimental corrections . it is neither surprising nor exciting that measurement quantum mechanics is able to explain the experimental results of ref . @xcite in terms of a strong zeno inhibition . not surprising because measurement quantum mechanics contains more naive approaches to the problem of quantum measurements as the von neumann postulate , for instance , which was already shown to reproduce the experimental results . not exciting because strong zeno inhibition as well as full rabi oscillations are two trivial extreme regimes . however , measurement quantum mechanics tells us that another interesting and unexplored regime exists . it is the regime which occurs when the measurement coupling is comparable to the critical value . in this case a strong competition between stimulated transitions and measurement inhibition takes place as clearly recognized from fig . 2 and also from the analysis of the corresponding quantum variance , @xmath248 , which gets maximum at @xmath249 . measurement quantum mechanics suggests us also how to explore this regime . in order to make @xmath250 we should decrease @xmath199 and/or increase @xmath251 with respect to the values used in @xcite . according to the discussion of section iii @xmath199 can be decreased by reducing the density of particles in the informational environment . in the experiment @xcite the `` particles '' of the informational environment are the excited modes of the photon vacuum . their density can be reduced by lowering the intensity of the optical radiation acting as a measurement probe . on the other hand , eq . ( [ kcrit ] ) shows that @xmath251 can be increased by means of the amplitude @xmath193 of the perturbation which stimulates the rabi oscillations . in the experiment @xcite this amounts to increase the intensity of the radiofrequency radiation . of course , by increasing @xmath193 the rabi angular frequency increases too so that the condition @xmath234 used in @xcite can be maintained only by decreasing the period @xmath114 and , consequently , the duration of the measurement pulses . the possibility to vary the strength of the coupling is also related to the opportunity of looking for decoherence effects in the same experiment . indeed , from eq . ( [ rho12 ] ) we see that the off - diagonal elements of the density matrix vanish exponentially with time constant @xmath252 . for an appropriate measurement coupling , i.e. , a proper intensity of the optical radiation , @xmath253 could be experimentally accessible by optical homodyne tomography techniques @xcite and decoherence phenomena could be investigated . measurement quantum mechanics has been introduced by showing the equivalence among formalisms developed with the aim of including the effect of the measurement on the dynamics of a quantum system . the theory contains one parameter , the measurement coupling , for each measured observable . the measurement couplings can be expressed in terms of the detailed properties of a measurement apparatus which extracts information from the measured system in objective way , i.e. , independently of the presence of an observer looking at the pointer of the apparatus . the definition of quantum and classical measurements then emerges naturally according to the dominance , in the uncertainty of the pointer , of the classical fluctuations due to the apparatus or of the quantum fluctuations due to the measured system . ordinary quantum mechanics is obtained as a limit case for vanishing measurement couplings . in the opposite limit of infinite measurement couplings the dynamics of the measured system is frozen and classical . both limits are not interesting , the first one because there is no measurement and the second one because there is no quantum dynamics to measure . interesting physics is in between and measurement quantum mechanics is a tool for designing experiments in this intermediate regime .

in 1998 , blaze , bleumer , and strauss suggested a cryptographic primitive named _ proxy re - signatures _ where a proxy turns a signature computed under alice s secret key into one from bob on the same message . the semi - trusted proxy does not learn either party s signing key and can not sign arbitrary messages on behalf of alice or bob . at ccs 2005 , ateniese and hohenberger revisited the primitive by providing appropriate security definitions and efficient constructions in the random oracle model . nonetheless , they left open the problem of designing a _ multi - use unidirectional _ scheme where the proxy is able to translate in only one direction and signatures can be re - translated several times . + this paper solves this problem , suggested for the first time years ago , and shows the first _ multi - hop _ _ unidirectional _ proxy re - signature schemes . we describe a random - oracle - using system that is secure in the ateniese - hohenberger model . the same technique also yields a similar construction in the _ standard model _ ( i.e. without relying on random oracles ) . both schemes are efficient and require newly defined but falsifiable diffie - hellman - like assumptions in bilinear groups . * keywords . * multi - use proxy re - signatures , unidirectionality , pairings .

0802.1113

in 1998 , blaze , bleumer and strauss @xcite proposed a cryptographic primitive where a semi - trusted proxy is given some information that allows turning s signature on a message into bob s signature on the same message . these _ proxy re - signatures _ ( prs ) not to be confused with proxy signatures @xcite require that the proxy be unable to sign on behalf of alice or bob on its own . the last few years saw a renewed interest in proxy re - cryptography @xcite . + this paper presents the first constructions of _ multi - use _ _ unidirectional _ proxy re - signature wherein the proxy can only translate signatures in one direction and messages can be re - signed a polynomial number of times . our constructions are efficient and demand new ( but falsifiable ) diffie - hellman - related intractability assumptions in bilinear map groups . one of our contributions is a secure scheme in the standard model ( _ i.e. _ without resorting to the random oracle model ) . related work . the delegator can easily designate a proxy translating signatures computed using bob s secret key the delegatee into one that are valid w.r.t . her public key by storing her secret key at the proxy . upon receiving bob s signatures , the proxy can check them and re - sign the message using alice s private key . the problem with this approach is that the proxy can sign arbitrary messages on behalf of alice . proxy re - signatures aim at securely enabling the delegation of signatures without fully trusting the proxy . they are related to proxy signatures , introduced in @xcite and revisted in @xcite , in that any prs can be used to implement a proxy signature mechanism but the converse is not necessarily true . + in 1998 , blaze _ @xcite gave the first example of prs where signing keys remain hidden from the proxy . the primitive was formalized in 2005 by ateniese and hohenberger @xcite who pinned down useful properties that can be expected from proxy re - signature schemes . + 1 . re - signature keys can only be used for delegation in one direction ; 2 . a message can be re - signed a polynomial number of times ; 3 . re - signature keys can be kept secret by an honest proxy ; 4 . a user may not even know that a proxy exists ; 5 . a re - signature can not be linked to the one from which it was generated ; 6 . a user is only required to store a constant amount of secret data ; 7 . the delegatee does not act in the delegation process ; 8 . the proxy can not re - delegate signing rights ; [ fig : properties ] blaze _ et al . _ s construction is _ bidirectional _ ( _ i.e. _ the proxy information allows `` translating '' signatures in either direction ) and _ multi - use _ ( _ i.e. _ the translation of signatures can be performed in sequence and multiple times by distinct proxies without requiring the intervention of signing entities ) . unfortunately , ateniese and hohenberger @xcite pinpointed a flaw in the latter scheme : given a signature / re - signature pair , anyone can deduce the re - signature key that has been used in the delegation ( _ i.e. _ the _ private proxy _ property is not satisfied ) . another issue in @xcite is that the proxy and the delegatee can collude to expose the delegator s secret . + to overcome these limitations , ateniese and hohenberger proposed two constructions based on bilinear maps . the first one is a quite simple multi - use , bidirectional protocol built on boneh - lynn - shacham ( bls ) signatures @xcite . their second scheme is unidirectional ( the design of such a scheme was an open problem raised in @xcite ) but single - use . it involves two different signature algorithms : _ first - level _ signatures can be translated by the proxy whilst _ second - level _ signatures can not . a slightly less efficient variant was also suggested to ensure the privacy of re - signature keys kept at the proxy . the security of all schemes was analyzed in the random oracle model @xcite . + our contributions . ateniese and hohenberger left as open challenges the design of multi - use unidirectional systems and that of secure schemes in the standard security model . the present paper solves both problems : * we present a simple and efficient system ( built on the short signature put forth by boneh _ @xcite ) which is secure in the random oracle model under a reasonable extension of the diffie - hellman assumption ; * using an elegant technique due to waters @xcite , the scheme is easily modified so as to achieve security in the standard model . to the best of our knowledge , this actually provides the first unidirectional prs that dispenses with random oracles and thereby improves a recent bidirectional construction @xcite . both proposals additionally preserve the privacy of proxy keys ( with an improved efficiency w.r.t . @xcite in the case of the first one ) . they combine almost all of the above properties . as in prior unidirectional schemes , proxies are not completely transparent since signatures have different shapes and lengths across successive levels . the size of our signatures actually grows linearly with the number of past translations : signatures at level @xmath1 ( i.e. that have been translated @xmath2 times if the original version was signed at level @xmath3 ) consist of about @xmath4 group elements . in spite of this blow - up , we retain important benefits : * signers may want to tolerate a limited number ( say @xmath5 ) of signature translations for specific messages . then , if at most @xmath6 translations are permitted in the global system , users can directly generate a signature at level @xmath7 . * the conversion of a @xmath8 level signature is indistinguishable from one generated at level @xmath9 by the second signer . the original signer s identity is moreover perfectly hidden and the verifier only needs the new signer s public key . the simplicity of our schemes makes them attractive for applications that motivated the search for multi - use unidirectional systems in @xcite . one of them was to provide a proof that a certain path was taken in a directed graph : for instance , u.s . customs only need one public key ( the one of the immigration agent who previously validated a signature on an e - passport ) to make sure that a foreign visitor legally entered the country and went through the required checkpoints . another application was the conversion of certificates where valid signatures for untrusted public keys can be turned into signatures that verify under trusted keys . as exemplified in @xcite , unidirectional schemes are quite appealing for converting certificates between _ ad - hoc _ networks : using the public key of network b s certification authority ( ca ) , the ca of network a can non - interactively compute a translation key and set up a proxy converting certificates from network b within its own domain without having to rely on untrusted nodes of b. + roadmap . in the forthcoming sections , we recall the syntax of unidirectional prs schemes and the security model in section [ sec : model ] . section [ sec : bilinear ] explains which algorithmic assumptions we need . section [ bls - version ] describes our random - oracle - using scheme . section [ waters - version ] details how to get rid of the random oracle idealization . we first recall the syntactic definition of unidirectional prs schemes from @xcite . a ( unidirectional ) proxy re - signature ( prs ) scheme for @xmath10 signers and @xmath6 levels ( where @xmath10 and @xmath6 are both polynomial in the security parameter @xmath11 ) consists of a tuple of ( possibly randomized ) algorithms @xmath12 where : @xmath13 : : : is a randomized algorithm ( possibly run by a trusted party ) that takes as input a security parameter @xmath11 and produces a set of system - wide public parameters @xmath14 . @xmath15 : : : is a probabilistic algorithm that , on input of public parameters @xmath14 , outputs a signer s private / public key pair @xmath16 . @xmath17 : : : on input of public parameters @xmath18 , signer @xmath3 s public key @xmath19 and signer @xmath20 s private key @xmath21 , this ( ideally non - interactive ) algorithm outputs a re - signature key @xmath22 that allows translating @xmath3 s signatures into signatures in the name of @xmath20 . @xmath23 : : : on input of public parameters @xmath14 , a message @xmath24 , a private key @xmath25 and an integer @xmath26 , this ( possibly probabilistic ) algorithm outputs a signature @xmath27 on behalf of signer @xmath3 at level @xmath1 . @xmath28 : : : given common parameters @xmath14 , a level @xmath29 signature @xmath27 from signer @xmath30 and a re - signature key @xmath22 , this ( possibly randomized ) algorithm first checks that @xmath27 is valid w.r.t @xmath19 . if yes , it outputs a signature @xmath31 which verifies at level @xmath9 under public key @xmath32 . @xmath33 : : : given public parameters @xmath14 , an integer @xmath26 , a message @xmath24 , an alleged signature @xmath27 and a public key @xmath19 , this deterministic algorithm outputs @xmath34 or @xmath35 . for all security parameters @xmath36 and system - wide parameters @xmath14 output by @xmath13 , for all couples of private / public key pairs @xmath37 , @xmath38 produced by @xmath15 , for any @xmath26 and message @xmath24 , we should have @xmath39 to lighten notations , we sometimes omit to explicitly include public parameters @xmath14 that are part of the input of all but one algorithms . + the security model of @xcite considers the following two orthogonal notions termed _ external _ and _ insider security_. external security : : : is the security against adversaries outside the system ( that differ from the proxy and delegation partners ) . this notion demands that the next probability be a negligible function of the security parameter @xmath11 : @xmath40},\\ & & ( i^\star , l , m^\star,\sigma^\star ) \leftarrow { \mathcal{a}}^{\mathcal{o}_{sign}(.),\mathcal{o}_{resign}(.)}(\{pk_i\}_{i\in [ 1,n ] } ) : \\ & & \mathsf{verify}(l , pk_{i^\star},m^\star,\sigma^\star ) \wedge ( i^\star , m^\star ) \not\in q ] \end{aligned}\ ] ] where @xmath41 is an oracle taking as input a message and an index @xmath42 to return a first level signature @xmath43 ; the oracle @xmath44 takes as input indices @xmath45 and a level @xmath1 signature @xmath27 and returns the output of @xmath46 ; and @xmath47 denotes the set of ( signer , message ) pairs @xmath48 queried to @xmath41 or such that a tuple @xmath49 , with @xmath50 , was queried to @xmath44 . this notion only makes sense if re - signing keys are kept private by the proxy . + internal security : : : the second security notion considered in @xcite strives to protect users , as much as possible , against dishonest proxies and colluding delegation partners . three security guarantees should be ensured . + + * * limited proxy security : * this notion captures the proxy s inability to sign messages on behalf of the delegatee or to create signatures for the delegator unless messages were first signed by one of the latter s delegatees . formally , we consider a game where adversaries have all re - signing keys but are denied access to signers private keys . the following probability should be negligible : @xmath51},~ \{r_{ij } \leftarrow \mathsf{rekeygen}(pk_i , sk_j ) \}_{i , j \in [ 1,n ] } , & & \\ ( i^\star , l , m^\star,\sigma^\star ) \leftarrow { \mathcal{a}}^{\mathcal{o}_{sign } ( . , . ) } \big(\{pk_i\}_{i\in [ 1,n ] } , \{r_{ij } \}_{i , j \in [ 1,n ] } \big ) : & & \\ \mathsf{verify}(l , pk_{i^\star},m^\star,\sigma^\star ) \wedge m^\star \not\in q \big]\end{aligned}\ ] ] where @xmath52 is an oracle taking as input a message and an index @xmath42 to return a first level signature @xmath43 and @xmath47 stands for the set of messages @xmath24 queried to the signing oracle . + * * delegatee security : * informally , this notion protects the delegatee from a colluding delegator and proxy . namely , the delegatee is assigned the index @xmath34 . the adversary is provided with an oracle returning first level signatures on behalf of @xmath34 and is also granted access to re - signature keys and @xmath25 , with @xmath53 , and the definition can be simplified . in the general case , they remain part of the adversary s input . ] @xmath54 for all @xmath53 ( but not @xmath55 for any @xmath3 ) . her probability of success @xmath56 } , \\ & & \{r_{ij } \leftarrow \mathsf{rekeygen}(pk_i , sk_j ) \}_{i \in \{0,\ldots , n\},j \in \{1,\ldots , n\ } } \\ & & ( l , m^\star,\sigma^\star ) \leftarrow { \mathcal{a}}^{\mathcal{o}_{sign}(0 , . ) } \big(pk_0,\{pk_i , sk_i\}_{i\in [ 1,n]},\{r_{ij } \}_{i \in \{0,\ldots , n\},j \in \{1,\ldots , n\ } } \big ) : \\ & & \mathsf{verify}(l , pk_{0},m^\star,\sigma^\star ) \wedge m^\star \not\in q ~\big],\end{aligned}\ ] ] where @xmath47 is the set of messages queried to @xmath57 , should be negligible . + * * delegator security : * this notion captures that a collusion between the delegatee and the proxy should be harmless for the honest delegator . namely , we consider a target delegator with index @xmath34 . the adversary is given private keys of all other signers @xmath42 as well as _ all _ re - signature keys including @xmath55 and @xmath54 for @xmath42 . a signing oracle @xmath57 also provides her with first level signatures for @xmath34 . yet , the following probability should be negligible , @xmath58 } , \{r_{ij } \leftarrow \mathsf{rekeygen}(pk_i , sk_j ) \}_{i , j \in [ 0,n ] } , & & \\ ( 1,m^\star,\sigma^\star ) \leftarrow { \mathcal{a}}^{\mathcal{o}_{sign}(0 , . ) } \big(pk_0,\{pk_i , sk_i\}_{i\in [ 1,n ] } , \{r_{ij}\}_{i , j \in [ 0,n ] } , \big ) : & & \\ \mathsf{verify}(1,pk_{0},m^\star,\sigma^\star ) \wedge m^\star \not\in q \big],\end{aligned}\ ] ] meaning she has little chance of framing user @xmath34 at the first level . an important difference between external and limited proxy security should be underlined . in the former , the attacker is allowed to obtain signatures on the target message @xmath59 for signers other than @xmath60 . in the latter , the target message can not be queried for signature at all ( knowing all proxy keys , the attacker would trivially win the game otherwise ) . bilinear groups . groups @xmath61 of prime order @xmath62 are called _ bilinear map groups _ if there is a mapping @xmath63 with the following properties : 1 . bilinearity : @xmath64 for any @xmath65 and @xmath66 ; 2 . efficient computability for any input pair ; 3 . non - degeneracy : @xmath67 whenever @xmath68 . in these groups , we assume the hardness of the well - known computational diffie - hellman ( cdh ) problem which is to compute @xmath69 given @xmath70 and @xmath71 . flexible diffie - hellman problems . our signatures rely on new generalizations of the diffie - hellman problem . to motivate them , let us first recall the definition of the _ 2-out - of-3 diffie - hellman _ problem @xcite . in a prime order group @xmath72 , the * @xmath73-out - of-@xmath74 diffie - hellman * problem ( @xmath75-@xmath76-cdh ) is , given @xmath77 , to find a pair @xmath78 with @xmath79 . we introduce a potentially harder version of this problem that we call @xmath35-flexible diffie - hellman problem : the * @xmath80-flexible diffie - hellman * problem ( @xmath35-flexdh ) is , given @xmath81 , to find a triple @xmath82 . the unforgeability of our multi - use unidirectional proxy re - signatures is proved assuming the intractability of a relaxed variant of this problem where more flexibility is permitted in the choice of the base @xmath83 for the diffie - hellman computation . the * @xmath84-flexible diffie - hellman * problem ( @xmath1-flexdh ) is , given @xmath81 , to find a @xmath85-uple @xmath86 where @xmath87 for @xmath88 . a given instance has many publicly verifiable solutions : a candidate @xmath89-tuple @xmath90 is acceptable if @xmath91 , @xmath92 for @xmath93 and @xmath94 . the @xmath1-flexdh assumption is thus falsifiable according to naor s classification @xcite . in generic groups , the general intractability result given by theorem @xmath35 of @xcite by kunz - jacques and pointcheval implies the generic hardness of @xmath1-flexdh . for completeness , appendix [ generic ] gives an adaptation of this result in generic _ bilinear _ groups . the _ knowledge - of - exponent assumption _ ( kea1 ) @xcite was introduced in 1991 by damgrd @xcite . roughly speaking , kea1 captures the intuition that any algorithm which , given elements @xmath95 , computes a pair @xmath96 must `` know '' @xmath97 . under kea1 , the intractability of the @xmath1-flexible diffie - hellman problem is easily seen to be boil down to the diffie - hellman assumption . given @xmath98 , an adversary outputting @xmath99 necessarily `` knows '' @xmath100 and thus also @xmath101 as well as @xmath102 , which in turn successively yields logarithms of @xmath103 . although the kea1 assumption is inherently non - falsifiable , it holds in generic groups @xcite and our results can be seen as resting on the combination cdh+kea1 . modified diffie - hellman problem . the second assumption that we need is that the cdh problem @xmath104 remains hard even when @xmath105 is available . the * modified computational diffie - hellman * problem ( mcdh ) is , given @xmath106 , to compute @xmath107 . in fact , we use an equivalent formulation of the problem which is to find @xmath108 given @xmath109 ( the equivalence is readily observed by defining @xmath110 , @xmath111 , @xmath112 ) . to provide a better intuition of the underlying idea of our scheme , we first describe its single - hop version before extending it into a multi - hop system . + our approach slightly differs from the one in @xcite where signers have a `` strong '' secret and a `` weak '' secret that are respectively used to produce first and second level signatures . in our scheme , users have a single secret but first and second level signatures retain different shapes . another difference is that our re - signature algorithm is probabilistic . + we exploit the idea that , given @xmath113 for some @xmath114 , one can hardly generate a diffie - hellman triple @xmath115 without knowing the corresponding exponent @xmath116 @xcite . a valid bls signature @xcite @xmath117 can be blinded into @xmath118 using a random exponent @xmath5 . an extra element @xmath119 then serves as evidence that @xmath120 actually hides a valid pair . this technique can be iterated several times by adding two group elements at each step . to translate signatures from signer @xmath3 to signer @xmath20 , the key idea is to have the proxy perform an appropriate change of variable involving the translation key during the blinding . the scheme is obviously not strongly unforgeable in the sense of @xcite ( since all but first level signatures can be publicly re - randomized ) but this `` malleability '' of signatures is not a weakness whatsoever . it even turns out to be a desirable feature allowing for the unlinkability of translated signatures w.r.t . original ones . + in this scheme , signers public keys consist of a single group element @xmath121 . their well - formedness is thus efficiently verifiable by the certification authority that just has to check their membership in @xmath72 . this already improves @xcite where public keys @xmath122 ( @xmath123 and @xmath124 being common parameters ) must be validated by testing whether @xmath125 . @xmath126 : : : this algorithm chooses bilinear groups @xmath61 of prime order @xmath127 . a generator @xmath128 and a hash function @xmath129 ( modeled as a random oracle in the security proof ) are also chosen . public parameters only consist of @xmath130 @xmath126 : : : user @xmath3 s public key is set as @xmath131 for a random @xmath132 . @xmath133 : : : this algorithm outputs the proxy key @xmath134 which allows turning signatures from @xmath3 into signatures from @xmath20 . + @xmath135 : : : to sign @xmath136 at level @xmath35 , compute @xmath137 . + @xmath138 : : : to sign @xmath136 at level @xmath75 , choose @xmath139 and compute @xmath140 @xmath141 : : : on input of @xmath136 , the re - signature key @xmath142 , a signature @xmath143 and public keys @xmath144 , check the validity of @xmath145 w.r.t signer @xmath3 by testing @xmath146 . if valid , @xmath145 is turned into a signature on behalf of @xmath20 by choosing @xmath147 and computing @xmath148 if we set @xmath149 , we have @xmath150 @xmath151 : : : this algorithm accepts if @xmath146 . + @xmath152 : : : a second level signature @xmath153 is accepted for the public key @xmath154 if the following conditions are true . @xmath155 relations ( [ yes - ca - marche-1 ] ) and ( [ yes - ca - marche-2 ] ) show that translated signatures have exactly the same distribution as signatures directly produced by signers at level @xmath75 . + in comparison with the only known unidirectional prs with private re - signing keys ( suggested in section 3.4.2 of @xcite ) , this one features shorter second level signatures that must include a schnorr - like @xcite proof of knowledge in addition to @xmath76 group elements in @xcite . on the other hand , signatures of @xcite are strongly unforgeable unlike ours . + it is also worth mentioning that the above scheme only requires the @xmath35-flexible diffie - hellman assumption which is more classical than the general @xmath1-flexdh . the above construction can be scaled up into a multi - hop prs scheme if we iteratively apply the same idea several times . to prevent the linkability of signatures between successive levels @xmath9 and @xmath156 , the re - signature algorithm performs a re - randomization using random exponents @xmath157 . @xmath158 : : : to sign @xmath136 at the @xmath159 level , user @xmath3 chooses @xmath160 and outputs @xmath161 where @xmath162 @xmath163 : : : on input of a message @xmath136 , the re - signature key @xmath142 , a valid @xmath164 level signature @xmath165 and public keys @xmath144 , check the validity of @xmath27 under @xmath154 . if valid , @xmath27 is turned into a @xmath166 level signature on behalf of @xmath20 by drawing @xmath167 and computing @xmath168 where @xmath169 if we define @xmath170 and @xmath171 for @xmath172 , we observe that @xmath173 @xmath174 : : : the validity of @xmath175 at level @xmath176 is checked by testing if these equalities simultaneously hold : @xmath177 [ bls - proof ] the @xmath6-level scheme is a secure unidirectional proxy re - signature under the @xmath178-flexdh and mcdh assumptions in the random oracle model . _ limited proxy security . _ we show that an adversary @xmath179 with advantage @xmath180 implies an algorithm @xmath181 solving an @xmath178-flexdh instance @xmath182 with probability @xmath183 , where @xmath184 is the number of signing queries made by @xmath179 . system parameters : : : @xmath179 is challenged on parameters @xmath185 where @xmath186 is the random oracle controlled by the simulator @xmath187 . public key generation : : : when @xmath179 asks for the creation of user @xmath188 , @xmath181 responds with a newly generated public key @xmath189 , for a random @xmath132 , which virtually defines user @xmath3 s private key as @xmath190 . for all pairs @xmath191 , re - signature keys @xmath22 are calculated as @xmath192 . + oracle queries : : : @xmath179 s queries are tackled with as follows . following a well - known technique due to coron @xcite , a binary coin @xmath193 with expected value @xmath194 $ ] decides whether @xmath181 introduces the challenge in the output of the random oracle or an element of known signature . for the optimal value of @xmath195 , this introduces the loss factor @xmath196 in the success probability . + + * _ random oracle queries _ : to respond to these queries , @xmath181 maintains a list ( referred to as the @xmath197-list ) of tuples @xmath198 as follows : if the query @xmath24 already appears in the @xmath197-list , then @xmath181 returns @xmath124 ; 2 . otherwise , @xmath181 generates a random bit @xmath199 such that @xmath200 = \zeta$ ] ; 3 . it picks uniformly at random @xmath201 and computes @xmath202 if @xmath203 and @xmath204 otherwise ; 4 . it adds the 4-uple @xmath198 to the @xmath197-list and returns @xmath124 as the answer to the random oracle query . * _ signing queries _ : when a signature of signer @xmath3 is queried for a message @xmath24 , @xmath181 runs the random oracle to obtain the 4-uple @xmath198 contained in the @xmath197-list . if @xmath205 then @xmath181 reports failure and aborts otherwise , the algorithm @xmath181 returns @xmath206 as a valid signature on @xmath24 . after a number of queries , @xmath179 comes up with a message @xmath207 , that was never queried for signature for any signer , an index @xmath208 and a @xmath209 level forgery @xmath210 . at this stage , @xmath181 runs the random oracle to obtain the 4-uple @xmath211 contained in the @xmath197-list and fails if @xmath212 . otherwise , if @xmath213 is valid , it may be written @xmath214 which provides @xmath181 with a valid tuple @xmath215 , where @xmath216 , so that @xmath217 for @xmath218 . a similar analysis to @xcite gives the announced bound on @xmath181 s advantage if the optimal probability @xmath219 is used when answering hash queries . + [ [ delegatee - security . ] ] delegatee security . + + + + + + + + + + + + + + + + + + + we also show how to break the @xmath178-flexdh assumption out of a delegatee security adversary @xmath220 . given an input pair @xmath221 , the simulator @xmath222 proceeds as @xmath181 did in the proof of limited proxy security . system parameters and public keys : : : the target delegatee s public key is set @xmath223 . for @xmath224 , other public keys are defined as @xmath131 for a random @xmath225 . to generate re - signature keys @xmath22 , @xmath222 sets @xmath142 when @xmath226 and @xmath227 for @xmath228 . + queries : : : @xmath220 s hash and signing queries are handled exactly as in the proof of limited proxy security . namely , @xmath222 fails if @xmath220 asks for a signature on a message @xmath24 for which @xmath229 and responds consistently otherwise . + when @xmath220 outputs her forgery @xmath230 at level @xmath6 , @xmath222 is successful if @xmath231 , for some @xmath232 , and extracts an admissible @xmath233-uple as done in the proof of limited proxy security . + [ [ delegator - security . ] ] delegator security . + + + + + + + + + + + + + + + + + + + this security property is proven under the mcdh assumption . given an adversary @xmath234 with advantage @xmath180 , we outline an algorithm @xmath235 that has probability @xmath183 of finding @xmath236 given @xmath237 . public key generation : : : as previously , the target public key is defined as @xmath223 . remaining public keys are set as @xmath131 for a random @xmath238 for @xmath224 . this time , @xmath234 aims at producing a first level forgery and is granted _ all _ re - signature keys , including @xmath239 and @xmath240 . for indexes @xmath191 s.t . @xmath226 , @xmath235 sets @xmath142 . if @xmath241 , it calculates @xmath227 . if @xmath242 ( and thus @xmath243 ) , @xmath235 computes @xmath244 to @xmath234 . hash and signing queries are dealt with exactly as for previous adversaries . eventually , @xmath234 produces a first level forgery @xmath245 for a new message @xmath59 . then , @xmath235 can extract @xmath236 if @xmath246 for some @xmath247 , which occurs with probability @xmath248 using coron s technique @xcite . otherwise , @xmath235 fails . + [ [ external - security . ] ] external security . + + + + + + + + + + + + + + + + + + we finally show that an external security adversary @xmath249 also allows breaking the @xmath178-flexdh assumption almost exactly as in the proof of limited proxy security . the simulator @xmath250 is given an instance @xmath182 . as previously , @xmath250 must `` program '' the random oracle @xmath197 hoping that its output will be @xmath231 ( where @xmath251 is known ) for the message @xmath59 that the forgery @xmath252 pertains to . the difficulty is that @xmath250 must also be able to answer signing queries made on @xmath59 for all but one signers . therefore , @xmath250 must guess which signer @xmath60 will be @xmath249 s prey beforehand . at the outset of the game , it thus chooses an index @xmath253 . signer @xmath60 s public key is set as @xmath254 . all other signers @xmath255 are assigned public keys @xmath131 for which @xmath250 knows the matching secret @xmath256 and can thus always answer signing queries . + hash queries and signing queries involving @xmath60 are handled as in the proof of limited proxy security . when faced with a re - signing query from @xmath3 to @xmath20 for a valid signature @xmath257 at level @xmath26 , @xmath250 ignores @xmath257 and simulates a first level signature for signer @xmath20 . the resulting signature @xmath258 is then turned into a @xmath164 level signature and given back to @xmath249 . a re - signing query thus triggers a signing query that only causes failure if @xmath259 differs from @xmath260 for a known @xmath201 . + when @xmath249 forges a signature at level @xmath6 , @xmath250 successfully extract a @xmath233-flexible diffie - hellman tuple ( as @xmath181 and @xmath222 did ) if @xmath261 and if it correctly guessed the identity @xmath60 of the target signer . if @xmath249 s advantage is @xmath180 , we find @xmath262 as a lower bound on @xmath250 s probability of success , @xmath263 and @xmath264 being the number of signature and re - signature queries respectively . several extensions of bls signatures have a standard model counterpart when waters technique supersedes random oracle manipulations ( e.g. @xcite ) . likewise , we can very simply twist our method and achieve the first unidirectional prs scheme ( even including single hop ones ) that avoids the random oracle model . _ mutatis mutandis _ , the scheme is totally similar to our first construction and relies on the same assumptions . + as in @xcite , @xmath265 denotes the length of messages to be signed . arbitrary long messages can be signed if we first apply a collision - resistant hash function with @xmath265-bit outputs , in which case @xmath265 is part of the security parameter . + the scheme requires a trusted party to generate common public parameters . however , this party can remain off - line after the setup phase . @xmath266 : : : given security parameters @xmath267 , this algorithm chooses bilinear groups @xmath61 of order @xmath127 , generators @xmath268 and a random @xmath269-vector @xmath270 . the latter defines a function @xmath271 mapping @xmath265-bit strings @xmath272 ( where @xmath273 for all @xmath274 ) onto @xmath275 . the public parameters are @xmath276 @xmath126 : : : user @xmath3 sets his public key as @xmath131 for a random @xmath132 . @xmath133 : : : given user @xmath20 s private key @xmath277 and user @xmath3 s public key @xmath154 , generate the re - signature key @xmath134 that will be used to translate signature from @xmath3 into signatures from @xmath20 . @xmath278 : : : to sign a message @xmath279 at the first level , the signer picks @xmath280 at random and computes @xmath281 @xmath282 : : : to generate a second level signature on @xmath279 , the signer chooses @xmath283 and computes @xmath284 @xmath285 : : : on input of a message @xmath286 , the re - signature key @xmath142 , a signature @xmath287 and public keys @xmath144 , check the validity of @xmath27 w.r.t signer @xmath3 by testing if @xmath288 if @xmath145 is a valid , it can be turned into a signature on behalf of @xmath20 by choosing @xmath289 and computing @xmath290 where @xmath291 . if we set @xmath149 , we observe that @xmath292 @xmath293 : : : the validity of a first level signature @xmath294 is checked by testing if relation ( [ validity-1 ] ) holds . + @xmath295 : : : a second level signature @xmath296 is accepted for the public key @xmath154 if the following conditions are true . @xmath297 to the best of our knowledge , the above scheme is the first unidirectional prs in the standard model and solves another problem left open in @xcite where all constructions require the random oracle model . like the scheme of section [ bls - version ] , this extension of waters signature @xcite is scalable into a multi - hop prs . at levels @xmath298 , algorithms @xmath299 , @xmath300 and @xmath301 are generalized as follows . + @xmath302 : : : to sign @xmath286 at level @xmath9 , user @xmath3 picks @xmath303 , @xmath160 and outputs @xmath304 where @xmath305 @xmath306 : : : on input of a message @xmath307 , the re - signature key @xmath142 , a purported @xmath164 level signature @xmath308 and public keys @xmath144 , check the correctness of @xmath309 under @xmath154 . if valid , @xmath309 is translated for @xmath310 by sampling @xmath311 , @xmath167 and setting @xmath312 where @xmath313 if we define @xmath170 , @xmath314 and @xmath171 for @xmath172 , we observe that @xmath315 @xmath316 : : : a candidate signature @xmath317 is verified by testing if the following equalities hold : @xmath318 [ waters - proof ] the scheme with @xmath6 levels ( and thus at most @xmath319 hops ) is a secure unidirectional prs under the @xmath178-flexdh and mcdh assumptions . the proof is very similar to the one of theorem [ bls - proof ] and replaces random oracle manipulations by the tricks of @xcite . we prove the limited proxy and delegatee security properties under the @xmath178-flexdh assumption . the delegator security is demonstrated under the mcdh assumption . [ [ limited - proxy - security . ] ] limited proxy security . + + + + + + + + + + + + + + + + + + + + + + + we consider an adversary @xmath179 with advantage @xmath180 . we describe an algorithm @xmath181 solving a @xmath178-flexdh instance @xmath221 with probability @xmath320 , where @xmath184 is the number of signing queries made by @xmath179 , within a comparable time . system parameters : : : the simulator @xmath181 prepares common public parameters as follows . it first sets @xmath321 . the @xmath269-vector @xmath322 is defined by choosing @xmath323 and @xmath324 for @xmath325 using random vectors @xmath326 , @xmath327 , where @xmath328 is randomly chosen and @xmath329 . for any message @xmath330 , we have @xmath331 for functions @xmath332 , @xmath333 respectively defined as @xmath334 and @xmath335 . as in @xcite , @xmath181 will be successful if @xmath336 for the message @xmath337 of the forgery stage whereas @xmath338 for all messages @xmath339 queried for signature . since @xmath340 , we have @xmath341 with non - negligible probability @xmath342 . the adversary @xmath179 is challenged on parameters @xmath343 . + key generation : : : for user @xmath42 , @xmath181 defines a public key as @xmath189 , for a random @xmath132 , which virtually defines user @xmath3 s private key as @xmath190 . for pairs @xmath191 , re - signature keys are chosen as @xmath192 . + signing queries : : : when a signature of signer @xmath3 is queried for a message @xmath344 , @xmath181 fails if @xmath345 . otherwise , following the technique of @xcite , it can construct a signature by picking @xmath346 and computing @xmath347 which is returned to @xmath348 . if we define @xmath349 , @xmath27 has the correct distribution as @xmath350 and @xmath351 . + after polynomially many queries , @xmath179 comes up with a message , that was never queried for signature for any signer , and index @xmath208 and a forgery @xmath352 at level @xmath6 . at this stage , @xmath181 fails if @xmath353 . otherwise , if @xmath354 is valid , @xmath355 which provides @xmath181 with a valid @xmath233-uple @xmath356 a completely similar analysis to @xcite shows that @xmath357 with probability @xmath358 , which yields the bound on @xmath181 s advantage . + [ [ delegatee - security.-1 ] ] delegatee security . + + + + + + + + + + + + + + + + + + + a delegatee security adversary @xmath220 also implies a breach in the @xmath178-flexdh assumption . the simulator @xmath222 is given @xmath221 and uses a strategy that is completely analogous to the one of simulator @xmath181 in the proof of limited proxy security . system parameters and public keys : : : @xmath222 prepares public parameters exactly as in the proof of limited proxy security . the public key of the target user is defined as @xmath223 . the attacker @xmath220 must be provided with private keys for all the delegators of that user . for @xmath224 , other public keys are therefore chosen as @xmath131 for randomly picked @xmath132 . the adversary @xmath220 then receives @xmath359 as well as re - signature keys @xmath22 for @xmath360 and @xmath361 . these are set as @xmath227 and @xmath142 if @xmath243 . + signing queries : : : for all signers @xmath53 , @xmath220 can generate signatures on her own . when a signature of the target signer is requested for a message @xmath344 , @xmath222 proceeds as @xmath181 did when facing the limited proxy adversary @xmath179 . it fails if @xmath345 and can answer the query otherwise . + when @xmath220 eventually outputs a forgery @xmath362 at level @xmath6 , @xmath222 is successful if @xmath357 and extracts an admissible @xmath233-uple as @xmath181 did . + [ [ delegator - security.-1 ] ] delegator security . + + + + + + + + + + + + + + + + + + + a delegator security adversary @xmath234 having advantage @xmath180 after @xmath184 signing queries is finally shown to imply an algorithm @xmath235 to solve a problem which is equivalent ( under linear time reduction ) to the mcdh problem with probability @xmath363 . given @xmath237 , this problem is to find out @xmath236 . public parameters and public key generation : : : again , system parameters are prepared as in the proof of limited proxy security . namely , @xmath235 defines @xmath364 and chooses @xmath365 so as to have @xmath366 for some functions @xmath367 where @xmath368 cancels with non - negligible probability . the public key of the target delegator is set as @xmath223 . for @xmath224 , remaining public keys are set as @xmath131 for a random @xmath132 . the adversary @xmath234 receives @xmath359 . this time , she is provided with _ all _ re - signature keys ( including @xmath239 and @xmath240 ) and attempts to produce a first level forgery . for pairs @xmath191 such that @xmath226 , @xmath235 sets @xmath142 . if @xmath241 , it defines @xmath227 . if @xmath242 ( and thus @xmath243 ) , @xmath235 calculates @xmath244 and hands @xmath369 to @xmath234 . + signing queries : : : when @xmath234 asks for a signature from the target delegator for a message @xmath344 , @xmath235 fails if @xmath345 and can answer the query exactly as in the proof of limited proxy security otherwise . + eventually , @xmath234 produces a first level forgery @xmath370 for a message @xmath371 that was never queried for signature . if @xmath372 , @xmath235 fails . otherwise , given that @xmath373 , @xmath235 finds out @xmath374 . + [ [ external - security.-1 ] ] external security . + + + + + + + + + + + + + + + + + + we consider an adversary @xmath249 with advantage @xmath180 . we describe an algorithm @xmath250 solving a @xmath178-flexdh instance @xmath221 with probability @xmath375 within comparable time , where @xmath184 and @xmath264 are is the number of signing and re - signing queries . system parameters : : : the simulator @xmath376 prepares common public parameters as in the limited proxy security proof . in addition , it picks at random an integer @xmath377 . + public key generation : : : when @xmath249 asks for the creation of user @xmath42 , @xmath250 responds + + * with a newly generated public key @xmath131 , for a random @xmath132 if @xmath378 ( s.t . @xmath256 , user @xmath3 s private key , is known to the simulator ) ; * with @xmath379 if @xmath380 ( which virtually defines user @xmath3 s private key as @xmath116 ) . + oracle queries : : : @xmath249 s queries are tackled with as follows . + + * _ signing queries _ : when a signature of signer @xmath3 is queried for a message @xmath344 , * * @xmath250 uses its knowledge of @xmath256 to produce the signature if @xmath378 ; * * @xmath250 uses the simulation from the limited proxy security proof if @xmath380 ( and therefore fails if @xmath345 ) . * _ re - signing queries _ : for such a query on input @xmath381 , @xmath250 checks if @xmath257 is a valid @xmath8 level signature on @xmath344 for some @xmath382 with respect to the public key @xmath3 . if yes , @xmath250 produces a first level signature on @xmath344 for user @xmath20 ( using the previous simulation strategy ) , increases its level up to @xmath9 ( for the same public key ) using the re - signing algorithm ( with re - signature key simply equal to @xmath123 ) and outputs the resulting @xmath164 level signature . the simulation only fails if @xmath383 and @xmath384 . after polynomially many queries , @xmath249 comes up with a message @xmath371 , an index @xmath385 and a forgery @xmath386 at level @xmath6 . recall that @xmath371 can not have been queried to signer @xmath387 . again , @xmath250 fails if @xmath353 or @xmath388 . otherwise , if @xmath389 is valid , @xmath250 produces a valid @xmath178-flexdh - tuple as in the limited proxy security proof . a completely similar analysis to this proof ends up with the announced bound on @xmath250 s advantage . we described the first multi - use unidirectional proxy re - signatures , which solves a problem left open in 2005 . the random - oracle - based proposal also offers efficiency improvements over existing solutions at the first level . the other scheme additionally happens to be the first unidirectional prs in the standard model . + two major open problems remain . first , it would be interesting to see if multi - level unidirectional prs have efficient realizations under more classical intractability assumptions . a perhaps more challenging task would be to find out implementations of such primitives where the size of signatures and the verification cost grow sub - linearly with the number of translations . 10 m. abe , s. fehr . perfect nizk with adaptive soundness . in _ tcc07 _ , _ lncs _ 4392 , pages 118136 . springer , 2007 . an , y. dodis , and t. rabin . on the security of joint signature and encryption . in _ eurocrypt02 _ , _ lncs _ 2332 , pages 83107 . springer , 2002 . g. ateniese , k. fu , m. green , s. hohenberger . improved proxy re - encryption schemes with applications to secure distributed storage . in _ ndss _ , 2005 . g. ateniese , k. fu , m. green , s. hohenberger . improved proxy re - encryption schemes with applications to secure distributed storage . in _ acm tissec _ , 9(1 ) : pp . 130 , 2006 . g. ateniese , s. hohenberger . proxy re - signatures : new definitions , algorithms , and applications . in _ acm ccs05 _ , pages 310319 , acm press , 2005 m. bellare , a. palacio : the knowledge - of - exponent assumptions and 3-round zero - knowledge protocols . proc . of crypto04 , springer lncs vol . 3152 , 273289 ( 2004 ) m. bellare , p. rogaway . random oracles are practical : a paradigm for designing efficient protocols . in _ acm ccs93 _ , pages 6273 , acm press , 1993 . m. blaze , g. bleumer , m. strauss . divertible protocols and atomic proxy cryptography . in _ eurocrypt98 _ , _ lncs _ 1403 , pages 127144 , 1998 . a. boldyreva , a. palacio , b. warinschi . secure proxy signature schemes for delegation of signing rights . cryptology eprint archive : report 2003/096 , 2003 . d. boneh , x. boyen . efficient selective - id secure identity based encryption without random oracles . in _ eurocrypt04 _ , _ lncs _ 3027 , pp . springer , 2004 . d. boneh , b. lynn , h. shacham . short signatures from the weil pairing . in _ asiacrypt01 _ , volume 2248 of _ lncs _ , pages 514532 . springer , 2002 . r. canetti , s. hohenberger . chosen - ciphertext secure proxy re - encryption . in _ acm ccs07_. pages 185194 . acm press , 2007 . j .- s . coron . on the exact security of full domain hash . in _ crypto00 _ , volume 1880 of _ lncs _ , pages 229235 . springer , 2000 . i. damgrd : towards practical public key systems secure against chosen ciphertext attacks . proc . of crypto91 , springer lncs vol . 576 , 445456 ( 1991 ) a. dent . the hardness of the dhk problem in the generic group model . cryptology eprint archive : report 2006/156 , 2006 . y. dodis , a .- proxy cryptography revisited . in _ ndss03 _ , 2003 . m. green , g. ateniese . identity - based proxy re - encryption . in _ acns07 _ , _ lncs _ 4521 , pages 288306 . springer , 2007 . s. hohenberger . advances in signatures , encryption , and e - cash from bilinear groups . ph.d . thesis , mit , may 2006 . s. hohenberger , g. n. rothblum , a. shelat , v. vaikuntanathan . securely obfuscating re - encryption . in _ tcc07 _ , _ lncs _ 4392 , pages 233252 . springer , 2007 . s. kunz - jacques , d. pointcheval . about the security of mti / c0 and mqv . in _ scn06 _ , _ lncs _ 4116 , pages 156172 , springer , 2006 . s. lu , r. ostrovsky , a. sahai , h. shacham , b. waters . sequential aggregate signatures and multisignatures without random oracles . in _ eurocrypt06 _ , volume 4004 of _ lncs _ , pages 465485 , springer , 2006 . t. malkin , s. obana and m. yung . the hierarchy of key evolving signatures and a characterization of proxy signatures . in _ eurocrypt04 _ , volume 3027 of _ lncs _ , pages 306322 , springer , 2004 . m. mambo , k. usuda , e. okamoto . proxy signatures for delegating signing operation . in _ acm ccs96 _ , pages 4857 . acm press , 1996 . m. naor . on cryptographic assumptions and challenges . in _ crypto03 _ , _ lncs _ 2729 pages 96109 . springer - verlag , 2003 . j. shao , z. cao , l. wang , x. liang . proxy re - signature schemes without random oracles . in _ indocrypt07 _ , _ lncs _ 4859 , pages 197209 . springer , 2007 . c. p. schnorr . efficient identification and signatures for smart cards . in _ crypto89 _ , volume 435 of _ lncs _ , pages 239252 . springer , 1989 . b. waters . efficient identity - based encryption without random oracles . in _ eurocrypt05 _ , _ lncs _ 3494 , pages 114127 . springer 2005 . to provide more confidence in the @xmath1-flexdh assumption we prove a lower bound on the computational complexity of the @xmath1-flexdh problem for generic groups equiped with bilinear maps . in @xcite , kunz - jacques and pointcheval define a family of computational problems that enables to study variants of the cdh problem in the generic group model . let @xmath390 be an adversary in this model and @xmath391 be a multivariate polynomial whose coefficients might depend on @xmath390 s behaviour . for values of @xmath392 chosen by the simulator , and knowing their encodings , the goal of @xmath390 is to compute the encodings of @xmath393 such that @xmath394 all elements manipulated by @xmath390 are linear polynomials in @xmath392 and some new random elements introduced through the group oracle . let us denote @xmath395 the polynomial corresponding to @xmath396 ( it is a random variable ) , kunz - jacques and pointcheval proved the following result . the choice @xmath402 implies the generic hardness of the problem @xmath1-flexdh in groups . the purpose of this section is to prove that kunz - jacques and pointcheval result also holds in generic bilinear groups and therefore that the problem @xmath1-flexdh is intractable in these groups . let @xmath397 and @xmath403 be an upper bound for the probability @xmath399\ ] ] then the probability that @xmath390 wins after @xmath400 oracle queries to the group operations in @xmath72 , @xmath404 to the bilinear map @xmath405 satisfies @xmath406 in the following @xmath407 and @xmath408 , denote the set @xmath409 and are used to represent elements of @xmath72 and @xmath404 respectively . following @xcite , in the generic bilinear group model , an adversary @xmath390 has access to * an oracle @xmath410 that , on input @xmath411 , answers with the representation of @xmath412 in @xmath407 , where @xmath413 is the representation of @xmath414 and @xmath415 the representation of @xmath416 . * an oracle @xmath417 that , on input @xmath418 , answers with the representation of @xmath412 in @xmath408 , where @xmath413 is the representation of @xmath414 and @xmath415 the representation of @xmath416 . * an oracle @xmath419 that , on input @xmath411 , answers with the representation of @xmath412 in @xmath408 , where @xmath413 is the representation of @xmath414 and @xmath415 the representation of @xmath416 . the connection between representations and elements of @xmath420 is managed by the simulator through two lists @xmath421 and @xmath422 of pairs @xmath423 associating an element with its representation . a representation @xmath413 in an oracle query input does not need to correspond to an element of @xmath420 in @xmath421 or @xmath424 ; if it does , the corresponding element is used , otherwise a random element @xmath414 is drawn by the simulator in @xmath420 and bound to @xmath413 , that is , @xmath425 is added to @xmath421 or @xmath424 . the same rule applies for the answer to the query : if @xmath426 with @xmath427 in @xmath421 or @xmath424 , @xmath428 is answered . otherwise , a random representation @xmath428 , is chosen and @xmath427 is added to @xmath421 or @xmath424 , and the answer to the oracle query is @xmath428 . overall , each oracle query adds at most 3 pairs to @xmath421 or @xmath424 . for our problem , initially we have @xmath429 and @xmath390 is given @xmath430 . @xmath390 s goal is to output @xmath431 corresponding to @xmath432 in @xmath420 that , together with the @xmath256 s , cancel @xmath433 . the last queries of @xmath390 are assumed to be of the form @xmath434 . @xmath390 has won if @xmath435 where @xmath436 . to prove the generic hardness of the problem , we consider a simulator @xmath437 where random values in @xmath420 are replaced by formal unknowns @xmath154 . represents of elements of @xmath72 ( _ resp . _ @xmath404 ) correspond to linear combinations ( _ resp . _ quadratic polynomials ) of these unknowns with coefficients in @xmath420 . the simulation is similar to the one given in @xcite and @xmath390 s goal is to output @xmath431 corresponding to linear polynomials @xmath438 in @xmath439 $ ] that , together with the unknowns @xmath154 s , cancel @xmath433 . the difference between @xmath390 s success probability in the two simulation occurs only if @xmath437 s simulation , the representations of different polynomials ( linear or quadratic ) @xmath440 and @xmath441 collide in @xmath442 s simulation . the number of polynomials in @xmath421 and @xmath422 is upper - bounded by @xmath443 and their degrees is at most two . therefore , the difference appears with probability at most @xmath444 . as in @xcite , the success criterion in @xmath437 s simulation is stricter than in @xmath442 s simulation and as above the probability that @xmath390 succeeds in @xmath442 s simulation but not in @xmath437 s simulation is upper - bounded by @xmath445 ( since @xmath433 is of degree @xmath446 and the @xmath395 s are linear polynomial ) .

in this report , we study within the context of general relativity with one extra dimension compactified either on a circle or an orbifold , how radion fluctuations interact with metric fluctuations in the three non - compact directions . the background is non - singular and can either describe an extra dimension on its way to stabilization , or immediately before and after a series of non - singular bounces . we find that the metric fluctuations transfer undisturbed through the bounces or through the transients of the pre - stabilization epoch . our background is obtained by considering the effects of a gas of massless string modes in the context of a consistent massless background ( or low energy effective theory ) limit of string theory . we discuss applications to various approaches to early universe cosmology , including the ekpyrotic / cyclic universe scenario and string gas cosmology .

hep-th0509043

the idea that the universe is comprised of any number of extra dimensions , in addition to the three non - compact spatial dimensions that we observe , goes back all the way to the work of kaluza and klein in the 1920 s @xcite , where extra dimensions were first proposed as a manner in which one could unify particle interactions with gravity . since then , theories with extra dimensions have evolved from being a mathematical curiosity , albeit one with remarkable consequences for particle physics @xcite , to a theoretical necessity with the advent of string theory . since the famous anomalies cancellation calculation of green and schwarz @xcite , the challenge has remained for cosmologists and phenomenologists alike to account for the presence of extra dimensions in physics . from a pessimist s point of view , one might view these extra dimensions as an unwanted surplus that has to be swept under the rug at presently accessible energies . in its most unimaginative form , one could view this as the gist of the moduli problem . from an optimist s point of view however , one could view the requirement of extra dimensions for the consistency of string theory as a new window of opportunity , through which one might be able to resolve some of the outstanding problems of particle physics and cosmology . taken as an essential part of the radically new picture of space - time that string theory offers us , extra dimensions , with strings and branes of various dimensions propagating through them , could offer us a potentially non - anthropic explanation for the observed dimensionality of space - time @xcite . in addition to offering us potentially non - singular cosmologies @xcite , they could very well be at the heart of the resolution of the dark matter problem @xcite . recently , several alternative stringy cosmologies have emerged as possible successors to the standard big bang model of the universe , such as the pre - big bang model @xcite , brane / string gas cosmology @xcite , and the cyclic / ekpyrotic scenarios @xcite , all of which have the ultimate aim of becoming complete and testable models of the early universe . a central question that arises in determining whether or not these models of the early universe reproduce observations is , how metric fluctuations in the presence of extra dimensions evolve in the backgrounds that these cosmologies propose . the importance of understanding this question can not be understated in the context of the abundance of available experimental data , against which we must compare our eventual predictions . for instance , in the ekpyrotic / cyclic scenario the effects of branes colliding in a 5-dimensional bulk are explored key to the success of this program is being able to follow the evolution of metric fluctuations through the bounces of the extra dimension . the eventual goal of this is to be able to explain the observed inhomogeneities of the cosmic microwave background as being seeded by radion fluctuations generated through the motion of the branes towards each other , rather than through the quantum fluctuations of some putative inflaton field . similarly , in the context of the pre - big bang scenario , where an expanding and a contracting phase of the universe s evolution are naturally related to each other by the scale factor duality symmetry inherent to low energy effective string theory , a universe undergoing a big bang - big crunch transition is naturally implemented . in this context , the evolution of metric fluctuations through the bounce becomes a question of central importance in understanding the way the big bang phase of our universe s evolution happened the way it did . as it turns out , although at first seemingly unrelated , an associated and as yet unexamined issue is how any tentative solution of the moduli problem will modify the spectrum of the metric fluctuations which we observe . given that the universe is likely to contain many extra dimensions whose shape and volume moduli are dynamical in the early universe , one might expect that the dynamics of these moduli fields will lead to a highly nontrivial modification of any initial pre - stabilization spectrum . it is the goal of this report to show that , in the context of a 5-d universe where an extra dimension undergoes non - singular bounces , or is en route to stabilization , the final spectrum for the bardeen potential corresponding to ( long wavelength ) fluctuations of the scale factor for the non - compact dimensions , is identical to the initial spectrum of the five - dimensional fluctuations ( in a sense which will be specified later ) . specifically , if we had an initial scale invariant spectrum for the five - dimensional fluctuations , we would end up with a scale invariant spectrum for the bardeen potential after the bounce , or after the transients of the stabilization had settled down . furthermore , we show that the variable corresponding to metric fluctuations of the compact dimensions decays , which corresponds to the stability of the radion degree of freedom to fluctuations in our non - singular setup . one application of our work concerns the evolution of fluctuations in ekpyrotic / cyclic type models . in the context of a four space - time dimensional effective field theory toy model of this scenario , the dynamics of perturbations has been investigated in detail . the initial analyses @xcite yielded the result that a scale - invariant spectrum before the bounce transfers to a scale - invariant spectrum after the bounce . these analyses , however , were questioned in @xcite . a serious complicating factor turned out to be the fact that the proposed background evolution was singular , thus requiring the use of `` matching conditions '' ( such as those derived in @xcite in the context of an expanding cosmology undergoing a sudden phase transition ) to compute the post - bounce spectrum of fluctuations . as discussed in @xcite , there is a very sensitive dependence of the final result on the choice of the matching surface . independently , there has been recent work on the evolution of fluctuations through a nonsingular bounce in four space - time dimensional cosmologies , in which the bounce is constructed by adding extra terms to the standard lagrangian . these analyses @xcite yield results showing a sensitive dependence on the nature of the bounce . the bottom line of this work is that a correct analysis of fluctuations in the ekpyrotic / cyclic scenario needs to be done in a five space - time dimensional context , a context in which the nature of the bounce is unambiguous . a first important step in this direction was taken in @xcite , confirming the result that a scale - invariant spectrum passes through the bounce ( in the five - dimensional context , the bounce means that the radius of the extra spatial dimension bounces , not that the four space - time dimensional scale factor bounces ) without change in the spectral index . however , the analysis of @xcite was done in the context of a singular background and assuming specific matching conditions for fluctuations applied at a point when the perturbations in fact blow up . thus , the results are open to doubt . in this work , we study the transfer of fluctuations through a cosmology in which two boundary branes approach each other and bounce without encountering a singularity ( see also @xcite for previous work done in the context of a particular nonsingular ekpyrotic - type bounce proposed in @xcite ) . we begin by introducing a non - singular bouncing model of a 5-dimensional universe where one dimension is compactified on a circle or an orbifold . as we shall see shortly in detail , our non - singular background is obtained by considering the effects of gas composed of massless string modes on the dynamics of space - time , in the context of a consistent low energy effective theory limit , or ` massless background ' , of string theory . we then set up the framework for studying cosmological perturbations in this model and study how these transfer through the various bounces that the extra dimension undertakes . although our setup is seemingly specific to string gas cosmology , the essence of our framework is that we have a non - singular bounce / stabilization mechanism that is affected by degrees of freedom that become massless at a certain point . hence , the hope is that the results obtained here can be generalized to other settings , an issue we will discuss in detail when we consider applications to different approaches to stringy cosmology . we now commence our paper with a few preliminaries . consider a five dimensional space - time with the topology of @xmath0 , described by the metric corresponds to cosmic time . ] @xmath1 from which we derive the following components of the einstein tensor : @xmath2\ , , \\ \label{55 } g^y_{\,\,y } & = & -3 \bigl [ { \frac{\stackrel{ .. }{{a}}}{{a } } } + \bigl ( { \frac{\stackrel{.}{{a}}}{{a } } } \bigr)^2 \bigr]\,.\end{aligned}\ ] ] here , the indices @xmath3 and @xmath4 run over the three large spatial dimensions and @xmath5 denotes the extra dimension . the form of the energy - momentum tensor which will couple to the einstein tensor is given by @xmath6 we can recast the einstein equations @xmath7 , where @xmath8 is the five - dimensional newton s constant , in the form : @xmath9 = 0\,,\\ \label{b } \ddot{b } & + & 3h\dot{b } + 8\pi g~b[p - \frac{2r}{3 } - \frac{\rho}{3 } ] = 0\,,\\ \label{hc } \rho & = & \frac{3}{8\pi g}h(h + \mathcal h)\,,\end{aligned}\ ] ] where @xmath10 and @xmath11 . the energy - momentum tensor for a string gas in this toroidally compactified background was derived in @xcite to which we refer the reader if any of what follows is unfamiliar . it was found to be @xmath12 where @xmath13 denotes the determinant of the metric , @xmath14 ( with @xmath15 being the inverse string tension ) and the subscripts @xmath16 refer to the momentum and winding quantum numbers of a closed string along the 5th dimension respectively , on which all of the subscripted quantities will depend . the density @xmath17 is the number density of the string gas with the metric dependence factored out ( @xmath18 ) , while @xmath19 is the energy of a single closed string in this background @xmath20 where @xmath21 denotes the center of mass momentum along the three non - compact directions , and @xmath22 is the number of right - moving string vibrational modes . for reasons made clear in @xcite and @xcite , string states that are massless at special symmetry points ( i.e. the self - dual radius ) should play a very distinguished role in any string gas cosmology ( see also @xcite ) . there are several reasons for this , among which are prominently the desire to obtain a viable late time phenomenology and a robust stabilization mechanism for the radion . however , the main motivation is that the low energy effective theory limits of string theory ( such as general relativity and dilaton gravity ) are also consistent backgrounds on which we can study the propagation of massless ( and only massless ) strings . for these reasons we do not further justify the focus on massless string modes , and refer the reader to @xcite for more details . for a general review of string gas cosmology , we refer the interested reader to @xcite@xcite@xcite . we wish to emphasize that the framework within which we chose to work assumes nothing other than the fact that nature is described by a string theory at high energies , and that the degrees of freedom that are likeliest to be excited ( namely massless string modes ) will be excited . the two particular geometries that we consider for the extra dimensions ( i.e. a toroidal geometry in this section and an orbifold geometry in the next section ) , are necessitated by the fact that they are the unique compact backgrounds on which one stays within the approximation of the low energy effective limit of string theory , when considering extra dimensions that are similar in size to the string scale . this is because in general , were the metric to depend on the extra dimension ( as is the case for warped extra dimensions ) , one will introduce curvatures that approach the string scale when the extra dimensions themselves approach the string scale , and hence the low energy approximation will break down . we discuss this point further in the next section . the particular states we are interested in are those for which @xmath23 . that these states are massless at the self - dual radius ( @xmath24 ) is easily checked from ( [ e ] ) . these states imply that ( [ b ] ) becomes @xmath25 we will see that this background stabilizes around the self - dual radius ( @xmath26 ) , whilst the non - compact directions persistently expand as they would in a radiation dominated universe . before we get to this , however , we wish to discuss a generalization of this background to the situation where the extra dimension is compactified on an orbifold . the consistency of the framework that we propose , namely that we remain within the low energy effective theory ( or massless background ) limit of string theory , requires the background we consider to satisfy the condition @xmath28 \ll \frac{1}{\alpha'}\,.\ ] ] that is , the ricci scalar should be considerably bounded from above by the string tension . this requirement translates into the statement that the metric should not change very much on the string scale , which ensures that this background remains a consistent background for the propagation of massless string modes . conversely , were one to consider backgrounds which do not satisfy ( [ approx ] ) , not only would the approximations made in deriving the low energy limit of string theory break down , one would also expect massive modes to be created @xcite . for our purposes , the requirement of ( [ approx ] ) means that were we to consider compactifications of the extra dimension on scales comparable to the self - dual radius , then there should be absolutely no dependence of the metric along the compactified direction . this is a remarkable feature of using a toroidal compactification : one can in fact study string scale processes without invalidating the approximations inherent in the low energy limit of string theory . it was shown in @xcite and @xcite that the background we are about to derive , does in fact satisfy ( [ approx ] ) throughout its dynamics . returning to the problem at hand , we see that since we are not allowed to consider metric dependencies along the extra dimension ( by homogeneity and isotropy of the non - compact dimensions , the metric can then only depend on time ) , the components of the einstein tensor ( [ 00])-([55 ] ) are unchanged after orbifolding the extra dimension . the only place where orbifolding might make a difference is in the energy momentum tensor of the string gas . however , because the massless states that we have focused on are in the so - called untwisted sector @xcite ( in general , twisted states are localized at the orbifold fixed points ) , it turns out that there is no difference in the energy - momentum tensor either ( [ tt])-([aa ] ) . this is a consequence of the ` inheritance ' principle of orbifold theories @xcite , and permits us to use the framework just presented in either situation , provided we only use the modes that we have indicated and we do not consider any variations in any metric quantities along the extra dimension . however , there is a caveat to this in that two orbifold fixed planes are present , which might have matter confined to live on them . the requirement not to induce any metric variations along the extra dimension translates into , via the israel junction conditions @xcite , the condition ( among others ) @xmath29 \propto 8\pi g \rho_{brane},\ ] ] where @xmath30 $ ] is the jump of the derivative of the scale factor @xmath31 along the extra dimension , evaluated around either of the orbifold fixed planes . hence , we see that if we require there to be no dependence of the metric along the extra dimensions , then any matter localized on the branes must be sufficiently dilute to render the branes to behave as test branes . bearing all of this in mind , we can now proceed to derive the background solution within our framework . we first note from ( [ b ] ) that any form of matter which satisfies a radiative equation of state ( @xmath32 , @xmath33 ) for all times ( as opposed to our wound string states which become massless , hence radiative , only at the self - dual radius ) drops out of the driving term in the equation of motion for @xmath34 ( [ b ] ) . hence , we can safely consider a situation where we have a radiation gas in addition to our gas of wound strings , which dominates the evolution of the hubble factor @xmath35 for the non - compact dimensions through ( [ hc ] ) after the extra dimension has stabilized , or is close to stabilization . in fact , massless unwound closed string states ( gravitons : @xmath36 , @xmath37 ) can provide exactly such a bath . henceforth , it is natural to include such a bath in our setup . we do not consider this matter further , simply taking as a given in what follows that the scale factor @xmath31 for the non - compact dimensions expands at the background level as it would in a radiation dominated universe , that is @xmath38 the analytic approximation ( [ analyticb ] ) ( solid line ) is compared with the numeric solution of ( [ eom ] ) ( circles ) . ] let us go back to the equation of motion ( [ blah ] ) for @xmath39 , which reads @xmath40 where @xmath41 with @xmath42 is the center of mass momentum along the three large dimensions . introducing the characteristic time scale @xmath43 and defining the dimensionless time variable @xmath44 such that @xmath45 , we can simplify ( [ eomsubodh ] ) to @xmath46 where a prime denotes a derivative with respect to @xmath47 and we also introduced the free dimensionless parameter @xmath48 this equation can easily be integrated numerically , but it will turn out to be useful to have a simple analytic approximation . if we are close to the self - dual radius , that is @xmath49 with @xmath50 , we can expand the driving term in ( [ eom ] ) so that @xmath51 results . the general solution is given by @xmath52 with @xmath53 and @xmath54 constants . this solution may either describe the stabilization of the extra dimension or a series of non - singular bounces . is small . ] see fig.[fig : b_a_n ] for a comparison of the numeric solution of ( [ eom ] ) with the analytic one . since they are barely discernable we will use the analytic approximation in the following and switch freely between bounce / stabilization - language . we will first perturb the metric , focusing only on scalar metric perturbations . the most general scalar perturbation in `` generalized '' longitudinal gauge @xcite can be characterized by four scalar functions , @xmath55 and @xmath56 . these functions can be viewed as a basis of gauge invariant variables . in this gauge , the metric is given by @xmath57 where the signs in front of the perturbations are a mere convention . would be called the newtonian potential in a four dimensional description . ] for reasons discussed in sections [ secii ] and [ seciii ] , we do not consider any @xmath5 dependence , that is we assume homogeneity in the extra dimension ( see also @xcite for a discussion of this and other assumptions of sgc ) . the corresponding einstein tensor can be computed to be @xcite ( in the following @xmath58 denotes the laplacian of the three large spatial dimensions ) : @xmath59\ , \,,\,\,\ i \neq j,\label{g^i_j;ineqj}\\ \delta g^y_{\,\,t}&= & \left[\frac{3}{b^2}\left ( \frac{\ddot{a}}{a}-\frac{\dot{a}\dot{b}}{ab}\right ) + \frac{1}{2b^2a^2}\triangle\right]w \,,\label{einsteinyt}\\ \delta g^{y}_{\,\,x_i}&=&\partial _ { x_i}\bigg[\frac{1}{2b^2}\left(\frac{\dot{b}}{b}+\frac{\dot{a}}{a } + \partial _ t\right)w\bigg]\label{einsteinyxi } \,,\end{aligned}\ ] ] @xmath60\left ( -\gamma + \psi + \phi \right ) + 2\left(\frac{\dot{b}}{b } + 3\frac{\dot{a}}{a}+\partial _ t\right ) \partial _ t\psi \label{einsteinxixi}\\ \nonumber & & -2\left(\frac{\ddot{b}}{b } + 2\frac{\dot{a}\dot{b}}{ab } + \frac{\dot{a}^2}{a^2}+2\frac{\ddot{a}}{a } + \frac{\dot{b}}{2b}\partial _ t + \frac{\dot{a}}{a}\partial _ t\right ) \phi-\left(2\frac{\dot{a}}{a } + 2\frac{\dot{b}}{b}+\partial _ t \right ) \partial _ t\gamma \,,\\ \delta g^{y}_{\,\,y}&=&\left[-\frac{2}{a^2}\triangle + 3\left(\partial _ t + 4\frac{\dot{a}}{a}\right)\partial _ + \left[-\frac{1}{a^2}\triangle -6\left(\frac{\dot{a}^2}{a^2 } + \frac{\ddot{a}}{a}+\frac{\dot{a}}{2a}\partial _ t\right ) \right]\phi \ , , \label{einsteinyy}\\ \delta g^t_{\,\,t}&=&\left [ 3\left(\frac{\dot{b}}{b}\partial _ t + 2\frac{\dot{a}}{a}\partial _ t\right)-\frac{2}{a^2}\triangle\right]\psi -6\left(\frac{\dot{a}\dot{b}}{ab}+\frac{\dot{a}^2}{a^2}\right ) \phi -\left(\frac{3\dot{a}}{a}\partial _ t -\frac{1}{a^2}\triangle\right)\gamma \ , , \label{einsteintt}\label{g^t_t}\\ \delta g^{x_i}_{\,\,t}&=&\partial _ { x_i } \bigg[\frac{2}{a^2}\partial _ t\psi -\frac{1}{a^2}\left(\frac{\dot{b}}{b}+2\frac{\dot{a}}{a}\right)\phi -\frac{1}{a^2}\left(\frac{\dot{b}}{b}-\frac{\dot{a}}{a}+\partial _ t\right)\gamma \bigg ] \ , . \label{einsteinxit}\end{aligned}\ ] ] one can check that the equations of motion involving @xmath61 decouple from the other ones for the matter content ( [ emt ] ) we consider . since @xmath61 would appear only squared in a four dimensional effective theory , we will not need to compute it at all . decouple from the other ones is a direct consequence of the homogeneity in the @xmath5-direction . ] to write down the perturbed einstein equations , we also need the perturbed energy - momentum tensor @xmath62 . it will include the thermal bath of radiation ( denoted by the subscript @xmath63 ) , and the stringy matter sources denoted by a tilde . to be specific , we have @xmath64 ( where @xmath65 and @xmath66 are the radiation energy density and pressure , respectively , and @xmath67 is the radiation three velocity potential ) and @xmath68 where @xmath69 and @xmath70 are the string gas energy density and pressure , respectively , and @xmath71 is the string gas three velocity potential . note that anisotropic stress does not feature in our setup ( see appendix [ appendix ] ) , but we will keep @xmath71 around for the time being . we will also focus on adiabatic perturbations of the radiation fluid only , that is @xmath72 , even though two ideal fluids are present so that iso - curvature perturbations could arise . the reason for neglecting those is simplicity . the @xmath73 einstein equations yield immediately @xmath74 and , after introducing @xmath75 we infer from ( [ tt])-([aa ] ) , that the perturbed energy momentum tensor is of the form @xmath76 where the background quantities are given by @xmath77 in performing this calculation , we started with the source action for this string gas @xcite , and obtained the perturbed energy momentum tensor through direct calculation . the string gas energy momentum tensor only depends on the metric , the number density of strings @xmath17 , and the center of mass momentum of these strings . however , the latter quantity is not perturbed , because for long wavelength perturbations ( compared to the string scale ) , which we are restricted to if we are to remain within the limits of low energy effective theory , the center of mass motion of a string propagating on a perturbed spacetime is unaffected to first order . hence the only quantities left to perturb are the metric dependencies , and the number density of the string gas . next , we can write the perturbed zero component of the conservation equation @xmath78 for the stringy matter as @xmath79 for simplicity we will set @xmath80 from now on , that is we neglect the scalar velocity potential of the string gas , consistent with the arguments of appendix [ appendix ] . combining the diagonal einstein equations yields @xmath81 the last three equations ( [ eomnagain])-([eomxiagain ] ) are the dynamical ones for the perturbation variables @xmath56 , @xmath82 and @xmath22 . the ( @xmath83 ) equation gives the radiation fluid perturbation @xmath84 and the ( @xmath85 ) equation gives the scalar velocity potential @xmath67 of the radiation fluid in terms of the other variables . these equations simplify if we write them in terms of @xmath47 defined in ( [ defeta ] ) and make use of the background equation @xmath45 . introducing @xmath86 we get @xmath87 \label{eomgamma}\\ & & \hspace{1cm}+\frac{k^{*2}}{4}\gamma_k=\eta^2c\left(2\delta r^{*}+\delta\rho^{*}-3\delta p^{*}\right)\,,\\ \nonumber & & \xi_k^{\prime\prime}+\xi_k^{\prime}2\left(h+\frac{2}{\eta}\right)-\gamma_k^{\prime}h+(\xi_k-\gamma_k)\left[\frac{2h}{\eta}+h^2+h^{\prime}\right]\\ & & \hspace{1cm}+\frac{k^{*2}}{12}\xi_k=\eta^2c\left(3\delta p^{*}-\delta\rho^{*}\right)\,,\\ & & \nonumber \delta\rho^{*\prime}+\frac{3}{\eta}(\delta\rho^{*}+\delta p^{*})+h(\delta\rho^{*}+\delta r^{*})-\frac{3}{2}\xi_k^{\prime}(\rho^{*}+p^{*})\\ & & \hspace{1cm}-\frac{1}{2}\gamma_k^{\prime}(\rho^{*}-3p^{*}+2r^{*})=0\,.\label{eomrhoeta}\end{aligned}\ ] ] here , we made the transition to fourier space and we performed the rescaling @xmath88 , @xmath89 . all other starred quantities are defined via @xmath90 the two dimensionless constants left are @xmath91 which are not independent but related via @xmath92 after @xmath93 got approached ( that is after the extra dimension got stabilized or , in the language of the ekpyrotic / cyclic scenario , the two approaching branes came to a halt ) , we can simplify the equations of the previous section to @xmath94 where we introduced the constant @xmath95 we note that the source term for @xmath96 vanishes because of the background equation of state for the gas of massless string modes . also , the equations for @xmath96 and @xmath97 decouple at late times , just as they should do . the solutions are given by @xmath98 with @xmath99 constant and @xmath100 we would like to evaluate the spectrum @xmath101 when a long wavelength mode enters the hubble radius again at @xmath102 . if we taylor expand ( [ apprxxi ] ) for small @xmath103 we see that there is an approximately constant mode present for @xmath96 . on the other hand , @xmath97 from ( [ apprxgamma ] ) is oscillating and decaying @xmath104 . therefore , whatever the spectrum for @xmath96 was at the initial time , it should persist till re - entry and determine the spectrum for @xmath105 because , by neglecting @xmath97 , we have @xmath106 . this is the same result one would conclude in a simple four dimensional universe , dominated by radiation . to be specific , we can approximate @xmath107 where @xmath108 is some irrelevant phase . of course this holds true only , if the transient stabilizing epoch leaves no strong imprints onto the spectrum ; hence , we will examine this crucial issue in the next section . before we do that , let us have a quick look at the physics after a mode entered the hubble radius : @xmath96 will start to oscillate with a decaying amplitude @xmath109 . if one waits long enough , @xmath97 will catch up and become larger than @xmath96 . now @xmath110 and thus @xmath97 will determine the spectrum , because for @xmath111 @xmath112 there is of course a transition region where neither approximation holds . nevertheless , this only occurs after hubble radius crossing and shall not concern us in the following . the analytic solution of the previous section is valid after @xmath39 got stabilized ( or the branes came to a halt ) . however , the main question we would like to address is the following : what effect on the perturbations , if any , does the process of trapping the modulus @xmath39 have ? in order to address this question we have to solve ( [ eomgamma])-([eomrhoeta ] ) numerically . as we shall see , the late time solution is approached rapidly and long wavelength perturbations are not affected in any significant way . hence , the naive intuition that the value of @xmath105 for modes outside the hubble radius stay frozen is indeed valid , with important consequences for cosmological model building , be it in the framework of the ekpyrotic / cyclic scenario , or within string gas cosmology ( sgc ) we will focus on concrete models in the next section . @xmath113 is plotted for different values of @xmath103 , with the initial conditions given in section [ num ] . black : analytic solution of ( [ analyticphi ] ) ; grey ( bending curve ) : numerical solution of ( [ eomgamma])-([eomrhoeta ] ) . ] the spectrum of ( a ) @xmath96 , ( b ) @xmath97 and ( c ) @xmath114 is evaluated at hubble radius crossing @xmath115 , with the initial conditions given in section [ num ] . if the long wavelength regime @xmath116 is approached , all oscillations are damped away such that only the constant mode of @xmath96 survives , as expected . ] firstly , let us specify constants and initial conditions : the only free parameter in our setup is @xmath117 defined in ( [ qstar ] ) , describing the initial momentum of the string gas in the three large dimensions . being a dimensionless parameter , we choose the most natural value @xmath118 our results are not very sensitive to its exact value . induces a phase shift in @xmath105 and an increase in @xmath119 , while an increase in @xmath120 leads to a decrease in @xmath119 and no phase shift ; however , the metric degrees of freedom decouple form @xmath119 quickly and @xmath119 itself is of no interest to us . ] we start close to the self - dual radius with some arbitrary velocity ( once again , its exact value does not affect our results - see fig.[fig : phi2_b ] ) , that is @xmath121 next , we have to specify initial conditions for @xmath122 and @xmath119 at @xmath123 . , because this is the characteristic time scale of the background oscillations . this means , no oscillation could occur before @xmath124 . ] since we did not compute how the universe came close to the self - dual radius ( e.g. within the ekpyrotic / cyclic setup or a sgc setup ) , we have no way of deriving those . however , we can make an educated guess : if the universe underwent some phase of inflation in the three dimension before ( we will make this point more concrete in the next section ) , one should expect similar values for all metric perturbations ; hence we will set @xmath125 with some small @xmath126 . for simplicity , we will use @xmath127 for all @xmath128 . furthermore , since the long wavelength modes we are interested in should be frozen once they cross the hubble radius , we set @xmath129 the initial string density perturbation @xmath130 will be set to zero , because these massless modes just got produced . naturally , @xmath119 will get sourced by @xmath97 and @xmath96 . no conclusion in the following is sensitive to the chosen initial conditions , hence we chose the most simple ones above ( other ones were of course also tested ) . the only thing left to specify is the wave - number @xmath131 : we are interested in long wavelength modes , that is modes outside the hubble radius with small @xmath103 . in addition , we shall require @xmath132 , so that we can compare our results with ( [ analyticphi ] ) . with all initial conditions specified , we can use the analytic solution ( [ analyticb ] ) to solve ( [ eomgamma])-([eomrhoeta ] ) numerically . let us first compare @xmath133 for different values of @xmath103 , fig . [ fig : phi ] ( a)-(c ) : there @xmath105 is plotted both numerically ( grey ) and analytic ( black ) . we see that the analytic late time solution is approached pretty fast in fact , the small difference at the beginning is not visible in this plot , but only if @xmath96 is plotted alone as in fig . [ fig : fields ] ( a ) . the visible decaying oscillation of @xmath105 is the decaying mode of @xmath97 form ( [ apprxgamma ] ) with frequency @xmath134 , which is plotted in fig . [ fig : fields ] ( b ) ( compare with fig . [ fig : phi ] ( b ) ) . in addition , there are strongly damped oscillations on top of the constant mode of @xmath96 , that carry the same frequency as @xmath135 , fig . [ fig : b ] . as mentioned before , these are not visible in fig . [ fig : phi ] but in fig . [ fig : fields ] ( a ) where @xmath96 is plotted alone . these oscillations are the impact of trapping the modulus @xmath39 , an impact that can safely be ignored in the long wavelength regime : if one decreases @xmath131 , all of these effects become less pronounced . an other feature visible in fig . [ fig : phi ] is a bending of the lower curve in comparison to the analytic solution ( [ analyticphi ] ) . this bending is the beginning of an oscillation of @xmath96 and hence an expected feature , since the constant mode is actually the beginning of an oscillation with a very small frequency due to @xmath136 . all features are not sensitive to the initial values we choose for @xmath97 and @xmath96 . last but not least , @xmath119 is plotted in fig . [ fig : fields ] ( c ) : it gets sourced quickly by @xmath97 and @xmath96 and oscillates with a nearly constant amplitude thereafter , with all frequencies entering . however , there is no mentionable back - reaction of @xmath119 on the metric perturbations . hence , one could give @xmath119 a non zero initial value without changing the evolution of @xmath97 and @xmath96 in any significant way . ( a ) and ( b ) : @xmath39 over @xmath137 is plotted for @xmath138 , with the initial conditions given in section [ num ] . ( c ) : @xmath96 is plotted over the same time range as @xmath34 ; note how the oscillations in the background scale factor source transient oscillations in @xmath96 . ] the perturbation variables @xmath96 , @xmath97 and @xmath119 plotted for @xmath138 , with the initial conditions given in section [ num ] . ] in the end , we are interested in the spectrum of @xmath105 . so far , there seems to be little to no impact of the transient stabilizing era of @xmath34 onto @xmath97 and @xmath96 . this translates directly to the spectrum , where @xmath105 should approach a horizontal line for small @xmath103 . this is indeed the case , as can be seen in fig . [ fig : spectrum ] ( c ) , where the spectrum is evaluated at horizon crossing : the oscillations that are present for relatively large @xmath103 get damped once the long wavelength regime is approached . this is expected , because long wavelength modes enter the hubble radius later and henceforth , the decaying modes responsible for the oscillations in the spectrum get damped more . the frequency in the spectrum of @xmath105 is proportional to @xmath139 , in accordance with @xmath140 and equation ( [ analyticphi ] ) . note that no shift of the spectral index is induced , leaving the overall index unaltered . the oscillations in @xmath96 in fig . [ fig : spectrum](a ) are the main impact of the transient era of a dynamical @xmath34 . they are clearly negligible , because the tiny amplitude gets even smaller with decreasing @xmath103 . to summarize , the main conclusion of our numerical study is that long wavelength metric perturbations quickly approach their asymptotic solution , leaving no trace of the many bounces @xmath34 experiences on its way to stabilization . more specifically , we have shown that the initial spectrum of the non - decaying mode of @xmath96 is preserved , i.e. equals the spectrum of @xmath96 at late times . at late times , the spectrum of @xmath96 equals the spectrum of the bardeen potential @xmath105 , the potential which is relevant for late time observations . speaking more loosely , we have shown that the initial pre - bounce spectrum of bulk perturbations is transferred to the late time cosmological perturbations without a change in the spectral index . @xmath113 is plotted for @xmath138 . this is the same plot as in fig.[fig : phi ] ( b ) , but with different initial conditions for @xmath39 : @xmath141 and @xmath142 are used instead of ( [ 63 ] ) and ( [ 64 ] ) . note that there is no discernable difference between this plot and fig.[fig : phi ] ( b ) . ] we saw in the previous section that there is no significant effect of trapping the modulus @xmath39 on the spectrum of long wavelength perturbations in the sense that the final spectrum of the bardeen potential @xmath143 equals the initial value of the non - decaying mode of the bulk perturbation @xmath82 . this comes about since the value of @xmath82 for small @xmath131 modes which have a wavelength larger than the hubble radius remains frozen during the phase of the trapping of @xmath39 , and since the final spectrum of @xmath82 coincides with the spectrum of @xmath143 . even though our result agrees with a `` naive '' intuition coming from the analysis of fluctuations in expanding four space - time dimensions , it could be viewed as an unexpected result : the `` naive '' intuition turned out to be wrong in the case of bouncing cosmologies in four space - time dimensions , for example in pre - big - bang models where it was shown @xcite that the growing mode of @xmath143 in the contracting phase couples almost exclusively to the decaying mode of @xmath143 in the expanding phase , leading to the result that there is a large change in the spectral index of the dominant mode between the contracting and the expanding phase . other analyses performed within the context of four - dimensional general relativity ( see @xcite and references therein ) yielded a similar result . it is now also generally accepted that the four space - time dimensional toy models which were proposed to describe the ekpyrotic / cyclic models have the same feature , namely that the dominant mode of @xmath143 in the contracting phase matches predominantly to the decaying mode of @xmath143 in the expanding phase ( see e.g. @xcite ) . our analysis , however , confirms the analysis of @xcite , which showed , in the context of a singular bounce , that in the case of a bounce of boundary branes in five space - time dimensions the spectrum of @xmath143 could be preserved . our result is an important step in constructing a viable alternative to standard scalar field - driven inflationary models ( next to the remarkable but highly nontrivial kklmmt construction , @xcite and follow - up papers ) in the context of cosmologies with extra dimensions . we will now outline two possible proposals which seem promising to us in that regard : firstly , one could envisage a modification of the ekpyrotic / cyclic scenario . in the ekpyrotic proposal @xcite , a test brane moves slowly through the bulk towards the boundary brane of an orbifold ( the horava - witten @xcite setup ) . in the cyclic scenario @xcite , it is the two boundary branes that approach each other . in either case , a scale - invariant spectrum of perturbations can be generated during the contracting phase given a suitable potential for the modulus field which describes the inter - brane distance @xcite . during the collision of the branes ( a singular event in the ekpyrotic / cyclic scenarios ) a hot universe is supposed to emerge on the boundary brane we live on today . our background construction can be viewed as a regularized version of the ekpyrotic / cyclic scenarios ( ekpyrotic in the context of a single bounce , cyclic in the case we consider the evolution until @xmath144 has stabilized ) . our work shows that in the context of such a regularized scenario , an initial scale - invariant spectrum could pass through the bounces and thus survive from the initial contracting phase to the final phase of expansion of our three spatial dimensions ( this scenario will be developed further in a follow - up paper ) . in our scenario , the branes do not actually hit each other , but come to a halt as a consequence of the appearance of new massless modes that get produced explosively at a certain brane - separation ( that is at a certain value of @xmath34 ) . our results show that such dynamics does not spoil a scale invariant spectrum generated during the initial brane movement . in this framework , one can still reheat the universe , since the `` kinetic energy '' of the branes will get transferred to the stabilizing massless modes and other light modes which act like radiation on the brane ( corresponding to the radiation bath that we already included in our setup ) . secondly , an incorporation of inflation into brane / string gas cosmology will rely on the stabilization mechanism described in this article : if any mechanism of inflating three dimensions is found , the internal dimensions will most likely have to deflate in some way . for example , one could employ the idea of anisotropic inflation ( studied in the context of vacuum solutions of higher dimensional general relativity in @xcite and @xcite ) . once the internal dimensions get close to the self - dual radius , certain string modes become massless and get produced explosively along with a radiation bath , in close analogy to reheating after standard scalar field - driven inflation . these modes will then stabilize the internal dimensions , curing the graceful exit problem of anisotropic inflation , while the spectrum of fluctuations produced during inflation in the metric degrees of freedom remains unaltered . a concrete realization of this proposal is in preparation by the authors of this article . in this report , we studied how radion , matter and metric fluctuations interact in a universe that exhibits a transient stabilizing epoch of its extra dimension , which could also be viewed as a series of bounces of the extra dimension . we were primarily interested in the imprints of this epoch on the spectrum of the bardeen potentials , motivated by the hope of combining an earlier phase , generating a scale invariant spectrum ( e.g. in the framework of a modified ekpyrotic scenario , or via an incorporation of inflation within string gas cosmology ) , with a successful late time stabilizing mechanism , provided by a gas of massless string states . we found that the spectrum of long wavelength perturbations remains unaltered by this epoch , which has important consequences for various approaches to stringy models of the early universe . in particular , specific realizations of the ekpyrotic / cyclic scenario or an incorporation of inflation into string gas cosmology become possible we provided two proposals in this report , but many variations are viable . we would like to thank scott watson for helpful comments and discussions , and also andrei linde , diana battefeld and claudia de rahm for comments on the draft . t.b . would like to thank the physics department of mcgill university for hospitality and support . s.p . wishes to thank the theory group at queen mary university in london and the cosmology group at damtp in cambridge for hospitality during the course of this work . this research is supported at mcgill by an nserc discovery grant , by funds from the canada research chair program , and by funds from mcgill university . t.b . has been supported in part by the us department of energy under contract de - fg02 - 91er40688 , task a. in this appendix , we will provide arguments as to why no anisotropic stress arises in the setup of @xcite , which we examined at the perturbative level in this article . in the following @xmath145 denote string world - sheet coordinates . first note that our unperturbed background is undergoing frw expansion in the non - compact dimensions while the compact direction is stabilized at the self dual radius by a string gas . as a preliminary , we wish to show that long wavelength perturbations of the metric do not affect the motion of strings to first order in the perturbation variables . we begin with the polyakov action for a closed string @xmath146 where we work in conformal gauge from the outset . consider now perturbations around the background : @xmath147 . it was shown in the appendix of @xcite how one can inherit the string spectrum and the constraint algebra provided the derivatives of the background metric @xmath148 ( i.e. the hubble factors ) are several orders of magnitude smaller than the string energy scales . we want to make sure that we stay within this regime while studying the dynamical compactification of the extra dimension . thus , when we consider the equations of motion for a closed string in this perturbed spacetime , we must neglect all derivatives of the unperturbed metric compared to the worldsheet derivatives of the string co - ordinates , as these will be of the order of the string scale whereas the metric derivatives are constrained to be much smaller @xcite . consider the equation of motion for a closed string @xmath149 from which follows @xmath150 now we expand @xmath151 as @xmath152 , where @xmath153 is the solution to the unperturbed equation of motion @xmath154 , see appendix of @xcite . our perturbed equation of motion then becomes @xmath155 notice that there are two vastly different scales in the equation above : the worldsheet derivatives and the spacetime derivatives . if we imagine expanding @xmath156 in terms of fourier modes , the derivatives would bring down a factor of the 4-d wave - vector . realizing that we need to stay within the domain of validity of our setup , namely that we can not introduce fluctuations in the metric that would invalidate the conclusion in the appendix of @xcite concerning the string spectrum , we need these fluctuations to vanish on energy scales several orders of magnitude smaller than the string scale . this still allows for metric fluctuations that are considerably short distance in nature , but nevertheless above the string scale . with this in mind , we can neglect all the terms in ( [ perteom ] ) that are multiplied by derivatives of the metric perturbation , since these will be orders of magnitude smaller than the other ones . henceforth , we obtain the result @xmath157 that is , the string motion is unchanged by long wavelength perturbations . ordinarily this would signal the end of any hopes to study perturbative physics ; this is not the case for us , since we are primarily interested in perturbing around a background which consists of a gas of such strings . as a result , if we consider the perturbed energy - momentum tensor for such a gas , it will contain no anisotropic stress . to see this , consider the off - diagonal spatial components which can be written as @xmath158 where @xmath159 and @xmath160 indicates an ensemble ( gas ) average . these components vanish , since we just determined that for individual strings @xmath161 , which means that there is no anisotropic stress to deal with in our setup . o. klein , `` quantum theory and five - dimensional theory of relativity , '' z. phys . * 37 * , 895 ( 1926 ) [ surveys high energ . phys . * 5 * , 241 ( 1986 ) ] . t. kaluza , `` on the problem of unity in physics , '' sitzungsber . berlin ( math . phys . ) * 1921 * , 966 ( 1921 ) . m. j. duff , b. e. w. nilsson and c. n. pope , `` kaluza - klein supergravity , '' phys . rept . * 130 * , 1 ( 1986 ) . m. b. green and j. h. schwarz , `` anomaly cancellation in supersymmetric d=10 gauge theory and superstring theory , '' phys . b * 149 * , 117 ( 1984 ) . r. h. brandenberger and c. vafa , `` superstrings in the early universe , '' nucl . b * 316 * , 391 ( 1989 ) . a. a. tseytlin and c. vafa , `` elements of string cosmology , '' nucl . b * 372 * , 443 ( 1992 ) [ arxiv : hep - th/9109048 ] . m. gasperini and g. veneziano , `` pre - big bang in string cosmology , '' astropart . phys . * 1 * , 317 ( 1993 ) [ arxiv : hep - th/9211021 ] . a. nusser , s. s. gubser and p. j. e. peebles , `` structure formation with a long - range scalar dark matter interaction , '' phys . d * 71 * , 083505 ( 2005 ) [ arxiv : astro - ph/0412586 ] . s. s. gubser and p. j. e. peebles , `` cosmology with a dynamically screened scalar interaction in the dark sector , '' phys . d * 70 * , 123511 ( 2004 ) [ arxiv : hep - th/0407097 ] . s. s. gubser and p. j. e. peebles , `` structure formation in a string - inspired modification of the cold dark matter model , '' phys . d * 70 * , 123510 ( 2004 ) [ arxiv : hep - th/0402225 ] . t. battefeld and s. watson , `` effective field theory approach to string gas cosmology , '' jcap * 0406 * , 001 ( 2004 ) [ arxiv : hep - th/0403075 ] . r. h. brandenberger , `` moduli stabilization in string gas cosmology , '' arxiv : hep - th/0509159 . r. h. brandenberger , `` challenges for string gas cosmology , '' arxiv : hep - th/0509099 . t. battefeld and s. watson , `` string gas cosmology , '' arxiv : hep - th/0510022 . j. khoury , b. a. ovrut , p. j. steinhardt and n. turok , `` the ekpyrotic universe : colliding branes and the origin of the hot big bang , '' phys . rev . d * 64 * , 123522 ( 2001 ) [ arxiv : hep - th/0103239 ] . p. j. steinhardt and n. turok , `` cosmic evolution in a cyclic universe , '' phys . d * 65 * , 126003 ( 2002 ) [ arxiv : hep - th/0111098 ] . p. j. steinhardt and n. turok , `` a cyclic model of the universe , '' arxiv : hep - th/0111030 . j. khoury , b. a. ovrut , n. seiberg , p. j. steinhardt and n. turok , `` from big crunch to big bang , '' phys . d * 65 * , 086007 ( 2002 ) [ arxiv : hep - th/0108187 ] . n. turok , m. perry and p. j. steinhardt , `` m theory model of a big crunch / big bang transition , '' phys . rev . d * 70 * , 106004 ( 2004 ) [ erratum - ibid . d * 71 * , 029901 ( 2005 ) ] [ arxiv : hep - th/0408083 ] . t. j. battefeld , s. p. patil and r. brandenberger , `` perturbations in a bouncing brane model , '' phys . d * 70 * , 066006 ( 2004 ) [ arxiv : hep - th/0401010 ] . r. kallosh , l. kofman and a. d. linde , `` pyrotechnic universe , '' phys . d * 64 * , 123523 ( 2001 ) [ arxiv : hep - th/0104073 ] . j. khoury , b. a. ovrut , p. j. steinhardt and n. turok , `` density perturbations in the ekpyrotic scenario , '' phys . d * 66 * , 046005 ( 2002 ) [ arxiv : hep - th/0109050 ] . d. h. lyth , `` the primordial curvature perturbation in the ekpyrotic universe , '' phys . b * 524 * , 1 ( 2002 ) [ arxiv : hep - ph/0106153 ] . r. brandenberger and f. finelli , `` on the spectrum of fluctuations in an effective field theory of the ekpyrotic universe , '' jhep * 0111 * , 056 ( 2001 ) [ arxiv : hep - th/0109004 ] . d. h. lyth , `` the failure of cosmological perturbation theory in the new ekpyrotic scenario , '' phys . b * 526 * , 173 ( 2002 ) [ arxiv : hep - ph/0110007 ] . j. c. hwang , `` cosmological structure problem in the ekpyrotic scenario , '' phys . d * 65 * , 063514 ( 2002 ) [ arxiv : astro - ph/0109045 ] . s. tsujikawa , `` density perturbations in the ekpyrotic universe and string - inspired generalizations , '' phys . b * 526 * , 179 ( 2002 ) [ arxiv : gr - qc/0110124 ] . j. martin , p. peter , n. pinto - neto and d. j. schwarz , `` passing through the bounce in the ekpyrotic models , '' phys . d * 65 * , 123513 ( 2002 ) [ arxiv : hep - th/0112128 ] . n. deruelle and v. f. mukhanov , `` on matching conditions for cosmological perturbations , '' phys . d * 52 * , 5549 ( 1995 ) [ arxiv : gr - qc/9503050 ] . r. durrer and f. vernizzi , `` adiabatic perturbations in pre big bang models : matching conditions and scale invariance , '' phys . d * 66 * , 083503 ( 2002 ) [ arxiv : hep - ph/0203275 ] ; + c. cartier , r. durrer and e. j. copeland , `` cosmological perturbations and the transition from contraction to expansion , '' phys . d * 67 * , 103517 ( 2003 ) [ arxiv : hep - th/0301198 ] . p. peter and n. pinto - neto , `` primordial perturbations in a non singular bouncing universe model , '' phys . d * 66 * , 063509 ( 2002 ) [ arxiv : hep - th/0203013 ] . j. martin and p. peter , `` parametric amplification of metric fluctuations through a bouncing phase , '' phys . d * 68 * , 103517 ( 2003 ) [ arxiv : hep - th/0307077 ] ; + j. martin and p. peter , `` on the properties of the transition matrix in bouncing cosmologies , '' phys . rev . d * 69 * , 107301 ( 2004 ) [ arxiv : hep - th/0403173 ] . s. tsujikawa , r. brandenberger and f. finelli , `` on the construction of nonsingular pre - big - bang and ekpyrotic cosmologies and the resulting density perturbations , '' phys . d * 66 * , 083513 ( 2002 ) [ arxiv : hep - th/0207228 ] . f. finelli , `` study of a class of four dimensional nonsingular cosmological bounces , '' jcap * 0310 * , 011 ( 2003 ) [ arxiv : hep - th/0307068 ] . l. e. allen and d. wands , `` cosmological perturbations through a simple bounce , '' phys . d * 70 * , 063515 ( 2004 ) [ arxiv : astro - ph/0404441 ] . v. bozza and g. veneziano , `` scalar perturbations in regular two - component bouncing cosmologies , '' arxiv : hep - th/0502047 . v. bozza and g. veneziano , `` regular two - component bouncing cosmologies and perturbations therein , '' arxiv : gr - qc/0506040 . t. j. battefeld and g. geshnizjani , `` perturbations in a regular bouncing universe , '' arxiv : hep - th/0503160 . t. j. battefeld and g. geshnizjani , `` a note on perturbations during a regular bounce , '' arxiv : hep - th/0506139 . a. j. tolley , n. turok and p. j. steinhardt , `` cosmological perturbations in a big crunch / big bang space - time , '' phys . d * 69 * , 106005 ( 2004 ) [ arxiv : hep - th/0306109 ] . s. rasanen , `` on ekpyrotic brane collisions , '' nucl . b * 626 * , 183 ( 2002 ) [ arxiv : hep - th/0111279 ] . s. p. patil and r. brandenberger , `` radion stabilization by stringy effects in general relativity , '' phys . d * 71 * , 103522 ( 2005 ) [ arxiv : hep - th/0401037 ] . s. p. patil and r. h. brandenberger , `` the cosmology of massless string modes , '' arxiv : hep - th/0502069 . s. watson and r. brandenberger , `` stabilization of extra dimensions at tree level , '' jcap * 0311 * , 008 ( 2003 ) [ arxiv : hep - th/0307044 ] . s. watson , `` moduli stabilization with the string higgs effect , '' phys . rev . d * 70 * , 066005 ( 2004 ) [ arxiv : hep - th/0404177 ] . l. kofman , a. linde , x. liu , a. maloney , l. mcallister and e. silverstein , `` beauty is attractive : moduli trapping at enhanced symmetry points , '' jhep * 0405 * , 030 ( 2004 ) [ arxiv : hep - th/0403001 ] . s. p. patil , `` moduli ( dilaton , volume and shape ) stabilization via massless f and d string modes , '' arxiv : hep - th/0504145 . w. israel , nuovo cim . b * 44s10 * , 1 ( 1966 ) [ erratum - ibid . b * 48 * , 463 ( 1967 nucia , b44,1.1966 ) ] . p. binetruy , c. deffayet and d. langlois , `` non - conventional cosmology from a brane - universe , '' nucl . b * 565 * , 269 ( 2000 ) [ arxiv : hep - th/9905012 ] . v. f. mukhanov , h. a. feldman and r. h. brandenberger , `` theory of cosmological perturbations . classical perturbations . quantum theory of perturbations . extensions , '' phys . rept . * 215 * , 203 ( 1992 ) . c. van de bruck , m. dorca , r. h. brandenberger and a. lukas , `` cosmological perturbations in brane - world theories : formalism , '' phys . rev . d * 62 * , 123515 ( 2000 ) [ arxiv : hep - th/0005032 ] . s. watson and r. brandenberger , `` linear perturbations in brane gas cosmology , '' jhep * 0403 * , 045 ( 2004 ) [ arxiv : hep - th/0312097 ] . s. watson , `` uv perturbations in brane gas cosmology , '' phys . d * 70 * , 023516 ( 2004 ) [ arxiv : hep - th/0402015 ] . r. brustein , m. gasperini , m. giovannini , v. f. mukhanov and g. veneziano , `` metric perturbations in dilaton driven inflation , '' phys . d * 51 * , 6744 ( 1995 ) [ arxiv : hep - th/9501066 ] . s. kachru , r. kallosh , a. linde , j. maldacena , l. mcallister and s. p. trivedi , `` towards inflation in string theory , '' jcap * 0310 * , 013 ( 2003 ) [ arxiv : hep - th/0308055 ] . p. horava and e. witten , `` heterotic and type i string dynamics from eleven dimensions , '' nucl . b * 460 * , 506 ( 1996 ) [ arxiv : hep - th/9510209 ] ; + p. horava and e. witten , `` eleven - dimensional supergravity on a manifold with boundary , '' nucl . b * 475 * , 94 ( 1996 ) [ arxiv : hep - th/9603142 ] . r. b. abbott , s. m. barr and s. d. ellis , `` kaluza - klein cosmologies and inflation , '' phys . d * 30 * , 720 ( 1984 ) ; + r. b. abbott , s. d. ellis and s. m. barr , `` kaluza - klein cosmologies and inflation . 2 , '' phys . rev . d * 31 * , 673 ( 1985 ) ; + e. w. kolb , d. lindley and d. seckel , `` more dimensions - less entropy , '' phys . rev . d * 30 * , 1205 ( 1984 ) . j. j. levin , `` inflation from extra dimensions , '' phys . b * 343 * , 69 ( 1995 ) [ arxiv : gr - qc/9411041 ] .

high - resolution spectra of the hot white dwarf g191-b2b , covering the wavelength region 9051187 , were obtained with the far ultraviolet spectroscopic explorer ( _ fuse _ ) . this data was used in conjunction with existing high - resolution hubble space telescope stis observations to evaluate the total h i , d i , o i and n i column densities along the line of sight . previous determinations of(d i ) based upon ghrs and stis observations were controversial due to the saturated strength of the d i lyman line . in the present analysis the column density of d i has been measured using only the unsaturated lyman and lyman lines observed by _ fuse_. a careful inspection of possible systematic uncertainties tied to the modeling of the stellar continuum or to the uncertainties in the _ fuse _ instrumental characteristics has been performed . the column densities derived are :d i ,o i , andn i quoted with uncertainties . the measurement of the h i column density by profile fitting of the lyman line has been found to be unsecure . if additional weak hot interstellar components are added to the three detected clouds along the line of sight , the h i column density can be reduced quite significantly , even though the signal - to - noise ratio and spectral resolution at lyman are excellent . the new estimate of(h i ) toward g191-b2b reads :h i ( uncertainty ) , so that the average ( d / h ) ratio on the line of sight is : ( d / h) ( uncertainty ) .

astro-ph0112180

deuterium is thought to be produced in significant amount only during primordial big bang nucleosynthesis ( bbn ) , and to be thoroughly destroyed in stellar interiors . deuterium is thus a key element in cosmology and in galactic chemical evolution ( see e.g. audouze & tinsley 1976 ; gautier & owen 1983 ; vidal madjar & gry 1984 ; boesgaard & steigman 1985 ; olive _ et al . _ 1990 ; pagel 1992 ; vangioni - flam & cass 1994 ; vangioni - flam _ _ 1995 ; prantzos 1996 ; scully _ et al . _ 1997 ; cass & vangioni - flam 1998 ; tosi _ et al . _ 1998 ) . indeed , its primordial abundance is the best tracer of the baryonic density parameter of the universe , @xmath11 , and the decrease in its abundance during galactic evolution traces the cosmic star formation rate at various epochs . the first ( indirect ) measurement of the deuterium abundance was carried out using @xmath12he in the solar wind , giving the presolar value d / h@xmath13 ( geiss & reeves 1972 ) . the first measurements of the d / h ratio in the interstellar medium ( ism ) were reported shortly thereafter by rogerson & york ( 1973 ) , and their value ( d / h)@xmath14 has remained a landmark average value for the interstellar d / h ratio . finally direct measurements of the primordial ( d / h ) ratio in low - metallicity material at high redshift have been successfully carried out these past few years ( e.g. , burles 2001 for a review , and references therein ) . the values derived cluster around ( d / h)@xmath15 although with significant dispersion , which may or may not be real . quite similarly the measurements of the ( d / h ) ratio in the galactic ism towards hot stars with the _ copernicus _ satellite lead to many evaluations of d / h ( see e.g. york and rogerson 1976 ; vidal madjar _ et al . _ 1977 ; laurent _ et al . _ 1979 ; ferlet _ et al . _ 1980 ; york 1983 ; allen _ et al . _ 1992 ) which also show dispersion around the above york & rogerson ( 1973 ) value . this dispersion has been recently confirmed by imaps observations ( jenkins _ et al . _ 1999 ; sonneborn _ et al . _ 2000 ) , indicating that the d / h ratio may vary by a factor @xmath16 in the solar neighborhood , i.e. , within a few hundred parsecs . in this paper we present a new determination of the d / h ratio on the line of sight to the nearby da white dwarf ( wd ) g191-b2b based on observations obtained with the far ultraviolet spectroscopic explorer ( _ fuse _ , moos _ et al . _ 2000 ; sahnow _ et al . _ 2000 ) . this paper is one of a series in this volume describing the first results of the _ fuse _ ( d / h ) program in the local ism ( lism ) . this program and its results are summarized in the overview paper by moos et al . ( 2002 ) . observing white dwarfs has many advantages over hot and cool stars for studying the d / h ratio , as explained in vidal - madjar _ et al . _ ( 1998 ) : these targets can be chosen close to the sun , in order to avoid a complex line of sight structure , and in the high temperature range , so that the interstellar absorption is superimposed on a smooth stellar continuum . the risk of contamination by low column density h i fluffs possibly present in the hot star winds ( gry , lamers & vidal madjar 1984 ) is negligible for wds , and their hot continuum offers the possibility of observing the numerous uv lines of n i and especially o i , which is a very useful tracer of h i and d i. the ( d / h ) ratio has already been measured toward four white dwarfs , using the hst : g191-b2b ( lemoine _ et al . _ 1996 ; vidal madjar _ et al . _ 1998 ; sahu _ et al . _ 1999 ) , hz43a ( landsman _ et al . _ 1996 ) , sirius b ( hbrard _ et al . _ 1999 ) and feige 24 ( vennes _ et al . for hz43a , sirius b and feige 24 the average d / h values obtained are compatible with the local ism ( lism ) d / h determination ( linsky 1998 ) made within the local interstellar cloud ( the lic in which the sun is embedded ) , although in the case of sirius b this compatibility is marginal . in the case of g191-b2b , it was found that the line of sight comprises one neutral region corresponding to the lic , and two more ionized absorbing components ( lemoine _ et al . _ 1996 ; vidal - madjar _ et al . the average ( d / h ) ratio ( defined as the ratio of the total column densities of d i and h i ) was found to be ( d / h)@xmath17 ( vidal madjar _ et al . _ 1998 ) , significantly different from the value measured toward capella ( d / h)@xmath18 ( linsky 1998 ) . the ( d / h ) ratio measured toward g191-b2b has been contested by sahu et al . ( 1999 ) , who used stis data and concluded to the presence of two interstellar components only , and to a d / h ratio compatible with that observed toward capella . the disagreement resides in the evaluation of the total d i column density ( vidal - madjar 2000 ; sahu 2000 ) , and arises presumably because the lyman @xmath1 d line is saturated and the column density is thus sensitive to the line profile . the number of components assumed on the line of sight may also introduce differences between the analyses of these groups ( see vidal - madjar 2001 for a detailed discussion ) . in the present work we re - examine these issues , making use of high quality _ fuse _ and stis observations of g191-b2b . we first measure the @xmath0(d i ) , @xmath0(o i ) and @xmath0(n i ) column densities using the unsaturated lines of these elements in the _ fuse _ datasets , notably lyman @xmath2 and lyman @xmath3for d i ( section 2 ) . we then analyze the recent high quality stis observations in section 3 , and provide explicit evidence for the presence of three absorbing components ( at least ) on the line of sight . we also provide a refined estimate of the total h i column density . all throughout this work , considerable effort has been put on quantifying possible systematic uncertainties related to fixed - pattern noise , detector artifacts , background uncertainties , wavelength calibration and modeling of the stellar continua , as well as to the velocity structure of the line of sight . in particular , we argue in section 3.3 that previous estimates of the total @xmath0(h i ) are subject to a large systematic uncertainty related to the possible presence of additional weak [ @xmath0(h i)@xmath19@xmath20 hot ( @xmath21k ) components . this effect may have a large impact on our understanding of the observed variations of the ( d / h ) ratio in the ism , as it may affect other lines of sight , and is the subject of a companion paper ( vidal - madjar & ferlet 2002 ) . we provide a summary of our results and a short discussion in section 4 ; an overall discussion of the _ fuse _ results is given by moos _ the _ fuse _ instrument ( moos _ et al . _ 2000 ; sahnow _ et al . _ 2000 ) gives access to the wavelength range 9051187 on eight detector segments , through three different entrance apertures ( the large , medium and narrow apertures lwrs , mdrs and hirs , respectively ) and four channels ( two lif and two sic ) . in particular , it gives access to the higher order unsaturated lyman lines of d i and can thus offer a reliable estimate of @xmath0(d i ) . similarly , many weak lines of o i and n i are present in this bandpass , as shown in table [ fuse - lines ] . the fuse spectrum of g191-b2b in the sic1b channel observed with the hirs aperture is shown in fig . [ fuse_hirs_1bsic ] , in which the lines of d i , n i , o i used in our study are indicated . the white dwarf g191-b2b has been observed several times with _ fuse _ through the three different entrance apertures ; the observation log is given in table [ fuse - gbb ] . since the target is point - like , the resolutions obtained with these apertures should not be too different , but there is a clear variation due to some off - axis effect when moving from one aperture to another . the main problem comes from the geocoronal lyman @xmath2 emission whose strength increases with the surface area of the aperture , and which blends with the interstellar h i absorption . this leads us to employ different approaches in the analysis of the data , depending upon the slit used , as discussed below . the observations were treated through the version 1.8.7 of the calfuse pipeline . the data were then collected channel by channel by series of subexposures . due to the sensitivity of _ fuse _ and the brightness of the source , the data were obtained in the histogram mode ( a spectral imaging mode , in which a two - dimensional spectrum was accumulated onboard ) . each subexposure has a s / n ratio that is high enough to clearly see the strong interstellar or photospheric features . the different subexposures could then be easily aligned on top of each other to compensate for the slow thermal drifts that displace the wavelength scale from one subexposure to another . these drifts never translates into more than @xmath22 pixels . a high s / n spectrum is thus reconstructed in each channel and for each aperture . this process preserves the ultimate _ fuse _ resolution and it also partly eliminates some of the detector s fixed pattern noise by acting like a random fp - split procedure . since the _ fuse _ sampling is @xmath23 pixels per resolution element , we rebinned the spectra by three . we measure column densities by profile fitting of the various lines observed , using the code owens.f developed by one of us ( m. l. ) and the french _ fuse _ team . our general procedure for data analysis is as follows . we split each spectrum into a series of small sub - spectra centered on absorption lines to be analyzed , whose width is a few depending on the line density and fit all lines contained in all sub - spectra simultaneously . between different sub - spectra , the zero - point velocity offset is left to vary to compensate for the wavelength calibration ripple effects in _ fuse _ data ; the line spread function is generally taken to be a gaussian whose fwhm varies freely during the fitting procedure and independently from sub - spectrum to sub - spectrum . it is also checked that considering a double gaussian lsf with free widths , free amplitude ratio and zero pixel offset between the two gaussian components , does not change the results . the background levels used for each line are those evaluated at the bottom of the closest lyman line ( see fig . [ fuse_lyg ] for an example ) , which leads to an upper limit for the derived column densities ; fixing the background level at the 0 value gives access to the lower limit of the evaluated column densities . this uncertainty is taken into account in our procedure ; however _ fuse _ is operating in first order grating mounts and consequently the stray light level is generally quite low , as can be seen in fig . [ fuse_lyg ] for lyman @xmath3 . this analysis procedure allows to reduce systematic uncertainties tied to fixed pattern noise effects or detector artifacts and uncertain calibration , since different lines of the same elements observed through different apertures and in different channels are fitted simultaneously . for instance , if a given line is subject to an artifact in the dataset , the @xmath24 of the fit in the region around this line , or equivalently the quality of the fit of this particular line should stand out in the overall fit if the number of lines analyzed is sufficiently large , as is the case here . moreover , by letting the unknown instrumental parameters vary freely , i.e. , the lsf shape and width , the background flux , the wavelength zero - point and the shape of the continuum , and be optimized simultaneously with the physical parameters that define the absorbers , one marginalizes the final result over these unknowns . therefore the errors attached to the uncertainty on these above characteristics are included in the final errors quoted for the column densities . finally , all lines are also analyzed one by one , _ i.e. _ each independently of all others , in order to check for overall consistency . a detailed discussion of this procedure and survey of possible systematic errors tied to _ fuse _ data is given in a companion paper by hbrard _ ( 2002 ) . since _ fuse _ can not resolve the velocity structure of the line of sight ( see section 3 ) , we have used only unsaturated lines of d i , n i and o i and assumed only one interstellar component to be present . therefore , all measured column densities in this section are integrated over the line of sight . it has been verified that using three interstellar components in the fit of the _ fuse _ did not change the conclusions as long as only non - saturated lines were considered ; an example of single _ vs _ multiple component fits is shown in fig . [ fuse_lyg ] . note that hst observations can not give a reliable measurement of the o i column density without a complete knowledge of the velocity and broadening structure of the line of sight since the only absorption line available at 1302 is strongly saturated . in contrast the n i triplet at 1200 in the ghrs or stis bandpass is only moderately saturated and may provide a precise estimate of the total @xmath0(n i ) . the various measurements of @xmath0(n i ) and @xmath0(o i ) are listed in table [ oi_pub ] . since the contamination of interstellar lines by weak photospheric lines is a likely possibility in the _ fuse _ spectral range due to the high density of atomic lines in this bandpass , we excluded from the fit all lines that presented an obviously excessive equivalent width or column density relative to the others . as an example as shown on figure [ fuse_ni ] the n i line at 954.1042 is very probably blended with a photospheric line which shows up at nearly the same wavelength in the nlte spectrum calculated with the tlusty / synspec codes ( described in section 3 ) . such blending becomes particularly important in the case of g191-b2b because the total o i or n i column densities are small , and their absorptions are relatively weak features which may be more easily blended with undetected photospheric ones . this is much less important for the other _ fuse _ ( d / h ) targets ( see the synthesis of _ fuse _ results by moos _ et al . _ 2002 , and the individual line of sights studies by friedman _ et al . _ 2002 , hbrard _ et al . _ 2002 , lehner _ et al . _ 2002 , sonneborn _ et al . _ 2002 and wood _ et al . _ 2002 ) which present larger column densities , except for the white dwarf hz43a ( kruk _ et al . _ 2002 ) which presents a pure h atmosphere ( barstow , holberg & koester 1995 ; dupuis _ et al . _ 1998 ) and thus no possible blend with photospheric lines . however note that this line selection process may induce in the case of g191-b2b a slight underestimation of the total o i or n i column densities . the o i and n i lines used in this study are listed in table [ fuse - lines ] . some of them were rejected either because they are saturated ( marked `` strong '' in table [ fuse - lines ] ) , or because of a blend with a photospheric feature ( marked `` blend '' ) , or because they are too weak and subject to systematic effects of the noise ( marked `` weak '' ) . some were kept in one channel but not in an other if a nearby detector defect was identified . for this reason , having the same spectrum observed in different independent channels was extremely helpful . by following the general method depicted above , the final column density determinations are : log @xmath0(o i)@xmath25 = 14.86 ( @xmath26 ) ( @xmath27 ) + log @xmath0(n i)@xmath25 = 13.87 ( @xmath26 ) ( @xmath28 ) where the errors are first statistical ( 2@xmath29 ) and second systematic ( ) . the statistical errors were evaluated through the @xmath30 technique , an example of which is given for the @xmath0(d i)determination to follow . however , one important comment is of order at this point . the total @xmath24 for all spectral lines fitted simultaneously ( h i , d i , o i and n i ) in the case of the hirs observations is equal to 3211.44 for 1843 degrees of freedom ( d.o.f . ) , even though the fit appears very satisfactory by eye . this @xmath24 value lies well above the upper 95.5% confidence level limit for statistical fluctuation at 1966.71 . we interpret this large @xmath24 value as a result of a systematic underestimate of the magnitude of the individual pixel errors by the instrument . in order to quantify this uncertainty , we have completed low order polynomial fits over relatively smooth and flat continuum sections close to each of the spectral lines analyzed . these simple fits gave a @xmath24 of 688 for 500 degrees of freedom , again above the 95.5% c.l . upper limit at 563.9 . this indicates that indeed the individual pixel errors are underestimated by instrument , probably due to the presence of fixed pattern noise in the data . the discrepancy in terms of ratio of measured to expected @xmath24 is not as strong as for the best - fit @xmath24 values , but this may be due to differences in the error levels for pixels located on the continuum and those located at the bottom of the saturated lyman lines . in order to compensate for these effects , we have thus decided to rescale all @xmath24 measured , which amounts to an overall rescaling of the individual pixel error estimates . in particular , we have chosen to divide all @xmath24 by @xmath31 , where @xmath32 corresponds to the lowest expected @xmath24 at the 95.5% confidence level for 1843 d.o.f . this rescaling rescales ( increases ) the error bars on the physical parameters that we derive since the @xmath30 contour levels are also rescaled . in effect , in order to derive the @xmath33 error bar around a single parameter minimum value , we search the values of this parameter that lead to an increase in ( non - rescaled ) @xmath24 of @xmath34 instead of @xmath35 . our choice of rescaling is thus very conservative with respect to the final error bars measured . all presented fuse statistical errors will be evaluated following this method . the systematic errors quoted above for o i and n i reflect the range of values obtained for the best - fit solutions for different apertures , given in table [ oi_pub ] . note that in this table , the errors quoted include the satistical error plus part of the above systematic errors , since it accounts for the differences measured between different observations , different channels and different spectral ranges where lines of the species concerned are seen , but observed through the same aperture ( see the discussion in hbrard _ et al . _ note that the total o i or n i column density evaluations made with the lwrs+mdrs are in both cases lower than those made with the other apertures . we do not know the cause of this apparent systematic effect , but included it in the overall uncertainty on our result . the quadratic sum of statistical and systematic uncertainties then leads to : log @xmath0(o i)@xmath25 = 14.86 ( @xmath36 ) + log @xmath0(n i)@xmath25 = 13.87 ( @xmath36 ) in the _ fuse _ bandpass , d i is clearly detected at lyman @xmath2 , weakly at lyman @xmath3 , and not at lyman @xmath37 . we list in table [ di_pub ] the different estimates of the total column density of d i that have been obtained from hst observations of lyman @xmath1 , and which we measure using the present _ fuse _ data . here as well , we assumed that only one component is present on the line of sight since the d i lines are optically thin . again we checked the impact of a more complex line of sight structure and found it to be negligible , as expected . we carried out the same investigations of systematic effects as for o i and n i , using the approach detailed above and in hbrard _ ( 2002 ) . there is however one essential difference between d i and o i or n i in terms of systematics . namely the continuum to the d i absorption is provided by the blue wing of the corresponding h i absorption , while the continuum around the o i and n i is very smooth . it is thus necessary to measure the possible systematic tied to the estimate of the stellar continuum on the measurement of @xmath0(d i ) . in order to do so , we have measured @xmath0(d i ) with and without beforehand correction of the data by a theoretical nlte stellar continuum ( described in section 3 ) . some corresponding fits are shown in fig . [ fuse_lyg ] , and the best - fit to the hirs data is shown in fig . [ aas_fig2 ] . the effect of any assumption on the stellar continuum was found to be negligible as compared to the statistical uncertainty . however , we have found that the value of @xmath0(d i ) measured if the h i lines are excluded from the fit and their blue wing modeled by a polynomial is higher than when the h i are included in the fit . this systematic effect may result from the following : in the latter ( standard ) approach , where both d i and h i are fitted together , the position of the d i absorption is tied to that of h i , i.e. , they have the same radial velocity and their zero point wavelength offsets are the same as both lines appear in the same sub - spectrum . however , in the former approach , the velocity of d i is not subject to this constraint . these two approaches are denoted in table [ di_pub ] as `` d i & h i '' ( standard approach ) and `` d i , no h i '' for the other one . one should note that a discrepancy in radial velocity between the d i line and its counterpart h i could be attributed to small scale ripples in the _ fuse _ wavelength calibration or to the presence of weak h i absorbers which would shift the position of the h i line with respect to that of the d i line in which these absorbers would not be felt . in this respect the approach in which d i is fitted independently of h i seems more adequate ; however the interpolation of the blue wing of the h i line by a polynomial is not always straightforward . we should mention that this systematic effect is not fully understood and does not seem to be always present on different lines of sight . in any case it is included in our estimate of the systematic error . finally a small additional systematic effect is found to be attached to the uncertainty in the shape of the lsf , i.e. whether it is a single gaussian or a double gaussian with wide wings ( indicated as `` double lsf '' in table [ di_pub ] ) . as mentioned previously , a single gaussian lsf has a free fwhm in the fit , while a double gaussian lsf has free amplitude ratio between both components , free fwhm for each , but separation between both components fixed to zero pixel . the impact of the shape of the lsf on the measured @xmath0(d i ) is shown in figure [ aas_fig3 ] where the @xmath38 is plotted as a function of @xmath39@xmath0(d i ) . in practice this curve is calculated by fixing @xmath39@xmath0(d i ) to a given value , finding the best fit @xmath24 for this value with all other parameters free , and plotting the difference between this @xmath24 and the best @xmath24 obtained for all possible values of @xmath39@xmath0(d i ) . in this figure , the curvature of a @xmath30 curve gives the statistical error , as usual , while the relative shift between two curves corresponding to two different sets of assumptions is interpreted as an estimate of the @xmath40 systematic error tied to the uncertainty on the assumptions . we thus conclude that the total d i column density on this line of sight measured using the hirs data is : @xmath41 i@xmath42 + where the errors are first statistical ( 2@xmath29 ) and second systematics ( ) . combining these errors leads to : @xmath41 i@xmath43 as can be seen from the comparison between the evaluations of @xmath0(d i)made for the different apertures , there is no significant systematic effect tied to the aperture , contrary to the analysis for o i and n i. the average @xmath0(d i ) measured through the different apertures then reads : @xmath41 i@xmath44 this value agrees with all previously measured values of @xmath0(d i ) shown in table [ di_pub ] , except with that of sahu _ et al . _ ( 1999 ) , using the stis - ech#1 data , which gave @xmath39@xmath0(d i)@xmath45 . it thus appears that the discrepancy between this value and the others should be attributed to the dataset used , in agreement with the conclusion of vidal - madjar ( 2001 ) , and the debate around the value of @xmath0(d i ) is now settled . in this section we analyze new high resolution high signal - to - noise stis data of g191-b2b in order to determine the number of components on the line of sight and provide a new estimate of the total neutral hydrogen column density . as mentioned in section 1 , there exists a controversy in the literature about the total number of absorbing components , as sahu _ ( 1999 ) claimed to see two only , whereas vidal - madjar _ et al . _ ( 1998 ) claimed that three at least were present . in this section , we present concrete evidence in favor of the latter . previous measurements of @xmath0(h i ) are compiled in table [ hi_pub ] ; these values scatter around a ( non - weighted ) mean @xmath39@xmath0(h i)@xmath46 ( the dispersion corresponds to the non - weighted dispersion of the individual measurements around the mean ) . individual measurements tend not to agree with each other , and therefore a new independent estimate is useful . moreover we argue in this section that the total @xmath0(h i ) toward g191-b2b is , as a matter of fact , much less well known than previously thought , since additional weak hot absorbing components may strongly affect the column density estimate from lyman @xmath1 . the new stis data analyzed here represent about three times the total exposure time cumulated hitherto toward g191-b2b . all stis observations were extracted following the method of howk and sembach ( 2000 ) . we selected spectral lines corresponding to n i , o i , si ii , si iii , c ii , s ii , s iii in the e140h echelle configuration and fe ii lines in the e230h one , both at @xmath47 resolving power , as reported in table [ stislines ] . all these species are seen in one or more spectral lines , strong and/or weak , and cover a wide range of atomic masses , so that this data should provide strong constraints on the structure of the line of sight as well as on the component to component physical state and thermal and non - thermal broadening of each . in order to verify that no photospheric line is blended with one of the interstellar lines analyzed , we have computed a synthetic spectrum of g191-b2b which includes all species observed in the atmosphere of g191-b2b , i.e. , c , n , o , si , p , s , fe , and ni ( lanz _ et al . _ 1996 ; vennes _ et al . we employ the program tlusty developed by hubeny & lanz ( 1995 ) to compute a nlte metal line - blanketed model by using the atmospheric parameters of barstow _ et al . _ ( 1998 ) , i.e. , @xmath48 and @xmath49k , and the abundances determined by barstow _ ( 2001 ) . from the comparison of the observations with the model calculation it was possible to select the lines where no or very weak photospheric features are coincident with the interstellar ones , as for the _ fuse _ data . furthermore it is possible that some photospheric lines are present but not predicted by the model ; we thus fitted simultaneously as many different spectral lines of each species as available , in order to identify and minimize the effect of unpredicted photospheric lines , as was done for the analysis of the _ fuse _ data . this approach also reduces the possible impact of instrumental defects which may be present in some area of the detector but not in others . following the general method of analysis described in the previous section , we split the stis data into sub - spectra , each of them centered on one spectral line and having typical width @xmath50 . all sub - spectra are then fitted simultaneously . the wavelength zero point offsets are left free to vary during the fit . note that shifts in the wavelength calibration from region to region could still be present even after a careful wavelength scale calibration . as an example , we detected an erroneous 4 km s@xmath51wavelength shift between the ghrs si iii line position and that in the stis data . after comparison with ghrs first order grating observations , we confirmed that the stis calibration was correct , and the ghrs calibration erroneous . we found that the average relative shift between two spectral regions in the stis data is @xmath52 km s@xmath51 ( @xmath8 error ) . rms _ error of 0.5 km s@xmath51 in the stis calibration is indeed compatible with the @xmath533.3 km s@xmath51 resolution of the instrument . the stellar continua in each region are interpolated by low order polynomials ( except for lyman @xmath1 where the theoretical continuum is used in some cases , see below ) , and the background level in each region is determined using the closest saturated line . the typical background flux is @xmath54 ( 2@xmath29 error ) of the continuum for the e140h data , and @xmath55 ( 2@xmath29 error ) for the e230h data . finally in some cases ( see below for details ) , the lsf is left free to vary and corresponds either to a simple gaussian or to a double gaussian . again , we emphasize that the uncertainty on these instrumental characteristics is contained in the final error bars given below , which we determine using a @xmath30 method , since the corresponding parameters are left free to vary during the fit . in the case of the stis observations made with the e140h and e230h echelle modes , we used the tabulated lsf , corresponding to the slit used for these observations ( [email protected] arc sec ) . this is an important issue because these lsfs possess wings that could have some impact on the precise determination of the physical parameters we are aiming at . to test the quality of the fits we also tried a single gaussian lsf with free fwhm . we found excellent fits with this freely variable width single gaussian lsf for both echelle datasets . we find for the e140h an average fwhm@xmath57 pixel ( 2@xmath29 ) and for e230h , fwhm@xmath58 pixel ( 2@xmath29 ) . these determinations are certainly compatible with the tabulated lsf in terms of `` average '' widths , but are different in terms of shape . they correspond to 80,000 and 100,000 resolving power respectively for the e140h and e230h spectra . finally , we also tested double gaussian lsfs but found no significant improvement . we thus decided to present results with both types of lsfs ( free gaussians and tabulated ) in order to show the stability and robustness of our column density determinations . as we will see , such a change has negligible impact on the derived total h i column density . following the analysis of the _ fuse _ data , we evaluate the accuracy of the intrument estimate of the individual pixel errors by completing low order polynomial fits over relatively smooth and flat continuum sections close to each of the fitted spectral lines . these simple fits give a total @xmath24=921.37 ( sum of all @xmath24 in the different portions of continuum analyzed ) for 661 degrees of freedom , which lies well above the upper 95.5% confidence level limit of 723.9 for a @xmath24 distribution with 661 d.o.f . therefore we rescale all @xmath24 by a common factor @xmath59 , where 600.63 corresponds to the lower 95.5% confidence level limit for 661 d.o.f . finally , in the course of the analysis , the physical characteristics of each absorbing component ( velocity @xmath60 , temperature @xmath61 and micro - turbulent broadening @xmath62 ) are determined ; however we will not provide a detailed analysis of these characteristics and their uncertainties since we are primarily interested in the total number of absorbing components and the total neutral hydrogen column density . in order to settle the debate on the total number of absorbing components , we have studied two and three component solutions . simple eye inspection of the data indicates that at least two components are present on the line of sight ; as mentioned previously , the red component can be identified with the lic , a moderately neutral region , while the blue component is clearly more ionized , and , as we argue , more complex than a single absorber ( see vidal - madjar _ et al . _ 1998 and sahu _ et al . _ 1999 ) . the result of this analysis is summarized in table [ chi2 ] , which gives the @xmath24 values obtained for the fits of various spectral regions , and for all regions ( excluding lyman @xmath1 ) , for two and three component solutions , using either the tabulated stis lsf or a freely varying single gaussian lsf . it is clear that the improvement in @xmath24 when going from two to three components is extremely high whichever lsf is used . this is particularly true for fe ii in which case the improvement can be seen directly on figure [ fe_2comp_3comp ] . indeed if one looks closely at the fe ii lines , one can see an asymmetry of the line profile of the bluer component which is common to all fe ii lines . this constitutes clear evidence for the presence of three absorbing components and not two . one sees this third component in fe ii and not in other lines since fe ii is the heaviest species , and its line widths are thus the smallest . moreover its lines have been observed through the echelle presenting the higher resolving power ( 100,000 ) . this suffices to reveal the double structure of component b : components b1 and b2 are separated by @xmath63 km s@xmath51 ( 2@xmath29 ) , _ i.e. _ of the order of the e140h spectral resolution but more than that of the e230h . the radial velocities of the three components is estimated as @xmath64 km s@xmath51 , @xmath65 km s@xmath51and @xmath66 km s@xmath51 , with @xmath40 errors due to the wavelength calibration . the lic velocity is in perfect agreement with its projected velocity on the line of sight . one can also see further evidence for the double structure b1-b2 in the s iii 1190.2 line shown in the upper right panel of fig . [ stis_3comp_fig3 ] . this line is seen only in the bluest component b1 , which reflects the difference in ionization between the various components ; its spectral position is quite precisely known as it is located close to the si ii line at 1190.4 and is seen in two different spectral orders ( see table [ stislines ] ) . note that the position of most lines is actually very well controled , which is important to determine the number of absorbing components . in effect , the sharp geocoronal lines of n i , o i and o i@xmath67 are in several locations ( see fig . [ stis_3comp_fig3 ] ) , notably close to interstellar lines of n i , o i and si ii . other si ii lines or fe ii lines are well constrained since in each of these species the absorption due to the lic is sharp and pronounced , and serves as a useful spectral reference . in order to test the hypothesis that the presence of a third component is not required by the fit , we perform an @xmath68test . this test uses the fisher - snedecor law which describes the probability distribution of the @xmath24 ratio . here we test the probability that the decrease of the @xmath24 due to the inclusion of a third component could be due to the increase of free parameters and not to actual information contained in the data . the result is shown species by species and for all species fitted simultaneously in table [ chi2 ] , and gives a probability @xmath69 that this third component is not required by the data . the @xmath68tests performed species by species also confirm our previous impression that si ii and fe ii are the two most sensitive species to the presence of this component . we also note that the need for a third component is present whichever lsf is used . we note that the @xmath24 values corresponding to the use of the tabulated stis lsf are too high , as they all lie above the upper 95.5% confidence level limit for statistical fluctuation ( an except being fe ii for three components ) . in contrast , the @xmath24 values for a freely varying single gaussian lsf are satisfactory , since only o i and si ii stand out with @xmath24 beyond the 95.5% confidence level for the three component solution . their high @xmath24propagates into the total @xmath24 summed over all windows . however the si ii 1190 , 1193 and o i 1302 lines are saturated , and small errors in the estimates of the error bars at the bottom of the line could potentially explain large differences in @xmath24 . in effect our procedure of rescaling the @xmath24 amounts to rescaling all error bars by a common factor , which may not be correct for pixels at the bottom of saturated lines for the following reason . error bars on flux values arise as a combination of background noise , fixed pattern noise and poisson noise . background noise dominates for pixels with flux values close to zero , while for high s / n data poisson noise or fixed pattern noise would dominate in the continuum . if the errors in the estimation of the noise array for the background component and for the continuum component ( fixed pattern or poisson ) are different , then the procedure of rescaling error bars with a common factor is , strictly speaking , incorrect . finally , since this rescaling factor was estimated from the continuum , this rescaling should be correct for all pixels whose error is not dominated by the other noise component ( background ) . we also note that this o i line has a complex profile , with two geocoronal and one nearby photospheric feature ( see figure [ stis_3comp_fig3 ] ) . nevertheless , we decided to keep this line in our analysis as it adds important constraints on the b1 and lic component t and @xmath62 parameter evaluations . using the @xmath68test , we can determine the probability that the use of the freely varying gaussian lsf is not required ( instead of using the stis lsf ) . in all cases , it is found that this probability is smaller than @xmath70 , and therefore we conclude that the gaussian lsf with a width that varies from spectral region to spectral region is required to fit the stis data . this result may be unexpected but it can find a simple explanation in the fact that the tabulated stis lsf is an average lsf given over the whole spectral range , while our gaussian lsf has a fwhm that varies with wavelength . for standard datasets it is probably sufficient to use the stis lsf but the very high quality of the present data necessitates higher order corrections to this lsf . in particular we noted a slight broadening of the lsf near the order edges . furthermore , the wings of the tabulated lsf are not so important and the single gaussian lsfs are sufficient to properly fit the data . we thus conclude that three absorbers are present on the line of sight to g191-b2b . the background level is directly measured at the bottom of the strongly saturated lyman @xmath1 line . it is found to be of the order of @xmath71 ( 2@xmath29 ) of the nearby continuum , i.e. within a few percent . variations in the photospheric continuum over the interstellar lyman @xmath1 line presents an additional difficulty . indeed the instrument sensitivity over this spectral region is evaluated by using model spectra of white dwarfs and comparing them to the data , and one of the calibration wds used is g191-b2b itself . therefore it is difficult to separate the instrumental effects from the photospheric intrinsic profile in the spectrum processed through the stis pipeline . for this reason , we follow the approach detailed in vidal madjar _ ( 1998 ) and further discussed by vidal madjar ( 2000 ) , and evaluate simultaneously all parameters and the stellar continuum , which we model in two different ways . in the first approach , we model this continuum as a polynomial whose parameters are adjusted during the fitting procedure , and which represents the real photospheric profile times the instrument sensitivity ; we denote this fitting model `` _ u _ '' , which stands for uncorrected ( as the data have not been corrected prior to the fit by a stellar model ) . in the second approach , we correct the data prior to the fit by a stellar model and keep a free low - order polynomial during the fit which models the inaccuracy of the stellar model and the variation of the instrument sensitivity ; we denote this approach by `` _ the calculated stellar model was shifted by 24.56 km s@xmath51 in order to be consistent with the velocity of the photospheric features clearly detected in n v and si iii ( see figure [ stis_3comp_fig3 ] ) . these two approaches have been followed in parallel to estimate the uncertainty in the modeling of the stellar continuum on the final solution . we use two models to compute the photospheric lyman @xmath1 line profile . first , we use the program synspec ( i. hubeny 2000 , private communication ) in conjunction with the tlusty nlte metal line - blanketed model described above . this version of synspec contains the lemke s stark broadening tables for hydrogen , which were computed by lemke ( 1997 ) within the framework of vidal , cooper , & smith ( 1973 ) . second , we use detlev koester s lte code and his best fit parameters derived under the assumption of pure - h lte ( @xmath72 and @xmath73 ) . figure [ lyaandmodel ] shows the nlte and lte lyman @xmath1 line profiles and illustrates that nlte effects are significant only in the core of the line , which is formed high in the atmosphere where departures from lte are important ( wesemael _ et al . _ 1980 ) . however , as is shown in fig . [ lyaandmodel ] the observations do not contain any information in that central region of the photospheric line simply because it is lost at the bottom of the saturated interstellar lyman @xmath1 line . our fitting procedure will however be able to test the slightly different slopes in the wings of the models . an additional correction made by a low - order polynomial is needed not only to take into account uncertainties in the instrument sensitivity but also because when one fits a section of the g191-b2b stellar continuum devoid of stellar or interstellar absorption , it is necessary to model the continuum with a polynomial of degree of order 1 to 3 for a spectral width of order 0.3 ( see for instance fig . [ stis_3comp_fig3 ] ) . for a spectral region of width @xmath74 as is the case for lyman @xmath1 , a 6@xmath75 order polynomial provides a satisfactory approximation to the stellar continuum . the best - fit solutions obtained for 3 components on the line of sight for all lines in the stis domain including lyman @xmath1 and for various models ( prior correction or not of the lyman @xmath1 stellar continuum , type of lsf ) are summarized in table [ 3complya ] . the zero point wavelength of the lyman @xmath1 spectral domain was found to be compatible with the other regions well within the @xmath40 uncertainty of 0.5 km s@xmath51 . concerning the fit with a free double gaussian ( model 5 in table [ 3complya ] ) , we found that the wings of the broader gaussian ( about 5 to 10 times larger than the narrow one ) contribute in amplitude ( relatively to the amplitude of the narrow core gaussian ) to less than 1@xmath76 in all but 5 spectral windows , less than 2@xmath76 in 3 of these 5 and to about 3@xmath76 in the 2 remaining ones . since this effect is marginal , we kept only as comparison tests the use of both the tabulated stis lsf or the simple gaussian lsf . all these effects were included in the evaluation of the statistical errors . it appears that different values of @xmath0(h i ) are obtained depending on the model used ; in particular the comparison of models 1 and 2 on the one hand , and of models 3 and 4 on the other hand shows that the value of @xmath0(h i ) derived depends whether the stellar continuum at lyman @xmath1 has been corrected or not beforehand by a theoretical stellar profile . one obtains @xmath39@xmath0(h i)=18.37 for an uncorrected profile , and @xmath39@xmath0(h i)=18.32 for a corrected stellar profile . since there is no significant difference in @xmath24 between these solutions , this discrepancy must be interpreted as a systematic uncertainty tied to the choice of modeling of the continuum . a @xmath30 analysis around each of these solutions give a statistical error of @xmath77 dex ( @xmath40 ) , so that the value of @xmath0(h i ) is dominated by the above systematic error . furthermore after correction of the stellar continuum by an lte calculation , which as shown in fig . [ lyaandmodel ] represents an intermediate situation between nlte and non - corrected stellar profiles , the value of @xmath0(h i ) derived was found to be an intermediate value between the above two . however a further systematic error appears if additional weak hot components are present on the line of sight , and its investigation is the subject of the following section . one can not exclude the presence of weak h i components which could perturb or bias the measurement of @xmath0(h i ) , but could not be detected in any other species due to their weak column density . such absorbers could arise as the signature of high velocity shocks ( cowie _ et al . _ 1979 ) , or as cloud interfaces with the hot gas within the local ism ( bertin _ et al . _ 1995 ) or as `` hydrogen walls '' , i.e. , the shock interaction between the solar wind ( or stellar wind ) and the surrounding ism ( linsky 1998 ) . the latter possibility has been modeled by wood _ et al . _ ( 2000 ) , and a prediction for the line of sight toward g191-b2b is shown in fig . [ hwall ] overlaid on the stis data . it shows that most of the expected absorption should take place in the saturated core of the interstellar line , but some weak absorption ( @xmath78 ) may be present , extending over several tenths of an angstrom on the red side of the line , due to the neutral hydrogen atoms seen behind the shock in the downwind direction where g191-b2b is located . we have thus investigated the possible impact of additional weak hot components on the determination of @xmath0(h i ) . in order to do so , we have added one or two additional components in h i only , and performed the fit of all lines as before with an nlte correction of the lyman @xmath1profile and using a free single gaussian lsf ( model 3 ) . in order to constrain the presence of these extra absorbers , we have added to the fit the four lyman @xmath2 lines observed through the _ fuse _ hirs aperture . for each new model characterized by the number of additional components , we have measured @xmath0(h i ) using a @xmath30 analysis . the best fit solutions with zero , one and two extra absorbers are shown for lyman @xmath1 and lyman @xmath2 in fig . [ lyalyb ] . we have found that this introduction of additional components leads to a substantial decrease of the total @xmath0(h i ) together with a significant improvement of @xmath24 . in particular , we found that the best fit solution for one extra component leads to @xmath79 ( 1961 d.o.f . ) and @xmath39@xmath0(h i)=18.24 , and two extra components lead to @xmath80 ( 1958 d.o.f . ) and @xmath39@xmath0(h i)=18.11 . the @xmath24values quoted have been rescaled by a factor 1.53 corresponding to the stis data obtained in the previous section , and which remains close ( 20% ) to the factor measured in the _ fuse _ data range . moreover , if a @xmath30 analysis is performed for each model , the curve obtained flattens for low values of @xmath0(h i ) down to @xmath39@xmath0(h i)=@xmath81 ; in other words , the space of solutions becomes degenerate in the low @xmath0(h i ) region , and consequently the value of @xmath0(h i ) measured from the profile fitting of the lyman @xmath1 region is subject to a large systematic uncertainty . this behaviour can be explained as follows . in the three component best fit solution , most of the hydrogen is contained in the lic whose absorption makes the red wing of lyman @xmath1 . in the solutions with extra absorbers , the combined profile of these hot components can replace the damping wing contribution of the lic , as shown in fig . [ lyalyb ] . in order to scan the parameter space , the @xmath0(h i)@xmath82 has been forced to take various values , since the profile fitting code can not force the total h i column density to take a particular value . when the lic @xmath0(h i ) becomes negligible with regards to the total @xmath0(h i ) , we force the next dominant component to take various values of @xmath0(h i ) . in this way it is possible to scan the parameter space of the total @xmath0(h i ) . one finds that once the lic has become negligible , component b2 makes most of the total @xmath0(h i ) , and that when the total @xmath0(h i ) is taken below @xmath81 , the @xmath24 rapidly increases , which gives the final error bar on the total @xmath0(h i ) estimate . at this stage , one should point out that these best - fit solutions are quite difficult to find . the impact of additional weak absorbers on the profile fitting of @xmath0(h i ) had already been studied by vidal - madjar _ et al . _ ( 1998 ) but no significant effect had been found , as the above solutions had eluded detection . the above solutions have actually been obtained only in a late stage of the present work . we interpret this as evidence for the fact that the @xmath24 surface becomes complex when additional unconstrained absorbers are introduced in the fit of the fully blended lyman @xmath1 profile and optimization is then delicate . one should also remark that we have modeled these extra absorbers using voigt profiles , which implicitly assumes a maxwellian distribution for the velocities of the atoms . however if these absorbers correspond to interstellar structures similar to hydrogen walls , the latter assumption is incorrect and the overall profile should be closer to that modeled by wood _ et al . _ ( 2000 ) and shown in fig . [ hwall ] . for this reason , the above modeling and solutions should be interpreted as a first approximation . for this reason as well , we do not discard the values for @xmath0(h i ) obtained in the previous section without additional absorbers , even though the presence of extra absorbers induce a significant gain in @xmath24 . at the worst the final value for @xmath0(h i ) will be more conservative . with these remarks in mind , we conclude that the total hydrogen content toward g191-b2b should be contained in the interval : @xmath83 i@xmath84 in the absence of any indication on the distribution of the errors for @xmath0(h i ) , we consider the above two extreme values as @xmath8 limits since they correspond to rather extreme solutions , and quote the following value for @xmath0(h i ) : @xmath39@xmath0(h i)@xmath85 ( @xmath40 error ) . one should note that this effect of additional absorbers had not been found in previous studies of the total hydrogen content toward g191-b2b . therefore it is important to remark that the above large uncertainty must also affect the previous values of @xmath0(h i ) . in other words , the above results supersedes any previous estimate of @xmath0(h i)made toward g191-b2b using profile fitting of the lyman @xmath1 line . one may also wonder whether extra absorbers may affect the determination of @xmath0(h i ) toward other stars ; this important question is the subject of a forthcoming paper ( vidal - madjar & ferlet 2002 ) . on the other hand , the above value for @xmath0(h i ) should be contrasted with the value derived from euve observations and modeling of the atmosphere of g191-b2b which gave @xmath86(h i)=@xmath87 ( dupuis _ et al . _ 1995 , 2@xmath29 error ) , @xmath86(h i)=@xmath88 ( lanz _ et al . _ 1996 , no error bar quoted ) and the recent detailed measurement @xmath89(h i)=@xmath90 ( barstow & hubeny 1998 ; this value also agrees with the more recent work of barstow , hubeny & holberg 1999 ) . these values would tend to indicate that the contamination of the lyman @xmath1 profile by weak hot absorbers ( if any ) is not important , as they agree with the values we obtained without including such extra absorbers however , as noted in barstow & hubeny ( 1998 ) , these fits of the euve spectrum of g191-b2b typically lead to very large and unexplained reduced @xmath24 , implying that the overall fit is not yet satisfactory in spite of the use of sophisticated wd atmosphere models , and that some unknown effects have yet to be accounted for . we have measured the total column densities of d i , n i and o i toward g191-b2b using unsaturated absorption lines of these elements in high quality _ fuse _ spectra . after a careful examination of the possible systematic uncertainties tied to the choice of the stellar continuum and to the instrumental configuration , we have derived the following column densities with @xmath8 uncertainties : we have also analyzed new high signal - to - noise ratio high resolution stis observations of g191-b2b and provided concrete evidence for the presence of at least three interstellar absorbing components on the line of sight by analyzing the interstellar absorption lines of n i , o i , si ii , si iii , s iii and fe ii present in the stis bandpass . we have also measured the total hydrogen column density on the line of sight using the velocity structure derived from the above metals . we have performed an exhaustive study of systematic effects on the value of @xmath0(h i ) . in particular we have discovered a new major source of uncertainty on @xmath0(h i ) tied to the possible presence of additional weak hot absorbers whose combined absorption profile can contribute significantly to the wings of the blended lyman @xmath1 profile . the column density of these absorbers is small compared to the other main components , and they would not be detected in any other species than h i , but their contribution to the lyman @xmath1 absorption profile can reduce significantly the total h i column density measured from the profile fitting . in order to constrain their impact , we have analyzed simultaneously lyman @xmath1 and the higher order lyman lines , and concluded that the best value of @xmath0(h i ) toward g191-b2b is : we emphasize that this uncertainty is a systematic uncertainty which had gone unnoticed before . therefore the above result supersedes previous estimates of @xmath0(h i ) toward g191-b2b obtained from the profile fitting of lyman @xmath1 . a detailed analysis of this uncertainty and its consequences on @xmath0(h i ) determinations toward other stars is discussed in a companion paper ( vidal - madjar & ferlet 2002 ) . we thus derive the following neutral abundance ratios toward g191-b2b , with @xmath8 uncertainties : most of the uncertainty in the above result results from the systematic uncertainty on the @xmath0(h i ) determination . this clearly shows the importance of measuring accurate ( d / o ) and ( d / n ) ratios in the interstellar medium instead of abundances relative to hydrogen , as emphasized by timmes _ interestingly if one uses the recent measurement of @xmath0(h i ) from the modeling of the atmosphere of g191-b2b and the fit of the euve spectrum , @xmath86(h i)@xmath99 ( @xmath8 ) , one finds @xmath100 ( @xmath8 ) . at this stage , however , due to the uncertainty inherent to the modeling of the white dwarf atmosphere , it is probably more conservative to use the interstellar determination for @xmath0(h i ) , and therefore the previous value of the ( d / h ) ratio . the above new value for the ( d / h ) ratio agree with the range of values measured by linsky ( 1998 ) toward a dozen stars of the lism and with the values previously derived toward g191-b2b . however the discrepancy between previous estimates of the ( d / h ) ratio toward g191-b2b and the lism average d / h ratio has disappeared due to a revision of the uncertainty on the estimation of the total h i content . a detailed interpretation of this ( d / h ) value and of the accompanying ( d / o ) and ( d / n ) ratios and their implications is provided in a companion paper by moos _ ( 2002 ) . rccccc 1025.4440 & d i & 0.264 @xmath102 & 0.190 @xmath103 & + 1025.4429 & d i & 0.527 @xmath102 & 0.190 @xmath103 & + 972.2725 & d i & 0.967 @xmath104 & 0.813 @xmath105 & + 972.2721 & d i & 0.193 @xmath102 & 0.813 @xmath105 & + 949.4848 & d i & 0.465 @xmath104 & 0.421 @xmath105 & weak + 949.4846 & d i & 0.929 @xmath104 & 0.421 @xmath105 & weak + 1134.9803 & n i & 0.435 @xmath102 & 0.150 @xmath103 & + 1134.4149 & n i & 0.297 @xmath102 & 0.154 @xmath103 & + 1134.1653 & n i & 0.152 @xmath102 & 0.158 @xmath103 & + 954.1042 & n i & 0.676 @xmath104 & 0.330 @xmath105 & blend + 953.9699 & n i & 0.348 @xmath102 & 0.170 @xmath103 & + 953.6549 & n i & 0.250 @xmath102 & 0.183 @xmath103 & + 953.4152 & n i & 0.132 @xmath102 & 0.193 @xmath103 & + 1039.2303 & o i & 0.920 @xmath104 & 0.947 @xmath105 & + 1026.4757 & o i & 0.246 @xmath104 & 0.111 @xmath105 & weak + 1026.4744 & o i & 0.187 @xmath106 & 0.118 @xmath107 & weak + 1025.7633 & o i & 0.201 @xmath106 & 0.212 @xmath107 & blend + 1025.7626 & o i & 0.302 @xmath104 & 0.191 @xmath105 & blend + 1025.7616 & o i & 0.169 @xmath102 & 0.765 @xmath105 & blend + 988.7734 & o i & 0.465 @xmath102 & 0.226 @xmath103 & strong + 988.6549 & o i & 0.830 @xmath104 & 0.566 @xmath105 & blend + 988.5778 & o i & 0.553 @xmath106 & 0.629 @xmath107 & weak + 976.4481 & o i & 0.331 @xmath104 & 0.386 @xmath105 & + 971.7382 & o i & 0.116 @xmath102 & 0.585 @xmath105 & strong + 971.7376 & o i & 0.207 @xmath104 & 0.146 @xmath105 & blend + 950.8846 & o i & 0.158 @xmath104 & 0.194 @xmath105 & + 948.6855 & o i & 0.631 @xmath104 & 0.100 @xmath103 & + 936.6295 & o i & 0.365 @xmath104 & 0.100 @xmath103 & + 929.5168 & o i & 0.229 @xmath104 & 0.100 @xmath103 & + 925.4460 & o i & 0.354 @xmath106 & 0.459 @xmath107 & weak + 924.9500 & o i & 0.154 @xmath104 & 0.100 @xmath103 & + 921.8570 & o i & 0.100 @xmath104 & 0.562 @xmath107 & weak + cccc hst ghrs - ech . & 14.84@xmath108 & 13.90@xmath109 & vidal madjar _ et al . _ 1998 + lwrs+mdrs & 14.79@xmath108 & 13.82@xmath36 & this work + mdrs & 14.84@xmath110 & 13.89@xmath111 & this work + hirs & 14.88@xmath111 & 13.84@xmath36 & this work + -all & 14.86@xmath36 & 13.87@xmath36 & this work + cll hst ghrs - ech . & 13.43@xmath109 & vidal madjar _ et al . _ 1998 + hst stis - ech.#1 & 13.55@xmath112 & sahu _ et al . _ 1999 + hst ghrs - ech . & 13.40@xmath108 & sahu _ et al . _ 1999 + lwrs+mdrs & 13.41@xmath113 & d i , no h i , this work + mdrs & 13.41@xmath114 & d i & h i , this work + hirs & 13.36@xmath110 & d i & h i ( fig . [ aas_fig2 ] ) , this work + hirs & 13.38@xmath111 & d i & h i , double lsf , this work + hirs & 13.42@xmath110 & d i , no h i , this work + all & 13.40@xmath36 & this work + lcccc 18.315@xmath115 & euve & & & dupuis _ et al . _ 1995 + 18.32 & euve & & & lanz _ et al . _ 1996 + 18.36@xmath108 & ghrs - g160 m & 3 & free & lemoine _ et al . _ 1996 + 18.38@xmath109 & ghrs - ech . & 3 & free & vidal madjar _ et al . _ 1998 + 18.30@xmath114 & euve & & & barstow & hubeny 1998 + 18.31@xmath116 & stis - ech . # 1 & 2 & fixed & sahu _ et al . _ 1999 + 18.34@xmath109 & ghrs - ech . & 2 & fixed & sahu _ et al . _ 1999 + lllllccc 1334.5320 & c ii & 0.128 & 0.288 @xmath103 & 316 + 1200.7098 & n i & 0.0430 & 0.398 @xmath103 & 350 , 351 + 1200.2233 & n i & 0.0862 & 0.399 @xmath103 & 350 , 351 + 1199.5496 & n i & 0.130 & 0.401 @xmath103 & 351 + 1242.8040 & n v & 0.500 & 0.336 @xmath103 & 339 + 1302.1685 & o i & 0.0519 & 0.340 @xmath103 & 323 + 1304.3702 & si ii & 0.0917 & 0.107 @xmath117 & 323 + 1193.2897 & si ii & 0.585 & 0.409 @xmath117 & 352 , 353 + 1190.4158 & si ii & 0.293 & 0.410 @xmath117 & 353 , 354 + 1206.5000 & si iii & 1.67 & 0.255 @xmath117 & 349 + 1259.5190 & s ii & 0.0166 & 0.465 @xmath105 & 334 + 1190.2030 & s iii & 0.0231 & 0.651 @xmath105 & 353 , 354 + 2382.7651 & fe ii & 0.320 & 0.313 @xmath103 & 324 + 2586.6499 & fe ii & 0.0691 & 0.272 @xmath103 & 298 , 299 + 2600.1729 & fe ii & 0.239 & 0.270 @xmath103 & 297 + lllcllc n i & 395.0/314 & 382.3/311 & 1.7% & 323.2/309 & 309.1/306 & 0.3% + o i & 214.3/61 & 193.1/58 & 11% & 113.9/60 & 101.4/57 & 8.3% + si ii & 772.0/468 & 651.4/465 & @xmath118% & 678.6/463 & 553.6/460 & @xmath118% + si iii & 210.6/152 & 192.8/149 & 0.4% & 185.2/150 & 170.8/147 & 0.75% + fe ii & 277.6/142 & 141.8/139 & @xmath118% & 227.9/139 & 123.3/136 & @xmath118% + all & 2677.8/1270 & 1908.8/1259 & @xmath118% & 2076.7/1250 & 1442.5/1239 & @xmath118% + cccccc 1 & 7.7 & 11.2 & 19.4 & 18.32 & 2469.9 + 2 & 7.7 & 11.3 & 19.4 & 18.37 & 2470.1 + 3 & 7.7 & 11.7 & 19.4 & 18.32 & 2006.8 + 4 & 7.7 & 11.6 & 19.4 & 18.37 & 2001.8 + 5 & 7.7 & 11.6 & 19.4 & 18.33 & 1987.3 +

we report on a series of three - dimensional magnetohydrodynamic simulations of active galactic nucleus ( agn ) jet propagation in realistic models of magnetized galaxy clusters . we are primarily interested in the details of energy transfer between jets and the intracluster medium ( icm ) to help clarify what role such flows could have in the reheating of cluster cores . our simulated jets feature a range of intermittency behaviors , including intermittent jets that periodically switch on and off and one model jet that shuts down completely , naturally creating a relic plume . the icm into which these jets propagate incorporates tangled magnetic field geometries and density substructure designed to mimic some likely features of real galaxy clusters . we find that our jets are characteristically at least efficient at transferring thermal energy to the icm . irreversible heat energy is not uniformly distributed , however , instead residing preferentially in regions very near the jet / cocoon boundaries . while intermittency affects the details of how , when , and where this energy is deposited , all of our models generically fail to heat the cluster cores uniformly . both the detailed density structure and nominally weak magnetic fields in the icm play interesting roles in perturbing the flows , particularly when the jets are non - steady . still , this perturbation is never sufficient to isotropize the jet energy deposition , suggesting that some other ingredient is required for agn jets to successfully reheat cluster cores .

1001.1747

over the past decade , x - ray observations of galaxy clusters have clearly illustrated the immense energies associated with interactions between active galactic nuclei ( agns ) and their environments . for example , discoveries of x - ray cavities associated with agn radio - emitting ` relic bubbles ' ( _ e.g. _ @xcite ) demonstrate that radio lobes associated with current or previous agn activity often completely displace the surrounding intracluster medium ( icm ) . likewise , x - ray observations of ` ghost cavities ' devoid of radio emission ( _ e.g. _ @xcite ) have shown that these cavities maintain some structural integrity even after the radio population has aged beyond the limits of detectability . the energies associated with these plasma bubbles are found to be upward of @xmath1erg @xcite , suggesting that they could play a significant role in stifling cooling flows and maintaining the @xmath2kev temperature floor observed in clusters ( _ e.g. _ @xcite ) . this idea is further supported by simulations of cluster formation and evolution with agn feedback , such as those conducted by @xcite and @xcite . observations of detached relic bubbles and evolved ghost cavities also suggest that a complete picture of agn interactions with the icm should incorporate a history of realistic jet lifetimes and duty cycles . some attempts to estimate agn intermittency are based largely on energetics . @xcite , for example , estimate jet lifetimes of @xmath3years with duty cycles of @xmath4 or less to stifle cooling while matching optical and radio observations of cooling cores . observations of bubbles in the absence of agn @xcite have also been used to estimate lifetimes of @xmath3years and duty cycles of at most @xmath5 . other approaches focus more on timescales associated with observed features in individual sources , such as those conducted by @xcite , @xcite , @xcite , and @xcite , all of whom find a rate of one event per @xmath3years . observations of the so - called double - double radio sources @xcite , in which two distinct episodes of jet activity are clearly visible , have implied rather short agn lifetimes of @xmath6years at most . @xcite take a completely different approach , using qso clustering to estimate agn intermittency , and their result suggests agn lifetimes of @xmath7years . most recently , @xcite use a complete sample of clusters to estimate duty cycles of at least @xmath0 for agn outbursts in cooling flow clusters . @xcite also examine a large sample of clusters and infer that the observed size distribution of x - ray cavities is most consistent with either continuous hydrodynamic inflation or discrete production of bubbles by magnetically dominated jets . although there is much observational evidence to support models of intermittent jets that feed energetic plasma bubbles to the icm , much still needs to be understood about the precise nature of energy flow in these systems . simple analytic models , such as those developed by @xcite , @xcite , and @xcite , are powerful tools for analyzing the basic dynamics and energetics in these systems , but are not designed to address the details of energy transfer to the icm . furthermore , these models tend to feature arguments based upon self - similarity or constant jet luminosities , and so may not apply to realistic icm atmospheres or intermittent jets . likewise , numerical studies that focused on jet dynamics in both the non - relativistic ( _ e.g. _ @xcite ) and relativistic regimes ( _ e.g. _ @xcite ) made great strides toward understanding how jets propagate , but were not necessarily interested in the details of energy transfer to the icm . @xcite provide a useful analysis of the primary differences between relativistic and non - relativistic simulations , pointing out that jets of equal thrust have similar ages and efficiencies in both cases , although relativistic jets typically develop smaller cocoons than their non - relativistic counterparts . recent numerical studies of jet energetics , on the other hand , have contributed greatly to the understanding of energy flow from agn to their environments . @xcite , for example , conducted an ensemble of two - dimensional simulations of jets in detailed cluster environments , finding that a significant fraction of jet energy is deposited irreversibly into the icm . likewise , our previous study of three - dimensional magnetohydrodynamic ( mhd ) jets @xcite found that energy transfer to the icm was efficient for both uniform and stratified environments . continuing work in three dimensions , @xcite pointed out that simple hydrodynamic models of jet feedback characteristically develop a relatively lossless low - density channel and subsequently fail to dissipate and distribute energy widely enough to prevent catastrophic cooling . @xcite , however , attempted to address this issue by simulating wobbling hydrodynamic jets in an evolving cluster atmosphere that served to isotropize the deposited energy . whether the jet and jet - blown cocoon remain characteristically stable or break apart or mix with the ambient medium ( as illustrated recently in @xcite , for example ) clearly influences the efficiency with which jets deliver energy to cluster cores . some recent numerical studies of relic bubble propagation also have had implications for energy transfer and transport in the icm . simulations conducted by @xcite and @xcite in two - dimensions and @xcite , @xcite , and @xcite in three dimensions , for example , illustrated that rising mhd bubbles lift a great deal of icm material in their wake . this material has the potential to mix and distribute energy within the icm . a recent paper by @xcite points out that this mixing at the bubble interface is further enhanced in the hydrodynamic case when turbulence is treated with a subgrid approach , which explicitly models the energetics and sizes of turbulent eddies that are not properly resolved by the numerical grid . this is in contrast to the usual method of treating turbulence in which the computational grid resolution sets the effective scale of dissipation in the system . it is currently unclear how this approach will affect the case of magnetized bubbles . @xcite , taking a different approach , showed that hydrodynamic bubbles could explicitly transform wave energy into heat that could be dissipated . several groups have also explored the effects of jet intermittency on flow dynamics and energetics . the earliest example of a simulated restarting jet was conducted by @xcite , who found that subsequent jets moved quickly through previous jet material , forming a new bow shock in the process . using a two - dimensional hydrodynamics code , @xcite found that plumes from dead radio galaxies lifted substantial amounts of icm material as they rose and that a large fraction of the total jet energy went into heating the icm . this was followed by @xcite , who demonstrated in three dimensions that a buoyant jet - blown bubble continues to dredge up material from the cluster core for much longer than the jet lifetime . also in three dimensions , @xcite found that most of the jet energy in an outburst ends up too far away from the cluster core . in a subsequent simulation that included cooling , @xcite found that the amount of dense , cool gas immediately above the central agn had a pronounced effect on where most of the jet energy was dissipated . while the previously mentioned work by @xcite showed that this effect was never sufficient to completely offset cooling in a simple icm atmosphere , recent simulations by @xcite suggest that the modeling of subgrid turbulence may alleviate some of these problems . our study combines several of these different investigations , exploring the interactions of intermittent jets with the icm . we seek to determine how intermittency affects energy flow , asking in what form , where , and how rapidly energy is transferred from jet - produced structures to the icm . these simulations specifically examine whether a combination of jet intermittency , magnetic stresses , and ambient density fluctuations alone can lead to broader distribution and dissipation of jet energy , or whether something like the more chaotic jet model of @xcite , or a very high reynolds number turbulence such as that proposed by @xcite , is required . furthermore , we seek to understand the basic dynamics and morphology of intermittent flows . we test how flow structures produced by intermittent and extinct jets evolve , how they transport and transfer energy , and how they interact with and potentially enhance ambient magnetic fields . to address these questions , we employ a set of mhd simulations of jets in realistic cluster environments . the jets are oppositely directed , launching into an icm with full three - dimensional gravity , tangled magnetic fields , and density fluctuations designed to mimic cluster substructure . section [ sec : calc ] describes the details of the numerical methods and the models employed . section [ sec : disc ] contains a discussion of the results and astrophysical implications , while section [ sec : conc ] lists the main conclusions of this work . our simulations employ a second - order eulerian total variation diminishing ( tvd ) non - relativistic ideal mhd code , described in @xcite and @xcite . the code explicitly enforces the divergence - free condition for magnetic fields through a constrained transport scheme detailed in @xcite . the numerical method conserves mass , momentum , and energy to machine accuracy . gravitational energy is handled in a fully conservative fashion by including this component in the total energy of the gas . additionally , energy fluxes across the boundaries and changes in energy on the grid are tracked and output for each energy type to facilitate analysis of energy flow ( see section [ sec : energy ] ) . model parameters ( discussed in section [ sec : ambient ] ) are selected such that the cluster cooling times are longer than the simulation times ( @xmath8 myr in the cluster cores ) , allowing us to neglect cooling in our simulations . all simulations described here are performed on a three - dimensional cartesian grid of full physical dimensions @xmath9 . each zone represents one cubic kiloparsec of volume with @xmath10 kpc . given that the total jet diameter is @xmath11 kpc ( see section 2.1 ) , this is sufficient to resolve shocks and multiple scales of turbulence within the jet and cocoon . with the exception of one model ( am , described in section [ sec : jets ] ) , these simulations feature sets of oppositely directed jets emanating from a cylindrical internal boundary located in the central region of the computational grid ( @xmath12 ) . the jet region is updated prior to each tvd step , and the settings in this region are used to control jet luminosity and intermittency . crrr [ table ] st & steady inflow for duration & @xmath13 + i13 & switch on / off every 13 myr & @xmath14 + i26 & switch on / off every 26 myr & @xmath15 + re & on for 26 myr , then off & @xmath16 + am & no jets included & @xmath17 + since the jets are launched from the center of the computational grid , the outer grid boundary conditions have little influence on the evolution of jet / cocoon structures . still , the outer boundaries are crucial in maintaining the dynamical stability of the constantly evolving ambient medium . in the outer grid boundaries , we use a set of modified continuous boundaries designed to maintain approximate hydrostatic equilibrium in the ambient medium while minimizing the impact of small - amplitude waves incident upon the boundaries . specifically , we apply the constraints that the sound speed be constant and that hydrostatic equilibrium apply to set the density and pressure conditions across a given boundary . we find that these boundaries maintain the total mass and energy of the grid to an accuracy greater than @xmath18 in the absence of a jet over three grid sound - crossing times , a time much longer than any of our jet simulations . a passive color tracer @xmath19 , representing the mass fraction of jet material , is introduced through jet flow to identify material that has passed through the jet orifice . additionally , we include a population of passive , nonthermal , relativistic cosmic ray particles in our models . these cosmic rays are transported , injected , accelerated , and aged in a self - consistent fashion to enable the construction of realistic synthetic observations ( see , _ e.g. _ , @xcite , @xcite , @xcite , @xcite ) . here , we describe five simulations in which the initial ambient conditions are precisely the same , but the time histories of jet activity vary . in the following two subsections , we describe the jets and ambient media in detail . the physical parameters of each simulation are described in table 1 . + jet inflow is introduced entirely along the @xmath20-axis without any initial transverse component . the previously mentioned internal jet boundary consists of a cylindrical region originating from @xmath21 and extending @xmath22 kpc along @xmath20 . the resulting jets have uniform cores of radius @xmath23 kpc , surrounded by a concentric transition annulus that smoothly connects to the ambient conditions at a radius of @xmath24 kpc . the nominal core density contrast is @xmath25 , and the jets initially enter in pressure equilibrium with their surroundings . the jet core magnetic field in this set of models is completely toroidal , of the form @xmath26 within the jet core , but that decreases quadratically to zero outside of the core region . this jet field was chosen to be strictly toroidal so that it could be initially divorced from the ambient field . the fiducial ratio of thermal pressure to magnetic pressure in the jet ( measured at the boundary of the core ) is @xmath27 , and the associated ( maximum ) inflowing jet field strength is @xmath28 . for this set of models , the jet luminosity can be expressed in terms of the steady jet values of the density @xmath29 , velocity @xmath30 , and pressure @xmath31 as @xmath32 { \rm~erg~s}^{-1 } , \label{eq : powerset2}\ ] ] ignoring the relatively minor contributions of magnetic and gravitational energy to the jet luminosity . since it is the velocity ( and not the luminosity ) that goes smoothly to zero in the transition annulus , the size of the annulus enters into the kinetic and thermal energy flux terms differently , resulting in the prefactor of 1.9 ( see the appendix for details ) . this analytic expression is more than @xmath33 accurate over the lifetime of a steady jet . the first jet model features two steady mach 3 jets ( labeled st for steady ) that correspond to a physical flow speed @xmath34 and jet luminosity @xmath35 per jet . the other runs in this series of simulations feature jets with variable activity . in all cases , the maximum injected jet speed and mach number match those of the st model , and it is worth noting explicitly that these assumed jet luminosities and speeds ( corresponding to a lorentz factor of 1.004 ) are consistent with a non - relativistic treatment . in two runs , the jets intermittently switch on or off at regular intervals . in these two cases , the jets switch states every 26 myr ( labeled i26 ) and 13 myr ( labeled i13 ) , respectively . this switch is accomplished by ramping the density , pressure , and momentum density exponentially over a 3.27 myr ( 1.64 myr ) timescale for the i26 ( i13 ) run . the exponential ramp changes the density and pressure to a volume - averaged value sampled from a spherical region immediately surrounding the jet , allowing for a smooth transition from active jets to a quiescent state . in the fourth model ( labeled re for relic ) , designed to mimic radio relic sources , the jets are completely shut down after 26 myr and are not restarted . additionally , there is a fifth model ( labeled am for ambient ) in which the jets are never activated . this final model allows us to separate the evolution of the ambient medium from jet - driven behaviors to correctly assess the role of the environment in these simulations . figure [ fig : stspvr ] shows volume renderings of flow speed at late times for each of the four jet - driven models . in each frame , the jets originate near the center of the image at an angle with respect to the viewer , with the approaching jet pointing toward the lower right and the receding toward the upper left . all four of the jet - driven simulations run until wave or shock disturbances from the jets reach a grid boundary . this timescale varies from model to model depending on how the flow is driven , and the total simulation times are given in table 1 . , @xmath36 , and @xmath37 are given in table 1 . the labels @xmath38 and @xmath39 represent the locations of the icm density core radii.,scaledwidth=50.0% ] in all five models , we employ a single type of ambient medium designed to incorporate the essential properties of galaxy cluster environments . the gas density base profile is a double-@xmath40 model ( _ e.g. , _ @xcite ) , similar to those typically used to observationally describe clusters , of the following form ( see figure [ fig : icmdenpret ] ) : @xmath41 , \label{eq : doublebetaden}\ ] ] where @xmath42 g @xmath43 is the density at @xmath44 . this profile features two density `` cores '' weighted differently : @xmath45 kpc with a weight @xmath46 and @xmath47 kpc with a weight @xmath48 . these values were chosen to be consistent with the best - fit values found through a statistical analysis conducted by @xcite on 79 x - ray clusters . for these models , @xmath49 , which generates an asymptotic profile only slightly steeper than @xmath50 when @xmath51 . on top of this profile , we have superposed a @xcite spectrum of density fluctuations with a maximum amplitude of @xmath52 locally . this amplitude of adiabatic fluctuations in density corresponds to the range of pressure fluctuations observed by @xcite in the coma cluster . this initial spectrum of fluctuations is spatially tapered by a gaussian envelope near the edges of the computational grid to impede the development of unphysical flows at the grid edges . for these simulations , gravity is derived from an nfw profile @xcite , intended to model the underlying cluster dark matter distribution : @xmath53 which generates a gravitational acceleration : @xmath54 \label{eq : dmgrav}\ ] ] where @xmath55 kpc is taken to be the characteristic dark matter core radius . another free parameter is the dark matter density , @xmath56 . rather than deriving this from estimated cluster masses , we instead selected the value that would produce a reasonable temperature distribution ( see below ) . in this case , the temperature profile selected corresponds to @xmath57 g @xmath43 . for a cluster with a virial radius @xmath58 mpc , this leads to a virial mass @xmath59 , which is within a factor of a few of the mass of the perseus cluster ( _ e.g. _ @xcite ) . the pressure profile is determined by the requirement that the base atmosphere ( _ i.e. _ , ignoring the density fluctuations ) be in hydrostatic equilibrium . the gravitational acceleration and gas density profile are sufficiently complex that the pressure profile is derived numerically and is shown in figure [ fig : icmdenpret ] . the core value of the pressure is @xmath60 . the pressure and density profiles can be combined to give a temperature profile ( figure [ fig : icmdenpret ] ) , which is used as mentioned to provide the normalization of gravity . the temperature profile was selected to resemble typical clusters ( _ e.g. , _ @xcite ) . as figure [ fig : icmdenpret ] shows , the atmosphere has a core temperature @xmath61 kev ( assuming a mean molecular weight @xmath62 ) , with a total temperature range of @xmath63 over the computational grid . the sound speed , with a profile identical to that of the temperature , features a core value @xmath64 . the ambient magnetic field ( see figure [ fig : fv_fieldt0 ] ) is designed to be a tangled , yet analytically specifiable , structure of the following form : @xmath65 where the two components are @xmath66 and @xmath67 the constants @xmath68 and @xmath69 are chosen to allow the field to vary in @xmath70 and @xmath71 . in our simulations @xmath72 , to allow for several full field reversals over either angle . @xmath73 and @xmath74 are , in general , arbitrary functions of @xmath75 that we use to approximate a constant @xmath40 atmosphere with fluctuations . in practice , we choose @xmath76 $ ] and @xmath77 $ ] to provide field variation with radius . the constants @xmath78 and @xmath79 are used to provide fields that vary on spatial scales comparable to or slightly larger than the width of the jet . these scales should be interpreted as the maximum length scales over which the field varies in our clusters rather than the scales on which magnetic power is concentrated . the strongest fields in our clusters vary over a few tens of kiloparsecs , closer to the values inferred for observed clusters in @xcite , for example . @xmath80 is simply the ambient field strength as derived from the pressure profile and the requirement that @xmath81 over a radial average , as shown in figure [ fig : alphaplot ] . ( solid line ) , plotted as a function of radius from the cluster center . the dotted line , shown for comparison , illustrates @xmath82 , corresponding to @xmath81.,scaledwidth=50.0% ] we begin our discussion with an examination of energy flows in these systems . by carefully tracking the jet power over time and comparing the integrated power to the measured change in energy on the grid for a given energy type , we can determine how energy is converted as the flow evolves . additionally , we can use the passive color tracer @xmath19 to indicate what fraction of this added energy enters the icm . we are particularly interested here in the amount of thermal energy transferred to the icm since this energy is most likely to be of relevance to the quenching of cooling flows . again , we neglect radiative cooling ; our primary goal is to understand energy transfer from the jets . figures [ fig : st_eplot ] - [ fig : re_eplot ] illustrate the energy flows for each of the four jet simulation models we use . in each figure , four plots are shown , corresponding to the four types of energy present : kinetic , thermal , magnetic , and gravitational . in each plot , the solid line represents the total amount of energy added to the grid by the jets , minus the energy measured to have exited the grid through the outer boundaries . in these models , the latter quantity is very small since the systems would be in hydrostatic equilibrium were it not for the evolution of the added density perturbations and magnetic field fluctuations . in practice , the amount of energy flux across the outer boundaries is further reduced by the fact that the density , pressure , and magnetic field are at their weakest in these regions . the dashed line in each plot shows the integrated jet power for that particular energy type . the dotted line shows the measured change in that energy type on the computational grid , excluding the jet inflow region that we treat as a boundary . , but for the i26 model.,scaledwidth=50.0% ] , but for the i13 model.,scaledwidth=50.0% ] , but for the re model.,scaledwidth=50.0% ] , but for the am model . in this case , no jet inflow is shown since the jets are never activated.,scaledwidth=50.0% ] since the systems were not in perfect equilibrium initially , the ambient medium relaxes as the simulations progress , causing some exchange of energy independent of jet action . really , we are interested in energy flows resulting from the jet - icm interactions rather than those established by the details of the icm initialization , so this background relaxation is subtracted away from the measured change in energy ( _ dotted lines _ ) in each of figures [ fig : st_eplot ] - [ fig : re_eplot ] . to calibrate this background energy flow , we used the results of the am model , which featured no jets . figure [ fig : am_eplot ] shows the energy flow evolution for this model . the solid line in each plot in figure [ fig : am_eplot ] illustrates the accuracy with which we account for energy changes on the grid ( the expectation value is zero ) , while the dotted lines show the change in each energy type as usual . no dashed lines are included since no jet inflow was introduced . as expected , the am model relaxes by gaining kinetic and thermal energy at the expense of gravitational and magnetic energy . the amounts of energy that change form in the am model are only approximately @xmath83 of the total added jet energy in the st model , but this amount is worth tracking since it exceeds the total added jet magnetic energy in that model , for example . focusing now on a discussion of figures [ fig : st_eplot ] - [ fig : re_eplot ] , we point out some general features of energy flow in these simulations . unsurprisingly , the inflowing jet energy in all cases is dominated by the kinetic energy component ( illustrated by the dashed line being just below the solid line in the upper - left plot in figure [ fig : st_eplot ] , for example ) . most of the remainder of the energy added by the jets is in the form of thermal energy , with relatively minor magnetic and gravitational components . the gravitational component ( _ dashed line in lower - right plot _ ) , in particular , is roughly constant for active jets , representing the introduction of jet material in the gravitational potential upon jet initialization . we note that the thermal component of the inflowing jet power ( _ dashed line in upper - right plot _ ) asymptotically reaches a somewhat larger fraction of the total jet luminosity for all active jets as time goes on . this is caused by the sides of the jet cylinder allowing material back across that boundary , and is thus an artifact of the way in which we launch our jets . the calculated jet power is reduced by this amount , however , so that the depictions of jet power are an accurate representation of the net inflow from the jet region . there are also several global trends present in the measured changes on the grid ( _ dashed lines _ ) for the jet - driven models . first , we note that in all models the measured thermal energy exceeds any of the other components . regardless of the details , all models generate substantially more thermal energy than is introduced through the jets . additionally , we note that all models feature appreciable increases in the gravitational energy . unlike our previous work @xcite , these simulations include a realistic model of gravity in which the total gravitational energy on the grid is comparable to the total thermal energy . since little gravitational energy is introduced at the jet orifice , the increase in gravitational energy is almost entirely a result of energy exchange in these systems . finally , the measured magnetic energy evolution also features an interesting shape , suddenly increasing in each figure . this is representative of the fact that the measured magnetic energy change on the grid is actually negative at early times and so is not shown on these plots ; a point we will discuss shortly . before we discuss the details of each model , we should first introduce another measure of energy flow in these systems . by incorporating the passive color tracer , we can determine what fraction of the measured change in energy on the grid takes place in the icm . this is accomplished by measuring @xmath84 , where @xmath85 for jet material and @xmath86 for the icm . figure [ fig : st_efrac ] shows , for the four jet models , this fraction for each different type of energy . again , the background change in energy due to the relaxation of the icm is removed so that the included changes are ultimately the result of jet - icm interactions . the most significant trend to note is that all models are efficient at transferring energy from jets to the icm . as we found in @xcite , efficient energy transfer seems to be a general characteristic of these systems , and all jet models have transmitted @xmath87 or more of their energy to the icm by the end of the simulations . significantly , thermal energy always represents the largest fraction of energy added to the icm for all models . unsurprisingly , given the large @xmath40 in the jet inflow , the magnetic energy always represents the smallest fraction of energy added . the exact distributions of energy vary according to model , however , so we will address them individually . -direction as measured by the passive variable . dotted lines indicate the position of the bow shock as a function of time . values are averaged over the two jets in each simulation.,scaledwidth=50.0% ] focusing for the moment on steady jets , represented by the st model , there are several interesting features to note . similar to what we found in @xcite , the active jets efficiently convert kinetic to thermal energy . the measured thermal energy is asymptotically a factor of @xmath88 larger than thermal energy added by the jet , which is similar to the amount of kinetic energy lost from the jet inflow . from the upper - left panel of figure [ fig : st_efrac ] , we see that @xmath89 of the total energy added to the grid ends up as thermal energy in the ambient medium , suggesting that these steady jets efficiently heat their environments . although it is clear that this energy is being added to the icm , we also seek to discover where in the icm this energy appears . as @xcite and @xcite demonstrated , the location of deposited energy is at least as significant as the amount of energy deposited for a system to offset cooling . figure [ fig : entropycomparisons ] shows illustrations of the entropy density @xmath90 for the st and i13 models . specifically , we apply the color tracer and show @xmath91 for @xmath92 , which removes the initial ambient entropy @xmath93 and focuses only on enhancements in the icm . looking at the upper - left panel , most of the visible @xmath94 forms a sheath around the jet . this is caused by a mixture of jet - icm mixing at the boundary ( jets have low density , so at a given pressure , high entropy ) and icm entropy increases resulting from shock heating . in the saturated image ( _ upper right _ ) , we can see the effects of the bow shock , although the shock discontinuity itself is not visible . these images suggest that much of the irreversibly added energy measured in the icm in the st model probably is transferred and remains near the jets themselves rather than uniformly filling the shocked icm region . figure [ fig : st_eplot ] also shows that the measured gravitational energy change represents an increasing fraction over time of the total added energy for steady jets , although it still comprises less than @xmath5 of the total energy by the end of the simulation . since the jets are light with respect to the ambient medium , they contain a very small fraction of the total gravitational energy . in fact , you would have to lift jet material @xmath95 times higher than core icm material for the same energy gain . rather , this measured change is caused by the jets driving icm material higher into the gravitational potential . figure [ fig : st_efrac ] also supports this interpretation since the fraction of gravitational energy added to the ambient medium ( @xmath96 ) is comparable to the total fraction estimated from figure [ fig : st_eplot ] . the increase in gravitational energy can either be accomplished mechanically through lifting , dragging , or entrainment or through the icm reacting to the added heat and local pressure . in the st model , dragging ( _ i.e. _ , as in a wake ) should not play a role since the jet is never disconnected from the source . since the passive variable appears to have a sharp boundary in this model ( which you can see from the mass - fraction - selected entropy in figure [ fig : entropycomparisons ] ) , the icm is also not significantly entrained . any lifted icm material would have to be located between the jet / icm contact discontinuity and the bow shock . in the direction of jet propagation for an active jet , this is a very small region and so not likely to contribute much . since we know that the icm is heated , this is the most likely source of the increase in gravitational energy for this model . the evolution of the magnetic energy in the st model is interesting since magnetic energy is initially lost from the system and later increases . in fact , this is a general characteristic of all four jet models . although the total magnetic energy budget is tiny with respect to the other forms of energy ( making the accurate subtraction of the icm more crucial ) , we still want to understand how magnetic fields evolve , particularly since these fields are an essential component of the synchrotron emission we observe . there are two possible physical scenarios one can imagine to achieve a net decrease in magnetic energy after the icm relaxation has been subtracted . one is that the stronger central icm fields get advected upward to where they can more easily expand , reducing their magnetic energy . this seems implausible since the st model does not advect much ambient material without compressing it . also , the observed effect is almost immediate , and so operates at early times . the second possibility is that the ambient fields reconnect with the jet fields and then relax . although reconnection as a process would not likely lead directly to a significant loss of magnetic energy , it has the potential to create new field geometries that could facilitate magnetic relaxation . this is the more plausible of the two scenarios since the jet fields certainly reconnect with the surrounding icm ( see the discussion in section 3.3 ) . unfortunately , it is less obvious that this process would necessarily lead to a net reduction in magnetic energy since the stretching and twisting of reconnected fields could also amplify the fields ( see , for example , figure [ fig : relicreconnection ] ) . after this initial dip , however , it is clear that the overall magnetic energy in the system is increasing . some of this is compression of the icm magnetic field by the jet / cocoon / bow shock , but figure [ fig : st_bm ] suggests that much of it is simply poynting flux introduced by the jet inflow . the animations associated with figure [ fig : st_bm ] further suggest that the magnetic energy in the cocoon is replenished by the jet since the field in the cocoon is reduced in magnitude during periods of jet quiescence , presumably as a result of adiabatic expansion . -axis , shows the change in magnetic energy on the grid in the st model . the dotted line ( scaled for comparison ) shows the magnitude of the icm field in the direction of jet propagation.,scaledwidth=50.0% ] interestingly , there is a correlation between changes in magnetic energy and jet encounters with the structure of the icm magnetic field . as figure [ fig : alphaplot ] shows , the radial distribution of magnetic field is not uniform , but rather peaks in magnitude every @xmath97kpc or so . using the measured jet lengths in figure [ fig : doublejetlengths ] , which we will discuss further in section [ sec : dm ] , we can translate the time evolution of magnetic energy to evolution as a function of jet length . figure [ fig : st_beplot ] shows the relationship between the change in magnetic energy and the icm magnetic field in the direction of jet propagation . the two quantities appear to be correlated or anti - correlated with a slight offset . the causal reasons for this are interesting since it is not obvious that the structure of the globally weak icm field should be detectable in the energy flow . animations associated with figure [ fig : st_bm ] show the magnetic field evolution for the st and i13 models . in each case , it is clear that the icm magnetic energy is enhanced each time the jet passes through a strong magnetic pressure gradient in the icm . these waves of enhancement lag slightly behind the jet intersection of a high - field region , suggesting that the solid line in figure [ fig : st_beplot ] is correlated with , but lags slightly behind , the dotted line . since a substantial fraction of the magnetic energy on the grid is located in these regions , it makes sense that their perturbations would manifest themselves in the measured magnetic energy changes . we turn now to a discussion of energetics in the two intermittent jet models , i26 and i13 . as figure [ fig : doublejetlengths ] and the animations associated with figure [ fig : stspvr ] illustrate , the appearances of the two intermittent jet models deviate from one another very early on . this is due mostly to the fact that the i26 model manages to form a well - defined low - density jet channel before the first ` off ' cycle , whereas the i13 model must struggle at first to break through the dense icm . as figures [ fig : i26_eplot ] and [ fig : i13_eplot ] show , however , the two models are energetically quite similar . the kinetic and thermal energy influxes closely follow the level of jet activity . as an example , the i26 jet is on for 26 myr , corresponding to a monotonic increase in added kinetic and thermal energy , just like the st model . when it switches off from @xmath98myr , the total added energy levels off while kinetic energy briefly continues to convert to thermal at the terminal shock . while the jet remains off from @xmath99myr , the system loses kinetic energy and gains gravitational energy . as the jet finishes restarting by @xmath100myr , the cycle begins again , and similarly for the i13 model . it is worth noting that the gravitational energy increases monotonically through the entire lifetime of the model . this is consistent both with the icm reacting to the added heat ( as in the st model ) and the dragging up of icm material in the wakes of the unpowered , momentum - driven jet plumes . this latter process almost certainly takes place in these models since we see the kinetic energy of the system decrease while the gravitational energy increases at @xmath99myr . the fractional energy plots ( figure [ fig : st_efrac ] ) show that both models are very efficient at heating the icm , although the fraction of energy in the icm at a given time can vary by as much as @xmath101 , depending on the level of jet activity . one interesting point concerning the i13 model is that there are enough cycles of jet activity that we can attempt to discuss asymptotic behaviors . compared to the st model , the i13 model has approximately the same total efficiency ( @xmath87 ) in transferring energy to the icm . the fraction of kinetic and thermal energy entering the icm is similar for the two runs , but the gravitational energy fraction is @xmath102 higher for the i13 model . this is slightly misleading , however , since the _ total amount _ of gravitational energy introduced into the system at a given time is higher for the st than the i13 model . the fractional gravitational energy differences in figure [ fig : st_efrac ] are thus more indicative of the difference in the total energy available to each model . additionally , we can again examine the entropy to see where the intermittent jets characteristically deposit their energy . the bottom two panels in figure [ fig : entropycomparisons ] show the color - selected @xmath94 for the i13 model . although the added icm entropy still forms a sheath around the jet cocoon , there is now a much more substantial amount of material nearer to the cluster core . likewise , as we will see in section [ sec : dm ] , the intermittent cocoons are characteristically fatter than those of the st model . this means that more high - entropy material is spread through a larger volume and nearer to the cluster core , both of which are needed to offset core cooling . we turn lastly to the re model , which evolves identically to the st model until the jet activity is shut off completely at @xmath103myr . in figure [ fig : re_eplot ] , we see the expected inflowing and total energy plateaus appearing at @xmath103myr . as in the i26 and i13 models , the system begins to exchange kinetic energy for gravitational energy as the thermal energy is approximately maintained . after @xmath100myr , however , we see the total thermal energy start to decrease as the kinetic energy mildly increases . this feature is also marginally present in the i26 model , but in that case the jet reactivates before it becomes very pronounced . in the re model , the slope of the added gravitational energy also begins to flatten . this behavior is consistent with the transition from a system actively driven by jets to one in which a passive plume consisting of a mix of ambient and jet / cocoon material is gradually slowed by the drag force associated with its motion through the icm ( see section [ sec : dm ] for details ) . as the relic plume rises , it should maintain approximate pressure equilibrium with the surrounding icm , expanding and losing thermal energy in the process . likewise , reversible heat added to the icm can be transformed back to kinetic energy through adiabatic expansion . ultimately , the total gravitational energy should decrease when the system becomes dominated by buoyancy , but there is still sufficient momentum in the plume through the duration of our simulations that we have not yet reached that regime . figure [ fig : st_efrac ] shows that approximately @xmath104 of the jet energy has entered the icm by the end of the re simulation . despite being more efficient at delivering what energy it has , however , figure [ fig : re_eplot ] shows that the re model introduces less total energy into the icm than any of the intermittent or steady models as a result of its lower integrated luminosity . here , we present a discussion of the dynamics and morphology of our simulated flows . as mentioned in the previous section , figure [ fig : stspvr ] and the associated animations clearly show the evolution of flow speed in each jet model over time from various angles . likewise , figure [ fig : st_en ] and its associated animations show volume renderings of jet entropy surrounded by more transparent isosurface renderings of icm entropy . animations such as these are typically the best way to quickly visualize causal relationships between flow structures and to convey the general character of the flow . first , we discuss the general properties of jet length evolution as a function of time , as illustrated by figure [ fig : doublejetlengths ] . in that figure , the solid lines represent the maximum distance between the cluster core and @xmath105 in the direction of jet / relic motion and averaged over the two jets in each model . the dotted lines show the position of the forwardmost shock , also in the direction of jet / relic motion and averaged over both jets . to locate the shocks , we used an identification strategy similar to that employed in @xcite in which the temperature and entropy gradients were of the same sign , the velocity divergence was negative , and the temperature jump was that of a mach 1.3 shock or stronger . in figure [ fig : doublejetlengths ] , we see that substantial separation develops between the bow shock and the jet material at some point in every non - steady jet model . when the dotted line and solid line are exactly coincident , this corresponds to a shock with @xmath106 or no shock at all . + + + there are two interesting evolutionary features in figure [ fig : doublejetlengths ] . first , we note that the st , i26 , and i13 models all asymptote to similar slopes , suggesting a common power - law evolution of @xmath107 for these models , where @xmath108 . a simple expectation for the advance speed , gotten from self - similar scaling arguments in @xcite , for example , is that @xmath109 assuming a homologous structure for the cocoon and that the ambient density follows a power law @xmath110 . although the ambient density in this case is not a strict power law , we can estimate @xmath111 around @xmath112kpc and @xmath113 around @xmath114kpc . these values produce an expected @xmath115 , which is substantially flatter than the measured value of @xmath108 . self - similarity is thus probably not a reasonable assumption for an atmosphere whose density dependence changes slope so dramatically . that the measured values of @xmath116 for the i26 and i13 models are slightly flatter than the st model is not surprising , however , since the jet momentum is probably distributed over a slightly larger area , as we will see momentarily . in contrast to systems in which jets actively drive the flow , the advance speed of the re model asymptotes to a power law of approximately @xmath117 as it decelerates in the icm . although the residual jet flows terminate quite rapidly after being shut off , we expect the overall momentum of the system to be distributed in a mix of cocoon material and icm that has been entrained , dragged , or otherwise pushed into motion . we can estimate the rate at which the plumes are decelerated , assuming that the dynamics are dominated by the effects of drag resulting from motion of the relics through the icm . the drag force is given by @xmath118 , where @xmath119 is the drag coefficient , @xmath120 is the cross - sectional area of the plume , and @xmath121 is the relic velocity . we assume , as before , that the ambient density follows a power law , this time specifying it more fully as @xmath122 , where the `` off '' subscript refers to the value of that quantity at the position of the jet head when the jets are shut off . if we assume that the radial expansion is adiabatic , then @xmath123 is conserved , assuming the volume of the relic is proportional to the cube of its radius . if the ambient pressure profile is proportional to that of the density ( _ i.e. , _ the icm is approximately isothermal ) , then @xmath124 , meaning that the relic area can be approximated as @xmath125 . under these assumptions , the relic deceleration resulting from drag can be written as @xmath126 where , writing the total mass of the relic as @xmath127 and assuming @xmath128 , @xmath129 these expressions can be simplified considerably when we realize that , measured at @xmath130kpc , @xmath131 . this reduces the drag deceleration expression to @xmath132 , which admits a solution @xmath133 . to estimate the value of @xmath134 , we can model the initial shape of the relic plume as a cylinder of mass @xmath135 . again using the fact that @xmath136 , this allows us to write @xmath137 . unfortunately , using the passive color tracer , @xmath19 , to estimate the relic density would fail to account for the any contribution to the total mass from the icm component of the relic . to more accurately estimate the relic density , we instead measure from our simulated data the average density of material moving upward with velocities above a certain threshold and positions between the jet origin and the top of the relic ( thus excluding the bow shock ) . choosing only material with @xmath138 , which would correspond everywhere to initially supersonic motion in the unperturbed icm , we measure an average relic density @xmath139 at @xmath140 myr , just after the jets are shut off . at that relic height ( @xmath141 kpc ) , @xmath142 , resulting in an expected value of @xmath143 . since @xmath144 , the expected value of @xmath116 is @xmath145 , which is similar to the observed @xmath117 . the closeness of the predicted and observed power laws indicates that this deceleration model is plausible , although one also assumes that this scaling would eventually break down when the relic has lost enough momentum that buoyancy becomes the dominant force in the system . moving on to a discussion of jet and flow morphology , we first describe steady jets as exemplified by the st model . our steady jets are narrow , stable structures , a common feature in such simulations . since we do not perturb them at the source , but instead allow the icm substructure to seed instabilities , they do not develop complicated shock structures like those observed in @xcite and @xcite , and the jets are not broken apart , as in @xcite . furthermore , the jet magnetic field is sufficiently weak that the jet does not become unstable to pinch instabilities , nor does it develop an obviously modified nose cone as has been seen in some strongly magnetized jet simulations ( _ e.g. , _ as shown in the lower - right panel of figure [ fig : doublejetshocks ] , the bow shock is a relatively smooth surface and the internal shock structures are regularly spaced . from the animations and image of the flow speed associated with figure [ fig : stspvr ] , we see that the jets are not completely axisymmetric , however , and these slight asymmetries become much more pronounced when the agns are shut down . the cocoons of shocked jet material in steady jets are relatively thin and do not develop a strong backflow . the cocoon structure is seen clearly in a volume rendering of the entropy shown in figure [ fig : st_en ] , where the low - density plasma is very bright . the icm pressure gradient in these models is sufficient to prevent much material from traveling very far opposite the direction of jet propagation . the i26 and i13 intermittent model jets feature less axisymmetry as the flows evolve . figure [ fig : stspvr ] shows that portions of the flow with moderate speeds at late times tend to bend in response to the icm substructure . likewise , we can see filaments of low - speed material that form in the cocoon . the shock structures in these two models are very different from the steady case . figure [ fig : doublejetshocks ] shows a volume rendering of the i13 shock structures at three different times . not only do the i13 shocks look different from those in the st model , but they look different at a given time from earlier or later shock structures in the same model . the animation associated with figure [ fig : doublejetshocks ] clearly shows the propagation of internal shocks and waves associated with restarted jet activity . typically , the subsequent jets form a bow shock in the old jet material that quickly advances forward through the low - density channel . this was seen even in the earliest models of restarting jets conducted by @xcite . a set of irregularly shaped reverse shocks that dampen into waves moves back through this material each time the jet turns off . compared to the st model , the two intermittent models feature much wider cocoons that are ` puffed out ' by subsequent episodes of jet activity . figure [ fig : st_en ] shows images of the entropy for these models . in the i26 model , we can see that the second episode of jet activity has run into the plume from the first episode , causing it to widen . likewise , the i13 model cocoon has several visible width enhancements associated with subsequent jet activity . these features are associated in time with the measured grid increases in thermal energy , representing energy deposition in the plumes . interestingly , these structures persist after they are formed , presumably because subsequent jets are easily redirected by the icm into the relatively low - density channels formed by previous jets . the animations of entropy for the i13 model in particular show these features lasting to the end of the simulation . there are at least two reasons , both having to do with where shocks form , that intermittent jets produce wider cocoons than their steady counterparts . the first is that the icm is drawn in behind the wake of the rising plume before the next intermittent jet launches . this refilling is incomplete , as seen in the animations associated with figure [ fig : st_en ] , but is sufficient to cause the next episode of jet activity to produce a shock at the base of the plume . the second reason is that the restarted jet forms a terminal shock and multiple reverse shocks of its own ( as seen in figure [ fig : doublejetshocks ] ) as it propagates through the existing plume . both of these episodes of shock formation deposit thermal energy in the plume and cause the cocoon to expand laterally as a result . finally , the relic model is distinct from all other models in that the jet plumes completely detach from the source . figures [ fig : stspvr ] and [ fig : st_en ] show the evolution of this model , in which two plumes of material rise in the icm as the deposited momentum is depleted . when the jets shut down , the flows are already well over @xmath146kpc from the cluster core , and their cocoons are prolate in shape . as the ends of the plumes decelerate , however , the relic bubbles grow less prolate . since observed relic and ghost bubbles are typically seen to be intact structures , we note that the entropy at late times shows relatively rounded morphology , particularly near the leading edge of the bubble . significantly , we do not observe the formation of a torus , in contrast with simulations of bubbles initialized as spherical cavities in clusters that would otherwise be in hydrostatic equilibrium ( see @xcite for an extensive discussion ) . finally , it is important to note that we have only explored a small part of the available parameter space with this particular relic model . to fully evaluate whether jet / icm magnetic fields stabilize jet - blown relic bubbles , as has been suggested by @xcite , for example , further variation of magnetic field strength and geometry would be required . finally , we devote some space to a discussion of magnetic field structures in these flows . section [ sec : energy ] addressed the evolution of magnetic energy in these systems , but the details of field structure evolution were not extensively discussed . employing a visualization package called fieldvis @xcite , we are able to graphically represent magnetic field lines and to map useful flow properties on them as needed . this is a tremendously powerful approach since the challenges of visualizing multiple three - dimensional vector and scalar fields can otherwise be overwhelming . figure [ fig : streconnection ] shows the evolution of magnetic field structures in the st model . at early times ( _ upper - left image _ ) , the jet is propagating toward an undisturbed icm field structure , similar to one of the structures seen in figure [ fig : fv_fieldt0 ] . in that image , the jet moves from the lower left toward the upper right of the image and is represented by a set of lightly colored velocity streaklines . the jet is surrounded by a toroidal field , represented by the darker set of coiled lines . as the jet disturbance first reaches the icm ( _ upper - right image _ ) , the icm field lines are pushed aside and stretched . the intensity of the magnetic field lines represents the field strength ( brighter = stronger ) , and we see that the field is enhanced as it wraps around the jet . this process continues in the next frame ( _ lower left _ ) , where the jet has completely penetrated the icm field . in that same image , we also note evidence of an internal shock within the jet corresponding to a concentration of field lines roughly halfway down the jet structure . in the last image ( _ lower right _ ) , we see that the jet field has interacted with the ambient field . to explain this image completely , we must first point out that the identified icm field lines were selected to lie ahead of the jet so that the jet field would necessarily intersect them . this final image shows that at least one of the icm field lines has become attached to the toroidal jet field . this is seen most easily by noting that there are four visible lines to the left of the jet , but only three emerge to the right while the other is attached to the larger toroidal structure near the end of the jet . we examined the values of the passive variable in this region to find that in fact the large toroidal field is associated with jet material and connects with a line that is associated with the icm . although reconnection in our numerical scheme does not represent a significant amount of direct energy exchange , the geometries that develop from it have the potential to affect the system energetics . as this system evolves , the reconnected field line will get stretched out with the jet flow , possibly leading to an alignment between the jet and field orientation . this radial realignment of icm fields could have important implications for cluster heating via thermal conduction along field lines , as has been suggested by @xcite and @xcite , for example . additionally , this alignment could be similar to that inferred from polarization maps of some stronger sources @xcite . we list this outcome only as a speculative possibility , however . constructing synthetic radio observations is the only rigorous way to determine the polarization properties of a simulated structure , and we defer that discussion to a later work . kpc from the cluster core . intensity reflects field strength ( light = strong ) . _ lower panel : _ the same region at @xmath147myr . now , the relic velocity is shown as streaklines with ` baseball diamond ' glyphs pointing in the direction of motion . the surrounding stretched lines represent icm field lines that have gotten stretched and amplified by the motion of the relic.,title="fig:",scaledwidth=48.0% ] + kpc from the cluster core . intensity reflects field strength ( light = strong ) . _ lower panel : _ the same region at @xmath147myr . now , the relic velocity is shown as streaklines with ` baseball diamond ' glyphs pointing in the direction of motion . the surrounding stretched lines represent icm field lines that have gotten stretched and amplified by the motion of the relic.,title="fig:",scaledwidth=48.0% ] interestingly , the icm magnetic field seems to interact strongly with the behavior of the intermittent jets . although the nominal mean magnetic pressure is a factor @xmath148 smaller than the gas pressure , local magnetic pressure gradients can still exert a substantial force on the flow . there are two reasons for this . first , the variations in field strength can lead to locally much smaller values of @xmath40 . second , it is important to remember that the dynamical consequence of magnetic pressure comes through its gradient , not its magnitude , per se . when the local length scale of magnetic field variations is smaller than the length scale of the gas pressure variations , the dynamical role of the magnetic pressure is stronger than the @xmath40 parameter implies . in the st model , these forces are swamped by the momentum of the jets and are not easily seen . in the intermittent cases , however , there is some evidence that the icm field influences the cocoon shape as the jets propagate . figure [ fig : st_bm ] and the associated animation illustrate this point well . as the jet restarts , the bright regions of enhanced field in the jet coincide with the icm enhancements . likewise , we can see that the cocoon width responds to these fields . since this is more difficult to see in other models , however , we conclude that it is the interaction between intermittency and the ambient field that makes these structures stand out . figure [ fig : relicreconnection ] shows another field geometry , this time for the re model . the upper image shows a set of undisturbed icm field lines at roughly @xmath149kpc from the cluster core . these have a structure much like those seen in figure [ fig : fv_fieldt0 ] and the upper - left panel of figure [ fig : streconnection ] . again , field strength is shown as a surface property of the lines and the structure of the constant-@xmath40 atmosphere can be seen from the darker background fields . the lower image shows the interaction of the relic bubble and the icm field , where the relic is represented by velocity streaklines using a different glyph . as the relic moves from the lower left to upper right of the image , it drags a significant amount of magnetic field with it . specifically , we see several very strong ( _ i.e. _ , brightly colored ) field structures that are stretched in the direction of relic propagation . these are icm fields that connected with the relic bubble fields as it was rising . once this happens , the continued propagation of the relic serves to stretch and amplify these fields . again , the passive variable confirms this explanation by showing that several of the strong field lines in the lower image are associated with both jet and icm material . this is a prime example of how magnetic fields can get enhanced and transported in a cluster even when the central agn engine is off . we have conducted a series of simulations designed to examine the role of intermittency and cluster substructure on the energetics , dynamics , and morphologies of jet - driven flows . the most important results from this work are summarized here as follows . all models of jet behavior , including steady , intermittent , and relic flows , are at least @xmath87 efficient at delivering thermal energy to their environments . while this efficiency is remarkable , none of the jet models are particularly adept at distributing this energy uniformly throughout their environments . specifically , the transfer of energy from steady jets to the icm takes place very near the jet cocoon , while intermittent jets spread the energy in only a slightly wider pattern . jet - blown relic bubbles are the most efficient at transferring what energy they have , but they also transmit less total energy than any other model . taken together , these results confirm that the lack of uniform heating in hydrodynamic systems described by @xcite is also a problem for mhd jets in realistic magnetized clusters . it is possible that the subgrid modeling of turbulence undertaken by @xcite could help to alleviate this problem in magnetized systems , too , but this will require a modification of the subgrid model that has been used thus far . the morphologies of intermittent jets retain evidence of previous episodes of jet activity . specifically , subsequent bouts of jet activity encounter previously formed structures , creating reverse shocks at their interfaces . the cocoons formed by such flows also evolve to be characteristically wider than those of steady jets . these structures can be associated with particular depositions of energy into the icm and may indicate where heating is taking place . a full exploration of whether such features are observationally detectable is deferred to the analysis of synthetic observations from our simulated data @xcite . the presence of cluster substructure can perturb initially axisymmetric jets and does so much more quickly when the jets are non - steady . this is a very natural way to perturb jets in clusters that have not undergone recent mergers and which appear to be relaxed . still , as noted , the density and magnetic structures included in these simulations are small and insufficient to significantly isotropize the energy distribution in cluster cores . \4 . magnetic structures in the icm are affected by reconnection and can get carried along with jet - driven and passively evolving flows . although the specific process of reconnection will never directly increase the magnetic energy , it can facilitate field geometries that operate in conjunction with the flow to amplify fields through field line stretching and twisting . additionally , such processes have the potential to advect core icm fields into the outer regions of clusters and could possibly orient such fields in the direction of jet motion , with potentially important implications for heat transfer via thermal conduction . interestingly , local icm magnetic pressure gradients are capable of affecting these flows even when the global field strengths are considered to be insignificant . while the magnetic energy increases associated with such interactions are minor , this is a specific way in which mhd jet flows with nominally weak fields differ from their hydrodynamic counterparts . \5 . the motion of relic plumes produced by agn jets is initially determined by a combination of the momentum resident in these systems and the drag force induced by moving through the icm . in our model , this mode of evolution persisted for 100 myr after the jet had switched off , suggesting that buoyant models of jet propagation in clusters may only be relevant at late times or for jets with less total momentum deposition . the resulting relic morphology and evolution is a useful reminder of the important differences between simulations of jet - formed bubbles and bubbles that are initialized as spherical cavities . this work was supported at the university of minnesota by nsf grant ast06 - 07674 and by the university of minnesota supercomputing institute . s.m.o . further acknowledges the support of a doctoral dissertation fellowship granted by the university of minnesota . we thank peter mendygral for a series of helpful discussions that improved the quality of this work , and we thank the anonymous referee for their useful feedback . visualizations presented in this paper were constructed using the software packages fieldvis and the hierarchical volume renderer ( hvr ) , both developed at the university of minnesota , and the commercial program amira . as noted in section [ sec : jets ] , the ratio of the unit kinetic to thermal luminosities contains a prefactor that is of order unity . to see how this prefactor depends upon the details of jet injection , let us first define our jet density , pressure , and velocity profiles : where @xmath153 is the outer radius of the jet core , @xmath154 is the outer radius of the transition sheath , and the `` zero '' subscript indicates a baseline value that could in principle be time - varying . to understand how these profiles affect the luminosity , we first compute the kinetic luminosity , @xmath155 : we see that each of the two luminosities has a dependence upon the size of the transition region as parameterized by @xmath159 . that the two luminosities vary differently with @xmath159 simply reflects the decision to have the velocity transition smoothly to zero , independent of the density or pressure . to recover the prefactor found in section [ sec : jets ] , we take the ratio of the two luminosities : @xmath160 plugging in the values from table 1 , @xmath161 g @xmath43 , @xmath162 dyne @xmath163 , and @xmath164 cm s@xmath165 , we obtain : @xmath166 for our assumed jet profile , @xmath167 , so the ratio of kinetic to thermal luminosity is @xmath168 . croom , s. m. , boyle , b. j. , shanks , t. , outram , p. , smith , r. j. , miller , l. , loaring , n. , kenyon , s. , & couch , w. 2004 , agn physics with the sloan digital sky survey , asp conference series , eds . g. t. richards & p. b. hall , 311 , 457 mcnamara , b. r. , wise , m. w. , nulsen , p. e. j. , david , l. p. , carilli , c. l. , sarazin , c. l. , odea , c. p. , houck , j. , donahue , m. , baum , s. , voit , m. , oconnell , r. w. , & koekemoer , a. 2001 , , 562 , l149 robinson , k. , dursi , l. j. , ricker , p. m. , rosner , r. , calder , a. c. , zingale , m. , truran , j. w. , linde , t. , caceres , a. , fryxell , b. , olson , k. , riley , k. , siegel , a. , vladimirova , n. 2004 , , 601 , 621

in this paper we derive the equations of motion for two - layer point vortex motion on the upper half plane . we study the invariants using symmetry , including the hamiltonian and show that the two vortex problem is integrable . we characterize all two vortex motions for the cases where the vortex strengths are both equal , and when they are opposite . we also prove that there are no equilibria for the two vortex problem when . we show that there is only one relative equilibrium configuration when and the vortices are in different layers . we also make observations concerning the finite - time collapse of two vortices in the half plane . we then compare the regimes of motion for both cases ( motion on the half plane ) with the case of the two - layer vortex problem on the entire plane . we also study several classes of streamline topologies for two vortices in different layers . we conclude with a hamiltonian study of integrable two - layer 3 vortex motion on the half plane by studying integrable symmetrical configurations and provide a rich class of new relative equilibria .

1308.1010

there is a vast literature on @xmath2-body vortex problems , much of which is reference in the book by newton @xcite . the geostrophic vortex models are described in many geophysics texts including @xcite . there has been some work done on two - layer point vortex dynamics on the entire plane . these include the works of young @xcite , and hogg and stommel @xcite@xcite , and which have been primarily numerical investigations . some experimental work has been done by griffiths and hopfinger @xcite . more analytic results can be found in the work of gryanik @xcite and zabusky and mcwilliams @xcite and flierl , polvani and zabusky @xcite . integrable two layer point vortex motion in the plane has been extensively studied by jamaloodeen and newton @xcite@xcite . more recent studies can also be found in the works of koshel et al @xcite@xcite . in these are studied the equilibrium solutions , the vortex collapse problem and the transition to chaotic advection through perturbations of known equilibrium solutions . the two layer vortex problem in domains other than the entire plane has not been as extensively studied as the one layer vortex problem on domains with boundaries . a good exposition of the one layer vortex problem on domains with boundaries can be found in work of flucher and gustafsson@xcite . the work in this study can be understood as applying the techniques for one layer integrable vortex dynamics on domains with boundaries @xcite , and the analytic techniques for integrable two layer vortex dynamics in the unbounded plane @xcite@xcite@xcite@xcite , to integrable two layer vortex dynamics in the upper half plane . this paper is organized as follows . we begin by deriving the equations of motion , by first obtaining the streamfunctions for an ensemble of point vortices . we also obtain the invariants through the use of symmetry and establish the integrability of the two vortex problem in the upper half plane . through analysis of the hamiltonian energy curves we characterize all 2 point vortex motion in the upper half plane for both cases @xmath3 @xmath4 and @xmath3 @xmath5 including conclusions about 2 vortex collapse . we then determine all equilibrium solutions for the 2 point vortex motion , again , for both cases . we proceed by comparing the hamiltonian energy curves for the two layer problem in upper half plane with the one layer problem in the upper half plane . we then present qualitative aspects of streamline topologies for the two layer problem , for both cases @xmath3 @xmath6 and @xmath3 @xmath5 . we conclude with a study of integrable 3 vortex motion by considering symmetrical initial configurations , and seeking conditions to maintain the symmetry of the initial configuration . by enforcing those conditions , which simplify by symmetry of the configuration , we are able to obtain , numerically , rich classes of relative equilibria in both cases @xmath3 @xmath6 and @xmath7 @xmath5 . we consider first an ensemble of two point vortices in the upper half plane where @xmath8 is in bottom layer and @xmath9 is in top . denoting the respective stream functions in the corresponding layers by @xmath10 ( @xmath11 ) the expession for determining the streamfunctions corresponding to unit ( delta ) point vortices @xmath12 in each of the two layers are [ coupled]@xmath13 where the subscripts identify the layer and @xmath14 is the internal radius of deformation . these equations apply here to the domain @xmath15 , the upper half plane with boundary @xmath16 , and boundary conditions @xmath17 . by introducing the sum and differences @xmath18 and @xmath19 the equations ( [ coupled ] ) uncouple [ uncup]@xmath20 the fundamental solution of ( [ uncup1 ] ) on the half plane is obtained , through the method of images @xcite , using the green s function , @xmath21 the fundamental solution of ( [ uncup2 ] ) on the half plane is obtained using the green s function @xmath22 with @xmath23 the modified bessel function using the complex variable notation , @xmath24 in this study we restrict the radius of deformation to be @xmath25 so that @xmath26 many regimes of motion can still be analyzed through suitable scaling coming from initial vortex separation and/or vortex strength ( circulation ) assignments . by superposition considering a vortex of strength ( circulation ) @xmath8 in layer 1 and @xmath9 in layer @xmath27 the motion of the vortices in each layer may be computed from the streamfunctions @xmath28 and @xmath29 arising from all the vortices ( except the one being advected in the case of 2 or more vortices ) . for example suppose there is nonzero @xmath8 in layer 1and @xmath30 , then stream functions induced by @xmath8 are , @xmath31 likewise with nonzero @xmath9 in layer 2 and @xmath32 , then stream functions induced by @xmath9 are , @xmath33 here , ( see fig . [ 2lfig1 ] ) @xmath34 and use is made of @xmath35 [ t ] by superposition then , with arbitrary @xmath36 @xmath37 and @xmath38 in layers 1 and 2 respectively the combined streamfunctions are @xmath39 now the dynamics of the point vortices can be obtained by differentiation of the streamfunctions as follows @xmath40 it is well know that the equations for point vortices are a hamiltonian system . it can be verified that the energy of the system @xmath41 is invariant . integrating by parts , substituting for @xmath42 using ( [ dynamic1 ] ) , @xmath10 using ( [ streams_combined])and @xmath43 using ( [ uncup ] ) as well as the @xmath44 boundary conditions shows that the hamiltonian can be simplified to , @xmath45 using delta distributed point vortices @xmath36 @xmath37and @xmath38 and the streamfunction for the ensemble ( [ streams_combined ] ) the invariant hamiltonian for @xmath8 at @xmath46 and @xmath9 at @xmath47simplfies to @xmath48 + 2\gamma_{1}\gamma_{2}\left [ \ln r_{1,2}-\ln r_{1,2^{\ast}}+k_{0}\left ( r_{1,2}\right ) -k_{0}\left ( r_{1,2^{\ast}}\right ) \right ] + \gamma_{2}^{2}\left [ k_{0}\left ( r_{2,2^{\ast}}\right ) -\ln r_{2,2^{\ast}}\right ] \text{,}\ ] ] again with reference to fig.([2lfig1 ] ) . image vortices are denoted with a @xmath49 . note also from the symmetry of the geometry of fig . ( [ 2lfig1 ] ) that @xmath50 and @xmath51 . when @xmath30 , @xmath52corresponding to a single vortex , the hamiltonian , @xmath53 simplifies to @xmath54 = c$ ] and by the monotonicity of @xmath55 we conclude that @xmath56 corresponding to the vortex @xmath8 translating parallel to the @xmath57-axis . this is the general solution then for the case of a single vortex . notice also in ( [ hamiltonian ] ) that the hamiltonian , @xmath53 , is invariant with respect to arbitrary displacements of both @xmath8 at @xmath46 and @xmath9 at @xmath47 by @xmath58 in the @xmath57-direction . this implies by noether s theorem the invariance of @xmath59 we provide an explicit proof of ( [ y_momentum ] ) in the appendix . the invariance of the hamiltonian , @xmath53 , ( [ hamiltonian ] ) and the momentum in the @xmath60-direction ( [ y_momentum ] ) imply that the 2 layer , 2 point vortex system in the half - plane is integrable . an easy way to see this is to notice that the hamiltonian depends only on @xmath61 and @xmath62 so that by ( [ y_momentum ] ) it depends only on either @xmath63 or @xmath64 . the other term in the hamiltonian is @xmath65 , from which we conclude that the hamiltonian is a function of the two variables , @xmath63 ( or @xmath64 ) and @xmath66 . we consider the cases @xmath3 @xmath6 and @xmath3 @xmath5 separately . in this case we use the invariant @xmath67 or @xmath68 with parameter @xmath69 . with reference to fig . ( [ 2lfig1 ] ) we designate @xmath70 so that @xmath71 ( so that @xmath72 and @xmath73 and denote @xmath74 with the hamiltonion ( [ hamiltonian ] ) simplifying to@xmath75 the hamiltonian level curves are shown for various parameter values @xmath76 in fig . ( [ 2lfig2 ] ) . there are no periodic solutions , and , in particular , no equilibria . we shall rigorously justifiy this in the sequel . there are two types of motion . the first corresponds to @xmath77 and @xmath78 approaching a nonzero value , or vice versa @xmath79 with @xmath74 approaching a nonzero value . which of these will depend on which side of the phase plane one begins ; either on the front side or the back side of the phase plane . this is similar to the only type of motion seen for the one layer problem . see fig ( [ 2lfig4 ] ) ( a ) , for the corresponding case @xmath80 for the one layer problem in the half plane . there is a second type of motion for the two layer problem seen , for all values of @xmath69 including @xmath81 . these are phase curves that cross the @xmath82 @xmath52 or @xmath83 line . in the case that @xmath81 this crossing corresponds to @xmath84 and @xmath83 , which corresponds to vortex collapsing configurations . we are still investigating the nature of these collapsing configurations , as to whether they are finite time or infinite time vortex collapse solutions . preliminary numerical results suggest that these are infinite time collapsing configurations . notice that these collapsing configurations are not admitted for @xmath3 @xmath5 , in the one layer case in the upper half plane . see fig ( [ 2lfig4 ] ) ( a ) , for the corresponding case @xmath80 for the one layer problem in the half plane and fig ( [ 2lfig4 ] ) ( b ) for the case @xmath85 . the second type of motion is seen in the one layer case , provide @xmath86 such as @xmath85 , however while @xmath87 , the @xmath60 values do not since @xmath85 . [ ptbh ] graph1.eps in this case we use the invariant @xmath67 or @xmath88 with parameter @xmath69 . with reference to fig . ( [ 2lfig1 ] ) we designate @xmath70 so that @xmath89 and denote @xmath90 with the hamiltonian ( [ hamiltonian ] ) simplifying to@xmath91 the hamiltonian level curves are shown for various parameter values @xmath69 in fig . ( [ 2lfig3 ] ) . the motion in this case is very similar in many ways to the one layer problem with @xmath3 @xmath6,(see fig ( [ 2lfig4 ] ) ( c ) ) with one minor difference . the similarities include the two types of motions . unlike the case @xmath3 @xmath92 , the case @xmath3 @xmath6 admits closed periodic solutions . also in the two layer case the center of these periodic curves corresponds to a fixed equilibrium . as previously mentioned the two layer problem in the upper half plane for @xmath3 @xmath92 does not admit any equilibrium solutions . what is interesting about these fixed equilibria is that they are centered at @xmath93 corresponding to @xmath94 and @xmath95 . the coordinates of both vortices are the same meaning they are stacked one on top of the other for the fixed equilibrium configuration . we will study , more carefully , in what follows this equilibrium solution and rigorously show that there are no other equilibrium solutions for the case @xmath3 @xmath6 in the upper half plane . clearly for the one layer problem we can not have both @xmath94and @xmath78 since there is freedom to stack the vortices , and so this would correspond to a collapsing configuration in which the vortices where initially located such that @xmath94and @xmath78 which is not feasible . this is the one major difference between the two layer and one layer problems in the upper half plane for the case @xmath3 @xmath6 . the second type of motion observed in fig . ( [ 2lfig3 ] ) , is a non periodic regime of motion in which either @xmath96 and @xmath63 approaches a nonzero number or where @xmath77 and @xmath63 approaches a nonzero number . which of these will again depend on which side of the phase plane one begins ; either on the front side or the back side of the phase plane . [ ptb ] graph2.eps in this section we consider equilibrium solutions for the 2 vortex problem . we begin by showing that there are no relative equilibrium solutions when @xmath3 @xmath5 with vortex @xmath25 , @xmath8 at @xmath97 in layer @xmath25 and vortex @xmath27 , @xmath9 at @xmath98 in layer @xmath27 . in this case the invariant @xmath99 , becomes @xmath68 . we consider the distance ( see fig . ( [ 2l2v_g1_negative_g2 ] ) ) between the two vortices @xmath8 and @xmath9 , @xmath100 and show that it is never constant . consider , @xmath101 where the last follows by the invariance of @xmath68 or @xmath102 . using the equations ( [ dynamicx1],[dynamicx2 ] ) and @xmath103 @xmath6 , we obtain , [ t ! ] 2lfig2.eps @xmath104 where the inequality @xmath105 follows from the positivity @xmath106 and the fact that @xmath107 which can easily shown using the geometry of the upper half plane . next we show that there is only one relative equilibrium solutions when @xmath3 @xmath6 with vortex @xmath25 , @xmath8 at @xmath97 in layer @xmath25 and vortex @xmath27 , @xmath9 at @xmath108 in layer @xmath27 . this corresponds to the case @xmath109 corresponding to two vortices lying exactly one on top of the other . note this equilibrium position is not feasible in the one - layer case . these relative equilibria are clear in fig . ( [ 2lfig3 ] ) in which @xmath110 , and @xmath95 and correspond to the center of the periodic orbits . in this case the invariant @xmath99 , becomes @xmath88 . we consider the distance ( see fig.([2l2v_g1_positive_g2 ] ) ) between the two vortices @xmath8 and @xmath9 , @xmath111 and show that it is never constant . [ h ] 2lfig3.eps consider , @xmath112 @xmath113 @xmath114 by considering the phase curves @xmath3 @xmath6 shown in fig . ( [ 2lfig3 ] ) we see that the curves all pass through @xmath115 . we show that in this case @xmath116 so that in eq . ( [ relativeequation_both_positive ] ) for a relative equilibrium @xmath117 now when @xmath110 , ( [ x1_minus_x2_both_positive ] ) becomes @xmath118 which can be shown to be nonzero except when @xmath119 which would be simultaneous with @xmath120 so in eq . ( [ relativeequation_both_positive ] ) we require ( [ relativeequation_both_positive_ypart ] ) . in this case when @xmath120 , ( [ relativeequation_both_positive_ypart ] ) simplifies to @xmath121 which again can be shown to be nonzero except when @xmath119 which would be simultaneous with @xmath120 . the case @xmath3 @xmath6 admits then the relative equilibria @xmath122 where the vortices lie exactly one on top of the other . we can see that the phase curves ( [ phase2 ] ) near this equilibrium position are closed periodic orbits so that the phase curves near to @xmath123 are almost relative equilibria , in the sense that the level curves ( [ phase2 ] ) close to @xmath122 are asymptotically elliptical @xmath124 or @xmath125 or @xmath126 corresponding to what would be a true relative equilibrium . this is a novel relative equilibrium solution , keeping in mind that the one - layer two vortex problem on the upper half plan admits no relative equilibrium solutions . we summarize the one layer two vortex regimes of motion in the upper half plane for completeness and to highlight the similarities and differences we mentioned earlier . in this case the green s functions using the method of images corresponds only to solving in one layer ( [ uncup1 ] ) and not ( [ uncup2 ] ) and is given by ( [ green1 ] ) . the general hamiltonian becomes,@xmath127 again , use is made of the invariant @xmath128 @xmath129 . we consider three representative cases : 1 . @xmath1 and @xmath130 @xmath131 the hamiltonian , using @xmath90 and @xmath132 , @xmath133 2 . @xmath1 and @xmath130 @xmath134 , so @xmath135 , with hamiltonian @xmath136 3 . @xmath0 and @xmath137so @xmath138 , with hamiltonian @xmath139 + the phase curves are shown in fig.([2lfig4 ] ) ( a ) , ( b ) and ( c ) respectively . + [ h ] + 2lfig4.eps + notice , in particular , that when @xmath30 @xmath52corresponding to a single vortex , the hamiltonian , @xmath53 simplifies to @xmath140 = c$ ] and by the monotonicity of @xmath141 we conclude that @xmath56 corresponding to the vortex @xmath8 translating parallel to the @xmath57-axis . this is the general solution then for the case of a single vortex . notice also in ( [ hamiltonian ] ) that the hamiltonian , @xmath53 , is invariant with respect to arbitrary displacements of both @xmath8 at @xmath46 and @xmath9 at @xmath47 by @xmath58 in the @xmath57-direction . this implies by noether s theorem the invariance of @xmath142 . we consider the cases @xmath3 @xmath6 and @xmath3 @xmath5 separately . we consider two basic configurations . one in which @xmath3 @xmath4 , with @xmath143 in layer @xmath25 , and @xmath144 in layer @xmath145 , with the same @xmath60-coordinate ( @xmath146 ) as shown in fig ( [ stream_configuration1 ] ) . we consider various values of the parameters @xmath69 and @xmath146 with reference to rossby radius of deformation @xmath147 . [ h ] , @xmath146 . the streamline is @xmath148 the streamlines are shown in fig . ( [ streamline1 ] ) for this case with @xmath149 and in fig . ( [ streamline2 ] ) for the case @xmath150 . for the case @xmath149 the streamlines are topologically similar as @xmath69 is varied . for the case @xmath150 the streamlines undergo topological change as @xmath69 changes through @xmath151 to @xmath152 with the introduction of a saddle stagnation point . [ h ] streamline1.eps [ h ! ] streamline2.eps the second basic configuration we consider is with 2 vortices placed in a vertical configuration , with @xmath8 in layer @xmath25 and @xmath9 in layer @xmath27 as shown in fig . ( [ stream_configuration2 ] ) . in this case @xmath153 . [ h ! ] 2lstream2.eps in this case 3 classes of streamlines are observed . notice that @xmath154 , and @xmath155 is invariant . the first streamline case corresponds to when @xmath156 ( when @xmath8 and @xmath157are relatively far apart ) and is shown in fig . ( [ streamline4](a ) ) . there is a saddle stagnation point observed for this case . the second case corresponds to when @xmath158 ( so that is close to the boundary ) and is shown in fig . ( [ streamline4](b ) ) . the third case corresponds to the intermediate case when @xmath69 is not much smaller than @xmath146 but not too similar in magnitute to @xmath146 either and is depicted in fig . ( [ streamline4](c ) ) . for the cases @xmath158 and the intermediate case when @xmath69 is not much smaller than @xmath146 but not too similar in magnitute to @xmath146 there are no stagnation points . while in both of these cases the streamlines are in many ways topologically similar , the two layer interaction does introduce distortion effects upon a closer examination of ( [ streamline4](b ) ) and ( [ streamline4](c ) ) . [ h ! ] streamline4_vertical.eps we consider the same two basic configurations . one in which @xmath3 @xmath5 , with @xmath143 in layer @xmath25 , and @xmath144 in layer @xmath27 , with the same @xmath60-coordinate as shown in fig ( [ stream_configuration1 ] ) . we consider various values of the parameters @xmath69 and @xmath146 with reference to rossby radius of deformation @xmath147 . note in this case @xmath159 is invariant . the streamlines are shown for @xmath150 in fig ( [ streamline5 ] ) . for all values of @xmath146 ( including @xmath160 ) a stagnation streamline emanating from the bounary is observed . there is a stagnation point at the boundary from which the stagnation streamline originates . as seen for the case @xmath150 in fig ( [ streamline5 ] ) the basic topology of the streamline does not change as @xmath69 is varied from @xmath161 through @xmath162 . [ h ] streamline5.eps [ thb ! ] in layer 1 at @xmath163 , @xmath157 in layer 2 at @xmath164.in this case @xmath165 is invariant . the typical case @xmath166 is shown . the streamfunction is @xmath167 the second configuration is as before with @xmath3 @xmath5 , with @xmath143 in layer @xmath25 and @xmath144 in layer @xmath27 . the 2 vortices are placed in a vertical configuration , with @xmath8 in layer @xmath25 at @xmath168 and @xmath9 in layer @xmath27 at @xmath169 shown in fig . ( [ stream_configuration2 ] ) . in this case @xmath153 . we consider various values of the parameters @xmath69 and @xmath146 with reference to rossby radius of deformation @xmath147 . note in this case @xmath170 is invariant . in this case there is a stagnation streamline joining two stagnation points on the boundary as seen in fig . ( [ streamline6 ] ) . we conclude with a hamiltonian study of integrable two - layer 3 vortex motion on the half plane by studying integrable symmetrical configurations and provide a rich class of new relative equilibria . we consider two basic symmetrical configurations of 3 vortices as depicted in fig.([int_3_vortex_config ] ) . we seek relative equilibrium solutions.in which the initial configuration is rigidly maintained . the method we adopt is similar to that as in the work of jamaloodeen and newton @xcite in which we seek relative equilbria base on a a symmetrical configuration and vary parameters that ensure the relative equilbrium or rigid shape is invariant . in the first symmetrical configuration we have @xmath3 @xmath6 , with @xmath143 in layer @xmath25 , and @xmath144 in layer @xmath27 , with the same @xmath60-coordinate as shown in fig ( [ int_3_vortex_config])(a ) . in this case it can be shown that @xmath176 and @xmath177 . the invariant @xmath178 , becomes @xmath179 , or @xmath180 . relative equilibrium solutions will then be admitted , for this configuration , by requiring that @xmath181 , and @xmath182 . these equations ( see appendix b for their explicit forms ) are solved numerically with @xmath69 and @xmath146 considered as parameters . these are summarized in in tables ( [ horizontal_3_equilibria_beta_negative])-([horizontal_3_equilibria_beta_positive ] ) , and exploiting the symmetry and the invariant @xmath179 , or @xmath183 only the coordinates @xmath184 are given . [ h ] int3.eps @xmath185{l|l}\hline\hline $ \left ( \alpha,\beta\right ) $ ( negative $ \beta)$ & relative equilibrium $ \left ( x , y\right ) $ \\\hline\hline $ \left ( 2.005,-0.25\right ) $ & $ \left ( 0.5639,4.524\right ) $ \\\hline $ \left ( 2.01,-0.25\right ) $ & $ \left ( 0.6827,4.328\right ) $ \\\hline $ \left ( 2.02,-0.25\right ) $ & $ \left ( 0.7252,4.114\right ) , \left ( 0.8457,12.43\right ) $ \\\hline $ \left ( 2.05,-0.5\right ) $ & $ \left ( 0.5847,3.291\right ) , \left ( 0.8497,9.977\right ) , \left ( 0.01254,9.941\right ) $ \\\hline $ \left ( 2.1,-0.5\right ) $ & $ \left ( 0.8317,4.851\right ) , \left ( 0.7775,3.466\right ) $ \\\hline $ \left ( 2.11,-0.5\right ) $ & $ \left ( 0.8558,3.857\right ) $ \\\hline $ \left ( 2.01,-0.75\right ) $ & $ \left ( 0.4542,3.737\right ) $ \\\hline $ \left ( 2.09,-0.75\right ) $ & $ \left ( 0.8156,8.314\right ) , \left ( 0.4843,2.773\right ) $ \\\hline $ \left ( 2.15,-0.75\right ) $ & $ \left ( 0.7855,4.941\right ) , \left ( 0.665,2.863\right ) $ \\\hline $ \left ( 2.19,-0.75\right ) $ & $ \left ( 0.8015,3.917\right ) $ \\\hline $ \left ( 2.1,-1\right ) $ & $ \left ( 0.5144,2.569\right ) , \left ( 0.8156,10.01\right ) $ \\\hline $ \left ( 2.25,-1\right ) $ & $ \left ( 0.7955,3.857\right ) , \left ( 0.663,2.622\right ) $ \\\hline $ \left ( 2.27,-1\right ) $ & $ \left ( 0.7775,3.315\right ) , \left ( 0.7172,2.773\right ) $ \\\hline $ \left ( 2.275,-1\right ) $ & $ \left ( 0.7654,3.104\right ) $ \\\hline $ \left ( 2.7,-5\right ) $ & $ \left ( 0.7252,7.11\right ) , \left ( 0.9661,1.177\right ) $ \\\hline $ \left ( 3,-5\right ) $ & $ \left ( 0.916,0.9963\right ) , \left ( 0.657,5.002\right ) , \left ( 0.04868,4.941\right ) $ \\\hline $ \left ( 4,-5\right ) $ & $ \left ( 0.5426,2.442\right ) , \left ( 0.7232,0.9059\right ) , \left ( 0.06072,2.442\right ) $ \\\hline $ \left ( 4.5,-5\right ) $ & $ \left ( 0.4522,1.779\right ) , \left ( 0.6028,0.9963\right ) $ \\\hline $ \left ( 4.825,-5\right ) $ & $ \left ( 0.4348,1.313\right ) $ \\\hline \end{tabular } \label{horizontal_3_equilibria_beta_negative}\ ] ] @xmath185{l|l}\hline\hline $ \left ( \alpha,\beta\right ) $ ( positive $ \beta)$ & relative equilibrium $ \left ( x , y\right ) $ \\\hline\hline $ \left ( 1.9855,0.1\right ) $ & $ \left ( 0.8845,6.198\right ) $ \\\hline $ \left ( 1.99,0.1\right ) $ & $ \left ( 0.8678,9.977\right ) , \left ( 0.8497,5.46\right ) $ \\\hline $ \left ( 1.9755,0.15\right ) $ & $ \left ( 0.8964,5.509\right ) $ \\\hline $ \left ( 1.98,0.15\right ) $ & $ \left ( 0.8608,7.385\right ) , \left ( 0.8548,4.987\right ) $ \\\hline $ \left ( 1.9533,0.25\right ) $ & $ \left ( 0.8798,4.791\right ) $ \\\hline $ \left ( 1.954,0.25\right ) $ & $ \left ( 0.8678,5.152\right ) , \left ( 0.8738,4.58\right ) $ \\\hline $ \left ( ( 1.96,0.25\right ) $ & $ \left ( 0.8738,6.206\right ) , \left ( 0.8558,4.399\right ) $ \\\hline $ \left ( 1.888,0.5\right ) $ & $ \left ( 0.91,3.917\right ) $ \\\hline $ \left ( 1.9,0.5\right ) $ & $ \left ( 0.8919,4.917\right ) , \left ( 0.8317,3.544\right ) $ \\\hline $ \left ( 1.95,0.5\right ) $ & $ \left ( 0.5907,3.448\right ) , \left ( 0.8738,10.03\right ) , ( 0.01856,10.03)$\\\hline $ \left ( 1.807,0.75\right ) $ & $ \left ( 0.9201,3.491\right ) $ \\\hline $ \left ( 1.82,0.75\right ) $ & $ \left ( 0.9201,4.052\right ) , \left ( 0.8608,3.158\right ) $ \\\hline $ \left ( 1.715,1\right ) $ & $ \left ( 0.9558,3.075\right ) $ \\\hline $ \left ( 1.75,1\right ) $ & $ \left ( 0.932,3.927\right ) , \left ( 0.8192,2.763\right ) $ \\\hline $ \left ( 1.82,1\right ) $ & $ \left ( 0.9201,5.569\right ) , \left ( 0.653,2.743\right ) $ \\\hline $ \left ( 1.9,1\right ) $ & $ \left ( 0.5546,2.93\right ) , \left ( 0.8919,9.941\right ) , \left ( 0.01254,9.941\right ) $ \\\hline $ \left ( 1.483,1.5\right ) $ & $ \left ( 1.009,2.541\right ) $ \\\hline $ \left ( 1.5,1.5\right ) $ & $ \left ( 1.027,2.802\right ) , \left ( 0.9617,2.375\right ) $ \\\hline $ \left ( 1.7,1.5\right ) $ & $ \left ( 0.9616,5.022\right ) , \left ( 0.7479,2.44\right ) $ \\\hline $ \left ( 1.85,1.5\right ) $ & $ \left ( 0.9039,10.01\right ) , \left ( 0.01254,10.01\right ) , \left ( 0.6871,2.858\right ) $ \\\hline $ \left ( 1.34,1.75\right ) $ & $ \left ( 1.07,2.334\right ) $ \\\hline $ \left ( 1.4,1.75\right ) $ & $ \left ( 0.9912,2.179\right ) , \left ( 1.082,2.79\right ) $ \\\hline $ \left ( 1.18,2\right ) $ & $ \left ( 1.181,2.161\right ) $ \\\hline $ \left ( 1.25,2\right ) $ & $ \left ( 1.098,2.048\right ) , \left ( 1.164,2.553\right ) $ \\\hline $ \left ( 1.5,2\right ) $ & $ \left ( 1.045,4.007\right ) , \left ( 0.9379,2.262\right ) $ \\\hline $ \left ( 0.8,2.5\right ) $ & $ \left ( 1.532,1.888\right ) $ \\\hline $ \left ( 0.85,2.5\right ) $ & $ \left ( 1.496,1.852\right ) , \left ( 1.472,2.066\right ) $ \\\hline $ \left ( 1.5,2.5\right ) $ & $ \left ( 1.051,4.987\right ) , \left ( 1.042,2.44\right ) $ \\\hline $ \left ( 1.75,2.5\right ) $ & $ \left ( 0.9494,3.03\right ) , \left ( 0.9403,10.03\right ) , \left ( 0.02784,10.03\right ) $ \\\hline \end{tabular } \ \label{horizontal_3_equilibria_beta_positive}\ ] ] notice that the numerical evidence suggest that there are no relative equilibria corresponding to this configuration with @xmath69 negative or @xmath186 . there also appears to be a complicated bifurcation of these relative equilibria . for example fixing @xmath146 and varying @xmath69 gives varying numbers of relative equilibria . consider , for example the case @xmath187 for which at @xmath188 , there is one relative equilibrium solution , then increasing @xmath69 through @xmath189 there are two and likewise two , again , after increasing @xmath69 through @xmath190 and finally increasing @xmath69 to @xmath191 there is again only one relative equilibrium solution . similarly when @xmath149 there is one equilibrium configuration when @xmath192 , two equilibrium configurations when @xmath193 and three equilibrium configurations when @xmath69 increases through @xmath194 . a scatterplot of all numerically found relative equilibria from tables ( [ horizontal_3_equilibria_beta_negative])-([horizontal_3_equilibria_beta_positive ] ) is shown in fig([s_plot1 ] ) , with associated @xmath195 for each configuration @xmath196 suppressed . [ h ! ] ) -([horizontal_3_equilibria_beta_positive ] ) with associated @xmath197 for each configuration @xmath198 suppressed . ] it is also possible that some of these relative equilibria are fixed equilibria , meaning that in fact not only are @xmath181 , and @xmath199 but @xmath181 , and @xmath200 . however preliminary numerical studies suggest that there are no fixed equilibrium configurations of the kind depicted in fig([int_3_vortex_config])(a ) . the bifurcation sequence suggested above and the fixed equilibria are topics for further study . in the second symmetrical configuration we have @xmath3 @xmath6 , with @xmath143 in layer @xmath25 , and @xmath144 in layer @xmath27 , with @xmath8 at @xmath201 , @xmath9 at @xmath202 ( in layer 1 ) and @xmath174 at @xmath175 ( in layer 2)with the same @xmath60-coordinate as shown in fig ( [ int_3_vortex_config])(b ) . in this case it can be shown that @xmath176 and @xmath203 . the invariant @xmath204 , becomes @xmath205 , or @xmath180 . in this case it can be shown that @xmath206 . relative equilibrium solutions will then be admitted , for this configuration by requiring that @xmath207 . again the invariant @xmath178 , becomes @xmath205 , or @xmath180 . these equations ( see appendix b for their explicit forms ) are solved numerically with @xmath69 and @xmath146 considered as parameters . these are summarized in in tables ( [ vertical_3_equilibria_beta_1])-([vertical_3_equilibria_beta_2 ] ) , and exploiting the symmetry and the invariant @xmath179 , or @xmath208 again it suffices to provide only the coordinates for @xmath209 . @xmath185{l|l}\hline\hline $ \left ( \alpha,\beta\right ) $ ( $ \alpha\leq2 $ ) & relative equilibrium $ \left ( y , y\right ) $ \\\hline\hline $ \alpha\leq1 $ & no numerical relative equilibrium configurations\\\hline $ \left ( 1.5,0.25\right ) $ & $ \left ( 4.132,1.549\right ) $ \\\hline $ \left ( 1.5,0.5\right ) $ & $ \left ( 3.859,1.306\right ) $ \\\hline $ \left ( 1.5,1\right ) $ & $ \left ( 3.687,1.211\right ) $ \\\hline $ \left ( 1.5,2\right ) $ & $ \left ( 3.758,1.353\right ) $ \\\hline $ \left ( 1.5,3\right ) $ & $ \left ( 4.369,1.749\right ) $ \\\hline $ \left ( 1.5,3.4\right ) $ & $ \left ( 4.72,1.923\right ) , \left ( 2.333,4.987\right ) $ \\\hline $ \left ( 1.5,3.5\right ) $ & $ \left ( 4.827,1.995\right ) , \left ( 2.832,5.379\right ) , \left ( 2.03,4.844\right ) $ \\\hline $ \left ( 1.5,5\right ) $ & $ \left ( 1.809,5.995\right ) , \left ( 5.393,8.073\right ) , \left ( 6.598,2.893\right ) $ \\\hline $ \left ( 1.5,10\right ) $ & $ \left ( 13.32,6.533\right ) , \left ( 13.92,17.06\right ) , \left ( 2.414,11.23\right ) $ \\\hline $ \left ( 1.75,-5\right ) $ & $ \left ( 1.037,8.751\right ) $ \\\hline $ \left ( 1.75,-2\right ) $ & $ \left ( 0.9649,4.791\right ) $ \\\hline $ \left ( 1.75,-1\right ) $ & $ \left ( 1.336,3.903\right ) $ \\\hline $ \left ( 1.75,-0.5\right ) $ & $ \left ( 1.899,4.102\right ) $ \\\hline $ \left ( 1.75,0.5\right ) $ & $ \left ( 3.68,1.736\right ) $ \\\hline $ \left ( 1.75,1\right ) $ & $ \left ( 3.703,1.745\right ) $ \\\hline $ \left ( 1.75,1.5\right ) $ & $ \left ( 3.963,1.959\right ) , \left ( 2.716,4.653\right ) $ \\\hline $ \left ( 1.75,1.75\right ) $ & $ \left ( 4.187,2.152\right ) , \left ( 1.846,3.936\right ) , \left ( 3.772,5.704\right ) $ \\\hline $ \left ( 1.75,2\right ) $ & $ \left ( 4.52,2.412\right ) , \left ( 1.659,3.857\right ) , \left ( 4.7,6.628\}\right ) $ \\\hline $ \left ( 1.75,3\right ) $ & $ \left ( 6.222,3.821\right ) , \left ( 1.436,4.303\right ) , \left ( 7.553,9.555\right ) $ \\\hline $ \left ( 2,0.25\right ) $ & $ \left ( 4.38,2.648\right ) , \left ( 2.365,4.107\right ) $ \\\hline $ \left ( 2,0.5\right ) $ & $ \left ( 4.589,2.837\right ) , \left ( 2.015,3.771\right ) $ \\\hline $ \left ( 2,0.75\right ) $ & $ \left ( 6.135,4.349\right ) , \left ( 1.745,3.562\right ) $ \\\hline $ \left ( 2,1\right ) $ & $ \left ( 1.603,3.491\right ) $ \\\hline $ \left ( 2,2\right ) $ & $ \left ( 1.3,3.651\right ) $ \\\hline $ \left ( 2,3\right ) $ & $ \left ( 1.235,4.349\right ) $ \\\hline $ \left ( 2,5\right ) $ & $ \left ( 1.302,6.296\right ) $ \\\hline $ \left ( 2,10\right ) $ & $ \left ( 1.708,11.66\right ) $ \\\hline $ \left ( 2,15\right ) $ & $ \left ( 2.201,17.14\right ) $ \\\hline \end{tabular } \ \ \label{vertical_3_equilibria_beta_1}\ ] ] @xmath185{l|l}\hline\hline $ \left ( \alpha,\beta\right ) $ ( $ \alpha>2 $ ) & relative equilibrium $ \left ( y , y\right ) $ \\\hline $ \left ( 2.35,-6\right ) $ & $ \left ( 14.69,13.01\right ) , \left ( 6.292,9.165\right ) , \left ( 0.5845,6.417\right ) $ \\\hline $ \left ( 2.35,-5\right ) $ & $ \left ( 12.34,10.67\right ) , \left ( 4.927,7.412\right ) , \left ( 0.578,5.416\right ) $ \\\hline $ \left ( 2.35,-3\right ) $ & $ \left ( 7.983,6.327\right ) , \left ( 0.665,3.405\right ) , \left ( 2.472,4.309\right ) $ \\\hline $ \left ( 2.35,-2\right ) $ & $ \left ( 5.784,4.219\right ) $ \\\hline $ \left ( 2.35,-1\right ) $ & $ \left ( 4.31,2.76\right ) $ \\\hline $ \left ( 2.35,-0.5\right ) $ & $ \left ( 4.615,3.011\right ) $ \\\hline $ \left ( 2.35,0.25\right ) $ & $ \left ( 2.012,3.651\right ) $ \\\hline $ \left ( 2.35,1\right ) $ & $ ( 1.365,3.28)$\\\hline $ \left ( 2.35,1.5\right ) $ & $ \left ( 1.223,3.408\right ) $ \\\hline $ \left ( 2.35,1.75\right ) $ & $ \left ( 1.202,3.547\right ) $ \\\hline $ \left ( 2.35,2\right ) $ & $ \left ( 1.175,3.704\right ) $ \\\hline $ \left ( 2.35,5\right ) $ & $ \left ( 1.211,6.567\right ) $ \\\hline $ \left ( 2.35,10\right ) $ & $ \left ( 1.538,11.87\right ) $ \\\hline $ \left ( 3,-10\right ) $ & $ \left ( 9.041,7.594\right ) , \left ( 32.93,21.49\}\right ) , \left ( 3.121,6.554\right ) $ \\\hline $ \left ( 3,-5\right ) $ & $ \left ( 4.898,3.509\right ) , \left ( 16.3,10.69\right ) , \left ( 1.246,3.206\right ) $ \\\hline $ \left ( 3,-4\right ) $ & $ \left ( 4.114,2.755\right ) , \left ( 12.78,8.437\right ) $ \\\hline $ \left ( 3,-3\right ) $ & $ \left ( 3.441,2.125\right ) , \left ( 8.962,6.012\right ) $ \\\hline $ \left ( 3,-2\right ) $ & $ \left ( 3.341,2.003\right ) , \left ( 5.525,3.685\right ) $ \\\hline $ \left ( 3,0.25\right ) $ & $ \left ( 1.852,3.954\right ) $ \\\hline $ \left ( 3,1\right ) $ & $ \left ( 1.247,3.488\right ) $ \\\hline $ \left ( 3,2\right ) $ & $ \left ( 1.086,4.096\right ) $ \\\hline $ \left ( 3,5\right ) $ & $ \left ( 1.131,7.039\right ) $ \\\hline $ \left ( 3,10\right ) $ & $ \left ( 1.387,12.53\right ) $ \\\hline $ \left ( 5,-10\right ) $ & $ \left ( 3.396,2.36\right ) , \left ( 6.055,4.096\right ) $ \\\hline $ \left ( 5,0.25\right ) $ & $ \left ( 4.301,17.17\right ) $ \\\hline $ \left ( 5,0.5\right ) $ & $ \left ( 2.729,10.86\right ) $ \\\hline $ \left ( 5,1\right ) $ & $ \left ( 1.796,7.29\right ) $ \\\hline $ \left ( 5,5\right ) $ & $ \left ( 1.176,8.581\right ) $ \\\hline $ \left ( 5,5\right ) $ & $ \left ( 1.248,13.86\right ) $ \\\hline \end{tabular } \ \ \ \label{vertical_3_equilibria_beta_2}\ ] ] notice again that the numerical evidence suggests that there are no relative equilibria corresponding to this configuration with @xmath69 negative or @xmath186 . there again appears to be a complicated bifurcation of these relative equilibria . for example fixing @xmath69 and varying @xmath146 gives varying numbers of relative equilibria . a scatterplot of all numerically found relative equilibria from tables ( [ vertical_3_equilibria_beta_1])-([vertical_3_equilibria_beta_2 ] ) is shown in fig([s_plot1 ] ) , with associated @xmath195 for each configuration @xmath210and @xmath211 suppressed . [ h ! ] ) -([vertical_3_equilibria_beta_2 ] ) with associated @xmath212 for each configuration @xmath213and @xmath214 or @xmath215 suppressed . ] we have presented results on integrable two - layer point vortex motion on the upper half plane and shown similarities and differences with integrable one - layer point vortex motion in the upper half plane and with integrable two - layer point vortex motion on the entire plane including the study of equilbrium solutions , the finite - time vortex collapse problem and a study of streamline topologies . at present we are also pursuing work to determine 3 or 4 vortex initial vortex configurations that lead to finite time collapse of the vortices . should such configurations exist , they will depend on the initial configuration of the vortices and the vortex strengths , and would possibly be self - similar . also of interest would be to perform a systematic bifurcation analysis of the 3 vortex relative equilibria we found shown in figs.([s_plot1]-[s_plot2 ] ) . the bifurcation analysis in both cases would depend on the parameter @xmath195 as depicted in fig . ( [ int_3_vortex_config ] ) . finally we are also pursuing work in the direction of establishing conditions under which 3 vortices may induce chaotic advection of fluid particles much along the lines for the two layer problem in the entire plane as done in the work of koshel et al@xcite . a good starting point are perturbations of the configurations used to obtain the streamlines depicted in figs . ( [ stream_configuration1 ] and [ stream_configuration2 ] ) . 99 flierl , g.r . , polvani , l.m . , zabusky , n.j . , _ two - layer geostrophic vortex dynamics . uper - layer v - states and merger , _ j. fluid . mech , * 205 * 325242 ( 1989 ) . flucher , m. , gustafsson , b. , _ vortex motion in two dimensional hydrodynamics _ , preprint , department of mathematics , royal institute of technology sweden , ( 1997 ) . griffiths , r.w . , hopfinger , e.j . , _ experiments with baroclinic vortex pairs in a rotating fluid , _ j. fluid . mech , * 173 * 501518 ( 1986 ) . gryanik v.m . , _ dynamics of singular geostrophic vortices in a two - level model of the atmosphere ( or ocean ) _ , bull . ( izv . ) , acad . ussr , atmospheric and oceanic physics * 29 * ( 3 ) 171 - 179 ( 1983 ) . hogg , n. , stommel , h. , _ the heton , an elementary interaction between discrete baroclinic geostrophic vortices , and its applications concerning eddy heat - flow _ , proc . a , * 397 * , 1 - 20 ( 1985 ) . hogg , n. , stommel , h. , _ hetonic explostions : the breakup and spread of warm pools as explained by baroclinic point vortices _ , j. atmos . , * 42 * ( 14 ) 1465 - 1476 ( 1985 ) . jamaloodeen , m. , _ hamiltonian methods for some geophysical vortex dynamics models _ , ph.d . thesis , dept . of mathematics , university of southern california ( 2000 ) . jamaloodeen , m.i . , newton , p.k . , _ the n - vortex problem on a rotating sphere . ii . heterogeneous platonic solid equilibria _ , journal of mathematical physics , proc . a , * 462 * , 3277 - 3299 ( 2006 ) . jamaloodeen , m.i . , newton , p.k . , _ two - layer quasigeostrophic potential vorticity model _ , journal of mathematical physics , * 48 * , 065601 ( 2007 ) . koshel , k.v . , sokolovskiy , m.a.,verrona , j. , _ three - vortex quasi - geostrophic dynamics in a two - layer fluid . analysis of relative and absolute motions _ , mech , * 717 * 232254 ( 2013 ) . koshel , k.v . , sokolovskiy , m.a.,verrona , j. , _ three - vortex quasi - geostrophic dynamics in a two - layer fluid . regular and chaotic advection around the perturbed steady states _ , j. fluid . mech , * 717 * 255280 ( 2013 ) . mcwilliams , j.c . , zabusky , n.j . , _ a modulated point - vortex model for geostrophic , _ @xmath146__-plane dynamics _ _ , phys . fluids * 25 * ( 12 ) 21752182 ( 1982 ) . melnikov yu . , a. , _ construction of green s functions for the two - dimensional static klein - gordong equation _ , j. part . , * 24 * ( 2 ) 114139 ( 1982 ) . newton , p.k . , _ the _ @xmath2__-vortex problem : analystical techniques _ _ , in applied mathematical science * 145 * ( 2001 ) . pedlosky , j. , _ geophysical fluid dynamics _ , springer -verlage , ny , ( 1987)*. * young , w.r . , _ some interactions between small numbers of baroclinic geostrophic vortice _ , geophys . fluid dynamics * 33 * 35 - 61 ( 1985 ) . we show that @xmath217 . use , @xmath218 @xmath219\end{gathered}\ ] ] @xmath220 \text{.}\ ] ] here @xmath221 it is clear that since @xmath222 and @xmath50 that @xmath223 . for completeness we present the dynamical equations for the @xmath57-components , for reference as needed when used . @xmath224 \ + \frac{\gamma_{2}\left ( y_{1}-y_{2}\right ) } { r_{1,2}}\left [ k_{1}\left ( r_{1,2}\right ) -\frac { 1}{r_{1,2}}\right ] \label{dynamicx1}\\ + \frac{\gamma_{2}\left ( y_{1}+y_{2}\right ) } { r_{1,2^{\ast}}}\left [ \frac { 1}{r_{1,2^{\ast}}}-k_{1}\left ( r_{1,2^{\ast}}\right ) \right ] \text{.}\ ] ] @xmath225 + \frac{\gamma_{2}\left ( y_{2}-y_{1}\right ) } { r_{1,2}}\left [ k_{1}\left ( r_{1,2}\right ) -\frac { 1}{r_{1,2}}\right ] \label{dynamicx2}\\ + \frac{\gamma_{2}\left ( y_{1}+y_{2}\right ) } { r_{1,2^{\ast}}}\left [ \frac { 1}{r_{1,2^{\ast}}}-k_{1}\left ( r_{1,2^{\ast}}\right ) \right ] \text{.}\ ] ] we provide the equations of motion for the two cases considered . [ [ the - case - gamma_1-gamma_21gamma_3-alpha - with - gamma_1-at - left - xyright - gamma_2-at - left -- xyright - in - layer-1-and - gamma_3-at - left-0y_3right - in - layer-2 - 1 ] ] the case @xmath3 @xmath171 with @xmath8 at @xmath172 , @xmath9 at @xmath173 ( in layer 1 ) and @xmath174 at @xmath175 ( in layer 2 ) ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ in this case it can be shown that @xmath176 and @xmath203 . relative equilibrium solutions will then be admitted , for this configuration by requiring that @xmath181 , and @xmath226 . in this case the invariant @xmath178 , becomes @xmath179 , or @xmath180 . @xmath227 [ [ the - case - gamma_1-gamma_21gamma_3-alpha - with - gamma_1-at - left-0y_1right - gamma_2-at - left0y_2right - in - layer-1-and - gamma_3-at - left-0y_3right - in - layer-2 - 1 ] ] the case @xmath3 @xmath171 with @xmath8 at @xmath201 , @xmath9 at @xmath202 ( in layer 1 ) and @xmath174 at @xmath175 ( in layer 2 ) ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ in this case it can be shown that @xmath206 . relative equilibrium solutions will then be admitted , for this configuration by requiring that @xmath207 . again the invariant @xmath178 , becomes @xmath179 , or @xmath180 .

we consider the estimation of an i.i.d . ( possibly non - gaussian ) vector from measurements obtained by a general cascade model consisting of a known linear transform followed by a probabilistic componentwise ( possibly nonlinear ) measurement channel . a novel method , called adaptive generalized approximate message passing ( adaptive gamp ) , that enables joint learning of the statistics of the prior and measurement channel along with estimation of the unknown vector is presented . the proposed algorithm is a generalization of a recently - developed em - gamp that uses expectation - maximization ( em ) iterations where the posteriors in the e - steps are computed via approximate message passing . the methodology can be applied to a large class of learning problems including the learning of sparse priors in compressed sensing or identification of linear - nonlinear cascade models in dynamical systems and neural spiking processes . we prove that for large i.i.d . gaussian transform matrices the asymptotic componentwise behavior of the adaptive gamp algorithm is predicted by a simple set of scalar state evolution equations . in addition , we show that when a certain maximum - likelihood estimation can be performed in each step , the adaptive gamp method can yield asymptotically consistent parameter estimates , which implies that the algorithm achieves a reconstruction quality equivalent to the oracle algorithm that knows the correct parameter values . remarkably , this result applies to essentially arbitrary parametrizations of the unknown distributions , including ones that are nonlinear and non - gaussian . the adaptive gamp methodology thus provides a systematic , general and computationally efficient method applicable to a large range of complex linear - nonlinear models with provable guarantees .

1207.3859

consider the estimation of a random vector @xmath0 from the measurement model illustrated in figure [ fig : model ] . the random vector @xmath2 , which is assumed to have independent and identically distributed ( i.i.d . ) components @xmath3 , is passed through a known linear transform that outputs @xmath4 . the components of @xmath1 are generated by a componentwise transfer function @xmath5 . this work addresses the cases where the distributions @xmath6 and @xmath5 have some unknown parameters , @xmath7 and @xmath8 , that must be learned in addition to the estimation of @xmath2 . such joint estimation and learning problems with linear transforms and componentwise nonlinearities arise in a range of applications , including empirical bayesian approaches to inverse problems in signal processing , linear regression and classification @xcite , and , more recently , bayesian compressed sensing for estimation of sparse vectors @xmath2 from underdetermined measurements@xcite . also , since the parameters in the output transfer function @xmath5 can model unknown nonlinearities , this problem formulation can be applied to the identification of linear - nonlinear cascade models of dynamical systems , in particular for neural spike responses @xcite . when the distributions @xmath6 and @xmath5 are known , or reasonably bounded , there are a number of methods available that can be used for the estimation of the unknown vector @xmath2 . in recent years , there has been significant interest in so - called approximate message passing ( amp ) and related methods based on gaussian approximations of loopy belief propagation ( lbp ) @xcite . these methods originate from cdma multiuser detection problems in @xcite , and have received considerable recent attention in the context of compressed sensing @xcite . see , also the survey article @xcite . the gaussian approximations used in amp are also closely related to standard expectation propagation techniques @xcite , but with additional simplifications that exploit the linear coupling between the variables @xmath2 and @xmath9 . the key benefits of amp methods are their computational simplicity , large domain of application , and , for certain large random @xmath10 , their exact asymptotic performance characterizations with testable conditions for optimality @xcite . this paper considers the so - called generalized amp ( gamp ) method of @xcite that extends the algorithm in @xcite to arbitrary output distributions @xmath5 ( many original formulations assumed additive white gaussian noise ( awgn ) measurements ) . however , although the current formulation of amp and gamp methods is attractive conceptually , in practice , one often does not know the prior and noise distributions exactly . to overcome this limitation , vila and schniter @xcite and krzakala _ et al . _ @xcite have recently proposed extension of amp and gamp based on expectation maximization ( em ) that enable joint learning of the parameters @xmath11 along with the estimation of the vector @xmath2 . while simulations indicate excellent performance , the analysis of these methods is difficult . this work provides a unifying analytic framework for such amp - based joint estimation and learning methods . the main contributions of this paper are as follows : * generalization of the gamp method of @xcite to a class of algorithms we call _ adaptive gamp _ that enables joint estimation of the parameters @xmath7 and @xmath8 along with vector @xmath2 . the methods are computationally fast and general with potentially large domain of application . in addition , the adaptive gamp methods include the em - gamp algorithms of @xcite as special cases . * exact characterization of the asymptotic behavior of adaptive gamp . we show that , similar to the analysis of the amp and gamp algorithms in @xcite , the componentwise asymptotic behavior of adaptive gamp can be described exactly by a simple scalar _ state evolution _ ( se ) equations . * demonstration of asymptotic consistency of adaptive gamp with maximum - likelihood ( ml ) parameter estimation . our main result shows that when the ml parameter estimation is computed exactly , the estimated parameters converge to the true values and the performance of adaptive gamp asymptotically coincides with the performance of the oracle gamp algorithm that knows the correct parameter values . remarkably , this result applies to essentially arbitrary parameterizations of the unknown distributions @xmath6 and @xmath5 , thus enabling provably consistent estimation on non - convex and nonlinear problems . * experimental evaluation of the algorithm for the problems of learning of sparse priors in compressed sensing and identification of linear - nonlinear cascade models in neural spiking processes . our simulations illustrate the performance gain of adaptive gamp and its asymptotic consistency . adaptive gamp thus provides a computationally - efficient method for a large class of joint estimation and learning problems with a simple , exact performance characterization and provable conditions for asymptotic consistency . as mentioned above , the adaptive gamp method proposed here can be seen as a generalization of the em methods in @xcite . in @xcite , the prior @xmath6 is described by a generic @xmath12-term gaussian mixture ( gm ) whose parameters are identified by an em procedure @xcite . the expectation " or e - step is performed by gamp , which can approximately determine the marginal posterior distributions of the components @xmath13 given the observations @xmath14 and the current parameter estimates of the gm distribution @xmath6 . a related em - gamp algorithm has also appeared in @xcite for the case of certain sparse priors and awgn outputs . simulations in @xcite show remarkably good performance and computational speed for em - gamp over a wide class of distributions , particularly in the context of compressed sensing . also , using arguments from statistical physics , @xcite presents state evolution ( se ) equations for the joint evolution of the parameters and vector estimates and confirms them numerically . as discussed in section [ sec : em - gamp ] , em - gamp is a special case of adaptive gamp with a particular choice of the adaptation functions . therefore , one contribution of this paper is to provide a rigorous theoretical justification of the em - gamp methodology . in particular , the current work provides a rigorous justification of the se analysis in @xcite along with extensions to more general input and output channels and adaptation methods . however , the methodology in @xcite in other ways is more general in that it can also study seeded " or spatially - coupled " matrices as proposed in @xcite . an interesting open question is whether the analysis methods in this paper can be extended to these scenarios as well . an alternate method for joint learning and estimation has been presented in @xcite , which assumes that the distributions on the source and output channels are themselves described by graphical models with the parameters @xmath7 and @xmath8 appearing as unknown variables . the method in @xcite , called hybrid - gamp , iteratively combines standard loopy bp with amp methods . one avenue of future work is to see if methodology in this paper can be applied to analyze the hybrid - gamp methods as well . finally , it should be pointed out that while simultaneous recovery of unknown parameters is appealing conceptually , it is not a strict requirement . an alternate solution to the problem is to assume that the signal belongs to a known class of distributions and to minimize the maximal mean - squared error ( mse ) for the class . this minimax approach @xcite was proposed for amp recovery of sparse signals in @xcite . although minimax approach results in the estimators that are uniformly good over the entire class of distributions , there may be a significant gap between the mse achieved by the minimax approach and the oracle algorithm that knows the distribution exactly . indeed , this gap was the main justification of the em - gamp methods in @xcite . due to its asymptotic consistency , adaptive gamp provably achieves the performance of the oracle algorithm . the paper is organized as follows : in section [ sec : gampreview ] , we review the non - adaptive gamp and corresponding state evolution equations . in section [ sec : gamp ] , we present adaptive gamp and describe ml parameter learning . in section [ sec : convprop ] , we provide the main theorems characterizing asymptotic performance of adaptive gamp and demonstrating its consistency . in section [ sec : numex ] , we provide numerical experiments demonstrating the applications of the method . section [ sec : concl ] concludes the paper . a conference version of this paper has appeared in @xcite . this paper contains all the proofs , more detailed descriptions and additional simulations . before describing the adaptive gamp algorithm , it is useful to review the basic ( non - adaptive ) gamp algorithm of @xcite . consider the estimation problem in fig . [ fig : model ] where the componentwise distributions on the inputs and outputs have some parametric form , [ eq : pxzlam ] p_x(x|_x ) , p_y|z(y|z , _ z ) , where @xmath15 and @xmath16 represent parameters of the distributions and @xmath17 and @xmath18 some parameter sets . the gamp algorithm of @xcite can be seen as a class of methods for estimating the vectors @xmath2 and @xmath9 for the case when the parameters @xmath7 and @xmath8 are _ known_. in contrast , the adaptive gamp method that is discussed in section [ sec : gamp ] enables _ joint _ estimation of the parameters @xmath7 and @xmath8 along with the vectors @xmath2 and @xmath9 . in order that to understand how the adaptation works , it is best to describe the basic gamp algorithm as a special case of the more general adaptive gamp procedure . the basic gamp algorithm corresponds to the special case of algorithm [ algo : gamp ] when the _ adaptation functions _ @xmath19 and @xmath20 output fixed values [ eq : hfix ] h^t_z(^t,,^t_p ) = ^t_z , h^t_x(^t,^t_r ) = ^t_x , for some _ pre - computed _ sequence of parameters @xmath21 and @xmath22 . by pre - computed " , we mean that the values do not depend on the data through the vectors @xmath23 , @xmath24 , and @xmath25 . in the oracle scenario @xmath21 and @xmath22 are set to the true values of the parameters and do not change with the iteration number @xmath26 . the _ estimation functions _ @xmath27 , @xmath28 and @xmath29 determine the estimates for the vectors @xmath2 and @xmath9 , given the parameter values @xmath30 and @xmath31 . as described in @xcite , there are two important sets of choices for the estimation functions , resulting in two variants of gamp : * _ sum - product gamp : _ in this case , the estimation functions are selected so that gamp provides a gaussian approximation of sum - product loopy bp . the estimates @xmath32 and @xmath33 then represent approximations of the mmse estimates of the vectors @xmath2 and @xmath9 . * _ max - sum gamp : _ in this case , the estimation functions are selected so that gamp provides a quadratic approximation of max - sum loopy bp and @xmath32 and @xmath33 represent approximations of the map estimates . the estimation functions of the sum - product gamp are equivalent to scalar mmse estimation problems for the components of the vectors @xmath2 and @xmath9 observed in gaussian noise . for max - sum gamp , the estimation functions correspond to scalar map problems . thus , for both versions , the gamp method reduces the vector - valued mmse and map estimation problems to a sequence of scalar awgn problems combined with linear transforms by @xmath10 and @xmath34 . gamp is thus computationally simple , with each iteration involving only scalar nonlinear operations followed by linear transforms . the operations are similar in form to separable and proximal minimization methods widely used for such problems @xcite . appendix [ sec : gampdetails ] reviews the equations for the sum - product gamp . more details , as well as the equations for max - sum gamp can be found in @xcite . in addition to its computational simplicity and generality , a key motivation of the gamp algorithm is that its asymptotic behavior can be precisely characterized when @xmath10 is a large i.i.d . gaussian transform . the asymptotic behavior is described by what is known as a _ state evolution _ ( se ) analysis . by now , there are a large number of se results for amp - related algorithms @xcite . here , we review the particular se analysis from @xcite which is based on the framework in @xcite . [ as : gamp ] consider a sequence of random realizations of the gamp algorithm , indexed by the dimension @xmath35 , satisfying the following assumptions : * [ as : gauss ] for each @xmath35 , the matrix @xmath36 has i.i.d . components with @xmath37 and the dimension @xmath38 is a deterministic function of @xmath35 satisfying @xmath39 for some @xmath40 as @xmath41 . * [ as : px0 ] the input vectors @xmath2 and initial condition @xmath42 are deterministic sequences whose components converge empirically with bounded moments of order @xmath43 as [ eq : thetaxinit ] _ n ( , ^0 ) ( x,^0 ) , to some random vector @xmath44 for some @xmath45 . for the precise definition of this form of convergence . * [ as : pyz ] the output vectors @xmath9 and @xmath1 are generated by [ eq : zax ] = , y_i = h(z_i , w_i ) i=1, ,m , for some scalar function @xmath46 and vector @xmath47 representing an output disturbance . it is assumed that the output disturbance vector @xmath48 is deterministic , but empirically converges as [ eq : wlim ] _ n w , where @xmath43 is as in assumption [ as : gamp ] ( b ) and @xmath49 is some random variable . we let @xmath5 denote the conditional distribution of the random variable @xmath50 . * the estimation function @xmath51 and its derivative with respect to @xmath52 , is lipschitz continuous in @xmath52 at @xmath53 , where @xmath54 is a deterministic parameter from the se equations below . a similar assumptions holds for @xmath55 . * the adaptation functions @xmath19 and @xmath20 are set to for some deterministic sequence of parameters @xmath56 and @xmath57 . also , in the estimation steps in lines [ line : zhat ] , [ line : shat ] and [ line : xhat ] of algorithm [ algo : gamp ] , the values of the @xmath58 and @xmath59 are replaced with the deterministic parameters @xmath60 and @xmath54 from the se equations defined below . assumption [ algo : gamp](a ) simply states that we are considering large , gaussian i.i.d . matrices @xmath10 . assumptions ( b ) and ( c ) state that the input vector @xmath2 and output disturbance @xmath48 are modeled as deterministic , but whose empirical distributions asymptotically appear as i.i.d . this deterministic model is one of key features of bayati and montanari s analysis in @xcite . assumption ( d ) is a mild continuity condition . assumption ( e ) defines the restriction of adaptive gamp to the non - adaptive gamp algorithm . we will remove this final assumption later . note that , for now , there is no assumption that the true " distribution of @xmath61 or the true conditional distribution of @xmath62 given @xmath63 must belong to the class of distributions for any parameters @xmath7 and @xmath8 . the analysis can thus model the effects of model mismatch . now , given the above assumptions , define the sets of vectors [ eq : thetaxz ] _ x^t & : = & \{(x_j , r_j^t,_j^1 ) , j=1, ,n } , + _ z^t & : = & \{(z_i,_i^t , y_i , p_i^t ) , i=1, ,m } . the first vector set , @xmath64 , represents the components of the the true , " but unknown , input vector @xmath2 , its gamp estimate @xmath32 as well as @xmath25 . the second vector , @xmath65 , contains the components the true , " but unknown , output vector @xmath9 , its gamp estimate @xmath33 , as well as @xmath23 and the observed output @xmath14 . the sets @xmath64 and @xmath65 are implicitly functions of the dimension @xmath35 . the main result of @xcite shows that if we fix the iteration @xmath26 , and let @xmath66 , the asymptotic joint empirical distribution of the components of these two sets @xmath64 and @xmath65 converges to random vectors of the form [ eq : thetabarxz ] _ x^t : = ( x , r^t,^1 ) , _ z^t : = ( z,^t , y , p^t ) . we precisely state the nature of convergence momentarily ( see theorem [ thm : stateevogamp ] ) . in , @xmath61 is the random variable in assumption [ as : gamp](b ) , while @xmath67 and @xmath68 are given by [ eq : rvxt ] @xmath69 for some deterministic constants @xmath70 , @xmath71 , and @xmath54 that are defined below . similarly , @xmath72 for some covariance matrix @xmath73 , and [ eq : yzpt ] y = h(z , w ) , ^t = g_z^t(p^t , y,_p^t,_z^t ) , where @xmath49 is the random variable in and @xmath73 contains deterministic constants . the deterministic constants @xmath70 , @xmath71 , @xmath54 and @xmath73 represent parameters of the distributions of @xmath74 and @xmath75 and depend on both the distributions of the input and outputs as well as the choice of the estimation and adaptation functions . the se equations provide a simple method for recursively computing these parameters . the equations are best described algorithmically as shown in algorithm [ algo : se ] . in order that we do not repeat ourselves , in algorithm [ algo : se ] , we have written the se equations for adaptive gamp : for non - adaptive gamp , the updates and can be ignored as the values of @xmath57 and @xmath56 are pre - computed . with these definitions , we can state the main result from @xcite . [ thm : stateevogamp ] consider the random vectors @xmath64 and @xmath65 generated by the outputs of gamp under assumption [ as : gamp ] . let @xmath74 and @xmath75 be the random vectors in with the parameters determined by the se equations in algorithm [ algo : se ] . then , for any fixed @xmath26 , the components of @xmath64 and @xmath65 converge empirically with bounded moments of order @xmath76 as [ eq : thetalimgamp ] _ n _ x^t _ x^t , _ z^t _ z^t . where @xmath74 and @xmath75 are given in . in addition , for any @xmath26 , the limits [ eq : taulimgamp ] _ n _ r^t = _ r^t , _ p^t = _ p^t , also hold almost surely . the theorem shows that the behavior of any component of the vectors @xmath2 and @xmath9 and their gamp estimates @xmath32 and @xmath33 are distributed identically to a simple scalar equivalent system with random variables @xmath61 , @xmath63 , @xmath77 and @xmath78 . this scalar equivalent model appears in several analyses and can be thought of as a _ single - letter characterization _ @xcite of the system . remarkably , this limiting property holds for essentially arbitrary distributions and estimation functions , even ones that arise from problems that are highly nonlinear or noncovex . from the single - letter characterization , one can compute the asymptotic value of essentially any componentwise performance metric such as mean - squared error or detection accuracy . similar single - letter characterizations can also be derived by arguments from statistical physics @xcite . [ algo : gamp ] [ line : taup ] [ line : phat ] [ line : lamz ] [ line : zhat ] [ line : shat ] [ line : taus ] [ line : taur ] [ line : rhat ] [ line : lamx ] [ line : xhat ] [ line : taux ] as described in the previous section , the standard gamp algorithm of @xcite considers the case when the parameters @xmath7 and @xmath8 in the distributions in are known . the adaptive gamp method proposed in this paper , and shown in algorithm [ algo : gamp ] , is an extension of the standard gamp procedure that enables simultaneous identification of the parameters @xmath7 and @xmath8 along with estimation of the vectors @xmath2 and @xmath9 . the key modification is the introduction of the two _ adaptation functions _ : @xmath79 and @xmath80 . in each iteration , these functions output estimates , @xmath81 and @xmath82 of the parameters based on the data @xmath23 , @xmath14 , @xmath25 , @xmath58 and @xmath59 . we saw the standard gamp method corresponds to the adaptation functions in which outputs fixed values @xmath57 and @xmath56 that do not depend on the data , and can be used when the true parameters are known . for the case when the true parameters are not known , we will see that a simple maximum likelihood ( ml ) can be used to estimate the parameters from the data . to understand how to estimate parameters via the adaptation functions , observe that from theorem [ thm : stateevogamp ] , we know that the distribution of the components of @xmath25 are distributed identically to the scalar @xmath67 in . now , the distribution of @xmath67 only depends on three parameters @xmath70 , @xmath71 and @xmath7 . it is thus natural to attempt to estimate these parameters from the empirical distribution of the components of @xmath25 . to this end , let @xmath83 be the log likelihood [ eq : phixll ] _ x(r,_x,_r,_r ) = p_r(r|_x,_r,_r ) , where the right - hand side is the probability density of a random variable @xmath84 with distribution @xmath85 then , at any iteration @xmath26 , we can attempt to perform a maximum - likelihood ( ml ) estimate + & & = _ _ x _ x _ ( _ r,_r ) s_x(_r^t ) \ { _ j=1^n _ x(r^t_j,_x,_r,_r)}. [ eq : hxml ] here , the set @xmath86 is a set of possible values for the parameters @xmath87 . the set may depend on the measured variance @xmath59 . we will see the precise role of this set below . similarly , the joint distribution of the components of @xmath23 and @xmath14 are distributed according to the scalar @xmath88 which depend only on the parameters @xmath89 and @xmath8 . thus , we can define the likelihood @xmath90 where the right - hand side is the joint probability density of @xmath91 with distribution @xmath92 then , we can attempt to estimate @xmath8 via the ml estimate @xmath93 again , the set @xmath94 is a set of possible covariance matrices @xmath89 . it is useful to briefly compare the above ml parameter estimation with the em - gamp method proposed by vila and schniter in @xcite and krzakala _ et . _ in @xcite . both of these methods combine the bayesian amp @xcite or gamp algorithms @xcite with a standard em procedure @xcite as follows . first , the algorithms use the sum - product version of the amp / gamp algorithms , so that the outputs can provide an estimate of the posterior distributions on the components of @xmath2 given the current parameter estimates . specifically , at any iteration @xmath26 , define the distribution + & = & p_x(x_j|_x^1 ) . [ eq : phatpost ] for the sum - product amp or gamp algorithms , it is shown in @xcite that the se equations simplify so that @xmath95 and @xmath96 , if the parameters were selected correctly . therefore , from theorem [ thm : stateevogamp ] , the conditional distribution @xmath97 should approximately match the distribution for large @xmath35 . if , in addition , we treat @xmath98 and @xmath59 as sufficient statistics for estimating @xmath13 given @xmath14 and @xmath10 , then @xmath99 can be treated as an approximation for the posterior distribution of @xmath13 given the current parameter estimate @xmath100 . some justification for this last step can be found in @xcite . using the approximation , we can approximately implement the em procedure to update the parameter estimate via a maximization + & : = & _ _ x _ x _ j=1^n , [ eq : lamem ] where the expectation is with respect to the distribution in . in @xcite , the parameter update is performed only once every few iterations to allow @xmath101 to converge to the approximation of the posterior distribution of @xmath13 given the current parameter estimates . in @xcite , the parameter estimate is updated every iteration . a similar procedure can be performed for the estimation of @xmath8 . we thus see that the em - gamp procedures in @xcite and in @xcite are both special cases of the adaptive gamp algorithm in algorithm [ algo : gamp ] with particular choices of the adaptation functions @xmath102 and @xmath103 . as a result , our analysis in theorem [ thm : stateevo ] below can be applied to these algorithms to provide rigorous asymptotic characterizations of the em - gamp performance . however , at the current time , we can only prove the asymptotic consistency result , theorem [ thm : consistent ] , for the ml adaptation functions and described above . that being said , it should be pointed out that em - gamp update is generally computationally much simpler than the ml updates in and . for example , when @xmath104 is an exponential family , the optimization in is convex . also , the optimizations in and require searches over additional parameters such as @xmath105 and @xmath106 . thus , an interesting avenue of future work is to apply the analysis result , theorem [ thm : consistent ] below , to see if the em - gamp method or some similarly computationally simple technique can be developed which also provides asymptotic consistency . given the distributions in assumption [ as : gamp ] , compute the sequence of parameters as follows : * _ initialization _ set @xmath107 with [ eq : seinit ] _ x^0 = ( x , ^0 ) , _ x^0 = _ x^0 , where the expectation is over the random variables @xmath108 in assumption [ as : gamp](b ) and @xmath109 is the initial value in the gamp algorithm . * _ output node update : _ compute the variables associated with the output nodes + [ eq : outse ] compute the variables _ p^t & = & _ x^t , _ p^t = _ x^t , + _ z^t & = & h_z^t(p^t , y , _ p^t ) , [ eq : lambarzse ] + _ r^t & = & -^-1 , [ eq : taubarrse ] + _ r^t & = & ( _ r^t)^2 , + _ r^t & = & _ r^t , [ eq : alphase ] + where the expectations are over the random variables @xmath72 and @xmath62 is given in . * _ input node update : _ compute + [ eq : inse ] _ x^t & = & h_x^t(r^t,_r^t ) , [ eq : lambarxse ] + _ x^1 & = & _ r^t , + ^1_x & = & ( x,^1 ) , + where the expectations are over the random variables in . before proving the asymptotic consistency of the adaptive gamp method with ml adaptation , we first prove the following more general convergence result . [ as : agamp ] consider the adaptive gamp algorithm running on a sequence of problems indexed by the dimension @xmath35 , satisfying the following assumptions : * same as assumption [ as : gamp](a ) to ( c ) with @xmath110 . * for every @xmath26 , * [ thm : stateevo ] consider the random vectors @xmath64 and @xmath65 generated by the outputs of the adaptive gamp under assumption [ as : agamp ] . let @xmath74 and @xmath75 be the random vectors in with the parameters determined by the se equations in algorithm [ algo : se ] . then , for any fixed @xmath26 , the components of @xmath64 and @xmath65 converge empirically with bounded moments of order @xmath110 as [ eq : thetalim ] _ n _ x^t _ x^t , _ z^t _ z^t . where @xmath74 and @xmath75 are given in . in addition , for any @xmath26 , the limits [ eq : taulamlim ] @xmath111 also hold almost surely . the result is a natural generalization of theorem [ thm : stateevogamp ] and provides a simple extension of the se analysis to incorporate the adaptation . the se analysis applies to essentially arbitrary adaptation functions . it particular , it can be used to analyze both the behavior of the adaptive gamp algorithm with either ml and em - gamp adaptation functions in the previous section . the proof is straightforward and is based on a continuity argument also used in @xcite . we cam now use theorem [ thm : stateevo ] to prove the asymptotic consistency of the adaptive gamp method with the ml parameter estimation described in section [ sec : mladapt ] . the following two assumptions can be regarded as _ identifiability _ conditions . [ def : pxident ] consider a family of distributions , @xmath112 , a set @xmath113 of parameters @xmath114 of a gaussian channel and function @xmath83 . we say that @xmath104 is _ identifiable with gaussian outputs _ with parameter set @xmath113 and function @xmath115 if : * the sets @xmath113 and @xmath17 are compact . * for any true " parameters @xmath116 , and @xmath117 , the maximization + & & , [ eq : lamxscore ] is well - defined , unique and returns the true value , @xmath118 . the expectation in is with respect to @xmath119 and @xmath120 . * for every @xmath7 and @xmath105 , @xmath106 , the function @xmath83 is pseudo - lipschitz continuous of order @xmath110 in @xmath52 . [ def : pzident ] consider a family of conditional distributions , @xmath121 generated by the mapping @xmath122 where @xmath123 is some random variable and @xmath124 is a scalar function . let @xmath125 be a set of covariance matrices @xmath89 and let @xmath126 be some function . we say that conditional distribution family @xmath127 is _ identifiable with gaussian inputs _ with covariance set @xmath125 and function @xmath128 if : * the parameter sets @xmath125 and @xmath18 are compact . * for any true " parameter @xmath129 and true covariance @xmath130 , the maximization + & & , [ eq : lamzscore ] is well - defined , unique and returns the true value , @xmath131 , the expectation in is with respect to @xmath132 and @xmath133 . * for every @xmath8 and @xmath89 , the function @xmath126 is pseudo - lipschitz continuous in @xmath134 of order @xmath110 . in addition , it is continuous in @xmath135 uniformly over @xmath136 and @xmath137 . definitions [ def : pxident ] and [ def : pzident ] essentially require that the parameters @xmath7 and @xmath8 can be identified through a maximization . the functions @xmath115 and @xmath128 can be the log likelihood functions and , although we permit other functions as well , since the maximization may be computationally simpler . such functions are sometimes called _ pseudo - likelihoods_. the existence of a such a function is a mild condition . indeed , if such a function does not exists , then the distributions on @xmath84 or @xmath138 must be the same for at least two different parameter values . in that case , one can not hope to identify the correct value from observations of the vectors @xmath25 or @xmath139 . [ as : agamp - ml ] let @xmath104 and @xmath140 be families of distributions and consider the adaptive gamp algorithm , algorithm [ algo : gamp ] , run on a sequence of problems , indexed by the dimension @xmath35 satisfying the following assumptions : * same as assumption [ as : gamp](a ) to ( c ) with @xmath110 . in addition , the distributions for the vector @xmath61 is given by @xmath141 for some true " parameter @xmath116 and the conditional distribution of @xmath62 given @xmath63 is given by @xmath142 for some true " parameter @xmath129 . * same as assumption [ as : agamp](c ) . * the adaptation functions are set to and . [ thm : consistent ] consider the outputs of the adaptive gamp algorithm with ml adaptation as described in assumption [ as : agamp - ml ] . then , for any fixed @xmath26 , * the components of @xmath64 and @xmath65 in converge empirically with bounded moments of order @xmath110 as in and the limits hold almost surely . * in addition , if @xmath143 , and the family of distributions @xmath144 , @xmath15 is identifiable in gaussian noise with parameter set @xmath86 and pseudo - likelihood @xmath115 ( see definition [ def : pxident ] ) , then [ eq : lamxcons ] _ n ^t_x = _ x^t = ^*_x almost surely . * similarly , if @xmath145 for some @xmath26 , and the family of distributions @xmath146 , @xmath16 is identifiable with gaussian inputs with parameter set @xmath94 and pseudo - likelihood @xmath128 ( see definition [ def : pzident ] ) then [ eq : lamzcons ] _ n ^t_z = _ z^t = ^*_z almost surely . _ proof : _ see appendix [ sec : consistentpf ] . the theorem shows , remarkably , that for a very large class of the parameterized distributions , the adaptive gamp algorithm is able to asymptotically estimate the correct parameters . moreover , there is asymptotically no performance loss between the adaptive gamp algorithm and a corresponding oracle gamp algorithm that knows the correct parameters in the sense that the empirical distributions of the algorithm outputs are described by the same se equations . there are two key requirements : first , that the optimizations in and can be computed . these optimizations may be non - convex . secondly , that the optimizations can be performed are over sufficiently large sets of gaussian channel parameters @xmath113 and @xmath125 such that it can be guaranteed that the se equations eventually enter these sets . in the examples below , we will see ways to reduce the search space of gaussian channel parameters . recent results suggest that there is considerable value in learning of priors @xmath6 in the context of compressed sensing @xcite , which considers the estimation of sparse vectors @xmath2 from underdetermined measurements ( @xmath147 ) . it is known that estimators such as lasso offer certain optimal min - max performance over a large class of sparse distributions @xcite . however , for many particular distributions , there is a potentially large performance gap between lasso and mmse estimator with the correct prior . this gap was the main motivation for @xcite which showed large gains of the em - gamp method due to its ability to learn the prior . here , we illustrate the performance and asymptotic consistency of adaptive gamp in a simple compressed sensing example . specifically , we consider the estimation of a sparse vector @xmath0 from @xmath148 noisy measurements @xmath149 where the additive noise @xmath48 is random with i.i.d . entries @xmath150 . here , the output " channel is determined by the statistics on @xmath48 , which are assumed to be known to the estimator . so , there are no unknown parameters @xmath8 . as a model for the sparse input vector @xmath2 , we assumed the components are i.i.d . with the gauss - bernoulli distribution , [ eq : gaussbern ] x_j ~\ { ll 0 & = 1- , + ( 0,_x^2 ) & = . where @xmath151 represents the probability that the component is non - zero ( i.e.the vector s sparsity ratio ) and @xmath152 is the variance of the non - zero components . the parameters @xmath153 are treated as unknown . in the adaptive gamp algorithm , we use the estimation functions @xmath154 , @xmath155 , and @xmath156 corresponding to the sum - product gamp algorithm . as described in appendix [ sec : gampdetails ] , for the sum - produce gamp the se equations simplify so that @xmath157 and @xmath96 . since the noise variance is known , the initial output noise variance @xmath158 obtained by adaptive gamp in algorithm [ algo : gamp ] exactly matches that of oracle gamp . therefore , for @xmath159 , the parameters @xmath70 and @xmath71 do not need to be estimated , and conveniently simplifies to @xmath160 where @xmath161 \times [ 0 , + \infty)$ ] . for iteration @xmath162 , we rely on asymptotic consistency , and assume that the maximization yields the correct parameter estimates , so that @xmath163 . then , in principle , for @xmath162 adaptive gamp uses the correct parameter estimates and we expect it to match the performance of oracle gamp . in our implementation , we run em update until convergence to approximate the ml adaptation . fig . [ fig : mseresults ] illustrates the performance of adaptive gamp on signals of length @xmath164 generated with the parameters @xmath165 . the performance of adaptive gamp is compared to that of lasso with mse optimal regularization parameter , and oracle gamp that knows the parameters of the prior exactly . for generating the graphs , we performed @xmath166 random trials by forming the measurement matrix @xmath10 from i.i.d . zero - mean gaussian random variables of variance @xmath167 . in figure [ fig : mseresults](a ) , we keep the variance of the noise fixed to @xmath168 and plot the average mse of the reconstruction against the measurement ratio @xmath169 . in figure [ fig : mseresults](b ) , we keep the measurement ratio fixed to @xmath170 and plot the average mse of the reconstruction against the noise variance @xmath171 . for completeness , we also provide the asymptotic mse values computed via se recursion . the results illustrate that gamp significantly outperforms lasso over the whole range of @xmath169 and @xmath171 . moreover , the results corroborate the consistency of adaptive gamp which achieves nearly identical quality of reconstruction with oracle gamp . the performance results indicate that adaptive gamp can be an effective method for estimation when the parameters of the problem are difficult to characterize and must be estimated from data . as a second example , we consider the estimation of the linear - nonlinear - poisson ( lnp ) cascade model @xcite . the model has been successfully used to characterize neural spike responses in early sensory pathways of the visual system . in the context of lnp cascade model , the vector @xmath0 represents the linear filter , which models the linear receptive field of the neuron . amp techniques combined with the parameter estimation have been recently proposed for neural receptive field estimation and connectivity detection in @xcite . as in section [ sec : gaussbern ] , we model @xmath2 as a gauss - bernoulli vector of unknown parameters @xmath172 . to obtain the measurements @xmath14 , the vector @xmath4 is passed through a componentwise nonlinearity @xmath173 to result in u(z ) = . let the function @xmath174 denote a vector ( z ) = ( 1 , u(z ) , , u(z)^r-1)^t . then , the final measurement vector @xmath14 is generated by a measurement channel with a conditional density of the form [ eq : pyzexp ] p_y|z(y_i|z_i,_z ) = ^-f(z_i ) , where @xmath175 denotes the nonlinearity given by @xmath176 adaptive gamp can now be used to also estimate vector of polynomial coefficients @xmath8 , which together with @xmath2 , completely characterizes the lnp system . . this plots illustrates near consistency of adaptive gamp for large @xmath35.,width=321 ] the estimation of @xmath8 is performed with ml estimator described in section [ sec : mladapt ] . we assume that the mean and variance of the vector @xmath2 are known at iteration @xmath107 . this implies that for sum - product gamp the covariance @xmath177 is initially known and the optimization simplifies to @xmath178 where @xmath179 . the estimation of @xmath7 is performed as in section [ sec : gaussbern ] . as before , for iteration @xmath162 , we assume that the maximizations and yield correct parameter estimates @xmath163 and @xmath180 , respectively . thus we can conclude by induction that for @xmath162 the adaptive gamp algorithm should continue matching oracle gamp for large enough @xmath35 . in our simulations , we implemented with a gradient ascend algorithm and run it until convergence . in fig . [ fig : nlresults ] , we compare the reconstruction performance of adaptive gamp against the oracle version that knows the true parameters @xmath181 exactly . we consider the vector @xmath2 generated with true parameters @xmath182 . we consider the case @xmath183 and set the parameters of the output channel to @xmath184 $ ] . to illustrate the asymptotic consistency of the adaptive algorithm , we consider the signals of length @xmath185 and @xmath186 . we perform @xmath187 and @xmath188 random trials for long and short signals , respectively , and plot the average mse of the reconstruction against @xmath169 . as expected , for large @xmath35 , the performance of adaptive gamp is nearly identical ( within @xmath189 ) to that of oracle gamp . we have presented an adaptive gamp method for the estimation of i.i.d . vectors @xmath2 observed through a known linear transforms followed by an arbitrary , componentwise random transform . the procedure , which is a generalization of em - gamp methodology of @xcite that estimates both the vector @xmath2 as well as parameters in the source and componentwise output transform . in the case of large i.i.d . gaussian transforms , it is shown that the adaptive gamp method is provably asymptotically consistent in that the parameter estimates converge to the true values . this convergence result holds over a large class of models with essentially arbitrarily complex parameterizations . moreover , the algorithm is computationally efficient since it reduces the vector - valued estimation problem to a sequence of scalar estimation problems in gaussian noise . we believe that this method is applicable to a large class of linear - nonlinear models with provable guarantees can have applications in a wide range of problems . we have mentioned the use of the method for learning sparse priors in compressed sensing . future work will include learning of parameters of output functions as well as possible extensions to non - gaussian matrices . as described in @xcite , the sum - product estimation can be implemented with the estimation functions [ eq : gsp ] g_x^t(r,_r,_x ) & : = & , [ eq : gxsp ] + g_z^t(p , y,_p,_z ) & : = & , [ eq : gzsp ] + g_s^t(p , y,_p,_z ) & : = & ( g_z^t(p , y,_p,_z ) - p ) , where the expectations are with respect to the scalar random variables [ eq : gspdist ] @xmath190 the estimation functions correspond to scalar estimates of random variables in additive white gaussian noise ( awgn ) . a key result of @xcite is that , when the parameters are set to the true values ( i.e. @xmath191 ) , the outputs @xmath32 and @xmath33 can be interpreted as sum products estimates of the conditional expectations @xmath192 and @xmath193 . the algorithm thus reduces the vector - valued estimation problem to a computationally simple sequence of scalar awgn estimation problems along with linear transforms . moreover , the se equations in algorithm [ algo : se ] reduce to a particularly simple forms , where @xmath54 and @xmath71 in are given by @xmath194,\ ] ] where the expectations are over the random variables @xmath72 and @xmath62 is given in . the covariance matrix @xmath73 has the form @xmath195 \beta\tau_{x0 } - \taubar_p^t & \beta\tau_{x0 } - \taubar_p^t \\[0.3em ] \end{bmatrix},\ ] ] where @xmath196 is the variance of @xmath61 and @xmath40 is the asymptotic measurement ratio ( see assumption 1 for details ) . the scaling constant becomes @xmath157 . the update rule for @xmath197 also simplifies to @xmath198,\ ] ] where the expectation is over the random variables in . bayati and montanari s analysis in @xcite employs certain deterministic models on the vectors and then proves convergence properties of related empirical distributions . to apply the same analysis here , we need to review some of their definitions . we say a function @xmath199 is _ pseudo - lipschitz _ of order @xmath200 , if there exists an @xmath201 such for any @xmath2 , @xmath202 , @xmath203 now suppose that for each @xmath204 , @xmath205 is a set of vectors [ eq : vsetapp ] ^(n ) = \{_i(n ) , i=1, ,(n ) } , where each element @xmath206 and @xmath207 is the number of elements in the set . thus , @xmath205 can itself be regarded as a vector with @xmath208 components . we say that @xmath205 _ empirically converges with bounded moments of order @xmath76 _ as @xmath209 to a random vector @xmath210 on @xmath211 if : for all pseudo - lipschitz continuous functions , @xmath212 , of order @xmath76 , @xmath213 when the nature of convergence is clear , we may write ( with some abuse of notation ) @xmath214 or @xmath215 finally , let @xmath216 be the set of probability distributions on @xmath211 with bounded @xmath76th moments , and suppose that @xmath217 is a functional @xmath216 to some topological space @xmath218 . given a set @xmath205 as in , write @xmath219 for @xmath220 where @xmath221 is the empirical distribution on the components of @xmath222 . also , given a random vector @xmath210 with distribution @xmath223 write @xmath224 for @xmath225 . then , we will say that the functional @xmath226 is _ weakly pseudo - lipschitz continuous _ of order @xmath76 if @xmath227 where the limit on the right hand side is in the topology of @xmath218 . the proof follows along the adaptation argument of @xcite . we use the tilde superscript on quantities such as @xmath228 , and @xmath229 to denote values generated via a non - adaptive version of the gamp . the non - adaptive gamp algorithm has the same initial conditions as the adaptive algorithm ( i.e. @xmath230 ) , but with @xmath82 and @xmath81 replaced by their deterministic limits @xmath231 and @xmath232 , respectively . that is , we replace lines [ line : zhat ] , [ line : shat ] and [ line : xhat ] with _ i^t & = & g_z^t(p_i^t , y_i,_p^t,_z^t ) , _ i^t = g_s^t(p_i^t , y_i,_p^t,_z^t ) , + _ j^1 & = & g_x^t(r_j^t , _ r^t,_x^t ) . this non - adaptive algorithm is precisely the standard gamp method analyzed in @xcite . the results in that paper show that the outputs of the non - adaptive algorithm satisfy all the required limits from the se analysis . that is , @xmath233 where @xmath234 and @xmath235 are the sets generated by the non - adaptive gamp algorithm : @xmath236 the limits are now proven through a continuity argument that shows that the adaptive and non - adaptive quantities must asymptotically agree with one another . specifically , we will start by proving that the following limits holds almost surely for all @xmath237 the proof of the limits and is achieved by an induction on @xmath26 . although we only need to show the above limits for @xmath110 , most of the arguments hold for arbitrary @xmath45 . we thus present the general derivation where possible . to begin the induction argument , first note that the non - adaptive algorithm has the same initial conditions as the adaptive algorithm . thus the limits and hold for @xmath159 and @xmath240 , respectively . we now proceed by induction . suppose that @xmath241 and the limits and hold for some @xmath26 and @xmath242 , respectively . since @xmath10 has i.i.d . components with zero mean and variance @xmath167 , it follows from the marenko - pastur theorem @xcite that that its @xmath243-norm operator norm is bounded . that is , there exists a constant @xmath244 such that [ eq : matrixbnd ] _ n _ k c_a , _ n ^_k c_a . this bound is the only part of the proof that specifically requires @xmath110 . from , we obtain @xmath245 where ( a ) is due to the norm inequality @xmath246 . since @xmath247 , we have that for any positive numbers @xmath248 and @xmath249 [ eq : pbndtwo ] ( a + b)^k 2^k(a^k + b^k ) . applying the inequality into , we obtain @xmath250 now , since @xmath251 and @xmath252 are the outputs of the non - adaptive algorithm they satisfy the limits to prove , we first prove the empirical convergence of @xmath257 to @xmath88 . towards this end , let @xmath258 be any pseudo - lipschitz continuous function @xmath212 of order @xmath76 . then @xmath259\right| } \nonumber \\ & \leq \frac{1}{m } \sum_{i = 1}^m \left|\phi(p_i^{t},y_i ) - \phi(\tilde{p}_i^{t},y_i ) \right| \nonumber\\ & + \left| \frac{1}{m } \sum_{i = 1}^m \phi(\tilde{p}_i^{t},y_i ) -\exp\left[\phi(p^t , y)\right]\right| \nonumber \\ & { { \overset{\mathrm{(a)}}{\leq } } } \frac{l}{m } \sum_{i = 1}^m \left(1 + |p_i^t|^{k-1 } + |\tilde{p}_i^t|^{k-1 } + |y_i|^{k-1 } \right)|p_i^t - \tilde{p}_i^t| \nonumber \\ & + \left| \frac{1}{m } \sum_{i = 1}^m \phi(\tilde{p}_i^{t},y_i ) -\exp\left[\phi(p^t , y)\right]\right| \nonumber \\ & { { \overset{\mathrm{(b)}}{\leq } } } lc \delta_p^t+\left| \frac{1}{m } \sum_{i = 1}^m \phi(\tilde{p}_i^{t},y_i ) -\exp\left[\phi(p^t , y)\right]\right| . \label{eq : phipybnd}\end{aligned}\ ] ] in ( a ) we use the fact that @xmath212 is pseudo - lipschitz , and in ( b ) we use hlder s inequality @xmath260 with @xmath261 and define @xmath262 as @xmath263^{k/(k-1 ) } } \nonumber \\ & \leq \frac{1}{m}\sum_{i=1}^m \left ( 1 + |p_i^t|^{k-1 } + |\tilde{p}_i^t|^{k-1 } + |y_i|^{k-1}\right)^{k/(k-1 ) } \nonumber \\ & \leq \mbox{const}\x\left [ 1 + \left(\frac{1}{m}\left\|{\mathbf{p}}^t\right\|_k^k\right)^{\frac{k-1}{k } } \right . \nonumber\\ & \left.+ \left(\frac{1}{m}\left\|\porc^t\right\|_k^k\right)^{\frac{k-1}{k } } + \left(\frac{1}{m}\left\|{\mathbf{y}}\right\|_k^k\right)^{\frac{k-1}{k } } \right ] , \label{eq : cpbnddef}\end{aligned}\ ] ] where the first step is from jensen s inequality . since @xmath264 satisfy the limits for the non - adaptive algorithm we have : [ eq : ptildeybnd ] @xmath265 < \infty \\ \lim_{n \rightarrow \infty } \frac{1}{m}\|{\mathbf{y}}\|_k^k & = \lim_{n \rightarrow \infty } \frac{1}{m}\sum_{i = 1}^m |y_i|^k = \exp\left[|y|^k\right ] < \infty\end{aligned}\ ] ] also , from the induction hypothesis , it follows that the adaptive output must satisfy the same limit @xmath266 < \infty.\ ] ] combining , , , , we conclude that for all @xmath237 [ eq : pyconv ] _ n ( ^t , ) ( p^t , y ) . the limit along with and the continuity condition on @xmath20 in assumption [ as : gamp](d ) prove the limit in . the limit together with continuity conditions on @xmath28 in assumptions [ as : gamp ] show that , and hold for @xmath26 . for example , to show , we consider the limit @xmath267 of the following expression @xmath268 where at ( a ) we used the lipschitz continuity assumption . similar arguments can be used for and . to show , we proceed exactly as for . due to the continuity assumptions on @xmath269 , this limit in turn shows that holds almost surely . then , and follow directly from the continuity of @xmath154 in assumptions [ as : gamp ] , together with and . we have thus shown that if the limits and hold for some @xmath26 , they hold for @xmath270 . thus , by induction they hold for all @xmath26 . finally , to show , let @xmath212 be any pseudo - lipschitz continuous function @xmath271 , and define @xmath272\right|,\end{aligned}\ ] ] which , due to convergence of non - adaptive gamp , can be made arbitrarily small by choosing @xmath35 large enough . then , consider @xmath273\right| \nonumber\\ & \leq \epsilon_n^t + \frac{1}{n } \sum_{j = 1}^n \left|\phi(x_j , \hat{r}_j^{t } , \xhat_j^{\tp1 } ) - \phi(x_j , \tilde{r}_j^{t } , \tilde{x}_j^{\tp1 } ) \right| \nonumber\\ & { { \overset{\mathrm{(a)}}{\leq } } } \epsilon_n^t + l\|{\mathbf{r}}^t - \rorc^t\|_1 + l\|\hat{{\mathbf{x}}}^{\tp1 } - \xorc^{\tp1}\|_1 \nonumber\\ & + \frac{l^\prime}{n}\sum_{j = 1}^n \left(|\hat{r}_j^t|^{k-1 } + |\tilde{r}_j^t|^{k-1 } \right)(|\hat{r}_j^t - \tilde{r}_j^t| + |\xhat_j^{\tp1 } - \tilde{x}_j^{\tp1}| ) \nonumber\\ & + \frac{l^\prime}{n}\sum_{j = 1}^n \left(|\xhat_j^{\tp1}|^{k-1 } + |\tilde{x}_j^{\tp1}|^{k-1}\right)(|\hat{r}_j^t - \tilde{r}_j^t| + |\xhat_j^{\tp1 } - \tilde{x}_j^{\tp1}| ) \nonumber\\ & { { \overset{\mathrm{(b)}}{\leq } } } \epsilon_n^t + l\left(\delta_r^t\right)^{\frac{1}{k } } + l\left(\delta_x^t\right)^{\frac{1}{k}}\nonumber\\ & + l^{\prime}\left(\delta_r^t\right)^{\frac{1}{k}}\left((\tilde{m}_x^{\tp1})^{\frac{k-1}{k}}+(\hat{m}_x^{\tp1})^{\frac{k-1}{k } } + ( \tilde{m}_r^{t})^{\frac{k-1}{k}}+(\hat{m}_r^{t})^{\frac{k-1}{k}}\right)\nonumber\\ & + l^{\prime}\left(\delta_x^t\right)^{\frac{1}{k } } \left((\tilde{m}_x^{\tp1})^{\frac{k-1}{k}}+(\hat{m}_x^{\tp1})^{\frac{k-1}{k } } + ( \tilde{m}_r^{t})^{\frac{k-1}{k}}+(\hat{m}_r^{t})^{\frac{k-1}{k}}\right ) \label{eq : xbound } \ ] ] where @xmath12 , @xmath274 are constants independent of @xmath35 and @xmath275 in ( a ) we use the fact that @xmath212 is pseudo - lipshitz , in ( b ) we use @xmath276-norm equivalence @xmath277 and hlder s inequality @xmath260 with @xmath261 . by applying of , and since , @xmath278 , @xmath279 , @xmath280 , and @xmath281 converge to a finite value we can obtain the first equation of by taking @xmath41 . the second equation in can be shown in a similar way . this proves the limits . also , the first two limits in are a consequence of and . the second two limits follow from continuity assumptions in assumption [ as : gamp](e ) and the convergence of the empirical distributions in . this completes the proof . part ( a ) of theorem [ thm : consistent ] is a direct application of the general result , theorem [ sec : stateevo ] . to apply the general result , first observe that assumptions [ as : agamp - ml](a ) and ( c ) immediately imply the corresponding items in assumptions [ as : agamp ] . so , we only need to verify the continuity condition in assumption [ as : agamp](b ) for the adaptation functions in and . we begin by proving the continuity of @xmath20 . fix @xmath26 , and let @xmath282 be a sequence of vectors and @xmath283 be a sequence of scalars such that [ eq : yptaulim ] _ n ( ^(n),^(n ) ) ( y , p^t ) _ n _ p^(n ) = ^t_p , where @xmath284 and @xmath285 are the outputs of the state evolution equations . for each @xmath35 , let [ eq : lamhatcons ] _ z^(n ) = h_z^t(^(n),^(n),_p^(n ) ) . we wish to show that @xmath286 , the true parameter . since @xmath287 and @xmath18 is compact , it suffices to show that , any limit point of any convergent subsequence is equal to @xmath288 . so , suppose that @xmath289 to some limit point @xmath290 on some subsequence @xmath291 . from @xmath291 and the definition it follows that @xmath292 now , since @xmath293 and @xmath289 , we can apply the continuity condition in definition [ def : pzident](c ) to obtain @xmath294 \geq 0.\end{aligned}\ ] ] also , the limit and the fact that @xmath128 is psuedo - lipschitz continuous of order @xmath76 implies that [ eq : phizcompc ] . but , property ( b ) of definition [ def : pzident ] shows that @xmath288 is the maxima of the right - hand side , so [ eq : phizcompd ] = . since , by definition [ def : pzident](b ) , the maxima is unique , @xmath295 . since this limit point is the same for all convergent subsequences , we see that @xmath296 over the entire sequence . we have thus shown that given limits , the outputs of the adaptation function converge as @xmath297 thus , the continuity condition on @xmath103 in assumption [ as : agamp](b ) is satisfied . the analogous continuity condition on @xmath102 can be proven in a similar manner . so , it remains to show parts ( b ) and ( c ) of theorem [ thm : consistent ] . we will only prove ( b ) ; the proof of ( c ) is similar . also , since we have already established , we only need to show that the output of the se equations matches the true parameter . that is , we need to show @xmath298 . this fact follows immediately from the selection of the adaptation functions : + & & _ _ x _ x _ ( _ r,_r ) s_x(_r^t ) + & & _ _ x _ x _ ( _ r,_r ) s_x(_r^t ) + & & + & & _ x^ * [ eq : lamxlim ] where ( a ) follows from the se equation ; ( b ) is the definition of the ml adaptation function @xmath299 when interpreted as a functional on a random variable @xmath67 ; ( c ) is the definition of the random variable @xmath67 in where @xmath300 ; and ( d ) follows from definition [ def : pxident](b ) and the hypothesis that @xmath301 . thus , we have proven that @xmath298 , and this completes the proof of part ( b ) of theorem [ thm : consistent ] . the proof of part ( c ) is similar .

a thermal activation model to describe the plasticity of bulk metallic glasses ( derlet and maa , phil . mag . 2013 , doi : 10.1080/14786435.2013.826396 ) which uses a distribution of barrier energies and some aspects of under - cooled liquid physics is developed further . in particular , a log - normal distribution is now employed to describe the statistics of barrier energies . a high temperature mean - field description of homogeneous macro - plasticity is then developed and is shown to be similar to a thermal activation picture employing a single characteristic activation energy and activation volume . in making this comparison , the activation volume is interpreted as being proportional to the average mean - square - value of the plastic shear strain magnitude within the material . also , the kinetic fragility at the glass transition temperature is shown to represent the effective number of irreversible structural transformations available at that temperature .

1311.3818

the deformation properties of bulk metallic glasses are characterised by two temperature regimes @xcite . at high temperatures , close to the glass transition temperature , bmgs deform homogeneously with the strain rate properties being reasonably well understood by thermally activated plastic flow . at lower temperatures , plasticity becomes heterogeneous and highly localized to shear - bands , resulting in limited ductility in tension and a few percent plastic strain prior to failure for some bmgs in compression . in this regime of temperatures both athermal @xcite and thermal @xcite theories of plasticity have been proposed . in the early work of turnbull and cohen @xcite these different regimes emerged as a result of two competing processes , one diffusional and one stress induced . later , spaepen @xcite attributed the homogeneous regime as resulting from a steady state of free volume creation and diffusional related free volume annihilation . heterogeneous flow on the other hand was dominated by free volume creation since annihilation became diffusion limited at low temperatures . that two distinct temperature scales exist was also recognised by argon @xcite , who from the perspective of thermal activation developed two microscopic theories where at high temperature , plasticity was mediated by small atomic reconfigurations that resulted in a localized shear strain event , and at lower temperature , a less local structural excitation that was akin to the nucleation of a dislocation loop occurs . such mechanisms could give insight into the observed high temperature homogeneous plasticity and the low temperature heterogeneous plasticity . some time after this , high - strain rate atomistic simulation revealed the generality of such local structural excitations , where the transition from homogeneous to heterogeneous plasticity ( as a function of decreasing temperature or increasing strain rate ) emerged from the degree of correlation between such localized activity @xcite . due to their high strain rates and stresses , both dynamic and static atomistic simulations primarily probe the athermal region of plasticity . out of the earliest of this work by falk and langer @xcite arose the concept of the shear transformation zone ( stz ) which was assumed to exist in regions of the material not so well relaxed the so - called liquid like regions @xcite . the corresponding structural transformations were assumed to be activated _ athermally _ once a critical stress had been reached . such ideas developed into what is now called the effective temperature theories in which the quenched disorder well below the glass transition temperature is characterised by an effective temperature that describes the structural fluctuations within the material , see for example @xcite . this approach has been quite successful in developing a quantitative understanding of experimental low temperature deformation data . it was recognised by bouchbinder , langer and falk @xcite that if one were to assume a thermal activation picture within the stz framework @xcite , the density of such local structural excitations would have to be `` improbably '' large if the corresponding activation energies are in the realistic range of hundreds of kilojoule per mole ( or electron volts per excitation ) . in going to an athermal picture , this difficulty was avoided and indeed stzs could now occur at quite dilute densities @xcite , a picture that is compatible with the distinct liquid like regions assumed to exist within the glass . understanding the nature of such stzs , or more generally localized structural excitations , has also been influenced by the physics of the under - cooled liquid regime , in which two distinct relaxation time - scales are evidenced : that of the slow @xmath0-relaxation and the fast @xmath1-relaxation processes first introduced by goldstein @xcite . at temperatures near the glass transition and below , the @xmath0-relaxation processes rapidly increase their time - scale with the traditional viewpoint being that they freeze out of the structural dynamics in the amorphous solid regime @xcite . despite such an assumed freezing , the terminology of @xmath0- and @xmath1-relaxation has also been applied to the amorphous solid regime where the microscopic @xmath1 processes mediate escape from the mega - basins of the elastic energy landscape of the @xmath0 processes @xcite . the present work also exploits this terminology and uses an additional fundamental feature of the under - cooled liquid regime , that the number of structural transformations available will scale exponentially with respect to a fundamental volume measure @xcite . such ideas have been used in the early thermodynamic theories of adam and gibbs @xcite and kirkpatrick _ et al _ @xcite see also the review article by heuer @xcite and provides a sufficient number of structural transformations to justify a thermal activation picture for temperatures that are well below the glass transition . in particular the present work develops further the ideas of refs . @xcite . in these works , a thermal activation picture was developed in which the characteristic plastic rate with respect to an internal heterogeneous volume scale was determined from a distribution of @xmath0-relaxation barriers whose number scales exponentially with the size of the heterogeneous volume and whose statistics is described by a distribution with extensive first and second cumulants . by doing this , the average time scale associated with plasticity at a particular temperature becomes intimately connected to the @xmath0-relaxation potential energy landscape ( mediated by @xmath1 relaxation processes ) which is thermally accessible at that temperature . at zero loading or any given shear stress , two critical temperature scales emerge : 1 ) the plastic transition temperature which indicates a significant drop in the amount of plastic activity as the temperature is reduced , and 2 ) the ( lower ) kinetic freezing temperature where ( on average ) a single structural transformation becomes available per heterogeneous volume . the transition between the higher temperature regime of 1 ) and the lower temperature region of 2 ) corresponds to a change of statistics , from that of the most probable to that of the extreme . importantly , before the critical freezing temperature is reached , there will exist a low temperature regime ( below the glass transition ) where a sufficient number of thermally activated events exist to mediate a non - negligible macroscopic plasticity . thus , in ref . @xcite , the low temperature heterogeneous and high temperature homogeneous deformation regimes are distinguished by a transition in the nature of the statistics of the @xmath0-relaxation energy landscape . put in other words , the freezing out of @xmath0-relaxation processes is only complete at temperatures far below that of the glass transition temperature , and until this extreme regime is reached thermal activation remains the underlying phenomenon controlling macroscopic plasticity a fact that has been established experimentally for a variety of deformation geometries @xcite including most recently that of shear band nucleation , propagation and arrest @xcite . the present work extends on the work done in ref . @xcite ( which will be referred to as paper i ) by using a log - normal distribution to describe the statistics of the @xmath0-relaxation barrier energy landscape to develop a mean - field description of thermally activated plasticity for the high temperature / low strain rate regime of homogeneous deformation . [ secsummary ] gives a brief overview of the current theory , secs . [ seclnd]-[sectheory ] develops the theory for the positive valued log - normal distribution of barrier energies , sec . [ secmfp ] develops a mean field description of plasticity suitable for the high temperature homogeneous regime of deformation , and sec . [ secexp ] applies the theory to well known experimental data for the well studied bmg vitroley-1 . in the discussion and conclusion , sec . [ secconc ] , two results are summarized : 1 ) when comparing with the usual arrhenius description of high temperature homogeneous plasticity , the corresponding activation volume parameter is found to be equivalent to the variance of the internal slipped volume distribution , and 2 ) the fragility of the material is a measure of the number of structural transformations available at the glass transition temperature . paper i considered a characteristic time scale of irreversible structural transformation activity , @xmath2 , whose inverse is seen as a plastic transition rate below the glass transition temperature . studying the resulting temperature and stress dependence of @xmath3^{-1}$ ] gave insight into the transition from elasticity to micro - plasticity . to determine @xmath3^{-1}$ ] it was recognised that , at a large enough length scale , bmgs are considered structurally homogeneous indicating that a sufficient amount of self - averaging occurs with respect to a shorter heterogeneous length scale . thus two length - scales naturally emerge which are realised through the homogeneous and heterogeneous volume elements . the plastic rate for a homogeneous representative volume element ( rve ) of volume @xmath4 is @xmath5^{-1}(t)=\sum_{n=1}^{n'}\left[\tau_{\mathrm{p , n}}\right]^{-1}(t)=n'\times\left[\tau_{\mathrm{p}}\right]^{-1}(t)\ ] ] in which @xmath6^{-1}(t)=\frac{1}{n'}\sum_{n=1}^{n'}\left[\tau_{\mathrm{p},n}\right]^{-1}(t ) . \label{eqntaup}\ ] ] in the above @xmath7^{-1}$ ] is the particular plastic rate for the @xmath8th heterogeneous volume element and @xmath9 is the number of such heterogeneous volume elements leading to a well - converged self - averaged @xmath3^{-1}$ ] . for the current work , the characteristic volume of the heterogeneous volume is labelled as @xmath10 giving @xmath11 . under the assumption of thermally activated plasticity , the plastic rate associated with one particular heterogeneity volume is written as a linear sum of the @xmath12 thermally active transition rates available to that volume element : @xmath13^{-1}(t)=\sum_{i=1}^{m}\left[\tau_{\mathrm{p0},ni}\right]^{-1}(t)\exp\left(-\frac{e_{\mathrm{p0},ni}}{k_{\mathrm{b}}t}\right ) , \label{eqntaupn}\ ] ] where @xmath14^{-1}(t)$ ] and @xmath15 are the attempt rate and barrier energy for the @xmath16th irreversible structural transformation within the @xmath8th heterogeneous volume element . [ eqntaup ] then becomes @xmath6^{-1}(t)=\left\langle\left[\tau_{\mathrm{p0}}\right]^{-1}(t)\exp\left(-\frac{e_{\mathrm{p0}}}{k_{\mathrm{b}}t}\right)\right\rangle \label{eqnavetaup}\ ] ] in which the average is performed with respect to @xmath17^{-1}(t)$ ] and @xmath18 . in terms of the underlying potential energy landscape ( pel ) , the simple ( first order ) expression of eqn . [ eqntaupn ] is only valid for thermally activated processes that do not multiply recross their energy barrier . this naturally leads to a coarse graining of the pel , in which the barriers and ( now diffusive ) pre - factors entering into eqn . [ eqntaupn ] underlie the collective microscopic activity that results in a lasting escape from a characteristic energy valley . by analogy to the under - cooled liquid pel framework @xcite where two distinct time scales occur the `` slow '' @xmath0-relaxation and `` fast '' microscopic @xmath1-relaxation modes the self - averaging of eqn . [ eqnavetaup ] reduces ( see paper i ) to @xmath6^{-1}=\left[\tau_{\mathrm{p00}}\right]^{-1}\exp\left(-\frac{e_{\mathrm{p00}}}{k_{\mathrm{b}}t}\right)\times m\left\langle\exp\left(-\frac{e}{k_{\mathrm{b}}t}\right)\right\rangle . \label{eqnplasticrate}\ ] ] eqn . [ eqnplasticrate ] consists of a diffusive attempt rate with a simple arrhenius temperature dependence , representing the mediating @xmath1 mode relaxation dynamics multiplied by a thermal factor whose temperature dependence will be derived from the statistical properties of the @xmath0-relaxation mode coarse grained pel @xcite . in paper i the number of structural transformations available to each heterogeneous volume was assumed to scale exponentially with the number of atoms within @xmath10 , giving @xmath19 where @xmath20 and @xmath21 is the characteristic number of atoms within a heterogeneous volume . the right most average in eqn . [ eqnplasticrate ] is then evaluated via a distribution of barrier energies defined by an extensive first and second cumulant , and written as @xmath22=\exp\left[-\frac{f(t)}{k_{\mathrm{b}}t}\right ] . \label{eqnfe}\ ] ] here @xmath23 is interpreted as the temperature dependent internal barrier energy and @xmath24 as the temperature dependent barrier entropy . within this context , @xmath25 is referred to as the free barrier energy . all of these quantities are extensive and @xmath26 gives the apparent number of structural transformations available at temperature @xmath27 . in paper i , a gaussian distribution was used giving closed form expressions for both @xmath23 and @xmath24 . an important property of the above is that for a large enough @xmath21 there exists a sharp crossover in behaviour where for , @xmath28 , eqn . [ eqnfe ] is negligible and for , @xmath29 , it is exponentially large . thus @xmath30 , where @xmath31 , defines a critical temperature at which the plastic transition rate rapidly rises . by assuming the application of a pure shear stress broadens the distribution via a quadratic increase in the variance ( with respect to stress ) , all mentioned quantities , including @xmath30 , gain a shear stress dependence . via the assumption that at zero applied shear stress @xmath30 corresponds approximately to the glass transition temperature , the stress dependence of @xmath30 rises rapidly from zero at @xmath32 as the temperature is lowered . at a temperature of approximately 0.8 - 0.9@xmath32 this temperature dependence weakens to an approximately linear behaviour upon further reduction in temperature . interpreting @xmath33 ( where @xmath34 is the temperature at which the deformation experiment is performed ) as being the stress at which plasticity begins , an analytical form of yield stress versus temperature that is valid in both the low and high temperature regimes could be developed , and could be fitted well to the available experimental data ( see fig . 8 of @xcite ) of lu _ et al _ @xcite and that shown in johnson and samwer @xcite . whilst a successful fit to low temperature yield data is not new , the current model has the virtue that the yield stress rapidly decreases as the temperature approaches @xmath32 due to a change in barrier energy statistics rather than a change in mechanism . in paper i a normal distribution of barrier energies was considered and also a modified version such that the distribution would limit to zero at zero barrier energy . the present work will consider a log - normal distribution of barrier energies . this distribution has the desirable features of being a positive valued distribution with independent first and second cumulants ( average and variance ) a requirement of the current model . the log - normal distribution is given as @xmath35 \label{eqlnd0}\ ] ] where the first moment , @xmath36 , is @xmath37\ ] ] and the second moment , @xmath38 , is defined via @xmath39.\ ] ] in terms of @xmath36 and @xmath38 , the distribution can be written as @xmath40}{4\log a}\right ] \label{eqlnd}\ ] ] with @xmath41 in paper i it was argued that the barrier energy distribution shall have extensive first and second cumulants , that is , the barrier energy mean @xmath42 and variance @xmath43 will be linear functions of the number of atoms , @xmath21 , within the heterogeneous volume element . as in paper i , an over - lined quantity will represent the corresponding intensive variable giving @xmath44 and @xmath45 . thus , the first and second moments of eqn . [ eqlnd ] will be @xmath46 resulting in @xmath47 the log - normal distribution in the form of eqn . [ eqlnd ] will now be used to obtain the average plastic rate per heterogeneous volume ( sec . [ sectheory ] ) and , through a mean - field description of the internal stress , the average plastic strain rate per heterogeneous volume ( sec . [ secmfp ] ) which can be considered as the flow equation for the high temperature homogeneous deformation regime . more generally , it will be shown that the same central results of paper i are also obtained using a log - normal distribution of barrier energies , instead of a normal distribution that was originally employed . the central mathematical challenge is to evaluate the final factor of eqn . [ eqnplasticrate ] : @xmath48 or , with respect to the barrier energy distribution , @xmath49 the integral in eqn . [ eqngf ] is the generating function of the distribution , and for the gaussian distribution considered in paper i , has a simple closed - form representation . for the log - normal distribution , such a straight forward avenue of calculation appears not to be available , however , the analogy to statistical mechanics derived in paper i provides a procedure for the evaluation of eqn . [ eqngf ] . indeed , eqn . [ eqnplasticrate1 ] can be viewed as @xmath50 where @xmath51 is an average over heterogeneous volumes . the right - hand - side term has the structure of an environmentally averaged partition function average allowing the mathematical problem to exploit the full apparatus of equilibrium statistical mechanics . to obtain a closed form expression of such a partition function , the micro - canonical approach of derrida @xcite can be used to construct the corresponding free ( barrier ) energy . this procedure is now applied when the barrier energy distribution is taken as a log - normal distribution . the average number of barrier energies between @xmath52 and @xmath53 , @xmath54 , is given by @xmath55 where @xmath56 must be small enough to ensure a well defined barrier energy but also large enough so that @xmath54 is a smooth function of @xmath52 ( see ref . @xcite for a related discussion ) . for a sufficiently large heterogeneous volume , fluctuations in eqn . [ eqnnumden ] become small and @xmath54 becomes a statistically meaningful quantity , allowing for a corresponding barrier entropy to be defined via @xmath57 . doing this for the log - normal distribution , eqn . [ eqlnd ] , results in the barrier entropy : @xmath58}{4\log a}+\log\left[\frac{1}{\sqrt{2\log a}}\right]+\log\left[\frac{de}{\sqrt{2\pi}e}\right]\right ) \label{eqnse}.\ ] ] to obtain the internal barrier energy and entropy as functions of temperature , an analogy to the thermodynamic definition , @xmath59 is applied to eqn . [ eqnse ] , giving @xmath60=-2\log a\left(1+\frac{e(t)}{k_{\mathrm{b}}t}\right).\ ] ] a solution to the above with respect to @xmath23 is obtained via the lambert w - function giving @xmath61 . \label{eqnet}\ ] ] eqn . [ eqnet ] gives the internal barrier energy as a function of temperature and its substitution into eqn . [ eqnse ] gives the barrier entropy as a function of temperature : @xmath62 + 3\log\left[\frac{e(t)}{\left\langle e\right\rangle}\right]+ \frac{1}{2\log a}\log\left[\frac{e(t)}{\left\langle e(t)\right\rangle}\right]^{2}\right)\right).\label{eqnst}\ ] ] in the above , the term containing @xmath56 in eqn . [ eqnse ] has been taken as @xmath63=\ln[\left\langle e\right\rangle / e]+\ln[de/(\sqrt{2\pi}\left\langle e\right\rangle)]\sim\ln[\left\langle e\right\rangle/(\sqrt{2\pi}e)]$ ] . this approximation will be discussed in the proceeding paragraphs . [ eqnet ] and [ eqnst ] give the free barrier entropy , @xmath64 , and therefore the final form of eqn . [ eqnpfapp ] as @xmath65.\label{eqnpfapp1}\ ] ] subsitution of this back into eqn . [ eqnplasticrate ] gives @xmath6^{-1}=\left[\tau_{\mathrm{p00}}\right]^{-1}\exp\left(\frac{s(t)}{k_{\mathrm{b}}}\right)\exp\left(-\frac{e_{\mathrm{p00}}+e(t)}{k_{\mathrm{b}}t}\right ) \label{eqnplasticrateapp}\ ] ] where @xmath66 can be viewed as the apparent barrier energy and the exponential involving @xmath24 as the number of thermally accessible structural transformations available to the heterogeneous volume . using eqns . [ eqne0 ] to [ eqna ] , the @xmath67 limit of @xmath25 , gives @xmath68\right]^{2}+2w\left[\frac{\delta\overline{e}_{0}^{2}}{k_{\mathrm{b}}t\overline{e}_{0}}\right]\right)-\overline{\alpha}\right)\label{eqnfben}\ ] ] demonstrating that for large enough heterogeneous volumes the free barrier energy is an extensive quantity as is the case of the a gaussian distribution . indeed by expanding those terms involving the lambert w - function as a taylor series to second order in @xmath69 recovers a central result of paper i : @xmath70 that is , for a narrow enough distribution ( set by @xmath71 ) the free barrier energy limits to that obtained for a gaussian distribution of barrier energies as derived in paper i. this is compatible with the fact that for a small enough @xmath71 , the log normal distribution limits to a gaussian distribution . [ eqnfbenlimit ] has the virtue that it leads to a simple analytic solution to @xmath31 for the critical temperature at which significant plastic activity can be expected ( see eqn . 11 of ref . @xcite ) . [ eqnfben ] will be used for further development since it is only at this limit that the right most term in eqn . [ eqnse ] becomes negligible and can be ignored . indeed derrida @xcite has argued that @xmath72 where @xmath73 . moreover eqn . [ eqnfben ] affords a simpler mathematical formalism which is more easily comparable to the normal distribution used in paper i. reabsorbing the factor @xmath21 into the free barrier energy gives @xmath64 as @xmath74\right]^{2}+2w\left[\frac{\delta e_{0}^{2}}{k_{\mathrm{b}}te_{0}}\right]\right)-\alpha\right)\label{eqnfben1}\ ] ] with @xmath75 \label{eqnben1}\ ] ] and @xmath76\right]^{2}\right ) . \label{eqnbsn1}\ ] ] thus the plastic strain rate ( eqn . [ eqnplasticrate ] ) becomes @xmath6^{-1}=\left[\tau_{\mathrm{p00}}\right]^{-1}\exp\left(-\frac{e_{\mathrm{p00}}+f(t)}{k_{\mathrm{b}}t}\right ) \label{eqnplasticrate2}\ ] ] withe @xmath25 being given by eqn . [ eqnfben1 ] . the above expression gains its external shear stress dependence via the assumed quadratic dependence of the variance of the barrier energy distribution @xcite : @xmath77 where @xmath78 becomes a fitting parameter . presently , the alternative form @xmath79 will be used . here @xmath80 is an appropriate value for the bulk shear modulus and @xmath81 is the corresponding fitting parameter with units of barrier energy . paper i considered only the characteristic escape rate from a mega - basin . for plastic strain evolution the new mega - basin into which the material enters is of importance . such a structural transformation from one mega basin to another will contribute to the bulk plastic strain . a natural consequence of a distribution of barrier energies is that any such event will have an associated plastic strain signature which , like the barrier energy , is a random variable . under zero load conditions , the material traverses ( at a long enough time - scale and high enough temperature ) a series of mega - basins and the internal plastic strain fluctuates accordingly . assuming the heterogeneous volumes interact via elasticity , this strain fluctuation corresponds to an internal stress fluctuation and therefore an elastic potential energy landscape . it is such a coarse grained potential energy landscape which underlies the alpha energy landscape and therefore the heterogeneity of the material . due to the assumed exponential number of available structural transformations there will also be an exponential number of mega - basins that the material can enter . hence upon exciting one mega - basin and entering another , the statistics will be negligibly affected , with the consequence being that the system has again access to all possible strain states . which one it chooses will be stochastically biased towards those that minimise the local elastic energy . this situation does not change upon application of a load , however those structural transformations that can reduce the elastic energy of a particular deformation geometry will become more likely . so far , the average properties of the non - interacting heterogeneous volumes have been considered , giving @xmath82 where @xmath83 is the externally applied pure shear stress tensor . from the above discussion , the primary interaction between heterogeneous volumes will be of elastic origin . the inclusion of such interactions may be done via the eshelby construction . here the goal is to calculate the elastic energy contribution arising from a local plastic strain increment within a particular heterogeneous volume ( the @xmath8th ) of volume @xmath10 . in the eshelby construction @xcite , this is done by removing from the the amorphous matrix the heterogeneous volume in which the plastic event has occurred , and allowing it to relax to a stress free state given by @xmath84 . the heterogeneous volume is then deformed elastically to return it to its original shape via the corresponding elastic strain @xmath85 and re - inserted into the matrix . at this point , the particular heterogeneous volume ( the inclusion ) has a stress derived from hookes law equal to @xmath86 whilst the surrounding matrix is stress free . the combined system is then allowed to relax to an equilibrium internal stress configuration reducing the inclusion stress by @xmath87 . eshelby has then shown that the total change in elastic energy is given by @xmath88 where @xmath89 is an additional external stress ( to the @xmath8th heterogeneous volume ) . here , and throughout the text , repeated indices are summed . assuming that this change in elastic energy modifies linearly the intrinsic barrier energy corresponding to this plastic event , one has @xmath90 the strain rate for the @xmath8th heterogeneous volume then becomes @xmath91^{-1}\exp\left[-\frac{e_{n , i}+\frac{1}{2}v_{0}\varepsilon_{\mathrm{t},ni}^{\mu\nu}\left(\sigma_{\mathrm{t},ni}^{\mu\nu}-\sigma_{\mathrm{c},ni}^{\mu\nu}-\sigma_{\mathrm{e},n}^{\mu\nu}\right)}{k_{\mathrm{b}}t}\right ] , \label{eqnpsr}\ ] ] where @xmath92 is the strain increment associated with the barrier energy @xmath93 and attempt rate @xmath94 . thus each barrier energy is modified from its ( local ) zero stress value by the total elastic strain energy associated with the plastic event occurring . the strain rate for the rve is therefore @xmath95 here @xmath4 is the volume of the rve volume giving @xmath96 and @xmath97^{-1}\exp\left[-\frac{e_{n , i}+\frac{1}{2}v_{0}\varepsilon_{\mathrm{t},ni}^{\mu\nu}\left(\sigma_{\mathrm{t},ni}^{\mu\nu}-\sigma_{\mathrm{c},ni}^{\mu\nu}-\sigma_{\mathrm{e},n}^{\mu\nu}\right)}{k_{\mathrm{b}}t}\right ] . \label{eqnavepsr0}\ ] ] the elastic relaxation associated with the @xmath16th plastic event , within and surrounding the @xmath8th heterogeneous volume , results in @xmath98 and an extended stress field field felt by other nearby and distant heterogeneous volumes ( through their own @xmath99 ) . from this latter perspective @xmath98 can be viewed as a self stress contribution to the elastic energy of the plastic event . the spatial dependence of these stress contributions depend in a non - trivial way on @xmath86 and must be calculated numerically ( see for example the book of t. mura @xcite and refs . @xcite ) . it is in this way , that elastic interactions between heterogeneous volumes may be formally introduced . in addition , the @xmath89 may contain a global applied external stress component written as @xmath83 giving @xmath100 . thus for each heterogeneous volume , @xmath89 arises from an external applied stress plus an internal stress arising from the elastic interaction between the plastic events from all other heterogeneous volumes . such a formalism has been considered by bulatov and argon @xcite in their development of a thermal activation model for the plasticity of glasses , and more recently using the finite - element - method by homer and schuh @xcite . in the present work a mean field approach will be taken which exploits the fact that the local contribution to @xmath89 arising from other heterogeneous volumes will on average equal zero , giving @xmath101 . for an assumed isotropic elasticity , @xmath102 will equal @xmath103 where @xmath104 will be the shear strain magnitude of the plastic event and @xmath80 is the appropriate bulk isotropic shear modulus for the system . @xmath105 embodies the elastic relaxation of the plastic strain event and will be different for each plastic event . within the current mean field approach this will be absorbed into @xmath80 resulting in the @xmath80 being an effective bulk shear modulus of the system . these approximations afford a simple average with respect to heterogeneous volumes resulting in eqn . [ eqnavepsr0 ] becoming @xmath106^{-1}\exp\left[-\frac{e_{\mathrm{p0}}+2v_{0}g\gamma_{\mathrm{t}}^{2}-\frac{1}{2}v_{0}\varepsilon_{\mathrm{t}}^{\mu\nu}\sigma^{\mu\nu}}{k_{\mathrm{b}}t}\right]\right\rangle,\ ] ] in which the average is performed with respect to @xmath84 , @xmath107 and @xmath18 . since there should be no correlation between the strain increment @xmath84 and the parameters @xmath107 and @xmath18 associated with the escape of a particular mega - basin ( since each strain state is always accessible ) , the above average with respect to @xmath84 may be performed independently from @xmath107 and @xmath18 , giving , @xmath108\right\rangle_{\varepsilon_{\mathrm{t}}^{\mu\nu } } m\left\langle\left[\tau_{\mathrm{p0}}\right]^{-1}\exp\left[-\frac{e_{\mathrm{p0}}}{k_{\mathrm{b}}t}\right]\right\rangle_{\tau_{\mathrm{p0}},e_{\mathrm{p0 } } } \nonumber \\ & = & \left\langle\varepsilon_{\mathrm{t}}^{\mu\nu}\exp\left[\frac{-2v_{0}g\gamma_{\mathrm{t}}^{2}+\frac{1}{2}v_{0}\varepsilon_{\mathrm{t}}^{\mu\nu}\sigma^{\mu\nu}}{k_{\mathrm{b}}t}\right]\right\rangle_{\varepsilon_{\mathrm{t}}^{\mu\nu}}\left[\tau_{\mathrm{p}}\right]^{-1}\left(t,\sigma^{\mu\nu}\right).\label{eqnavepsr1}\end{aligned}\ ] ] in the last equality , @xmath3^{-1}\left(t,\sigma^{\mu\nu}\right)$ ] is given by eqn . [ eqnplasticrate ] . a consequence of such a lack of correlation is that the magnitude of a barrier energy is , on average , not correlated with the magnitude of the resulting plastic strain increment . this scenario , again , appeals to the notion of an exponentially large diversity of possible structural excitation . moreover , numerical evidence in simple lennard jones structural glasses exists , demonstrating that the strain associated in traversing a nearby local pel minima does not at all correlate with the height of the intermediate saddle - point ( barrier ) energy @xcite . the plastic strain average in eqn . [ eqnavepsr1 ] may be written as @xmath109\right\rangle_{\varepsilon_{\mathrm{t}}^{\mu\nu}}=\frac{\partial}{\partial\left[\frac{\frac{1}{2}v_{0}\sigma^{\mu\nu}}{k_{\mathrm{b}}t}\right]}\left\langle\exp\left[\frac{-2v_{0}g\gamma_{\mathrm{t}}^{2}+\frac{1}{2}v_{0}\varepsilon_{\mathrm{t}}^{\mu\nu}\sigma^{\mu\nu}}{k_{\mathrm{b}}t}\right]\right\rangle_{\varepsilon_{\mathrm{t}}^{\mu\nu}}\label{eqngfidentity0}\ ] ] and therefore the goal is to calculate @xmath110\right\rangle_{\varepsilon_{\mathrm{t}}^{\mu\nu}}. \label{eqnstrainav}\ ] ] to evaluate this average further , the two dimensional case of plane strain will be considered involving only an external pure shear stress component : @xmath111 under these constraints , the characteristic shear strain matrix takes the form @xmath112 giving @xmath113 thus all possible plastic strain increments are defined by an orientation angle @xmath114 and a plastic shear strain magnitude @xmath115 . due to the exponentially large number of plastic strain increments available and the lack of a preferred direction , the random variable @xmath114 is assumed to be derived from a uniform distribution ranging from @xmath116 to @xmath117 . on the other hand , for reasonable values of the shear strain magnitude , the distribution for @xmath115 is expected to be bounded . presently ( and for mathematical simplicity ) the distribution is assumed to be a normal distribution centred on zero with standard deviation @xmath118 . it is acknowledged that other distributions are also feasible . with these assumptions eqn . [ eqnstrainav ] becomes @xmath110\right\rangle_{\theta_{\mathrm{t}},\gamma_{\mathrm{t}}}=\exp\left[\left(\frac{v_{0}\delta\gamma'\sigma}{2k_{\mathrm{b}}t}\right)^{2}\right]\mathrm{i}_{0}\left[\left(\frac{v_{0}\delta\gamma'\sigma}{2k_{\mathrm{b}}t}\right)^{2}\right ] . \label{eqngfidentity1}\ ] ] where @xmath119 is the modified bessel function of order @xmath8 and @xmath120 in sec . [ secexp ] , the latter approximation will be shown to be valid for temperatures close to the glass transition temperature . using eqn . [ eqngfidentity0 ] this finally gives the strain rate per heterogeneous volume as @xmath121\nonumber \\ & & \times\left(\mathrm{i}_{0}\left[\left(\frac{v_{0}\delta\gamma\sigma}{2k_{\mathrm{b}}t}\right)^{2}\right]+\mathrm{i}_{1}\left[\left(\frac{v_{0}\delta\gamma\sigma}{2k_{\mathrm{b}}t}\right)^{2}\right]\right)\nonumber \\ & & \times\left[\tau_{\mathrm{p}}\right]^{-1}\left(t,\sigma^{\mu\nu}\right)\end{aligned}\ ] ] using eqn . [ eqnplasticrate2 ] and writing , @xmath122 , this reduces to @xmath123\left(\mathrm{i}_{0}\left[\left(\frac{\delta\omega\sigma}{2k_{\mathrm{b}}t}\right)^{2}\right]+\mathrm{i}_{1}\left[\left(\frac{\delta\omega\sigma}{2k_{\mathrm{b}}t}\right)^{2}\right]\right)\nonumber \\ & & \times\exp\left[-\frac{e_{\mathrm{p00}}+f(t,\sigma)}{k_{\mathrm{b}}t}\right ] \label{eqnavepsr3}\end{aligned}\ ] ] where @xmath124 and @xmath125 eqn . [ eqnavepsr3 ] , written explicitly in terms of the internal barrier energy and barrier entropy , becomes @xmath126\left(\mathrm{i}_{0}\left[\left(\frac{\delta\omega\sigma}{2k_{\mathrm{b}}t}\right)^{2}\right]+\mathrm{i}_{1}\left[\left(\frac{\delta\omega\sigma}{2k_{\mathrm{b}}t}\right)^{2}\right]\right ) \nonumber \\ & & \times\exp\left[-\frac{e_{\mathrm{p00}}+e(t,\sigma)}{k_{\mathrm{b}}t}\right].\label{eqnavepsr4}\end{aligned}\ ] ] the above equation should be compared to the early thermal activation models grounded on the work of spaepen @xcite and argon @xcite , which for homogeneous plastic deformation in the vicinity of the glass transition temperature is usually taken as @xcite @xmath127\sinh\left[\frac{v_{\mathrm{ta}}\sigma}{k_{\mathrm{b}}t}\right ] \label{eqnsimplet}\ ] ] where @xmath128 contains various quantities including the fraction of volume available to deformation , @xmath129 is the attempt frequency , @xmath130 is the characteristic strain , @xmath131 is the activation energy and @xmath132 is the activation volume . comparison of eqn . [ eqnavepsr4 ] to [ eqnsimplet ] suggests that @xmath133 can be viewed as a characteristic strain rate ( @xmath134 of eqn . [ eqnsimplet ] ) , @xmath135 $ ] as a temperature and stress dependent number of available structural transformations , @xmath136 as a temperature and stress dependent activation energy , and @xmath137 as an activation volume . this latter interpretation is motivated by the term @xmath138+\mathrm{i}_{1}[x^{2}])$ ] being qualitatively similar to @xmath139 , with both limiting to a linear function for small @xmath140 . thus , although the present theory has a number of quite different physical assumptions when compared to earlier high - temperature thermal activation models , the mean - field prediction of the average strain rate as a function of temperature and stress in the high temperature / low strain rate regime does not operationally differ from these early models . this observation will be shown in more detail in the next section , when the model is fitted to actual experimental data . inspection of eqn . [ eqnavepsr3 ] reveals the free parameters of the model are @xmath133 , @xmath137 , and those associated with @xmath141 which are @xmath142 , @xmath42 , @xmath143 , @xmath0 and @xmath81 . the latter shear stress dependence gets its definition via eqn . [ eqsd ] where here @xmath143 is equal to @xmath144 . as in paper i , @xmath42 and @xmath143 , which define the barrier energy distribution can be entirely determined from experimentally accessible parameters , in particular via the viscosity @xmath145 and fragility @xmath146 at the glass transition temperature @xmath32 . linear ( newtonian ) viscosity may be formally given as @xmath147.\label{eqnviscocity}\ ] ] one common method to define the glass transition temperature is when the viscosity reaches @xmath148 pa - sec at a cooling rate of 20 k / min . that is , @xmath149.\label{eqnviscocity1}\ ] ] in addition to the value of the viscosity at @xmath32 , a common material parameter to structural glasses is the fragility the rate at which viscosity changes with respect to temperature at @xmath32 , @xmath150 substitution of eqn . [ eqnviscocity ] into eqn . [ eqnfragility ] gives @xmath151 or the internal barrier energy at @xmath32 @xmath152 this physically appealing result is general and arises because eqn . [ eqnter ] requires @xmath153 or equivalently the expression , @xmath154 , which is analogous to the gibbs - helmholtz equation in thermodynamics . further substitution of eqn . [ eqnconstraint1 ] into [ eqnviscocity1 ] , gives the barrier entropy at @xmath32 @xmath155\right ) . \label{eqnconstraint2}\ ] ] thus the experimental parameters @xmath145 , @xmath146 define the free barrier energy at @xmath32 for a given choice of @xmath137 , @xmath133 and @xmath0 . [ eqnconstraint1 ] and [ eqnconstraint2 ] in conjunction with eqns . [ eqnben1 ] and [ eqnbsn1 ] also uniquely determine @xmath42 and @xmath143 , @xmath156\ ] ] and @xmath157}.\ ] ] it is noted that the time rate of change of viscosity at the glass transition is given as @xmath158 where cr is the cooling rate from the under - cooled liquid regime . thus the cooling rate does not give additional information about the theory and emphasizes the fact that the current thermal activation approach is not applicable to the physics of the under - cooled liquid regime . the remaining free parameters are @xmath133 , @xmath137 , @xmath142 , @xmath0 and @xmath81 . @xmath142 can be determined from dynamical mechanical analysis or differential scanning calorimetry experiments @xcite and the rest from steady state strain rate versus stress data as a function of temperature , a common data set used to fit to the arrhenius form of eqn . [ eqnsimplet ] . for vitreloy-1 such a data set is found in the work of lu _ et al _ @xcite . for this structural glass initial estimates of @xmath159 k , @xmath160 and @xmath161 ev are taken from both lu _ et al _ @xcite and wang @xcite . [ figexpcurves]a plots the experimental steady state strain rate versus stress for a number of temperatures close to and above the glass transition temperature . following schuh _ @xcite and wang @xcite it is this data to which the current model ( eqn . [ eqnavepsr3 ] ) will be fitted with all parameters allowed to vary . the experimental data is taken from uniaxial compression tests and for conversion between the strain rate and uni - axial stress to the appropriate shear quantities , the former is multiplied by @xmath162 and the latter is divided by @xmath162 . this is motivated by fig . [ figexpcurves]b , which plots lu s linear viscosity ( fig . 8 of ref . @xcite ) data multiplied by the shear strain rate and divided by the shear stress . such numerical factors are common and represent an effective angle of dominant slip activity in a uniaxial deformation experiment , see for example kawamura _ et al _ @xcite . as expected , for a wide range of low shear stresses , the experimental data in fig . [ figexpcurves]b has converged to approximately unity indicating a stress independent newtonian viscosity coefficient . the solid curves in figs . [ figexpcurves]a - b indicate a fit of the model obtained via the simulated annealing global minimisation algorithm @xcite . the corresponding parameters of this fit are @xmath163 k , @xmath164 , @xmath165 ev , @xmath166 sec@xmath167 , @xmath168 m@xmath169 , @xmath170 and @xmath171 ev , with @xmath145 fixed at @xmath172 pa - sec . it is noted that for these parameters , the numerator of eqn . [ eqnapprox ] evaluated at @xmath32 equals @xmath173 for @xmath174 nm@xmath169 , justifiying the ensuing approximation of eqn . [ eqnapprox ] . when interpreted as an activation volume , the parameter @xmath137 has a numerical value that is similar to that found in refs . @xcite when using the arrhenius form of eqn . [ eqnsimplet ] . when assuming that the volume of the heterogeneous volumes is one to three orders of magnitude larger than this `` activation volume '' , eqn . [ eqnactivationvolume ] and [ eqncharacteristicsr ] together with the numerical value of @xmath133 indicate a value for @xmath175 ranging somewhere between @xmath176 to @xmath172 a range of values compatible with it being some multiple of the debye frequency of the system . the numerical value of @xmath81 indicates a weak stress dependence of the distribution of barrier energies , a consequence of which is that the apparent barrier energy has a weak stress dependence and therefore a value ( @xmath177 ev ) differing little from the activation energy obtained when using eqn . [ eqnsimplet ] , which is @xmath178 ev @xcite . the corresponding log - normal barrier energy parameters have values of @xmath179 ev and @xmath180 ev indicating that in the high temperature regime , the relevant part of the distribution ( @xmath181 remains that of low barrier energy tail . [ figexpcurves]c displays experimental uniaxial deformation curves at a temperature of 643k for a number of different strain rates . the data is taken directly from fig . 2 of ref . the higher strain rate curves are characterised by an initial rise in stress that may be loosely associated with the elastic deformation regime , a peak stress indicating the onset of a material instability and an eventual stead state flow regime . appendix b of paper i used a simple method to obtain a constant strain rate stress strain curve from eqn . [ eqnavepsr3 ] , resulting in the stress at time @xmath182 being given by @xmath183.\ ] ] for a small enough time interval this may be approximated as @xmath184\ ] ] and iterated to generate a stress - strain curve . in the above the plastic strain rate , @xmath185 is given by @xmath186 ( eqn . [ eqnavepsr3 ] ) , @xmath187 is the appropriate bulk youngs modulus and @xmath188 the chosen constant total strain rate . using this model , fig . [ figexpcurves]c also displays the resulting model stress - strain curves . whilst the steady state stress regime is well described by the model , particularly for low values of strain rate , the peak stress regime associated with the emergence of a stress overshoot is not at all present a result due to the mean field nature of the current plasticity model , which necessarily can only describe homogeneous plasticity . fig . [ figexpcurves]d now shows the experimental peak stress values of lu _ et al _ @xcite for different strain rates as a function of temperature ( shown as open circles in the figure ) . with @xmath32 at approximately 623k , such data clearly shows the transition from a high temperature / low strain rate homogeneous plasticity to a low temperature / high strain rate regime of deformation . whilst the model is unable to reproduce the peak stress behaviour it is revealing to plot on this curve the temperature dependence of the steady state stresses for comparable strain rates ( solid curves ) . for the lowest strain rate , good agreement is seen in the high temperature regime where experimentally the peak stress regime is absent . for larger strain rates , the quantitative agreement reduces due to the mean field nature of the current plasticity model however qualitatively the trends are similar . for this very same reason , the rapid change in experimental behaviour seen as the temperature reduces is not evident in the model , although at higher stresses ( beyond the range of the figure ) the predicted steady state stress does limit to a plateau with a weak temperature dependence a regime that was investigated in paper i ( see fig . 8 of @xcite ) . having said the above , fig . [ figexpcurves]d also includes ( the dashed lines ) the consequences of the model when fitted to the peak stress data , as in paper i , rather than the steady state data . the emerging plateau region at low temperatures now becomes evident , although the sharper transition to this temperature regime seen when using the modified gaussian of paper i is no longer evident when using the more restrictive log - normal distribution of barrier energies . the parameters of this fit are @xmath189 , @xmath190 , @xmath165 ev , @xmath191 sec@xmath167 , @xmath192 m@xmath169 , @xmath193 and @xmath194 ev . these values do not differ substantially from those associated with the fit to steady state flow stress data . the corresponding log - normal barrier energy parameters have values of @xmath195 ev and @xmath196 ev which represents a distribution that is closer in position to the the fitted modified gaussian distribution used in paper i ( see fig . 7c of @xcite ) . whilst such an approach is certainly not justified in the temperature regime where there is a cross - over from homogeneous to heterogeneous plasticity , and a strong difference between the peak stress and flow stress , well below the glass transition temperature there is little experimental distinction between these stresses . thus one could argue ( as was done in paper i ) that mean field would be applicable both to the quite low temperature and high temperature deformation modes for the onset of plastic deformation ( that is yield ) , with the intermediate cross - over regime not being well described . the dashed curves in fig . [ figexpcurves]d tend to support this conclusion . the present thermal activation theory provides a entirely kinetic interpretation of the fragility index at the glass transition temperature . this may be seen by writing the barrier entropy , eqn . [ eqnconstraint2 ] , as @xmath197=k_{\mathrm{b}}\left(m\ln10 + 1\right).\ ] ] or rather as @xmath198=k_{\mathrm{b}}\left(m\ln10 + 1\right).\label{eqnconstraint3}\ ] ] with @xmath199 being viewed as the characteristic experimental time scale associated with the deformation experiment ( see paper i ) . if @xmath200 where equal to the fundamental time scale of the @xmath1-relaxation modes , @xmath175 , then @xmath201 would equal @xmath202 where @xmath146 is the correspondingly measured fragility . in this limit , the fragility is therefore a direct measure of the apparent number of structural transformations available to each heterogeneous volume element at @xmath32 , that is , @xmath203 . an analogous identification has also been proposed between fragility and the configurational entropy of the under - cooled liquid pel by sastry @xcite . this present result immediately implies that @xmath0 can not be less than @xmath202 since @xmath204 ( eqn . [ eqnst ] ) . note that , like the barrier entropy , the fragility is an extensive quantity . for a realistic deformation experiment , @xmath205 , resulting in the above equality between @xmath201 and @xmath146 being an over estimation of the barrier entropy . this is because over the period of time , @xmath200 , significant _ experimentally unresolvable _ activity occurs which also contributes to the correspondingly measured fragility . through the term @xmath206 $ ] this activity contributes to an effective reduction of the needed barrier entropy . thus when the glass transition is viewed as an entirely kinetic phenomenon ( as is done here ) , the fragility becomes a direct measure of the apparent number of available @xmath0-mode structural transformations . thus eqn . [ eqnconstraint3 ] ( and eqn . [ eqnconstraint2 ] ) is considered more fundamental than the more familiar eqn . [ eqnconstraint1 ] which ( here ) arises from the requirement that the temperature @xmath27 defined via eqn . [ eqnter ] corresponds to the thermodynamic temperature of the system . using the fitted parameters of viteroly-1 obtained in sec . [ secexp ] , the value of the ratio @xmath207 suggests a reasonable choice of @xmath175 ( in the range of @xmath208 to @xmath209 sec ) will give a @xmath200 that is within the domain of a typical deformation experiment . the current mean - field description of macroscopic plasticity is only suited to purely homogeneous plasticity , where within one rve , it is valid to describe the plasticity as arising from the average response of a heterogeneous volume . whilst this might be valid for a broad range of temperatures in the flow stress regime , figs . [ figexpcurves]c - d demonstrate that for high enough strain rate or low enough temperature , the mean field approximation is unable to describe the presumably heterogeneous transition from elasticity to macroscopic plasticity seen in experiment . indeed at quite low temperatures , where the available number of structural transformations per heterogeneous volume approaches unity a new regime of strongly heterogeneous statistics emerges associated with that of the extreme value . the average temperature at which this occurs is referred to as the kinetic freezing temperature and may be found as the solution to the barrier entropy equalling zero , @xmath210 : the case when there exists , on average , one available structural transformation per heterogeneous volume . using eqn . [ eqnbsn1 ] and the known properties of the lambert w - function , this temperature is found to be @xmath211 . \label{eqnft}\ ] ] for the log - normal distribution . a gaussian distribution gives a similar result , but without the exponential factor . for @xmath212 the free barrier energy becomes independent of temperature equalling @xmath213 . when sampling the @xmath9 heterogeneous volumes of the rve , the kinetic freezing temperature will fluctuate around this value due to fluctuations in the corresponding kinetic freezing barrier energy , the statistics of which is set by the extreme value weibull distribution @xcite . in this temperature regime , a heterogeneous volume element may contain no accessible barriers below a particular energy threshold allowing for the possibility that no thermally activated structural transformation occurs . it is only in this temperature regime that a single structural transformation associated with a single barrier energy is formally valid . this scenario is quite different from that of the statistics of the most probable , occurring at higher temperatures , where on average there exists a large number of accessible barrier energies resulting in a statistically meaningful average plastic transition rate for each heterogeneous volume . for viteroly-1 , the parameters of sec . [ secexp ] give @xmath214 k indicating that below room temperature this strongly heterogeneous regime of statistics begins to dominate . when using the numerical values of the model parameters derived from a fit to the peak stress , the freezing temperature rises to approximately 203 k , which is somewhat smaller than that obtained when using a gaussian distribution fitted also to the peak stress ( as done in paper i ) . [ figfreeenergy ] plots the temperature dependence of the free barrier energy , apparent barrier energy and barrier entropy for the flow stress parameters of viteroly-1 . both the apparent barrier entropy and barrier entropy reduce with temperature , until the latter becomes zero at the freezing temperature , @xmath215 . below this temperature range both quantities become constants independent of temperature . the corresponding free barrier energy shown in fig . [ figfreeenergy]a is negative above a temperature regime associated with the glass transition , and rises when the temperature is lowered . according to eqn . [ eqnfe ] , in the temperature regime where the free barrier energy is positive , the plastic rate reduces many orders of magnitude when compared to the value associated with the glass transition temperature regime . at @xmath216 the free barrier energy also becomes constant , and it is in this regime a simple temperature independent value for both the number of available structural and the energy barrier is formally justified . despite this , the work of sec . [ secexp ] demonstrates that temperature independent quantities can be used to describe reasonably well the high temperature / homogeneous deformation regime . indeed fig . [ figfreeenergy ] demonstrates that both barrier energy and entropy do not vary greatly over the 550 k to 650 k temperature range . , below which the barrier entropy is zero and on average there exists one available structural transformation within each heterogeneous volume , resulting in the free energy and barrier energy becoming temperature independent quantities for @xmath216.,scaledwidth=90.0% ] the mean - field form for the shear strain rate , eqn . [ eqnavepsr4 ] , is in practice quite similar to the well known arrhenius form , eqn . [ eqnsimplet ] . indeed it is operationally similar justifying the so - called critical barrier energy alluded to by argon @xcite and johnson and samwer @xcite . indeed , in paper i , the fluctuations around the apparent barrier energy where shown to scale as @xmath217 . despite this similarity , there exist some fundamental differences between the models . the arrhenius shear strain rate is assumed to be proportional to the fraction of volume in which structural transformations can take place @xcite . the present theory and eqn . [ eqnavepsr4 ] also gives a plastic shear rate that is proportional to the number of available structural transitions , @xmath26 , however this number scales exponentially with atom number and therefore volume . since this number can be rather large , the traditional athermal concept of a dilute density of ( liquid like ) regions existing within the structural glass which are amenable to plastic deformation has been abandoned in the present work , any particular region of the material can admit a very large number and variety of structural transformations . of course , at low enough @xmath27 the apparent number of structural transformations drastically reduces as the entire @xmath0-relaxation pel gradually freezes out and the kinetic freezing regime is entered . another similarity between eqns . [ eqnavepsr4 ] and [ eqnsimplet ] is a shear stress dependent term which for the present work has the form @xmath218+\mathrm{i}_{1}[x^{2}])$ ] and for the arrhenius form is @xmath139 . in both cases @xmath219 . for the arrhenius form , the volume term is seen as an activation volume associated with the barrier enthalpy of the volume distortion needed for the structural transformation to occur . in the present work , this quantity originates from the variance of the distribution of available local plastic strains , @xmath118 , via @xmath220 . the right - hand - side of this equality may be seen as a measure of the variance of the available slipped volume within the system , which itself is coming from a characteristic internal slipped area multiplied by a characteristic slip distance . in eqn . [ eqnsimplet ] , the @xmath139 factor arises from reversibility of the single characteristic structural transformation of the system and that the sign of the enthalpy term changes for the inverse process . for the present work , the factor @xmath218+\mathrm{i}_{1}[x^{2}])$ ] in eqn . [ eqnavepsr4 ] is due to a quite different reason where each transition to a new mega - basin results in the material on average having once again access to all possible plastic transitions . thus although the reverse transition is allowed , it will occur with negligible probability , since it must compete with all other accessible @xmath0 structural transformations . the factor @xmath218+\mathrm{i}_{1}[x^{2}])$ ] then arises from the mean - field average ( integration ) over all possible plastic transitions . whilst the currently developed mean - field picture affords some insight into the consequences of the present theory to macro - plasticity , it certainly is limited to the low strain rate and high temperature range , a regime that is experimentally insensitive to fluctuations away from homogeneity , and also to the thermal and loading history of the material . indeed , inherent to the above picture is that the material has no memory of its past state . this is direct result of the mean field theory variant presently used since the elastic interaction between heterogeneous volumes is not at all considered . when such interactions are taken into account the pre - history of the material will play an integral role in its response to an external condition . in this regard the internal stress field may be viewed as the state - variable of the model . to address these aspects of low temperature , high strain rate and material history , the mean field approach must be abandoned and an approach which considers correlated spatial variations in plastic activity must be considered . how to develop a theory of macroscopic plasticity beyond mean - field , and which spans the regime of @xmath215 up to @xmath32 for experimentally accessible strain rates , will be the subject of paper iii in this series of work . the authors wish to thank d. rodney and k. samwer for helpful discussions . 99 c.a . schuh , t.c . hufnagel , u. ramamurty , acta . mater . 55 ( 2007 ) p.4067 . m.l . falk and j.s . langer , annual review of condensed matter physics 2 ( 2011 ) p.353 . f. spaepen , acta metall . 25 ( 1977 ) a. argon , acta metall . 27 ( 1979 ) p.47 . w.l . johnson and k.a . samwer , phys . 95 ( 2005 ) p.195501 . m. h. cohen and d. turnbull , j. phys . ( 1959 ) p. 16 d. turnbull and m. h. cohen , j. phys . 34 ( 1961 ) p. 120 d. turnbull and m. h. cohen , j. phys . 52 ( 1970 ) p. 3038 m.l . falk and j.s . langer , phys . e 57 ( 1998 ) p.7192 . schuh and a.c . lund , nat . . 2 p.449 . maloney and a. lematre , phys . 93 ( 2004 ) p.016001 . m.j . demkowicz and a.s . argon , phys . 91 ( 2004 ) p.025505 . y. shi and m.l . falk , phys . b 73 ( 2006 ) p.214201 . p. guan , m. chen and t. egami , phys . 104 ( 2010 ) p.205701 . m.j . demkowicz and a.s . argon , phys . b 72 ( 2005 ) p.245205 . m.j . demkowicz and a.s . argon , phys . b 72 ( 2005 ) p.245206 . langer , phys . e 77 ( 2008 ) p.021502 . bouchbinder , j.s . langer and i. procaccia , phys . e 75 ( 2007 ) p.036107 . falk , j.s . langer and l. pechenik , phys . rev . b 70 ( 2004 ) p.011507 j.s . langer , phys . e 70 ( 2004 ) p.041502 . m. goldstein , j. chem . ( 1969 ) p.3728 . angell , k.l . ngai , g.b . mckenna , p.f . mcmillan and s.w . martin , j. appl . 88 ( 2000 ) p.3113 . harmon , m.d . demetriou , w.l . johnson and k. samwer , phys . 99 ( 2007 ) p.135502 . wang , j. appl . ( 2011 ) p.053521 . m. scott shell , p.g . debenedetti and a.z . panagiotopoulous , phys . 92 ( 2004 ) p.035506 . fyodorov , phys . 92 ( 2004 ) p.240601 . g. adam and j.h . gibbs , j. chem . 43 ( 1965 ) p.139 . kirkpatrick , d. thirumalai and p.g . wolynes , phys . a 40 ( 1989 ) p.1045 . a. heuer , j. phys . : condens . matter 20 ( 2008 ) p.373101 . p.m. derlet and r. maa , phys . b ( rc ) 84 ( 2011 ) p.220201 . p.m. derlet and r. maa , mater . proc . vol . 1520 , doi : 10.1557/opl.2012.1689 ( 2012 ) . p.m. derlet and r. maa , phil . mag . , doi : 10.1080/14786435.2013.826396 ( 2013 ) . d. klaumnzer , r. maa , f. h. dalla torre and j. f. lffler , appl . 96 ( 2010 ) p.061901 . r. maa , d. klaumzer and j. f. lffler , acta mater . 59 ( 2011 ) p.3205 . r. maa , d. klaumzer , g. villard , p. m. derlet and j. f. lffler , appl . 100 ( 2012 ) p. 071904 . j. lu , g. ravichandran and w.l . johnson , acta mater . 51 ( 2003 ) p.3429 . b. derrida , phys . ( 1980 ) p.79 . eshelby , proc . soc . a 241 ( 1957 ) v.v . bulatov and a.s . argon , model . eng . 2 ( 1994 ) p.167 . v.v . bulatov and a.s . argon , model . eng . 2 ( 1994 ) v.v . bulatov and a.s . argon , model . eng . 2 ( 1994 ) t. mura , _ micromehanics of defects in solids _ , martinus hijhoff publishers , the hauge , 1982 . homer and c.a . schuh , acta mater . 57 ( 2009 ) p.2823 . e.r . homer and c.a . schuh , modelling simul . ( 2010 ) p.065009 . l. li , e.r . homer and c.a . schuh , acta mater . 61 ( 2013 ) p.3347 . d. rodney and c.a . schuh , phys . 102 ( 2009 ) p.235503 . d. rodney and c.a . schuh , phys . b. 80 ( 2009 ) p.184203 . y. kawamura , t. nakamura and a. inoue , scripta mat . 39 ( 1988 ) corana a , marchesi m , martini c and ridella s acm trans . math . ( 1987 ) p.262 . s. sastry , nature 409 ( 2001 ) p.164 . gumbel , statistics of extremes ( dover , 2004 )

in order to test the expansion of the universe and its geometry , we carry out an alcock & paczyski cosmological test , that is , an evaluation of the ratio of observed angular size to radial / redshift size . the main advantage of this test is that it does not depend on the evolution of the galaxies , but only on the geometry of the universe . however , the redshift distortions produced by the peculiar velocities of the gravitational infall do also have an influence , which should be separated from the cosmological effect . we derive the anisotropic correlation function of sources in three surveys within the sloan digital sky survey ( sdss ) : galaxies from sdss - iii / baryon oscillation spectroscopy survey data release 10 ( boss - dr10 ) , and qsos from sdss - ii and sdss - iii / boss - dr10 . from these , we are able to disentangle the dynamic and geometric distortions and thus derive the ratio of observed angular size to radial / redshift size at different redshifts . we also add some other values available in the literature . then , we use the data to evaluate which cosmological model fits them . we used six different models : concordancecdm , einstein - de sitter , open friedman cosmology without dark energy , flat quasi - steady state cosmology , a static universe with a linear hubble law , and a static universe with tired light redshift . only two of the six models above fit the data of the alcock & paczyski test : concordancecdm and static universe with tired light redshift ; whereas the rest of them are excluded at a% confidence level . if we assume thatcdm is the correct one , the best fit with a free is produced for .

1312.0003

there are several ways to test the geometry of a cosmological model and its expansion , i.e. , that redshifts of the galaxies are cosmological and not due to an alternative mechanism ( narlikar 1989 ; lpez - corredoira 2003 , section 2.1 ; lpez - corredoira 2006 ) . however , almost all of the cosmological tests are entangled with the evolution of galaxies and/or other effects . hubble diagrams ( laviolette 1986 ; schade et al . 1997 ; marosi 2013 ) , tolman surface brightness tests ( lubin & sandage 2001 ; andrews 2006 ; lerner 2006 ) , angular size tests ( kapahi 1987 ; kellerman 1993 ; lpez - corredoira 2010 ) are all affected by the evolution of galaxies at high redshifts . so far , there are no standard rods in physical objects . the ultra - compact radio sources which were claimed in the past to be free of evolutionary effects are now thought to present some evolution as well ( lpez - corredoira 2010 ; pashchenko & vitrishchak 2011 ) . lpez - corredoira ( 2010 ) thinks that the huge size evolution necessary to fit an angular size test with an expanding universe is not understood , but it still depends on our understanding of the galaxies rather than on pure cosmological approaches . baryonic acoustic oscillations ( baos ) in the cosmic microwave background radiation ( cmbr ) or in the large scale structure are usually thought to be standard rods to measure the cosmological expansion ( e.g. , rassat & refregier 2012 ) ; however their features in the power spectrum are not a direct observable quantity free of assumptions ( one must assume at least that bao peaks represent a faint imprint of the sound waves in the clustering of galaxies and matter today excited by the initial inflationary perturbations , i.e. the assumption that the scenario of standard cosmology is correct ) and these peaks could be generated with other cosmological assumptions different from the standard interpretation of acoustic peaks ( lpez - corredoira & gabrielli 2013 ) . time dilation tests in type ia supernovae ( snia ) look like one of the most successful tests in favor of the expansion of the universe ( goldhaber et al . 2001 ; blondin et al . 2008 ) , but they do not serve to test the geometry of the universe , and there are still some problems in the interpretation . the fact that sne ia light curves are narrower when redder ( nobili & goobar 2008 ) is an inconvenience for a clean test free of selection effects . other selection effects and the possible compatibility of the results with a wider range of cosmological models , including static ones , were also pointed out by lpez - corredoira ( 2003 , section 2 and references therein ) , brynjolfsson ( 2004b ) ; leaning ( 2006 ) ; crawford ( 2009b , 2011 ) ; holushko ( 2012 ) , and laviolette ( 2012 , section 7.8 ) . moreover , neither gamma - ray bursts ( crawford 2009a ) nor qsos ( hawkins 2010 ) present time dilation , which is puzzling . there are , of course , the cmbr anisotropies as a way to test cosmological models ; they are , perhaps , the most important support for the standard model . however , one should find an independent confirmation , because it is possible to generate / modify cmbr anisotropies by mechanisms different than the standard cosmology ( narlikar et al . 2007 ; angus & diaferio 2011 ; lpez - corredoira & gabrielli 2013 ; lpez - corredoira 2013 ) and/or contamination ( lpez - corredoira 2007 ) . the microwave background temperature measured from rotational excitation of some molecules as a function of redshift ( molaro et al . 2002 ; noterdaeme et al . 2011 ) is another possible test and was quite successful in proving the expansion : results from noterdaeme et al . ( 2011 ) with the exact expected dependence of @xmath4 are impressive . nonetheless , there are other results which disagree with that dependence ( krelowski et al . 2012 ; sato et al . . it might be due to a dependence on collisional excitation ( molaro et al . 2002 ) or bias due to unresolved structure ( sato et al . . the temperature of the intergalactic medium as a function of redshift might also be useful to constrain more directly the geometry of the universe . at present , it is typically inferred to be 20,000 k ; there is no evidence of evolution with redshift ( zaldarriaga et al . 2001 ) , which is puzzling in an expanding universe . here , we propose an alcock & paczyski ( 1979 ) cosmological test : an evaluation of the ratio of observed angular size to radial / redshift size ( see detailed explanation in section [ .method ] ) . the main advantage of this test is that it does not depend on the evolution of the galaxies , but only on the geometry of the universe . however , the redshift distortions ( kaiser 1987 ; hamilton 1998 ) do also have an influence . although there already have been many attempts to carry out this test ( see references in section [ .2pcf ] ) , here we adapt the method ( sections [ .method ] and [ .2pcf_calc ] ) and focus on the application as a test for any cosmological model , whereas other authors assume the standard cosmology and just fit some of their parameters . we apply it to three spectroscopic surveys with a total of @xmath5 galaxies up to redshift 0.8 and @xmath6 qsos at higher redshifts ( sections [ .application ] and [ .fitcosmo ] ) . as will be discussed in section [ .conclusions ] , this test will give us information about which cosmological model fits better . hence , this is a cosmological test free of entanglement with evolution , for checking both the geometry and the expansion of the universe . certainly , this exercise is not going to solve the question once and for all , but the method is a possible way to do it with the increasing possibility of future huge spectroscopic surveys . the key point of the test is based on the fact that the two - point correlation function in a distribution of galaxies in the real space must have spherical symmetry , that is , it only depends on the distance between the sources and not on any of the two angles which define the relative position between the two sources . this means that the fall of the two - point correlation function as a function of the distance must be the same either along the line of sight ( @xmath7 ) or perpendicular to it ( @xmath8 ) . the distance of two sources along the line of sight is given by the difference of redshift , @xmath9 , and depends on the cosmological model . the distance of two sources in the direction perpendicular to the line of sight is given by the angular separation , @xmath10 , and also depends on the cosmological model ; this is precisely what gives meaning to the angular - size tests . the comoving distance between two sources at redshift @xmath11 separated by a relatively small @xmath9 and @xmath10 ( radians ) in the redshift - space is ( see appendix [ .cosmomodels ] ) @xmath12 where @xmath13 is the speed of light , @xmath14 is the hubble constant , and @xmath15 and @xmath16 are functions which depend on the cosmological model . for the standard concordance model ( with the equation of state for the dark energy @xmath17 ) : @xmath18 @xmath19 see appendix [ .cosmomodels ] for their dependence in other cosmological models ; they are plotted in figure [ fig : models ] . all models have in common that @xmath20 , but there are already some differences for low z there are some differences ; in appendix [ .cosmomodels ] , we give the limits of @xmath21 for @xmath16 . for high @xmath11 , the differences are even higher . the possibility of deriving @xmath16 free of selection effects , evolution , redshift distortion , etc . , would tell us which theoretical model is correct . furthermore , note that even assuming the standard model as correct , the derivation of @xmath16 would allow us to determine @xmath2 and @xmath22 . for low @xmath11 in the concordance model ( equation ( [ lowz_1 ] ) ) : @xmath23 , so a high precision measurement of @xmath16 in the local universe could give us a direct measurement of @xmath2 ( and consequently @xmath22 , assuming @xmath24 ) . this measurement of @xmath16 is possible , as we will see next . as said , the two - point correlation function of a distribution of galaxies ( @xmath25 ) should only depend on the comoving distance , @xmath26 , in the real space for a given redshift , i.e. , @xmath27 . for the redshift - space ( @xmath28 ) , there is no spherical symmetry because we also have the redshift - space distortions in the field @xmath25 produced by the peculiar velocities of the gravitational infall . in the linear regime of fluctuations , assuming @xmath29 ( this is an acceptable approximation for scales @xmath30 @xmath31mpc , e.g. , ross et al . 2007 ; sylos labini et al . 2009 ; see also figure [ fig : xi0 ] ) with distortion parameter @xmath32 , @xmath33 , \ ] ] where @xmath34 is the cosine of the angle between @xmath35 and @xmath36 ; @xmath37 ; and the function @xmath38 is ( matsubara & suto 1996 ) : @xmath39 hence , in polar coordinates @xmath40 , @xmath41 : @xmath42@xmath43 , \]]@xmath44 where @xmath45 stands for certain amplitude . note that we do not use @xmath46 , like most of the references that talk about the anisotropic two - point correlation function , because we are not assuming any cosmology a priori . the advantage of the expression in equation ( [ xi ] ) is that it relates an observable function totally free of any modeling to the cosmology implicit in the values of @xmath16 , and the redshift - space distortions implicit in @xmath32 . if we wanted to calculate @xmath46 , we would have to obtain @xmath47 and @xmath48 using a given cosmological model and change / rescale this calculation every time we wanted to test another model . in figure [ fig : elipses ] , the effect of the redshift distortion is illustrated . if we had no redshift distortions ( @xmath49 ) , @xmath50 would not depend on @xmath34 and , for a fixed @xmath11 , when we plotted @xmath25 in the plane of @xmath51 , @xmath9 , the isocontours would be ellipses with axial ratios @xmath16 of the second / first axes ( figure [ fig : elipses]c ) . for the general case , we must consider redshift distortions . the parameter @xmath32 takes into account , as said , the large - scale effects of linear @xmath11-space distortions ( hamilton 1998 ) . usually , in a standard scenario , it is related to cosmological information as @xmath52 ( kaiser 1987 ) , where @xmath53 is the bias parameter , but here , in general and for any cosmological scenario , @xmath54 will be simply a parameter from which we will not extract any further information . we must be aware that kaiser s formalism was developed for the standard cosmology including expansion ; however , this remains valid for any model , provided that the meaning of @xmath54 is changed . in general , it is valid when the gravitational infall is given such that @xmath55 , where @xmath56 is the velocity field and @xmath57 the matter overdensity ( in the linear regime ) ; this yields even for a static universe . the continuity and euler equation remain the same thing in all models , but the poisson equation changes with the gravitational model as well as the evolution of the linear growth of overdensities , thus giving rise to different equivalences of @xmath54 with the cosmological parameters . the small - scale random motion of the galaxies should also be included in an analysis with small bin size : we should convolve equation ( [ xi ] ) with a gaussian distribution of @xmath9 ( e.g. , ross et al . 2007 , section 4.2 ) to get the correct @xmath50 ; however , we will neglect this correction since the dispersion of velocities is @xmath58 300 km s@xmath59 ( ross et al . 2007 ) , equivalent to @xmath60 , which is much smaller than our bin sizes . the low values of @xmath10 in which the features known as the `` fingers of god '' ( jackson 1972 ) are observed are not explored here . obtaining @xmath16 for different values of @xmath11 and comparing it with theoretical predictions of different cosmological models ( figure [ fig : models ] ) , perhaps allowing for some variation of @xmath2 instead of a fixed value , is theoretically a good way to constrain cosmology . nonetheless , a difficulty arises from the quasi - degeneracy between the variations of @xmath16 and @xmath32 , i.e. between geometric and dynamic distortions , respectively . the range of possible cosmological models compatible with the data for some free @xmath54 is too large ( e.g. , ross et al . 2007 ) , thus making it difficult to disentangle geometric and dynamic distortions . there are some methods to obtain @xmath32 from @xmath25 ( e.g. , hamilton 1992 , 1998 ; tocchini - valentini et al . 2012 ) or from @xmath61 in a sample of galaxies ( patiri et al . 2012 ) , but they depend on @xmath16 . nonetheless , there are ways to attack the problem . some approaches to our problem try to fit @xmath54 for the different cosmological models [ that is , fixing @xmath16 ] and adopt as valid cosmological the one which gives better fit to the function @xmath25 ( e.g. , marulli et al . 2012 ) . here we carry out something similar : a weighted fit , but as a function of @xmath9 and @xmath51 , instead of @xmath48 and @xmath47 , and , in order to solve better the quasi - degeneracy of @xmath54 and @xmath63 , we set a constraint for @xmath54 to follow @xmath64an amount which is independent of @xmath63 . we exclude from the fit the pixel ( 1,1 ) , because it contains higher nonlinear effects of the correlation and errors from binning . in practice , instead of taking strictly @xmath65 ( equivalent to @xmath66 ) , we take @xmath67 , i.e. , the first column ( for the galaxies samples ) . the effect of random velocities or the `` finger of god '' is negligible in the calculation of this last quantity for our bin sizes of @xmath68 ( for the galaxies samples ) : numerical calculations give errors of @xmath69% in @xmath70 with respect to the inclusion of velocity dispersions in @xmath9 with a gaussian distribution of @xmath71 , which is negligible in comparison with other sources of errors . if we obtained @xmath72 , we would set @xmath73 and recalculate the error bar to get 68% probability in the range @xmath74 within 1@xmath75 . the error bars of @xmath76 , @xmath63 and @xmath54 are given by the values which produce a @xmath77 + 3.5 $ ] ( the value of 3.5 defines the region with a 68% confidence level ( cl ) in a three - free - parameter fit ; avni 1976 ) . note that the error bars of @xmath54 and @xmath63 should include both the transmitted errors for a fixed value of @xmath76 and the variations of those values due to the variation of @xmath76 within the range of its error bar ; we will do so throughout this paper . tests have been carried out with monte carlo simulations and we could see that we recover approximately the introduced parameters . for instance , in figure [ fig : montec ] , we show the recovered values of @xmath78 versus the introduced values of the fit @xmath79 in a monte carlo simulation . the conditions are @xmath80 , different values of @xmath81 , and noise with relative rms equal to @xmath82 with @xmath83 over an area subdivided into 8x8 bins of size 0.0025 . as can be observed , we recover roughly the value of @xmath63 plus some noise . many authors have already realized on the advantages of using the anisotropic correlation function to measure the ratio of observed angular size to radial / redshift size and applied it as a possible way to constrain cosmological models . it was analyzed in ballinger et al . ( 1996 ) , marulli et al . ( 2012 ) , phillips ( 1994 ) for quasar pairs , ryden ( 1995 ) for voids , kim & croft ( 2007 ) for clusters , and nusser ( 2005 ) for radio emission in the epoch of reionization , among others . it was applied to real data of several surveys with galaxies ( blake et al . 2011 ) , luminous red galaxies ( ross et al . 2007 ; okumura et al . 2008 ; chuang & wang 2012 ; reid et al . 2012 ) , lyman - break galaxies ( da ngela et al . 2005b ) , qsos ( hoyle et al . 2002 ; outram et al . 2004 ; da ngela et al . 2005a ; ivashchenko et al . 2010 ) , galaxy pairs ( marinoni & buzzi 2010 ) , cross - correlation of qsos and luminous red galaxies ( mountrichas et al . 2009a ) , cross - correlation of galaxies and cluster of galaxies ( mountrichas et al . 2009b ) , and voids ( sutter et al . 2012 ) , among others . most of these authors analyzed the correlation on small scales ( @xmath84 @xmath85mpc ) and very few of them ( okumura et al . 2008 ; chuang & wang 2012 ) use larger scales . let us suppose we have a survey completely covering some area with some homogeneous criterion . the method to calculate the anisotropic two - point correlation function is as follows . we generate another mock survey in the same area with a random ( poisson ) distribution of galaxies and with the same distribution of redshifts as the data . then we use a generalization of the landy & szalay ( 1993 ) algorithm to get the anisotropic two - point correlation function , @xmath86 @xmath87 where dd , rr , dr and rd are the number of data data , random random , data random and random data pairs , and @xmath88 denotes their normalization to data data , i.e. , @xmath89 ; @xmath90 and @xmath91 are the total number of sources in the data and in the random sample ; @xmath92 and @xmath93 are the total number of sources in the data and in the random sample which contribute as first object in the pair for a given @xmath11 ( within some interval of redshifts ) . notice that @xmath94 and @xmath95 because we remove as first object of the pair the sources which are within an angular distance from a border of the survey lower than @xmath96 . consequently , dr is slightly different from rd : the parent galaxies ( the first character @xmath97 or @xmath98 ) are limited to a zone away from the borders of the selected area , whereas the second character includes all the galaxies of the catalog within a given distance from the parent source . in our case , we take a number of random points equal to the number of sources in the data . however , the result of the algorithm is independent of that number , provided that it is at least as large as the number of data points ( sylos labini et al . 2009 ) and it is high enough to avoid the production of further noise due to sampling errors . the relative error of @xmath99 is equal to the inverse of the root square of the number of pairs ( betancort - rijo 1991 ; ross et al . 2007 ) but , since we manage a huge number of pairs both in the real data and in the random sample , this error is small in comparison to other errors . this number must not be very high in order to have a reasonable computation time to calculate the number of pairs . other estimators of the two - point correlation function give very similar results even at large scales ( e.g. , landy & szalay 1993 ; porciani & norberg 2006 ; ross et al . 2007 ; sylos labini et al . 2009 ) . the error of @xmath50 is evaluated through a field - to - field method ( ross et al . 2007 , section 2.4 ) : @xmath100 ^ 2 } , \ ] ] where @xmath101 is the total number of subsamples ( fields ) , and the index @xmath102 indicates the statistics within the subsample @xmath102 . in our case , we take by default @xmath103 , dividing the covered region of the sky into 10 subregions . although landy & szalay ( 1993 ) claim that the error given by their estimator should be nearly poissonian , the poissonian error significantly underestimates the total error at large distances ( ross et al . 2007 , section 2.4 ) . the jackknife or bootstrap methods provide errors of the same order ( ross et al . 2007 , section 2.4 ) ; however , if there were important large - scale variations of the correlation , the jackknife or bootstrap algorithms would underestimate the total rms measured in the field - to - field algorithm ( lpez - corredoira et al . 2010 ) . for point distributions , there is a component of the total error that is caused by the finiteness of the number of points ( betancort - rijo 1991 ) and is closely related to that given by the resampling techniques such as jackknife or bootstrap . the field - to - field fluctuations also include intrinsic fluctuations in the large - scale structure of galaxies . this technique of field - to - field is equivalent to carrying out several independent realizations , one for each of the subsamples in which is divided the whole , equivalent to analyzing the rms in multiple mock catalogs generated randomly , with the advantage that here we can be pretty sure that the generated subsamples obey perfectly the statistical distribution of the real sky since they are part of the whole real sky , whereas mock catalogs may depend on how the algorithm generates the samples . one may wonder whether the correlations between the different subsamples can change the rms of equation ( [ errxi ] ) with respect to the real one , i.e. , whether the terms of the covariance matrix are significant . the answer is that this effect is negligible provided that @xmath104 ( see appendix [ .2subsamples_tpcf ] ) , which is true in our case . therefore , the field - to - field algorithm seems an approximate method to take all the errors into account in a simple and direct way . using all available galaxies in a catalog would be better in order to gain higher signal - to - noise in the statistics , provided that the criterion of selection of sources does not vary with the position . in principle , this is possible for cosmological purposes , since cosmological parameters are totally independent of the evolution of galaxies or their clustering . the function @xmath16 will only depend on the cosmological model . also , @xmath16 does not depend on the kind of sources examined or the wavelength ; we can even mix different types of sources for different @xmath11 without @xmath105-corrections , so the selection effects ( malmquist bias and others ) will not have any influence on the result . in a limiting - magnitude survey covering some area , we will have more luminous sources at higher redshifts , different observed `` at - rest '' wavelengths , etc . , but it does not matter in the measurement of the geometry of the universe . nonetheless , one must take into account the effect of a variable @xmath106 in the calculation of the two - point correlation function . if the maximum comoving radius , @xmath28 , were very small , and the galaxies within a sphere of radius @xmath107 were similar in their properties , @xmath25 might be calculated as usual in a homogeneous space . but in our case , we must consider some effect of variation for larger @xmath28 due to the variability of @xmath106 within the considered effects , mainly because of the malmquist bias . this variability of the mean density produces a contribution to the two - point correlation function , which must be subtracted in order to see only the contribution from the clustering of galaxies . this correction is @xmath108 where @xmath25 is the measured correlation , and @xmath109 is the correlation given by a random ( i.e. , uncorrelated , locally poisson ) distribution of sources within a field @xmath106 . in appendix [ .varrho ] , we demonstrate that , neglecting terms of @xmath110 with @xmath111 , @xmath112 this correction will be applied to our calculations . the sloan digital sky survey iii ( sdss - iii)/baryon oscillation spectroscopic survey ( boss ; eisenstein et al . 2011 ; ahn et al . 2012 ; bolton et al . 2012 ; smee et al . 2013 ; dawson et al . 2013 ) will target , when finished , 1.5 million galaxies in ten thousand square degrees , @xmath113 , selected in color magnitude space to be high - luminosity systems at large distances . the selection criteria are the union of two cuts designed to select targets in two different redshift intervals . cut i , aimed at the interval @xmath114 , is defined by @xmath115 , @xmath116 , and @xmath117 . cut ii , aimed at redshift @xmath118 , is defined by @xmath119 , @xmath120 , and @xmath121 , where the colors @xmath122 , @xmath123 , and @xmath124 are defined to track a stellar population passively evolving with redshift : @xmath125 , @xmath126 , and @xmath127 . these selection criteria produce a roughly constant comoving space density @xmath128 @xmath129mpc@xmath130 to @xmath131 with a slight peak at @xmath132 followed by a declining space density to @xmath133 . the average accuracy in the redshift determination of these sources is @xmath134 ( dawson et al . the survey has started in the fall of 2009 and is planned to be finished in approximately 2014 july . at present , we use the boss data release 10 version 5 to do statistical analyses of large - scale structure ( anderson et al . 2012 ) with uniform coverage . some masks were applied to avoid regions with some saturated sources or low completeness ( anderson et al . this subsample contains 830,089 galaxies . in this sample , we use the galaxies within @xmath135 ( 793,573 galaxies ) . given the considerations in section [ .vardens ] , we do not need a homogeneous sample with respect to the variation in @xmath11 . some malmquist bias is expected to be introduced , even if the sources were selected to be kept almost constant for some range of redshifts , but our procedure can use samples with variable @xmath106 . nonetheless , in order to assure homogeneity in the angular scale and avoid border effects , we select as first object in pairs which contribute to the two - point correlation function only those of them whose circles of angular radii @xmath136 radians are totally covered ; indeed , for more security , we require them to be totally covered within a distance of @xmath137 in the sdss sample . thus , the number of parent galaxies ( first one of the galaxies in a pair for the correlation calculation ) is 456,451 . these are divided into three redshift bins , as is described in table [ tab : ybg ] . we also take into account that the minimum separation between fibers is sep@xmath138 radians , so we must add @xmath139 to the value of the mean @xmath51 for the lowest @xmath140 bin , which is negligible in our scales . previous analyses of the anisotropic correlation function with previous releases of boss data were also carried out by white et al . ( 2011 ) and reid et al . ( 2012 ) , although with different scales , methods and purposes from the present paper . [ cols="^,^,^,^,^,^,^,^",options="header " , ] we observe in table [ tab : chi2 ] that some @xmath141 are much lower than unity for @xmath142 , more than would be expected statistically . this indicates two things : ( 1 ) the previously mentioned fact that the 18 points are not totally independent , and ( 2 ) the well - known fact that the scientific community tends to publish results with a dispersion much lower than expected statistically from the error bars , which means that either the error bars were overestimated , or there is a bias in the publication of results towards the preferred value ( croft & dailey 2011 ) . we also find some probabilities which are lower than 0.05 ( excluded at a 95% cl ) or even lower than 0.01 ( excluded at a 99% cl ) . in figure [ fig : models2 ] , we plot all the values of @xmath16 , including our values and those ones from the literature , as well as the different cosmological models . in tables [ tab : chi2_7 ] and [ tab : chi2_11 ] , we also show the same analysis with the subsamples of data from our @xmath143 points ( table [ tab : ybg ] ) and separately the analysis derived from the @xmath144 points from the literature ( table [ tab : yrefs ] ) . we see that the probabilities are very low for some cases in the subsamples of 11 points or the whole sample of 18 points ; the subsample of 7 points is not conclusive but it shows the same trends in the probabilities , which indicates that the exclusion of some cosmological models is robust . the application of the alcock paczyski test using the anisotropic two - point correlation function was shown here to be a useful tool for discriminating among different cosmological models without any a priori assumptions . we have seen that the major caveat is the disentanglement with the redshift distortions produced by the peculiar velocities of gravitational infall , but it is still possible to separate both effects . the error bars are large for the determination of the ratio of observed angular size to radial / redshift size , @xmath16 ; however , with large collections of data from many surveys , it is possible to get some constraints in cosmological models . we have applied our method to three surveys within sdss : galaxies from sdss - iii / boss , and qsos from both sdss - ii and sdss - iii / boss . we have not used the survey of galaxies from sdss - ii because most of their galaxies are at @xmath145 , and the condition of completeness within a radius @xmath136 would reduce enormously the sample of parent galaxies , producing a result much poorer than with boss galaxies . the alternative of exploring smaller distances in the correlation would alleviate this constraint of completeness areas ; however , it would introduce the problem of the `` fingers of god '' , which we have not taken into account here because they are expected to produce a negligible effect at the scales we were using . furthermore , we have used some estimations of @xmath16 given directly or indirectly by the literature , leading to the results in table [ tab : chi2 ] or in figure [ fig : models2 ] . we observe that there are four cosmological models excluded within @xmath1% cl : einstein de sitter , open friedman cosmology without dark energy , flat quasi - steady - state cosmology ( qssc ) , and a static universe with a linear hubble law . two models fit the data with a higher probability : concordance cosmology , and static universe with tired light redshift . for the standard concordance model ( @xmath146 ) , we have set constraints on @xmath2 to produce a better fit to the data ; they are : @xmath3 , which is in agreement with values derived from other independent methods . note , however , that , since we have not used any a priori assumptions , neither about the redshift distortion model nor other constraints including other data , the error bars here are larger than those derived from other methods e.g . , a recent analysis of cmbr data with _ planck _ gives @xmath147 ( planck collaboration 2013 ) . if we also allow a free value for the equation of state of the dark energy , @xmath148 , instead of a fixed value of -1 , we get as a best fit : @xmath149 , @xmath150 . certainly , our method here is not the best one to explore the possible values of the cosmological parameters once the cosmology is established ; this could be much improved with further constraints , as ross et al . ( 2007 ) and okumura et al . ( 2008 ) did , but this was not our goal in this paper . we have focused on a method to test cosmological models without any a priori assumption . the qssc model is excluded at a @xmath1% cl , but the fit is better ( excluded with a probability of 98.0% instead of 99.56% ) with parameters @xmath151 and @xmath152 , which lead to @xmath153 . this is compatible with the previously established value for the @xmath154-field density of @xmath155 ( banerjee & narlikar 1999 , with the angular redshift test ) or @xmath156 ( banerjee et al . 2000 , with sne data ) . the main problem with these fitted values is that they lead to the maximum redshift of an observable galaxy at @xmath157 , and we know there are galaxies beyond that . anyway , a flat qssc appears to have a very low probability to be a valid model according to only the present analysis . all these considerations are for a flat qssc universe ; if we allowed curvature ( @xmath158 ) , a wider range of probabilities could be obtained . the static models are not totally excluded from the present analysis . the one with a linear hubble law is excluded ( assuming , as we did , that the linear regime applies and that the peculiar velocity of the galaxies is irrotational , i.e. , with zero vorticity , as predicted by the usual gravitational growth theories ) , but not the one with tired - light hypothesis to explain the redshift . this tired - light model is also supported by some other tests , such as the angular size test ( lpez - corredoira 2010 ; laviolette 2012 , 7.4 ) or differential galaxy number counts versus magnitude ( laviolette 2012 , 7.7 ) . fitting the hubble diagram as @xmath159 , with @xmath160 being the luminosity distance , is not good with sne data using a `` simple '' tired light assumption ( where @xmath161 ; the factor @xmath162 stems from the fact that the luminosity is proportional to @xmath163 and is inversely proportional to @xmath164 due to the redshift without expansion , i.e. , without time dilation ) . however , the fit is acceptable using a `` plasma '' tired light redshift ( @xmath165 ; lpez - corredoira 2010 ) , and it is quite good with gamma - ray bursts even up to @xmath166 with a tired - light model of type @xmath167 ( marosi 2013 ) . future spectroscopic surveys with significantly more sources , coverage , and depth , will be able to improve the present results . in particular , at high redshift we still have few spectroscopic qsos , so the error bars are large ( see how noisy the correlations are in figs . [ fig : elipses2q ] , [ fig : elipses2q2 ] ) . one exception is the point from da ngela et al . ( 2005b ) at @xmath168 : @xmath169 ( note , however , that , looking at figure 5 of da ngela et al . ( 2005b ) , one could say that @xmath170 is quite likely within 1@xmath75 , and consequently the error bars might be somewhat larger ) . as it can be observed in figure [ fig : models ] , accurate values of @xmath63 at @xmath171 will be able to discern between the two successful cosmological models in this paper : @xmath0cdm ( concordance ) and static model with tired - light redshift . also , very accurate measurements of @xmath63 at low @xmath11 would be able to test which one is the correct , since @xmath23 for the concordance model , whereas @xmath172 for the tired - light one . it is also worth noting that the most important contribution stems from the already available data from previous analyses of the anisotropic correlation function , combining the results of six references . the contribution of the data ( @xmath143 ) analyzed in this paper does not change significantly the confidence levels of exclusion of the different cosmological models because we get higher error bars . note , however , that we can not guarantee that the results of those six references with somewhat lower error bars are correct . in principle , from the experience developing this paper , the conclusion that comes up is that the quasi - degeneracy between the values of @xmath54 and @xmath63 for the fit of @xmath50 is a severe problem which produces huge error bars of @xmath63 and , consequently , the alcock & paczyski method only becomes powerful when we join the results from different surveys . thanks are given to : juan e. betancort - rijo ( iac , tenerife , spain ) for discussions on the methodology of the alcock - paczyski test , donald schneider ( sdss - iii scientific publications coordinator ) for his recommendations to bring the paper into compliance with the sdss - iii publications policy , and to the anonymous referee for helpful comments . funding for the 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 web site is http://www.sdss.org/. the sdss is managed by the astrophysical research consortium for the participating institutions . the participating institutions are the american museum of natural history , astrophysical institute potsdam , university of basel , university of cambridge , case western reserve university , university of chicago , drexel university , fermilab , the institute for advanced study , the japan participation group , johns hopkins university , the joint institute for nuclear astrophysics , the kavli institute for particle astrophysics and cosmology , the korean scientist group , the chinese academy of sciences ( lamost ) , los alamos national laboratory , the max - planck - institute for astronomy ( mpia ) , the max - planck - institute for astrophysics ( mpa ) , new mexico state university , ohio state university , university of pittsburgh , university of portsmouth , princeton university , the united states naval observatory , and the university of washington . funding for sdss - iii has been provided by the alfred p. sloan foundation , the participating institutions , the national science foundation , and the u.s . department of energy office of science . the sdss - iii web site is http://www.sdss3.org/. sdss - iii is managed by the astrophysical research consortium for the participating institutions of the sdss - iii collaboration including the university of arizona , the brazilian participation group , brookhaven national laboratory , university of cambridge , carnegie mellon university , university of florida , the french participation group , the german participation group , harvard university , the instituto de astrofsica de canarias , the michigan state / notre dame / jina participation group , johns hopkins university , lawrence berkeley national laboratory , max planck institute for astrophysics , max planck institute for extraterrestrial physics , new mexico state university , new york university , ohio state university , pennsylvania state university , university of portsmouth , princeton university , the spanish participation group , university of tokyo , university of utah , vanderbilt university , university of virginia , university of washington , and yale university . 999 ahn , c. p. , alexandroff , r. , allende prieto , c. , et al . 2012 , apjs , 203 , 21 alcock , c. , & paczyski , b. 1979 , nature , 281 , 358 anderson , l. , aubourg , e. , bailey , s. , et al . 2012 , mnras , 427 , 3435 andrews , t. b. 2006 , in : first crisis in cosmology conference ( aip conf . 822 ) , e. j. lerner , j. b. almeida , eds . , american institute of physics , p. 3 angus , g. w. , & diaferio , a. 2011 , mnras , 417 , 941 ashmore , l. 2011 , intrinsic plasma redshifts now reproduced in the laboratory a discussion in terms of new tired light , http://vixra.org/pdf/1105.0010 avni , y. 1976 , apj , 210 , 642 ballinger , w. e. , peacock , j. a. , & heavens , a. f. 1996 , mnras , 282 , 877 banerjee , s. k. , & narlikar , j. v. 1999 , mnras , 307 , 73 . erratum : banerjee , s. k. , & narlikar , j. v. 1999 , mnras , 310 , 912 banerjee , s. k. , narlikar , j. v. , wickramasinghe , n. c. , hoyle , f. , & burbidge , g. 2000 , aj 119 , 2583 baryshev y. , & teerikorpi p. 2012 , fundamental questions of practical cosmology , springer , dordrecht betancort - rijo , j. 1991 , a&a , 245 , 347 blake , c. , glazebrook , k. , davis , t. m. , et al . 2011 , mnras 418 , 1725 blanchard , a. 2006 , in : current issues in cosmology , j .- c . pecker , and j. v. narlikar , ed . , cambridge university press , cambridge ( u.k . ) , p. 76 blondin , s. , davis , t. m. , krisciunas , k. , et al . 2008 , apj , 682 , 724 bolton , a. s. , schlegel , d. j. , aubourg , . , 2012 , aj , 144 , 144 brynjolfsson , a. 2004a , arxiv : astro - ph/0401420 brynjolfsson , a. 2004b , arxiv : astro - ph/0406437 chuang , c. h. , & wang , y. 2012 , mnras , 426 , 226 crawford , d. 2009a , arxiv:0901.4169 crawford , d. 2009b , arxiv:0901.4172 crawford , d. 2011 , j. cosmology , 13 , 3875 croft , r. a. c. , & dailey , m. 2011 , arxiv:1112.3108 da ngela , j. , outram , p. j. , shanks , t. , boyle , b. j. , croom , s. m. , loaring , n. s. , miller , l. , & smith , r. j. 2005a , mnras , 360 , 1040 da ngela , j. , outram , p. j. , & shanks , t. 2005b , mnras , 361 , 879 dawson , k. s. , schlegel , d. j. , ahn , c. p. , et al . 2013 , aj , 145 , i d . 10 eisenstein , d. j. , weinberg , d. h. , agol , e. , et al . 2011 , aj , 142 , 72 goldhaber , g. , groom , d. e. , kim , a. , et al . 2001 , apj , 558 , 359 gunn , j. e. , carr , m. , rockosi , c. , et al . 1998 , aj , 116 , 3040 gunn , j. e. , siegmund , w. a. , mannery , e. j. , et al . 2006 , aj , 131 , 2332 hamilton , a. j. s. 1992 , apj , 385 , l5 hamilton , a. j. s. 1998 , in : the evolving universe . selected topics on large - scale structure and on the properties of galaxies , d. hamilton , ed . , kluwer , dordrecht , p. 185 hawkins , m. r. s. , 2010 , mnras , 405 , 1940 holushko , h. 2012 , tired light and type ia supernovae observations , http://vixra.org/pdf/1203.0062 hoyle , f. , outram , p. j. , shanks , t. , boyle , b. j. , croom , s. m. , & smith , r. j. 2002 , mnras , 332,311 ivashchenko , g. , zhdanov , v. i. , & tugay , a. v. 2010 , mnras , 409 , 1691 jackson , j. c. 1972 , mnras , 156 , 1 kaiser , n. 1987 , mnras , 227 , 1 kapahi , v. k. 1987 , observational cosmology ( iau symp . 124 ) , a. hewitt , g. burbidge , and l. z. fang ( eds . ) , reidel , dordrecht , 251 kellerman , k. i. 1993 , nature , 361 , 134 kim , y .- r . , & croft , r. a. c. 2007 , mnras , 374 , 535 krelowski , j. , galazutdinov , g. , gnacinski , p. 2012 , an , 333 , 627 landy , s. d. , & szalay , a. s. 1993 , apj , 412 , 64 laviolette , p. a. 1986 , apj 301 , 544 laviolette , p. a. 2012 , subquantum kinetics : the alchemy of creation , 4th ed . , starlane publ . , niskayana ( ny , us ) leaning , s. p. 2006 , in : 1st crisis in cosmology conference ( aip conf . 822(1 ) ) , e. j. lerner , j. b. almeida , eds . , aip , melville , p. 48 lehnert , m. d. , nesvadba , n. p. h. , cuby , j .- , et al . 2010 , nature , 467 , 940 lerner , e. j. 2006 , in : 1st crisis in cosmology conference ( aip conf . 822(1 ) ) , e. j. lerner , j. b. almeida , eds . , aip , melville , p. 60 lpez - corredoira , m. 2003 , in : recent research developments in astronomy and astrophysics i , ed . s. g. pandalai , research signpost , kerala , p. 561 lpez - corredoira , m. 2006 , apeiron , newswire/2006 [ arxiv : astro - ph/0701061 ] lpez - corredoira , m. 2007 , j. astrophys . astron . , 28 , 101 lpez - corredoira , m. 2010 , int . d , 19 , 245 lpez - corredoira , m. 2013 , int . d , 22 , 1350032 lpez - corredoira , m. , & gabrielli , a. 2013 , physica a , 392 , 474 lpez - corredoira , m. , sylos - labini , f. , & betancort - rijo , j. 2010 , a&a , 513 , a3 lubin , l. m. , & sandage , a. 2001 , aj , 122 , 1084 marinoni , c. , & buzzi , a. 2010 , nature , 468 , 539 marosi , l. a. 2013 , advances in astronomy , 2013 , i d . 917104 marulli , f. , bianchi , d. , branchini , e. , guzzo , l. , moscardini , l. , & angulo , r. e. 2012 , mnras , 426 , 2566 matsubara , t. , & suto , y. 1996 , apj , 470 , l1 molaro , p , levshakov , s. a. , dessauges - zavadsky , m. , & dodorico , s. 2002 , a&a , 381 , l64 mountrichas , g. , sawangwit , u. , shanks , t. , croom , s. m. , schneider , d. p. , myers , a. d. , & pimbblet , k. 2009a , mnras , 394 , 2050 mountrichas , g. , sawangwit , u. , & shanks , t. 2009b , mnras , 398 , 971 narlikar , j. v. 1989 , space science reviews , 50 , 523 narlikar , j. v. , vishwakarma , r. g. , & burbidge , g. 2002 , pasp , 114 , 1092 narlikar , j. v. , burbidge , g. , & vishwakarma , r. g. 2007 , j. astrophys . astron . , 28 , 67 noterdaeme , p. , petitjean , p. , srianand , r. , ledoux , c. , & lpez , s. 2011 , a&a , 526 , l7 nusser , a. 2005 , mnras , 364 , 743 okumura , t. , matsubara , t. , eisenstein , d. j. , kayo , i. , hikage , c. , szalay , a. s. , & schneider , d. p. 2008 , apj , 676 , 889 outram , p. j. , shanks , t. , boyle , b. j. , croom , s. m. , hoyle , f. , loaring , n. s. , miller , l. , & smith , r. j. 2004 , mnras , 348 , 745 pris , i. , petitjean , p. , aubourg , . , et al . 2012 , a&a , 548 , a66 pris , i. , petitjean , p. , aubourg , . , et al . 2013 , arxiv:1311.4870 pashchenko , i. n. , & vitrishchak , v. m. 2011 , astronomy reports , 55(4 ) , 293 patiri , s. g. , betancort - rijo , j. , & prada , f. 2012 , a&a , 541 , l4 phillips , s. 1994 , mnras , 269 , 1077 planck collaboration 2013 , arxiv:1303.5076 porciani , c. , & norberg , p. 2006 , mnras , 371 , 1824 rassat , a. , & refregier , a. 2012 , a&a , 540 , a115 reid , b. a. , samushia , l. , white , m. , et al . 2012 , mnras , 426 , 2719 ross , n. p. , da ngela , j. , shanks , t. , et al . 2007 , mnras , 381 , 573 ryden , b. s. 1995 , apj , 452 , 25 sato , m. , reid , m. j. , menten , k. m. , & carilli , c. l. 2013 , apj , 764 , 132 schade , d. , barrientos , l. f. , & lpez - cruz , o. 1997 , apj , 477 , l17 schneider , d. p. , richards , g. t. , hall , p. b. 2010 , aj , 139 , 2360 smee , s. , gunn , j. e. , uomoto , a. , et al . 2013 , aj , 146 , 32 sutter , p. m. , lavaux , g. , wandelt , b. d. , & weinberg , d. h. 2012 , apj , 761 , 187 sylos labini , f. , vasilyev , n. l. , baryshev , y. v. , & lpez - corredoira , m. 2009 , a&a , 505 , 981 tocchini - valentini , d. , barnard , m. , bennett , c. l. , & szalay , a. s. 2012 , apj , 757 , 131 vauclair , s. c. , blanchard , a. , sadat , r. , et al . 2003 , a&a , 412 , l37 vishwakarma , r. g. 2002 , mnras , 331 , 776 white , m. , blanton , m. , bolton , a. , et al . 2011 , apj , 728 , 126 white , m. , myers , a. d. , ross , n. p. , et al . 2012 , mnras , 424 , 933 york , d. g. , adelman , j. , anderson , j. e. , jr . , 2000 , aj , 120 , 1579 zaldarriaga , m. , hui , l. , & tegmark , m. 2001 , apj , 557 , 519 zehavi , i. , weinberg , d. h. , zheng , z. , et al . 2004 , apj , 608 , 16 the comoving distance between two sources separated by a relatively small @xmath9 and @xmath10 ( radians ) in the redshift space is @xmath173}{dz}\ \delta z\right]^2\ + \ [ ( 1+z)^md_{\rm a}(z)\ \delta \theta ] ^2}\]]@xmath174 where @xmath175 and @xmath176 are our angular and comoving distances , respectively , from the first source , @xmath177 with expansion or 0 without expansion , and @xmath178}{dz } , \ ] ] @xmath179 with @xmath14 as the hubble constant , and @xmath13 as the speed of light . here we give these dependences for some cosmological models . 1 . standard concordance model : @xmath177 , @xmath180 @xmath181 where @xmath2 and @xmath22 are , respectively , the density of matter and dark energy and @xmath148 is the dimensionless number associated with the equation of state of the dark energy . we assume that these parameters do not vary with @xmath11 and that the universe is flat , i.e. , @xmath24 . hence , + @xmath182 + @xmath183 + for very low @xmath11 , @xmath184z \label{lowz_1 } .\ ] ] + for the standard model , we adopt the values of @xmath185 , @xmath186 , and @xmath17 . einstein de sitter model ( equation ( [ concordance ] ) with @xmath187 , @xmath188 , and @xmath189 ) : @xmath177 , @xmath190 , \ ] ] @xmath191 @xmath192 .\ ] ] for very low @xmath11 , @xmath193 + although this is not the standard model nowadays , there are some researchers who still consider it more appropriate than the concordance model ( e.g. , vauclair et al . 2003 ; blanchard 2006 ; andrews 2006 ) . regarding the compatibility with type ia sne data , see discussion in lpez - corredoira ( 2010 , 5.3 ) . 3 . friedmann model of negative curvature with @xmath185 and @xmath194 , which implies a curvature term @xmath195 ( baryshev & teerikorpi 2012 , 7.4.1 ) : @xmath177 , @xmath196 @xmath197@xmath198 @xmath199 @xmath200@xmath201 .\ ] ] for very low @xmath11 , @xmath202 + this model might be considered for the case where we accept that @xmath203 , but we do not want to include any @xmath0 term . quasi - steady state cosmology ( qssc ) , @xmath204 , @xmath205 , and @xmath206 ( @xmath154-field density ; banerjee & narlikar 1999 ) : @xmath177 , @xmath207@xmath208 @xmath209 @xmath210 for very low @xmath11 , @xmath211 + this cosmology is not the standard model , but it can also fit many data on some cosmological tests ( banerjee & narlikar 1999 ; banerjee et al . 2000 ; narlikar et al . 2002 ; vishwakarma 2002 ) . the expansion with an oscillatory term gives a dependence of the luminosity and angular distance similar to the standard model , adding the effect of matter creation ( @xmath154-field ) with slight changes depending on the parameters . the parameters of this cosmology are not as well constrained as those in the standard model . here , i use the best fit for a flat ( @xmath212 ) cosmology given by banerjee & narlikar ( 1999 ) : @xmath204 , @xmath213 , and @xmath155 , which corresponds to @xmath214 ( amplitude of the oscillation relative to 1 ) , @xmath215 ( ratio between the actual size of the universe and the average size in the present oscillation ) , and the maximum allowed redshift of a galaxy @xmath216 ( note , however , that the maximum observed redshift has risen above 8 nowadays according to some authors , e.g. , lehnert et al . other preferred sets of parameters give results that are close . the values most used are @xmath212 , @xmath217 , @xmath218 , and @xmath219 ( banerjee et al . 2000 ; narlikar et al . 2002 ; vishwakarma 2002 ; narlikar et al . 2007 ) , which imply @xmath220 and @xmath156 , but i avoid them because they do not allow galaxies to be fitted with @xmath221 . parameters with a curvature different from zero ( @xmath158 ) also give results that are very close in the angular size test ( banerjee & narlikar 1999 ) . static euclidean model with linear hubble law for all redshifts : @xmath222 , @xmath223 @xmath224 @xmath225 + these simple relations indicate that redshift is always proportional to angular distance , i.e. , a hubble law . we assume in this scenario that the universe is static . the caveat is to explain the mechanism as different from the expansion or the doppler effect , which gives rise to the redshift . this cosmological model is not a solution that has been explored theoretically or mathematically . however , from a phenomenological point of view , we can consider this relationship between distance and redshift as an ad hoc extrapolation from the observed dependence on the low - redshift universe . tired - light static euclidean model : @xmath222 , @xmath226 + @xmath227 + @xmath228 .\ ] ] + for very low @xmath11 , @xmath229 + this is again a possible ad hoc phenomenological representation which stems from considering that the photons lose energy along their paths due to some interaction , and the relative loss of energy is proportional to the length of that path ( e.g. laviolette 1986 ; brynjolfsson 2004a , section 5.8 ; ashmore 2011 ; laviolette 2012 , section 7.3 ) , i.e. , @xmath230 . given two rectangular adjacent regions where we want to calculate the two - point correlation function of parent objects embedded in each region within an angular distance @xmath10 , we want to calculate the number of common pairs in both subsamples . both regions have a width , @xmath231 , and height , @xmath232 , and they are in contact through the vertical divisory line , as shown in figure [ fig:2subsamples_tpcf ] . given any parent source in the left box at any height with completely covered circle , the probability that the second object of the pair within an angular distance @xmath10 is in the right box is : @xmath233 where @xmath234 is the fraction of the circle in the right box , according to figure [ fig:2subsamples_tpcf ] : @xmath235@xmath236 hence , @xmath237 therefore , the number of common pairs separated by an angular distance @xmath10 is @xmath238 . this gives us the degree of dependence of the calculation of the two - point correlation function in both adjacent subsamples . if we took the horizontal direction as the right ascension and the vertical direction as the declination ( as will be done in this paper ) , and the whole sky is divided into @xmath101 subsamples , each pair of adjacent subsamples would have an average @xmath239 . the subregions would not be rectangular but slices of constant r.a . width ; this can be translated into an average @xmath240 ( @xmath241 stems from @xmath242 ) . thus , the average number of common pairs in adjacent regions would be @xmath243 for @xmath244 , there are only common pairs among the adjacent subsamples , not with the regions that are not in contact ; therefore if we select two random subsamples , the probability of a pair to have a common pair in both subsamples is @xmath245 multiplied by a factor @xmath246 ( 2 is the number of adjacent regions , and @xmath247 are the total number of subsamples except the first selected one ) . we neglect the pairs near the polar caps , which are very few in comparison with the rest of the pairs ; in the application of the present paper , there are no sdss galaxies near the polar cap so their contribution is almost null . therefore , the degree of correlation among subsamples ( which would generate some terms in the covariance matrix ) is roughly ( we approximate @xmath248 for high values of @xmath101 ) @xmath249 an amount which is much lower than the one for low @xmath10 ; it can be considered negligible ( @xmath250 ) for @xmath251 . from the definition , a poisson distribution of sources with constant mean density in the space gives a null two - point correlation function . but if there is some variation of the mean density with redshift , even locally preserving the random / poisson distribution of sources , it will give a non - null contribution :

we study mutual dissociation of heavy nuclei in peripheral collisions at ultrarelativistic energies . earlier this process was proposed for beam luminosity monitoring via simultaneous registration of forward and backward neutrons in zero degree calorimeters at relativistic heavy ion collider . electromagnetic dissociation of heavy ions is considered in the framework of the weizscker - williams method and simulated by the reldis code . photoneutron cross sections measured in different experiments and calculated by the gnash code are used as input for the calculations of dissociation cross sections . the difference in results obtained with different inputs provides a realistic estimation for the systematic uncertainty of the luminosity monitoring method . contribution to simultaneous neutron emission due to grazing nuclear interactions is calculated within the abrasion model . good description of cern sps experimental data on au and pb dissociation gives confidence in predictive power of the model for auau and pbpb collisions at rhic and lhc . e - mail : [email protected] + e - mail : [email protected] + enea guest researcher + pacs : 25.75.-q , 25.20.-x , 29.27.-a + key words : ultrarelativistic heavy ions , photonuclear reactions , beams in particle accelerators +

nucl-th0101035

the study of a new form of strongly interacting matter , the so - called quark - gluon plasma , is at the core of current and future experimental programs at relativistic heavy ion collider ( rhic ) at brookhaven national laboratory ( bnl ) @xcite and large hadron collider ( lhc ) at cern @xcite . although colliders give well - known advantages compared to the fixed target experiments , the kinematics of ultrarelativistic heavy - ion collisions at colliders creates certain complications in the beam monitoring as well as in the identification of collision events . due to the geometrical factor @xmath3 , where @xmath4 is the impact parameter , the number of central nuclear collisions ( @xmath5 ) is relatively very small in the whole set of the collisions with nuclear overlap , @xmath6 ( @xmath7 and @xmath8 are the nuclear radii ) . moreover , in peripheral collisions without direct overlap of nuclear densities , @xmath9 , one or both nuclei may be disintegrated by the long - range electromagnetic forces . this process of electromagnetic dissociation ( ed ) is a well - known phenomenon @xcite . the properties of central and peripheral collisions are very different and then should be studied separately . the ed events are less violent than the collisions with the participation of strong interactions . namely , the average particle multiplicities are essentially lower @xcite and the main part of nucleons and mesons is produced in projectile and target fragmentation regions , very far from the mid - rapidity region . calculations show @xcite that the ed cross section in collisions of heavy nuclei at rhic and lhc by far exceeds the dissociation cross section due to the direct nuclear overlap . in @xmath10 and @xmath11 collisions at such energies many neutrons can be produced in the ed process @xcite . among other interesting phenomena one may expect a complete disintegration of nuclei induced by the electromagnetic fields of collision partners @xcite . this phenomenon is very well known in nuclear reactions under the name of `` multifragmentation '' @xcite . several operational problems of heavy - ion colliders are connected with the high rate of the ed process . on the one hand , the ed process reduces the lifetimes of heavy ion beams in colliders as compared with the proton - proton accelerator mode @xcite . on the other hand , the process of simultaneous neutron emission from the collision partners , where the ed process plays a dominant role , can be useful for the luminosity monitoring @xcite . the luminosity monitoring method based on mutual dissociation has several advantages @xcite . in particular , the beam - residual - gas interaction events can be strongly suppressed in favour of the beam - beam events by the condition that a pair of neutrons should be detected in coincidence by each arm of the calorimeter . the cross section of mutual neutron emission can be calculated in the framework of conventional theoretical models designed for describing the heavy ion disintegration in peripheral collisions . corresponding nuclear data , especially photoneutron emission cross sections , may be used as numerical input for such calculations . therefore , the neutron counting rates in zero degree ( very forward ) calorimeters may provide an accurate measure for the heavy - ion collider luminosity . in the present paper the neutron emission in peripheral collisions of ultrarelativistic heavy ions is considered with the aim of providing the theoretical basis for the luminosity monitoring method proposed in refs . the uncertainties in results originating from uncertainties in input nuclear data and in the theoretical model itself are carefully examined . a brief review of corresponding photonuclear data is given with special attention to the publications describing data evaluation and re - measurement . model predictions for the @xmath12 and @xmath13 fragmentation cross sections are compared with recent experimental data obtained in fixed target experiments at cern sps with the highest energies available thus far . this serves as an important test before extrapolating our methods to the rhic and lhc energies . the electromagnetic excitation of one of the collision partners , @xmath14 , followed by its dissociation is schematically shown in fig . [ em ] . in such a process another partner , @xmath15 , emits a photon , but remains in the ground state without any nuclear excitation . besides this `` classical '' process , one can consider a non - classical process where the emission of a photon is accompanied by the nuclear excitation ( see fig . [ emfirst ] ) , particularly , the giant resonance excitation . such lowest - order contribution to the simultaneous ( mutual ) excitation of the nuclei @xmath15 and @xmath14 was considered in refs . also the correction to photon - photon luminosity function due to the inelastic photon emission was considered in ref . @xcite for @xmath16 fusion reactions . as shown in refs . @xcite ( see also the discussion in ref . @xcite ) , the lowest order process of simultaneous excitation of the collision partners has a small cross section . for the cases of interest , i.e. for @xmath10 and @xmath11 collisions , the cross sections for the simultaneous dipole - dipole excitation of such nuclei are 0.49 and 0.54 mb , respectively @xcite . using a rough estimation of ref . @xcite , @xmath17 mb , one can get for the same nuclei 0.39 and 0.43 mb , respectively . from the following discussion one will see that for heavy nuclei these first order contributions ( fig . [ emfirst ] ) are negligible compared to the second order ones ( fig . [ emsecond ] ) . the latter leading order process with exchange of two photons is a classically allowed mechanism . it is considered in the next section where the formalism previously used in ref . @xcite is extended to the case of mutual excitation . let us consider a collision of heavy ultrarelativistic nuclei at impact parameter @xmath9 . the masses and charges of these nuclei are denoted as @xmath15 , @xmath18 and @xmath14 , @xmath19 , respectively . hereafter the case of equal nuclei ( @xmath20 , @xmath21 and @xmath22 ) is investigated . nevertheless , in some cases the indexes are used to show explicitly which of the collision partners emits or absorbs photons . according to the weizscker - williams ( ww ) method @xcite , the impact of the lorentz - boosted coulomb field of the nucleus @xmath15 on @xmath14 is treated as the absorption of an equivalent photon by the nucleus @xmath14 ( see fig . [ em ] ) . in the rest frame of this nucleus the spectrum of virtual photons from the collision partner @xmath15 at impact parameter @xmath4 is expressed as : @xmath23 here @xmath24 is the fine structure constant , @xmath25 is an argument of the modified bessel functions of zero and first orders , @xmath26 and @xmath27 , @xmath28 and @xmath29 is the lorentz factor of the moving charge @xmath18 . if the lorentz factor of each heavy - ion beam is @xmath30 , then @xmath31 for the case of collider . hereafter the natural units are used with @xmath32 . the mean number of photons absorbed by the nucleus @xmath14 in the collision at impact parameter @xmath4 is defined by : @xmath33 where the appropriate total photoabsorption cross section , @xmath34 is used . for @xmath35 one usually takes the neutron emission threshold , while the upper limit of integration is @xmath36 . we assume that the probability of multiphoton absorption is given by the poisson distribution with the mean multiplicity @xmath37 defined by eq . ( [ mb ] ) . following refs . @xcite , we express the cross section for the electromagnetic dissociation of _ one _ of the nuclei due to the absorption of a _ single _ photon ( fig . [ em ] ) leading to a certain dissociation channel @xmath38 as : @xmath39 where the probability of dissociation at impact parameter @xmath4 is defined as : @xmath40 and @xmath41 is the branching ratio for the considered channel @xmath38 in the absorption of a photon with energy @xmath42 on nucleus @xmath14 . the choice of a critical impact parameter @xmath43 , which separates the domains of nuclear and electromagnetic interactions , will be discussed in sec . [ parameters ] . let us turn now to the mutual dissociation process shown in fig . [ emsecond ] . the corresponding graph may be constructed from two graphs of the single dissociation by interchanging the roles of `` emitter '' and `` absorber '' at the secondary photon exchange . several assumptions have to be made to obtain an expression for the mutual dissociation cross section . firstly , we suppose that the emitted photon with energy @xmath44 does not change essentially the total energy , @xmath45 , of the emitting nucleus , where @xmath46 is the nuclear mass . this can be justified by estimating the ratio @xmath47 which is close to @xmath48 for heavy nuclei . therefore , the kinematical conditions for the secondary photon exchange are very similar to those for the primary one , and there are no correlations between the energies of the primary and secondary photons , @xmath42 and @xmath49 . in other words , the primary and secondary photon exchanges may be considered as independent processes even if they take place in the same collision during a short - term overlap of the lorentz - contracted coulomb fields of the colliding nuclei . secondly , the equivalent photon spectrum from the excited nucleus , @xmath50 in the notations of fig . [ emsecond ] , is the same as the spectrum from the nucleus in its ground state , @xmath14 . this follows from the fact that at ultrarelativistic energies the collision time is much shorter then the characteristic deexcitation time during which a nucleus changes its initial charge via proton emission or fission . following these assumptions , one can express the cross section for the _ mutual _ dissociation of nuclei @xmath15 and @xmath14 ( fig . [ emsecond ] ) to channels @xmath38 and @xmath51 , respectively , as : @xmath52 substituting eq . ( [ p1 ] ) for each of the nuclei and changing the order of integration one obtains : @xmath53 where the spectral function @xmath54 for mutual dissociation is introduced : @xmath55 conditions @xmath20 and @xmath21 were used in eqs . ( [ smfull ] ) and ( [ nm ] ) , as it is usually in heavy - ion colliders , and therefore @xmath56 . nevertheless , the dissociation channels @xmath38 and @xmath51 may be different for each of the nuclei even in such case . several remarks may be made concerning eqs . ( [ smfull ] ) and ( [ nm ] ) . compared to another process of the second order dissociation of a single nucleus ( see fig . [ em2 ] ) some points of similarity may be found . indeed , expressions given in ref . @xcite for the corresponding cross section @xmath57 of the second order process are as following : @xmath58 @xmath59 however , there is an important difference in the definitions of branching ratios @xmath41 , @xmath60 compared with @xmath61 , since the former are for the absorption of two photons by _ two different _ nuclei leading to certain dissociation channels @xmath38 and @xmath51 , while the latter is for the absorption of two photons by a _ nucleus leading to a channel @xmath38 . another difference is due to an additional factor of @xmath62 in eq . ( [ smfull ] ) compared with eq . ( [ sd ] ) . it comes from the fact that eq . ( [ smfull ] ) contains the product of the poisson probabilities for the single photon absorption for the collision with impact parameter @xmath4 : @xmath63 while @xmath64 is the poisson probability for the double photon absorption @xcite . results for partial cross sections of mutual dissociation will be presented and discussed in sec . [ lhc ] . the second order process , which is shown in fig . [ emsecond ] and considered in sec . [ emd_main ] , is the leading order mechanism of mutual electromagnetic dissociation . this is confirmed by the calculations of the corresponding total cross sections for the leading order ( lo ) , fig . [ emsecond ] and next - to - leading - order ( nlo ) processes , figs . [ emnlo12 ] and [ emnlo22 ] . following the assumption that the probability of multiphoton absorption is given by the poisson distribution with the mean multiplicity @xmath65 of eq.([mb ] ) , one has for the lo process shown in fig . [ emsecond ] : @xmath66 here the case of equal masses and charges of collision partners is considered , i.e. @xmath20 and @xmath21 . for the nlo process with exchange of three photons ( @xmath67 ) , [ emnlo12 ] , the total cross section is given by : @xmath68 a complementary process ( @xmath69 ) with the excitation of another nucleus , @xmath14 , via double photon absorption is equally possible and has the same cross section . another nlo process of mutual dissociation is due to exchange of four photons ( @xmath70 ) , fig . [ emnlo22 ] , and its cross section can be written as : @xmath71 finally , one can calculate the sum of all contributions using the prescription of ref . @xcite : @xmath72 since for each of the ions the probability of collision without photon exchange is equal to @xmath73 , eq.([smtot ] ) is evident . calculations of @xmath74 , @xmath75 , @xmath76 and cross sections for specific dissociation channels , @xmath77 , with and without nlo corrections were performed by the modified reldis code , which contains now a special simulation mode for the mutual electromagnetic dissociation process . our results for the total dissociation cross sections in electromagnetic and nuclear interactions are given in tab . these cross sections values are in good agreement with results of other authors . the total cross sections for mutual electromagnetic dissociation given in tab . [ t1 ] are much lower than the cross sections for single dissociation . however , even the former values are found to be comparable to the total nuclear dissociation cross sections , see tab . [ t1 ] . as one can see , the ratios between the first and second order processes are very different for single and mutual dissociation . the first order dissociation process ( with exchange of a single photon ) can be safely neglected in considering the mutual dissociation of heavy nuclei at ultrarelativistic energies . tab . [ t1 ] contains also the cross sections for lo process , @xmath75 . as one can see , the lo mechanism gives @xmath78% and @xmath79% of the @xmath74 at rhic and lhc energies , respectively . the sum of the nlo contributions to the total cross section gives additional @xmath8027% . therefore , at rhic energies , for example , the remaining contribution of @xmath81% ( or @xmath82 b ) is due to exotic triple excitations . finally , @xmath83 and @xmath84 written for the lo mutual dissociation process , eqs . ( [ smfull ] ) and ( [ nm ] ) , can be easily generalized to the case of different nlo processes . however , the resulting expressions are lengthy and we do not give it here for brevity s sake . several nucleons can be abraded from collision partners in grazing nuclear collisions . we are interested in a situation when only few nucleons are removed . this is the case when nuclear densities overlap weakly and mainly nuclear periphery is involved in the interaction . the cross section for the abrasion of @xmath85 nucleons from the projectile @xmath86 in a collision with the target @xmath87 may be derived from the glauber multiple scattering theory @xcite : @xmath88 here @xmath89 is calculated as the overlap of projectile , @xmath90 , and target @xmath91 densities in the collision with impact parameter @xmath4 : @xmath92 where the nuclear thickness functions , @xmath93 are introduced . in our calculations the nuclear density functions are approximated by fermi functions : @xmath94 where @xmath95 is a parameter which defines the nuclear half - density radius , @xmath96 , and @xmath97 fm is the diffuseness parameter . the choice of these and other important parameters of the model , the integration cutoff parameter , @xmath98 , and the total nucleon - nucleon cross section , @xmath99 , is discussed in sec . [ parameters ] . the above expressions determine only the number of nucleons removed from the projectile and do not specify how many protons or neutrons were knocked out . further assumptions are needed to determine the charge - to - mass ratio of the residual nucleus and hence the numbers of protons , @xmath100 , and neutrons , @xmath101 , abraded from the initial nucleus ( see also the discussion in ref . @xcite ) . in the present work we use the so - called hypergeometrical model @xcite , assuming that each removed projectile nucleon has a @xmath102 probability ( @xmath103 ) of being a neutron : @xmath104 in other words this means that there is no correlation at all between the proton and neutron distributions and the abrasion process removes protons and neutrons from the projectile nucleus in a random way . several physical processes , which might be important in heavy - ion collisions , were neglected in this model . the excited residual nucleus created due to the abrasion process should undergo its de - excitation on the second ablation step . on this step more neutrons may be emitted via evaporation . however , as it was noticed in ref . @xcite , the excitation energies obtained on abrasion step due to removal of one or two nucleons are generally not sufficient for intensive particle evaporation . therefore , for the cases of interest , i.e. @xmath105 and @xmath106 dissociation channels , the ablation step can be neglected . the abrasion of nucleons from projectile and target proceeds via high - energy collisions between nucleons . nucleon - antinucleon pairs may be created in such interactions and neutrons may be presented in these pairs . however , as one can find in a compilation @xcite , even at high energies @xmath107 agev , the relative rate of such pair production is not so high , @xmath108 . because of this , we do not consider such processes in calculations of the neutron emission cross sections . knocked - out nucleons can also suffer a final state interaction with spectators @xcite . we believe that this process is less important at high energies compared with intermediate energies of @xmath109 agev . for the latter case the escape probability is estimated to be @xmath110 for peripheral collisions of heavy nuclei @xcite . the momenta of recoil nucleons may be comparable with the fermi momentum of intranuclear nucleons and their angular distribution is very wide so that they can be easily captured by one of the spectators . the situation is different at high energies , where the transverse momenta of collided nucleons are typically large , of order of 0.5 gev , and therefore their subsequent capture is less probable . other effects like a finite hadronization length may further reduce the secondary interaction probability . therefore , we assume that in peripheral collisions at rhic and lhc the probability for each of the collided nucleons to escape the residual nuclei is close to unity , @xmath111 . as we will show in sec . [ comparison ] , the above - mentioned simplifications do not lead to noticeable disagreements with experimental data when the removal of one , two or three nucleons is considered . it means that either above mentioned physical effects are negligible , or they compensate each other in peripheral nuclear collisions with removal of only few nucleons . however , the predictions of the present model for more central collisions with removal of many nucleons should be taken with care . this simple abrasion model can be easily extended to the case of mutual dissociation . the cross section for the removal of @xmath112 neutrons and @xmath113 protons from the projectile @xmath114 _ simultaneously _ with the removal of @xmath115 neutrons and @xmath116 protons from the target @xmath117 , ( @xmath118 may be written as : @xmath119 since the number of nucleon - nucleon collisions in such a process is assumed to be equal to @xmath85 , the condition @xmath120 holds . using this condition for the process of mutual dissociation with given numbers of neutrons , @xmath112 and @xmath115 , removed from the projectile and target nuclei , respectively , one has finally : @xmath121 here the cross section for single abrasion process , @xmath122 , given by eqs . ( [ abr_main ] ) and ( [ hyper ] ) was used . as shown in sec . [ emtheory ] , the photonuclear cross sections are used as input data in calculations of the electromagnetic dissociation cross sections . this is verified by the coherent nature of the photon emission by the collision partner as a whole . since these photons represent the lorentz - boosted coulomb fields of original nuclei , their virtuality is very small , @xmath123 , i.e. these photons are almost real . therefore , one can use the photonuclear reaction data obtained in experiments with real monoenergetic photons and apply theoretical models describing such photonuclear reactions . the accuracy of the mutual dissociation cross section calculation depends heavily on the quality of the data and parameters used as input . as we found , for example , the mutual dissociation cross section is more sensitive to the proper choice of the critical impact parameter , @xmath43 , and to the input photonuclear cross sections than the single dissociation cross section . the input data and model parameters are discussed in detail in the following sections . over the years , the photoneutron cross sections for different nuclei have been measured with monoenergetic photons at saclay @xcite and livermore @xcite . data on different cross sections obtained in these and other laboratories were collected in compilations of refs . @xcite . concerning the nuclei of interest , @xmath124 and @xmath125 , the detailed data were obtained mainly for @xmath126 and @xmath127 reactions , while less detailed data exist for @xmath128 and @xmath129 reactions , see refs . the measurements were performed in the photon energy region @xmath130 mev , where the excitation of giant resonances plays a dominant role , see figs . [ fig : au ] and [ fig : pb ] . at such energies , the emission of charged particles @xmath131 is suppressed by a high coulomb barrier of heavy nuclei . therefore , the sum of partial cross sections for all neutron multiplicities : @xmath132 nearly coincides with the total photoabsorption cross section . each of the inclusive cross sections @xmath133 includes a small contribution from the channels with charged particles , @xmath134 , @xmath135 , ... at the same time such channels as @xmath136 , @xmath137 were neglected at all . according to ref . @xcite , this leads to a small systematic error @xmath138 in the total photoabsorption cross section measured at low energies via the neutron detection . above the giant resonance region , at @xmath139 mev , the quasideuteron mechanism of photon absorption dominates . only average characteristics of photon absorption by @xmath125 were measured in saclay @xcite in this energy region . neutron yields , @xmath140 , and the cross sections for emission of at least @xmath51 neutrons , @xmath141 , were obtained in addition to the mean value and the width of the neutron multiplicity distribution . to the best of our knowledge there are no direct measurements of neutron emission cross sections and multiplicities at @xmath142 mev in photoabsorption on @xmath124 . the only attempt to deduce the average photoneutron multiplicities from the experimentally obtained average excitation energies has been made at @xmath143 mev in a model - dependent way @xcite . for @xmath126 channel the overall agreement between livermore and saclay data is good . some inconsistency exists only in the giant resonance peak height ( @xmath144 for @xmath124 and @xmath145 for @xmath125 ) and on the right side of the peak , where @xmath126 and @xmath127 process compete with each other , see figs . [ fig : au ] and [ fig : pb ] . large discrepancies are present in @xmath127 cross sections measured in different experiments , both in shape and normalization , up to @xmath146 for @xmath125 target . several attempts of data evaluation and re - measurements have been made to reduce these discrepancies . based on the observation that the total photoneutron yields , @xmath147 , obtained in livermore and saclay experiments agree well , an explanation for the discrepancy was put forward in ref . it was attributed to the different neutron multiplicity sorting procedures adopted in different laboratories . as was concluded in ref . @xcite , the neutron multiplicity sorting procedure adopted at saclay was not correct since some of the @xmath148 events were interpreted as pairs of @xmath126 events . in 1987 new measurements were made in livermore @xcite , where it was found that the previously reported livermore @xcite and saclay @xcite results have to be rescaled . as was recommended in ref . @xcite , the saclay data of ref . @xcite for both @xmath124 and @xmath125 nuclei have to be used with the correction factor of 0.93 . we follow this prescription in using photonuclear cross section data in our reldis code . however , such a correction is not widely accepted and the authors of refs . @xcite use uncorrected saclay data . some new @xmath127 cross section data were obtained in ref . @xcite for @xmath124 and @xmath125 . for @xmath124 nucleus , @xmath127 data are nearly the same as the saclay @xmath127 data of ref . @xcite and the conclusion of ref . @xcite concerning the neutron multiplicity sorting procedure seems to be not confirmed . unfortunately , the recent data of ref . @xcite are available only up to a few mev above the @xmath127 threshold and the findings of ref . @xcite can not be completely ruled out . one of the most recent measurements of @xmath126 cross section for photoabsorption on @xmath125 was performed in russia at the saratov university @xcite . a fine structure of the low - energy wing of the giant dipole resonance ( gdr ) has been investigated in detail . the photoneutron cross sections were obtained from the photoneutron yield curves by means of the statistical regularisation method . an evaluation of @xmath149 cross section for @xmath125 has been made at the moscow state university @xcite by applying a statistical reduction method . because of systematic uncertainties in calibration and normalisation , the general characteristics of the measured @xmath126 cross section ( the energy integrated cross sections , weighted - mean values ) are different in different measurements @xcite . in the reduction method of ref . @xcite the renormalisation corrections were introduced for both the energy and the cross section scales in order to obtain the best agreement between the general characteristics of @xmath126 cross section measured in different experiments . the data obtained in refs . @xcite for @xmath126 reaction on @xmath125 target are also plotted in fig . [ fig : pb ] . good agreement with re - scaled saclay data of ref . @xcite is found up to @xmath127 threshold . unfortunately , the @xmath127 reaction was beyond the scope of investigations in refs . @xcite . it is evident from our consideration that the calculations of electromagnetic dissociation of ultrarelativistic @xmath124 and @xmath125 nuclei can not be based exclusively on the photoneutron cross sections measured in experiments with real photons . additional information on photoneutron cross sections for the whole energy domain of equivalent photons @xmath150 can be obtained by using theoretical models of photoabsorption . this is particularly indispensable for dissociation channels with emission of many neutrons @xmath151 and charged particles , @xmath152 , ... one of the two photonuclear reaction models used in the present work is the gnash code @xcite . it is very precise in describing low energy neutron emission data @xcite , however , it can be used only up to the pion production threshold , at @xmath153 mev . within this model the photoabsorption process is modelled through the excitation of the giant dipole resonance ( gdr ) at lower energies and the quasideuteron ( qd ) mechanism at higher energies . the photoabsorption cross section in the whole energy range from the threshold for neutron emission up to 140 mev is thus written in the form @xcite : @xmath154 where @xmath155 is given by a lorentzian curve with parameters taken from gdr systematics @xcite and corrected according to ref . the latter term , @xmath156 , is related by a levinger - type model to the experimental deuteron photodisintegration cross section , @xmath157 @xcite : @xmath158 where @xmath159 , @xmath160 and @xmath161 are , respectively , the neutron , proton and mass number of the corresponding nucleus . the levinger parameter , @xmath162 , is equal to 6.5 , and @xmath163 is a pauli - blocking factor , which reduces the free deuteron cross section , @xmath164 , by taking into account the pauli blocking of the excited neutron and proton in the nuclear medium . in ref . @xcite , @xmath165 was derived in the form of a multidimensional integral , approximated in the energy range 20140 mev by a polynomial expression : @xmath166 and by an exponential one outside the considered energy range : @xmath167 thus , @xmath163 tends to zero if @xmath168 goes to zero , and to unity if @xmath168 goes to infinity and is continuous with eq.([ff ] ) at 20 and 140 mev @xcite . finally , the experimental deuteron photodisintegration cross section is given by a simple parameterisation : @xmath169 where @xmath168 is expressed in mev , as in the previous formulae , and @xmath157 in mb . due to the correlation between intranuclear nucleons in the absorption on a quasideuteron pair , the initial particle - hole configuration is assumed to be @xmath170 rather than @xmath171 , see ref . @xcite . in the gnash code the initial interaction characterized by the total cross section of eqs . ( [ sabs])-([ff ] ) is followed by the preequilibrium emission of fast nucleons described by the exciton model @xcite . finally , when the nuclear system comes to equilibrium , sequential evaporation of particles is considered within the hauser - feshbach formalism @xcite . gnash code results for @xmath126 , @xmath127 and @xmath128 cross sections are shown in figs . [ fig : au ] and [ fig : pb ] for @xmath124 and @xmath125 nuclei , respectively . calculations describe @xmath126 and @xmath128 data very well . taking into account existing disagreements between the results of different measurements of @xmath127 cross sections , one can conclude that the gnash results fall in between the saclay @xcite and livermore data @xcite for @xmath125 , and very close to livermore data for @xmath124 , that seems to be satisfactory for both cases . therefore , one can use in eq . ( [ smfull ] ) the photonuclear cross sections , @xmath172 , and branching ratios , @xmath173 , calculated by the gnash code to estimate the mutual electromagnetic dissociation cross section @xmath174 . the influence of a constraint , @xmath175 mev , will be discussed in sec . [ lhc ] . branching ratios for neutron emission in photonuclear reactions , @xmath176 , can be calculated by means of the extended cascade - evaporation - fission - multifragmentation model of photonuclear reactions @xcite in the whole range of equivalent photon energies . some details of the calculation method as well as numerous comparisons with experimental data used for the model verification were given in refs . @xcite . here we describe only the general calculation scheme along with the modifications and advancements made in the model since the time when the works @xcite have been published . in the reldis model the values of the total photoabsorption cross section to be used in eq . ( [ smfull ] ) are taken from corresponding approximations of experimental data . in the gdr region the lorentz curve fits were used for this purpose with parameters from ref . @xcite corrected according to the prescription of ref . @xcite , as described in sec . [ photo_data ] . above the gdr region , where the quasideuteron absorption comes into play , the total cross section is taken from the quasideuteron model of ref . @xcite : @xmath177 here @xmath178 is the meson exchange part of the cross section for the deuteron photodisintegration , @xmath179 , and @xmath180 is an empirical constant @xcite . above the pion production threshold , at @xmath181 mev , a universal behaviour @xmath182 is observed ( see ref . @xcite for the latest experimental data ) . this means that the total photoabsorption cross section per bound nucleon @xmath183 has almost the same energy dependence for light , medium - weight and heavy nuclei , c , al , cu , sn , pb , at least up to @xmath184 gev . therefore , having the data for one nucleus one can calculate the cross section for other nuclei . however , in this energy region the universal curve @xmath183 is very different from the values extrapolated from the cross sections on free nucleons , @xmath185 , which are deduced from proton and deuteron data @xcite . at @xmath186 gev the universal behaviour breaks down , and the ratio @xmath183 for lead is 20 - 25% lower than for carbon @xcite due to the nuclear shadowing effect . in order to approximate the total photonuclear cross sections at @xmath186 gev we used recent results obtained within the framework of the glauber - gribov scattering theory and the generalized vector dominance model @xcite . such calculations describe well the general trend of experimental data obtained for high energy photon absorption , although the data have very large uncertainties at @xmath187 gev . by comparing tab.ii of ref . @xcite and tab . [ t1 ] of the present paper one can find minor differences in the total ed cross sections due to using different parameterisations of @xmath183 at @xmath186 gev . compared with the total ed cross sections , the single or double neutron emission cross sections are even less affected by the choice of the parameterisation . the reldis code performs the monte carlo simulation of the mutual dissociation process according to the following steps . first , a pair of the energies @xmath42 and @xmath49 of the photons exchanged between the colliding nuclei is selected according to the spectral function @xmath54 . second , the photoabsorption process is generated in both nuclei leading to the formation of excited residual nuclei . third , the de - excitation of both of the thermalized residual nuclei is simulated according to the statistical evaporation - fission - multifragmentation model , ( the smm model ) @xcite . the evaporation of neutrons from excited residual compound - like nucleus is the main process responsible for the @xmath149 , @xmath188 , @xmath189 channels of photoabsorption . the quality of description of such channels is very important for precise calculations of neutron emission in the mutual dissociation process . in the present paper the standard weisskopf evaporation scheme is used @xcite with several modifications taking into account the microscopic effects of nuclear structure in the nuclear mass and level density formulae . such effects reveal themselves in the noticeable difference , up to @xmath190 mev for heavy closed - shell nuclei , between the values of the nuclear mass measured in experiments and those predicted by the macroscopic liquid - drop model . moreover , this difference in mass , the so - called shell correction , and the level density parameter are strongly correlated . for closed - shell nuclei the actual values of the level density parameter are essentially lower than the average values expressed as @xmath191 mev@xmath192 , and these values depend noticeably on the excitation energy . proper accounting for these effects , as well as pairing effects , is important at low excitations , @xmath193 mev , i.e. in the region where @xmath105 and @xmath106 photoemission processes occur . the above - mentioned shell effects are most pronounced at low excitation energies , but almost disappear at @xmath194 mev , see ref . @xcite for details . several phenomenological systematics of the level density parameter were proposed to account for such behaviour , refs . our calculations are based on results of ref . @xcite where all existing data on the level densities , decay widths and lifetimes of excited nuclei have been analyzed in the framework of the statistical model . however , the creation and subsequent decay of an excited compound nucleus formed after the photoabsorption in gdr region is not the only process responsible for the neutron emission . indeed , a giant resonance is a coherent superposition of ( one - particle - one - hole ) @xmath195 excitations . a particle or a hole can interact with another nucleon and create a @xmath171 state . further spreading to @xmath196 states etc . finally leads to a statistically equilibrated system , the compound nucleus . instead of such evolution to equilibrium , a collective @xmath195 state can decay directly by the emission of one nucleon leading to a low - lying hole state in the residual nucleus , see among others ref . after such direct @xmath105 emission , the emission of a second neutron is generally impossible , even though the initial photon energy exceeds the @xmath106 emission threshold . in such a way @xmath127 channel is suppressed in comparison with the pure statistical decay . although the gdr state in @xmath125 nucleus decays mainly statistically , the existence of direct neutron emission has been clearly demonstrated in ref . the process where a fast nucleon is emitted and the final state @xmath197 nucleus is left with low excitation energy @xmath198 mev was identified in this experiment @xcite . an evidence of direct neutron emission in photoabsorption on @xmath12 and @xmath13 nuclei was given in ref . @xcite based on the analysis of competition between @xmath105 and @xmath106 emission channels . this was an independent confirmation of the findings of earlier works @xcite and @xcite devoted to the measurements of the neutron spectra in photoabsorption on the same nuclei . an excess of fast neutrons ( kinetic energy @xmath199 mev ) with respect to the predictions of the statistical evaporation model has been demonstrated and attributed to the direct emission . in ref . @xcite a non - statistical contribution in excited @xmath125 nucleus with @xmath200 mev was successfully extracted . out of this region the non - statistical contribution was found to be negligible . in our calculations we used the total fractions of the non - statistical neutron emission from @xmath12 and @xmath13 nuclei as @xmath201 and 0.26 , respectively , evaluated from experimental data in ref . such values are in line with modern theoretical expectations @xcite that the ratio of intensities of the direct and statistical neutron emission from photoexcited gdr in @xmath125 nucleus is about @xmath202 . in the reldis code the emission angles @xmath203 of non - statistical fast photoneutrons are generated according to the approximation @xmath204 which is found in ref . we assumed that the direct @xmath105 emission takes place at @xmath205 mev . since the adopted @xmath206 values have some uncertainties , we have investigated the sensitivity of results to these values . a part of calculations was made with @xmath207 , i.e. without accounting for direct emission , see figs . [ fig : au ] , [ fig : pb ] and tab . [ thill ] . as shown in fig . [ fig : au ] , the @xmath188 cross sections on gold calculated by the reldis code with @xmath208 are very close to saclay measurements @xcite , while livermore results @xcite are better described with @xmath209 . therefore the difference in calculation results obtained with @xmath209 and @xmath208 reflects the level of experimental uncertainties . since the nucleon - nucleon interaction has an isovector component , the interference of nuclear and electromagnetic amplitudes can not be excluded . such interference term was considered in ref . @xcite and found to be small . even for @xmath124 nucleus colliding with heavy targets the interference correction to the single neutron removal cross section was found to be less than 0.50.6 % of the corresponding nuclear or electromagnetic contributions . following this result , nuclear and electromagnetic parts of the dissociation cross section may be safely treated separately . in other words , one can add probabilities instead of coherent summation of amplitudes . let us consider the way how the probabilities of the nuclear and electromagnetic contributions should be added to obtain the total dissociation probability . at grazing impact parameters relativistic nuclei are partly transparent to each other . hard @xmath210 collisions may be absent at all in the case of a peripheral event with a weak overlap of diffuse nuclear surfaces , while the electromagnetic interaction may take place in this event leading to the electromagnetic dissociation . generally , at a grazing collision either the nuclear , or electromagnetic interaction , or even both of them may occur . as an example of the latter case , a single neutron - neutron collision in the participant zone may lead to the neutron removal , while a photon may be emitted and absorbed by charged spectators in the same event . therefore , in a detailed theoretical model a smooth transition from purely nuclear collisions at @xmath211 to electromagnetic collisions at @xmath212 should take place . such kind of transition was considered in a `` soft - sphere '' model of ref . a similar approach was adopted in ref . @xcite , where the cross section for at least one type of dissociation , either nuclear , or electromagnetic , or both , was written as : @xmath213 where @xmath214 and @xmath215 are , respectively , the probabilities of the nuclear and electromagnetic dissociation at given impact parameter @xmath4 . putting explicitly the integration limits for each term one obtains : @xmath216 here the impact parameter cutoff values , @xmath217 and @xmath218 , were used for the nuclear and electromagnetic interactions , respectively . however , due to several reasons a more simple expression is widely used @xcite : @xmath219 where a single cutoff parameter , @xmath43 , was chosen as : @xmath220 . the first reason is that with the latter condition one can effectively reduce the first and second terms of eq . ( [ fullp ] ) without subtracting the third nuclear - plus - electromagnetic term . numerical results based on eqs . ( [ fullp ] ) and ( [ truncatedp ] ) become similar as it was found for `` sharp - cutoff '' and `` soft - spheres '' models of ref . second , for heavy nuclei the difference between reasonable values of @xmath221 , @xmath222 and @xmath43 turns out to be less than 1 fm . as a result , the last nuclear - plus - electromagnetic term of eq . ( [ fullp ] ) turns out to be small . third , with eq . ( [ truncatedp ] ) one can study the nuclear and electromagnetic contributions separately . therefore , independent parameterisations may be found in experiments for the nuclear and electromagnetic parts , @xmath223 and @xmath224 . this is useful for studying nuclear and electromagnetic dissociation at ultrarelativistic colliders , where the products of nuclear and electromagnetic interactions populate very different rapidity regions , namely the central rapidity region and close to the beam rapidity , respectively . in the present paper the traditional form given by eq . ( [ truncatedp ] ) is adopted with a common impact parameter cutoff , @xmath43 , for nuclear and electromagnetic contributions . at relativistic energies , according to the widely used bcv parameterisation of ref . @xcite , @xmath43 is estimated as : @xmath225 the values @xmath226 fm and @xmath227 were found from a fit to glauber - type calculations of the total nuclear reaction cross sections @xcite . the evidences in favour of the @xmath43 choice according to the bcv parameterisation were given in refs . @xcite,@xcite , which we mention among others . as argued in ref . @xcite , using the bcv parameterisation one can perfectly describe experimental data on fragment angular distributions which are very sensitive to @xmath43 . for calculations within the abrasion model we used the following values for the total nucleon - nucleon cross section given in ref . @xcite : @xmath228 , 50 and 90 mb at sps , rhic and lhc energies respectively . some problems are connected with the choice of the nuclear density parameters , @xmath96 , and @xmath97 fm , sec . [ abrmodel ] . only the nuclear charge distributions are measured in electron scattering experiments , while the neutron densities are available only from calculations . we used @xmath229 fm as an average value between the proton and neutron distributions similar to one used in ref . the total nuclear reaction cross sections calculated with these @xmath95 and @xmath230 values within the abrasion model are in good agreement with the approximation of experimental data found in ref . @xcite . numerical results showing the sensitivity of the nuclear and mutual electromagnetic dissociation cross sections to the variations of the above - discussed parameters are given in sec . the calculated charge changing cross sections of the _ single _ dissociation of 158a gev @xmath125 ions are given in fig . [ fig : zdis ] . the lo mechanism with single photon exchange , fig . [ em ] , and the nlo mechanism with double photon exchange , fig . [ em2 ] , were taken into account . each proton removal process can be accompanied by neutron loss as well . the calculated cross sections given for @xmath231 correspond to the interaction where only neutrons are emitted . however , experimental data are absent for this channel . as one can see , the electromagnetic contribution dominates for the processes with removal of one , two and three protons where good agreement with the experimental data of ref . @xcite is found . another important check of the model becomes possible with recent experimental data for au fragmentation by ultrarelativistic pb ions @xcite . in this case the neutron emission is investigated directly . the experimental exclusive cross sections for emission of one or two neutrons are compared with theory in tab . [ thill ] . also in this case the contributions from the lo and nlo processes , figs . [ em ] and [ em2 ] , were taken into account . the calculations were made with and without accounting for direct neutron emission process , i.e. with @xmath208 and @xmath209 , respectively . the results with @xmath208 are in better agreement with experiment and this is especially true for @xmath106 emission channel . therefore , we use this value in further calculations . the predictions of abrasion model are also in good agreement with data @xcite . therefore , our choice of the critical impact parameter , @xmath43 , is justified by such a comparison . as was stressed in sec . [ abrmodel ] , the interaction of knocked out nucleons with spectators and spectator de - excitation process themselves were neglected in this version of the abrasion model aimed at considering only few nucleon removal processes . such good agreement indicates that a simple abrasion model proposed for describing the data at @xmath23210 agev can be used successfully at much higher energies as well . on the basis of the successful verifications at lower energies described in sec . [ comparison ] , the model can now be extrapolated to rhic and lhc energies . in a collider , the mutual heavy - ion dissociation process takes place at the crossing point of two beams . downstream from this point the dissociation products can be separated by the magnetic field according to their @xmath233 ratio . protons and nuclear fragments move close to the beam trajectories , while free neutrons leave the beam pipe after a dipole magnet . at rhic , zero degree ( very forward ) calorimeters ( zdc ) for each beam are located after the magnets and they are well designed for the neutron registration @xcite . therefore , in the following we consider mainly semi - inclusive mutual neutron emission cross sections , @xmath234 , where @xmath38 and @xmath51 denote corresponding channels , @xmath235 , @xmath236 , @xmath237 , ... besides the emission of a given number of neutrons , such dissociation channels contain any number of other particles denoted as @xmath238 or @xmath239 . protons can be found most often amongst the particles emitted along with neutrons . the proton emission rates predicted by the reldis code were found to be in agreement with the data on pb dissociation @xcite , see sec . [ comparison ] . therefore we believe that our model is accurate in estimating @xmath240 , @xmath241 and @xmath242 cross sections . if the dissociation channel of one of the collision partners in mutual dissociation is not exactly known , one can define inclusive mutual dissociation cross sections , @xmath243 , @xmath244 , @xmath245 etc . in such notations @xmath246 denotes an arbitrary dissociation mode of one of the nuclei . the cross sections for some specific channels of mutual dissociation at rhic energies are given in tab . [ tlonlo ] . compared to the total cross sections , the relative contributions of nlo processes are very different for the cross sections of @xmath105 , @xmath106 and @xmath247 emission . as one can note , @xmath248 has a small nlo correction , @xmath249% , while @xmath250 becomes almost twice as large as its lo value if nlo correction is included . this is due to the fact that the nlo processes shown in figs . [ emnlo12 ] and [ emnlo22 ] include nuclear excitation due to double photon absorption and , particularly , the double gdr excitation process . since the average gdr energy for au and pb nuclei is about 1314 mev , the double gdr excitation introduces , on average , 2628 mev excitation which is above the @xmath247 emission threshold . therefore , as a rule , @xmath105 and @xmath106 emission cross sections are less affected by including of nlo corrections than @xmath247 cross sections . the double gdr excitation process is a well - known phenomenon studied in low and intermediate energy heavy ion collisions , see , for example , ref . however , nothing is known about the multiple photon excitations of heavy ions at ultrarelativistic energies , where the key question is to what extent the distribution of probabilities for multiple excitations follows the poisson distribution ( the harmonic picture of excitations ) . in order to estimate the rates of multiple excitations , including those well above the giant resonance region , the ratios @xmath251 and @xmath252 should be measured at rhic . such measurements do not require preliminary determination of the collider luminosity . as it follows from tab . [ tlonlo ] , the above - mentioned ratios are remarkably reduced if the multiple excitations , and hence nlo processes , are suppressed in mutual electromagnetic dissociation at rhic . there remains still some freedom in choosing several parameters of our model . to check the sensitivity of our predictions to their variations we performed the calculations for a reasonable span of input parameters . tab . [ tpdir ] demonstrates the sensitivity of the mutual electromagnetic dissociation cross sections to the photonuclear cross sections used as the input . in order to demonstrate such sensitivity we used two different models to calculate such cross sections , namely the gnash code @xcite and the photonuclear reaction model implemented in the reldis code @xcite itself . additionally , in the latter model we used two different values for the probability of the direct neutron emission in @xmath105 channel , @xmath206 . besides the variations of the cross sections for emission of one or two neutrons , the variations of a cumulative value , the low multiplicity neutron ( lmn ) emission cross section defined as @xmath253 were evaluated for several sub - regions of equivalent photon energies , @xmath254 mev , @xmath255 mev and for the full range . in the latter case the nlo processes were taken into account along with the lo mutual dissociation process . therefore , the importance of nlo corrections can be also estimated from tab . [ tpdir ] . by examining tab . [ tpdir ] , one can draw several conclusions . first , the semi - inclusive cross section @xmath256 mb calculated for the photoabsorption in the giant resonance region is very close to the exclusive value @xmath257 mb obtained in ref . @xcite with the same condition : @xmath254 mev . second , the calculations based on the photonuclear cross sections predicted by the ghash code are very close to the reldis results for the photon energy region @xmath255 mev . the difference between the reldis results for @xmath254 mev and @xmath255 mev is explained by the contribution of the quasideuteron photoabsorption mechanism to the @xmath105 and @xmath106 emission . third , the calculations which take into account the quasideuteron photoabsorption and photoreactions above the pion production threshold give about 25% enhancement in @xmath258 if the whole energy region of equivalent photons is considered and the nlo corrections are properly taken into account . at the same time , the cross sections @xmath259 and @xmath260 increase up to two and four times , respectively , compared with the values calculated at the gdr region . the cross sections for these dissociation channels are large and such channels can be easily measured in experiments . although the lo process of photoabsorption in the gdr region gives an important contribution , the whole range of the equivalent photon energies should be considered . also the contributions from the nlo processes should be taken into account to obtain the realistic values of the dissociation cross sections . one more conclusion follows from the results presented in tab . [ tpdir ] . calculations with @xmath207 and @xmath201 give 10 - 40% difference in specific dissociation cross sections , but the values of the lmn cross section , @xmath261 , practically coincide . this cross section is very high , @xmath262 mb , and thus can be used for luminosity monitoring . as was shown above in sec . [ photo_data ] , an inevitable systematic error of @xmath108 should be assigned to this value due to uncertainties in the photoneutron cross sections measured in experiments . tab . [ tro ] shows the sensitivity of the mutual dissociation cross section in grazing nuclear collisions to the variations of the @xmath95 parameter in the nuclear density distribution , eq . ( [ fermi ] ) . the parameters of neutron density distributions are not well determined and this table demonstrates possible ambiguities in nuclear dissociation cross sections caused by this fact . the cross section variations are smaller in tab . [ tro ] compared to tab . [ tpdir ] , about 3 - 8% . a small decrease in the @xmath95 parameter leads to a decrease in correlated @xmath263 emission , but , on the contrary , it leads to an increase in @xmath264 and @xmath265 emission . however , the lnm cross section , @xmath266 , turns out to be more stable , within @xmath267% variations , compared to the cross sections for specific channels . the sensitivity of the dissociation cross section in grazing nuclear collisions to the variations of the total nucleon - nucleon cross section is investigated in tab . the variations of the @xmath99 in the range of 4060 mb have only a small effect , within 4% , on the cross sections of the specific neutron emission channels . since such variations have different signs , the influence on the cumulative value , @xmath266 , is less noticeable , below 2% . the cross sections given in tabs . [ tro ] and [ tnn ] for grazing nuclear collisions were found to be lower compared with the electromagnetic dissociation cross sections of tab . [ tpdir ] . moreover , it should be stressed , that only a part of the neutrons in grazing nuclear collisions is emitted at the forward or backward angles covered by the zdcs . therefore , an exact relation between nuclear and electromagnetic dissociation channels in each heavy - ion experiment should be only obtained by taking into account detection acceptances and trigger conditions of the corresponding experimental setup . tab . [ tbc ] demonstrates the sensitivity of the mutual dissociation cross section in electromagnetic and nuclear interactions to the impact parameter cutoff , @xmath43 . this is an important input parameter which has a noticeable influence on the final result . by changing this parameter by 5% , within the range of 14.516 fm , one obtains the variations of the electromagnetic dissociation cross sections within 38% . such variations of @xmath43 shift the point , which delimits the regions of nuclear and electromagnetic interactions , below or above the domain where the overlap of diffuse nuclear boundaries takes place . in other words , assuming first @xmath268 and then @xmath269 and considering the difference in final results , one can prove the possibility to use eq.([truncatedp ] ) instead of eq.([fullp ] ) . for example , if the cutoff value , @xmath43 , becomes lower , all the ed cross sections , @xmath263 , @xmath264 , @xmath265 and @xmath261 become lower . the variations of the nuclear cross sections are more noticeable and have the opposite trend : such cross sections become higher by 515% . finally , as one can see in tab . [ tbc ] , @xmath270 variations for both types of interactions are weaker than the variations of the corresponding individual cross sections , within 38% , while the sum @xmath271 is altered by 1 - 4% only . concluding the investigation of the sensitivity of the final results to the model parameters , one can note that for auau collisions at rhic @xmath272 mb and especially @xmath273 mb values are more stable with respect to such variations in comparison with the individual cross sections , @xmath240 , @xmath241 and others . the same tendency was found for pbpb collisions at lhc energies where such cross sections were found to be @xmath274 mb and @xmath275 mb . for both rhic and lhc cases the overall uncertainty of the @xmath270 calculation method may be estimated at the level of 57% . the condition for heavy ion dissociation to be mutual leads to some specific features for nuclear and electromagnetic interactions . the former interaction causes mainly symmetric or quasi - symmetric dissociation . the latter makes very probable asymmetric dissociation like @xmath276 or even @xmath277 . this feature is shown in figs . [ fig : rhic2d ] and [ fig : lhc2d ] where the cross sections with one , two or three neutrons in one arm of the zdc setup are presented . as one can see , @xmath277 dissociation is almost absent in nuclear collisions . this is not true for electromagnetic dissociation where the number of @xmath277 events is approximately 15% of the main dissociation channel , @xmath278 . in the reldis model nuclei undergo dissociation independently in the electromagnetic fields of each other . therefore there is no correlation between the numbers of neutrons , @xmath112 and @xmath115 , emitted by each of the nuclei , and asymmetric dissociations are possible along with symmetric ones . the extreme case of asymmetric dissociation is , of course , the single dissociation process . the nuclear dissociation considered in the framework of the abrasion model has different characteristic features . namely , the numbers of emitted neutrons and protons are correlated due to the condition @xmath279 , see sec . [ abrmodel ] , and nuclear dissociation is always mutual . the latter condition was also adopted in ref . @xcite . the main results of our study are presented in figs . [ fig : rhic ] and [ fig : lhc ] . they show the electromagnetic and nuclear dissociation cross sections . since @xmath105 and @xmath106 emission in electromagnetic collisions is enhanced due to the gdr and qd absorption mechanisms , the corresponding strips are prominent in the plots . simultaneous gdr excitation in both of the nuclei is a dominant process leading to the mutual dissociation , but it is responsible only for a part of the total mutual dissociation cross section , @xmath280 at rhic , for example . the rest is provided mainly by asymmetric processes , when one of the nuclei is excited in a gdr state , while another nucleus absorbs a photon with the energy above the gdr region , which leads to emission of many neutrons . as seen in figs . [ fig : rhic ] and [ fig : lhc ] , the probabilities of the simultaneous emission of three and more neutrons are small and such processes with participation of high energy photons are distributed over the large area in the plots . for the sake of completeness the cross sections of the mutual dissociation without neutrons emitted from one or both of the collision partners are also included . the rates of such processes , when mainly the proton emission takes place , are small . this is another difference between the electromagnetic and nuclear dissociation . the nuclear interaction events , when only a proton is removed from one or both of the nuclei , are very probable , see figs . [ fig : rhic ] and [ fig : lhc ] . since its experimental discovery , the process of electromagnetic dissociation of heavy ions has been studied only in fixed target experiments . in the year 2000 the relativistic heavy ion collider ( rhic ) became operational at bnl and among other experiments the electromagnetic dissociation of ultrarelativistic heavy ions can now be investigated in collider kinematics by means of the zero degree calorimeters ( zdc ) @xcite . this makes possible to study the mutual heavy ion dissociation process for the first time . in the present paper the equivalent photon method and the abrasion model for grazing nuclear collisions were extended to the case of mutual dissociation of collision partners . our numerical results for the total heavy ion dissociation cross sections are very close to the results of related studies , ref . @xcite . restricting ourselves to the domain of equivalent photon energies below 24 mev , where only the giant resonance excitations are possible in electromagnetic dissociation , we obtained the cross section of correlated 1n-1n emission close to one of ref . however , as we have found , at collider energies the neutron emission process in mutual electromagnetic dissociation is not entirely exhausted by the simultaneous excitation and decay of the giant resonances in both of the colliding nuclei . apart from the mutual gdr excitation , asymmetric processes with the gdr excitation in one of the nuclei accompanied by a photonuclear reaction in the other nucleus are very probable . a wide set of photonuclear reactions should be taken into account to obtain a realistic estimation of the mutual dissociation rate . as we found , the cross sections of emission of two and especially three neutrons in mutual dissociation are very sensitive to the presence of the next - to - leading - order processes where a collision partner ( or even both of the nuclei ) absorbs a pair of photons in a collision event . the contribution of multiple photon absorption processes to the total _ single _ electromagnetic dissociation cross section at ultrarelativistic energies is 12% only . this is very different from the case of _ mutual _ electromagnetic dissociation where 30 - 40% of events are due to multiple photon absorption . indeed , the main part of mutual dissociation events takes place in close collisions with small @xmath281 , where the probability to absorb a virtual photon is large , and hence , two or more photons can be absorbed by each of the collision partners . we have examined the reliability of our results by studying their sensitivity to the variation of input data and parameters . trying to answer a key question on whether the mutual dissociation cross section can be calculated with high accuracy , we have critically reviewed our model assumptions and the results of previous theoretical and experimental studies of photonuclear reactions and heavy ion dissociation processes . the ambiguity in the calculations of @xmath263 correlated emission cross section alone , @xmath282 , is found to be up to 15% . this is mainly due to the difference in the values of the photoneutron cross sections measured in different experiments . however , the ambiguity is lower , @xmath283% , if the sum of one and two neutron emission channels , @xmath284 , is considered . therefore , it is a kind of cumulative neutron emission rate which should be used as the luminosity measure at colliders . we have found several distinctive features of the mutual electromagnetic dissociation process which are helpful for its experimental identification . beside the enhancement of @xmath105 and @xmath106 emission channels , the electromagnetic interaction leads to very asymmetric mutual dissociation channels where only one neutron is lost by one collision partner while many neutrons are lost by another partner . such dissociation pattern is very unlikely in grazing nuclear collisions with participation of the strong nuclear forces . the correlated emission of one or two neutrons in both the forward and backward directions without any additional particles in the mid - rapidity region can be used as a clear sign of the electromagnetic dissociation of ultrarelativistic heavy ions . the identification of such mutual dissociation events and counting their rates in both arms of zdc along with the calculation results of the present paper can provide a basis for an absolute luminosity calibration at rhic . similar methods can be used for the alice heavy ion experiment @xcite planned at the future large hadron collider ( lhc ) to be build at cern . we are grateful to a.j . baltz , c.a . bertulani , a.s . botvina , m.b . chadwick , g. dellacasa , g. giacomelli , j.j . gaardhje , a.s . iljinov , a.b . kurepin and m. murray for useful discussions . special thanks are due to s.n . white who pointed our attention to the subject of the paper and encouraged us in the present study . i.a.p . thanks enea and the niels bohr institute for the warm hospitality and financial support . i.n.m . acknowledges the support from the niels bohr institute and from the grant rfbr-00 - 15 - 96590 . a.m. van den berg , d. chmielewska , j.a . bordewijk , s. brandenburg , a. van der woude , y. blumenfeld , n. frasaria , j.c . roynette , j.a . scarpaci , t. suomijrvi , n. alamanos , f. auger , a. gillibert , p. roussel - chomaz , j. blomgren , l. nilsson , n. olsson , r. turcotte , nucl . phys . * a578 * , 238 ( 1994 ) . a. grnschloss , k. boretzky , t. aumann , c.a . bertulani , j. cub , w. dostal , b. eberlein , th.w . elze , h. emling , j. holeczerk , r. holzmann , m. kaspar , j.v . kratz , r. kulessa , y. leifels , a. leistenschneider , e. lubkiewicz , s. mordechai , i. peter , p. reiter , m. rejmund , h. simon , k. stelzer , a. surowiec , k. smmerer , j. stroth , e. wajda , w. walus , s. wan and h.j . wollersheim , phys . c * 60 * , 051601 ( 1999 ) .

the nonequilibrium stationary state of an exclusive genetic switch is considered . the model comprises two competing species and a single binding site which , when bound to by a protein of one species , causes the other species to be repressed . the model may be thought of as a minimal model of the power struggle between two competing parties . exact solutions are given for the limits of vanishing binding / unbinding rates and infinite binding / unbinding rates . a mean field theory is introduced which is exact in the limit of vanishing binding / unbinding rates . the mean field theory and numerical simulations reveal that generically bistability occurs and the system is in a symmetry broken state . an exact perturbative solution which in principle allows the nonequilibrium stationary state to be computed is also developed and computed to first and second order .

1105.0301

genetic networks are interacting , many - component systems of genes , rna and proteins , that control the functions of living cells @xcite . they exhibit complex , nonlinear behaviour due to the feedback between the expression of different genes . at a simplistic level of description the gene sequence , through transcription by rna and translation by mrna , leads to the production of proteins . these proteins may themselves act as transcription factors which may switch on or off the activity of other genes @xcite . a particularly simple example of a genetic network is a toggle switch formed from pairs of genes that mutually repress each other s expressions @xcite . such switches serve as microscopic models for bistable and oscillatory states @xcite and importantly may be synthesized @xcite . genetic switches may exhibit bistability where there are two possible dynamically stable long - lived states for the switch . there has been considerable interest in how bistability may be maintained and how it is effected by stochastic fluctuations due to small numbers of proteins and intrinsic noise @xcite . in particular the switching time between the two states has been measured and numerical techniques have been devised to study the switching time within theoretical models @xcite . in this paper we consider the simplest toggle switch introduced by warren and p.r . ten wolde @xcite which we will refer to as the exclusive switch ( this model is referred to in @xcite as the exclusive switch without cooperative binding ) . the model comprises two genes labelled 1,2 each leading to the production of proteins @xmath0 and @xmath1 . these proteins also degrade stochastically . there is a single mutual binding site to which either an @xmath0 or @xmath1 may bind and when bound repress the production of the other protein . thus when an @xmath0 is bound , the @xmath0 population fluctuates around some steady state value determined by the balance of production and degradation while the @xmath1 population degrades towards zero . this switch may be considered more generally as a minimal model for the power struggle between two competing parties . this is illustrated by the following charicature of `` mob dynamics '' . there are two competing parties or gangs of individuals and room for only one individual to wield absolute power . when an individual of one party is in power his own party membership may grow but the other party membership dwindles . random influences imply that the control of power is occasionally lost and power may be seized by any individual . thus in a temporary power vacuum the membership of the minority party will increase and there is a small chance that a member of this party will seize power and that the minority will eventually become the majority . in the context of statistical physics the exclusive switch is an example of a nonequilibrium system . this is because the microscopic stochastic dynamics do not obey detailed balance . for example when an @xmath0 protein is bound the degradation of an @xmath1 protein is irreversible . the structure of nonequlibrium stationary states has been of considerable interest and is generally characterised by the existence of probability currents in the stationary state ( which do not exist in equilibrium stationary states due to the presence of detailed balance)@xcite . for example it has been shown that spontaneous symmetry breaking may occur in non - equilibrium systems under conditions where it is precluded from their equilibrium counterparts e.g. in one spatial dimension @xcite . the bistability exhibited in the exclusive switch may be thought of as symmetry breaking where although the microscopic dynamics is symmetric between the two proteins , the stationary state comprises two possible long - lived dynamical states in which the symmetry is broken and one protein dominates . in an equilibrium system the switching time between the two symmetry - broken states may be estimated by the arrhenius law @xmath2 where the free energy barrier @xmath3 is extensive in the system size . for a nonequilibrium system on the other hand the free energy or indeed the stationary state is not known a priori and one is required to construct the stationary state on a model by model basis . for genetic switches there has been recent interest in developing analytical approaches to describe the stationary states @xcite . in the present work we study analytically the nonequilibrium stationary state of the exclusive switch . previous analytical studies have concentrated on systems with only one gene @xcite . our aim is to understand whether symmetry breaking occurs and , if so , the nature of the symmetry broken state . to this end we develop two analytical approaches . first we construct a mean field theory . then we develop a perturbative approach that in principle allows the nonequilibrium stationary state to be computed exactly and we present analytical results to first order . a complementary approach to this system has been developed in @xcite , where an approximation scheme based on effective interactions is used . the paper is organised as follows . in section 2 we define the exclusive switch model and write down the system of master equations that describe the system . we also present some numerical simulations which illustrate the nature of the symmetry breaking and consider exactly solvable limits . in section 3 we present a mean field theory and compare to stochastic simulations . in section 4 we develop a perturbative approach , compute the results to first order and compare with simulation results . conclusions are drawn in section 5 . the state of the system is defined by : the number of free proteins of type 1 , @xmath4 ; the number of free proteins of type 2 , @xmath5 , and the state of the switch , @xmath6 , which takes value 0 if no protein is bound , value 1 if a protein of type 1 is bound , and value 2 if a protein of type 2 is bound . the stochastic dynamical processes are as follows : a protein degrades ( leaves the system ) with rate @xmath7 ; when the switch state is 0 proteins of both type 1 and 2 are produced with rate @xmath8 ; when the switch state is 1 proteins of type 1 produced with rate @xmath8 and when the switch state is 2 proteins of type 2 produced with rate @xmath8 ; if the switch state is 0 a protein binds with rate @xmath9 and when binding occurs the switch state changes to the type of bound protein and the number of free proteins is reduced by 1 ; a bound protein unbinds with rate @xmath10 and when unbinding occurs the switch state changes to 0 and the number of free proteins is increased by 1 . we shall consider the joint probabilities @xmath11 of the protein numbers @xmath4 , @xmath5 , switch state @xmath6 . following the stochastic dynamical processes described above the system of master equations that defines the model can be written as follows @xmath12\nonumber \\ & + & d[(n_1 + 1)p_0(n_1 + 1,n_2)+(n_2 + 1)p_0(n_1,n_2 + 1)\nonumber\\ & & -(n_1+n_2)p_0(n_1,n_2)]-b(n_1+n_2)p_0(n_1,n_2)\nonumber \\ & + & u[p_1(n_1 - 1,n_2)+p_2(n_1,n_2 - 1)]\label{master0 } \\ \frac{\partial p_1}{\partial t}(n_1,n_2 ) & = & g[p_1(n_1 - 1,n_2)-p_1(n_1,n_2)]\nonumber \\ & + & d[(n_1 + 1)p_1(n_1 + 1,n_2)+(n_2 + 1)p_1(n_1,n_2 + 1)\nonumber\\ & & -(n_1+n_2)p_1(n_1,n_2)]\nonumber\\ & + & b(n_1 + 1)p_0(n_1 + 1,n_2)-up_1(n_1,n_2 ) \label{master1 } \\ \frac{\partial p_2}{\partial t}(n_1,n_2 ) & = & g[p_2(n_1,n_2 - 1)-p_2(n_1,n_2)]\nonumber\\ & + & d[(n_1 + 1)p_2(n_1 + 1,n_2)+(n_2 + 1)p_2(n_1,n_2 + 1)\nonumber\\ & & -(n_1+n_2)p_2(n_1,n_2)]\nonumber\\ & + & b(n_2 + 1)p_0(n_1,n_2 + 1)-up_2(n_1,n_2 ) \label{master2}\end{aligned}\ ] ] where @xmath8 is the generation rate of a protein ( when the generation is not suppressed by the switch state ) , @xmath7 is the degeneration rate of a single protein , @xmath9 is the binding rate of a single protein and @xmath10 is the unbinding rate of the bound protein . the whole problem is clearly symmetric with respect to the variables @xmath13 and @xmath14 . also note that , the degeneration term is the same for the three equations , since it does not depend on the value of @xmath6 . in order to illustrate the qualitative behaviour , we first present stochastic simulations of the exclusive switch . these simulations are performed with a gillespie algorithm @xcite , in which the reactions described in the model are given a certain probability , depending on the state of the system and the value of the parameters . these probabilities are used to determine stochastically which reaction is going to happen next , and when it will happen . the time that the system spends in a given state @xmath15 is normalized to obtain the probability distributions . the reference values for the simulations are the typical values for bacteria such as escherichia coli @xcite @xmath16 in the system there is a clear symmetry between the two proteins species , since they undergo the same microscopic reactions . however , at any given time the system is typically dominated by one of the proteins ; thus the symmetry is broken . the reason for this is that , once a protein , e.g. of type @xmath0 , binds to the promoter site , proteins @xmath1 start disappearing , while the number of @xmath0 fluctuates around a steady value . that means that when the bound protein unbinds ( as it will do eventually due to stochasticity of the system ) , it is much more probable for proteins @xmath0 to bind to the promoter site again , since there are more of them . at the same time it is more difficult for proteins @xmath1 to bind to the promoter site . however the @xmath1 will not disappear permanently from the system ( there is no absorbing state ) , and will be produced again the moment the bound protein @xmath0 unbinds . thus , with a small probability , proteins @xmath1 will be able to bind again to the promoter site , and become the dominant species , as proteins @xmath0 start to degenerate . this means that there are two symmetry - broken states , in which one species is much more abundant than the other . with regard to the probability distribution @xmath17 , this bistability is translated into two peaks , concentrated around the axes , i.e , where one of the protein numbers is almost zero . this is illustrated in a contour plot in figure [ fig : contour ] for @xmath18 . however , this bistability depends strongly on the value of the parameters : the bigger the value of @xmath10 , the more irrelevant is the switch state for the dynamics of the protein , and the less important is the bistability we have described . for example in figure [ fig : contour ] , when @xmath8,@xmath9 and @xmath7 are kept constant and @xmath10 is increased , the peaks move together and eventually merge at some value of @xmath19 . thus there is an apparent transition from a distribution with two symmetry - related peaks to a distribution with one symmetric peak . we wish to study the nature of this transition i.e. is there an underlying phase transition at a finite value of @xmath10 where the system changes from symmetry - broken behaviour to symmetric behaviour , or is the transition simply due to two peaks coming closer together and no longer being resolved ? the latter would correspond to a ` geometrical transition ' in the form of @xmath17 but would not correspond to any underlying phase transition . obtained from stochastic simulations . a transition from a two peak regime to a one peak regime occurs as the unbinding parameter @xmath10 increases . @xmath8,@xmath9 and @xmath7 are kept equal to the e. coli values ( [ ecv]),title="fig : " ] * a * ) @xmath18 . obtained from stochastic simulations . a transition from a two peak regime to a one peak regime occurs as the unbinding parameter @xmath10 increases . @xmath8,@xmath9 and @xmath7 are kept equal to the e. coli values ( [ ecv]),title="fig : " ] * b * ) @xmath20 . obtained from stochastic simulations . a transition from a two peak regime to a one peak regime occurs as the unbinding parameter @xmath10 increases . @xmath8,@xmath9 and @xmath7 are kept equal to the e. coli values ( [ ecv ] ) ] obtained from stochastic simulations . a transition from a two peak regime to a one peak regime occurs as the unbinding parameter @xmath10 increases . @xmath8,@xmath9 and @xmath7 are kept equal to the e. coli values ( [ ecv ] ) ] although the whole probability distribution is always symmetric , distributions @xmath21 and @xmath22 will be a priori asymmetric , since they describe the probability of the number of proteins when a protein @xmath13 or @xmath14 are bound , which are not symmetric situations . let us now define @xmath23 and @xmath24 as the probabilty masses of @xmath21 on either sides of the diagonal @xmath25 : @xmath26 we can now study how @xmath23 , @xmath24 change with @xmath10 , and whether there is a clear transition between the situation in which they are different , and the one in which they are equal to each other ( if there is any ) . figure [ fig : rarb ] shows that these two quantities approach each other in a continuous way , and they are equal only when @xmath27 , that is , when the only possible state of the switch is @xmath28 and it has no longer any effect on the protein dynamics . therefore , the probability distributions @xmath29 and hence @xmath30 tend to zero , all the probability being concentrated in the distribution @xmath31 , which is always symmetric . defined in ( [ rarb])@xmath23 for @xmath32 and @xmath24 for @xmath33as @xmath10 increases . ] despite the fact that a change in the shape of the distribution is observed ( figure [ fig : p1dis ] ) , there is no evidence of the typical singularities that appear in a phase transition . instead @xmath21 appears to deform continuously into a symmetric distribution when @xmath34 . we conclude that ( at least for these parameter values ) @xmath35 remains asymmetric for @xmath36 , and as a consequence so does @xmath22 , and there is no transition to a symmetric state in this marginal distribution . as figure [ fig : p1dis ] shows , @xmath21 has only one peak for different values of @xmath10 , and even if it becomes smaller as @xmath10 increases , it does not become symmetric at any point . we deduce that , even though @xmath17 appears to become symmetric at some finite value of @xmath10 ( see figure [ fig : contour ] ) , there is no true phase transition between symmetric and asymmetric regimes , since @xmath21 and @xmath22 always remain asymmetric . thus , the bistability of the switch is always present , with the asymmetric distributions @xmath37 decreasing as the switch state becomes less important , that is , as @xmath10 increases . , for different values of @xmath10 . as @xmath10 increases , the peak of the distribution moves towards the diagonal @xmath25 , but the distribution is not completely symmetric as long as @xmath10 is finite . also , the probability mass of the distribution decays as @xmath10 increases , and is transferred to the symmetrical distribution @xmath31.,title="fig : " ] * a * ) @xmath38 . , for different values of @xmath10 . as @xmath10 increases , the peak of the distribution moves towards the diagonal @xmath25 , but the distribution is not completely symmetric as long as @xmath10 is finite . also , the probability mass of the distribution decays as @xmath10 increases , and is transferred to the symmetrical distribution @xmath31.,title="fig : " ] * b * ) @xmath18 . , for different values of @xmath10 . as @xmath10 increases , the peak of the distribution moves towards the diagonal @xmath25 , but the distribution is not completely symmetric as long as @xmath10 is finite . also , the probability mass of the distribution decays as @xmath10 increases , and is transferred to the symmetrical distribution @xmath31 . ] , for different values of @xmath10 . as @xmath10 increases , the peak of the distribution moves towards the diagonal @xmath25 , but the distribution is not completely symmetric as long as @xmath10 is finite . also , the probability mass of the distribution decays as @xmath10 increases , and is transferred to the symmetrical distribution @xmath31 . ] we now consider the system in the limit @xmath10 , @xmath39 with the binding constant @xmath40 held fixed . in this limit the system will equilibrate between binding / unbinding events . when the switch is in state 1 the number of type 2 proteins decays to zero and the number of type 1 is given by poissonian statistics for the generation / degeneration processes : @xmath41 where @xmath42 is the probability of the switch being in state 1 ( @xmath43 ) . similarly , in state 2 the number of type 2 is given by poissonian statistics @xmath44 and in state 0 both @xmath4 and @xmath5 are given by poissonian statistics @xmath45 in order to fix the probabilities @xmath46,@xmath42,@xmath47 we consider the master equation for @xmath42 which can be obtained from summing ( [ master1 ] ) over the variables @xmath48 : @xmath49 where the quantity @xmath50 is the mean value of @xmath4 given the system is in switch state @xmath28 , multiplied by the probability of being in switch state @xmath28 , i.e , @xmath51 . in the stationary state we have @xmath52 and ( [ r1dot ] ) becomes @xmath53 a similar equation holds for @xmath47 and the normalisation of probability yields @xmath54 thus we see that the system has three long - lived states : when @xmath55 the system is dominated by type 1 proteins ; when @xmath56 the system is dominated by type 2 proteins , and when @xmath28 the system is in a symmetric state . when @xmath57 the symmetric state has zero weight ( @xmath58 ) and the system is either in the @xmath55 state or the @xmath56 state . the switching time between the two states is expected to diverge in this limit therefore the system exhibits symmetry breaking . note that the transition from the @xmath55 state to the @xmath56 state is through the symmetric @xmath28 state . in this limit the unbinding and binding events happen on a faster time scale than the growth and degradation of proteins . therefore the switch state becomes decoupled from the numbers of protein and the probabilities obey @xmath60 where @xmath17 is the probability of @xmath4,@xmath5 proteins being in the system regardless of switch state . that is , for given number of proteins @xmath4 and @xmath5 ( which include any bound protein ) the switch probabilities equilibrate and obey @xmath61 which may be solved to obtain @xmath62^{-1 } \quad r_1= k n_1 r_0 \quad r_2= k n_2 r_0\ ] ] where the binding constant @xmath63 is given by ( [ bc ] ) . a master equation for the evolution of @xmath17 on the slower timescale on which generation and degeneration events occur may then be written down : @xmath64p(n_1 - 1,n_2 ) -d[n_1-r_1(n_1,n_2)]p(n_1,n_2 ) \label{meubinf } \\ & -g[1-r_2(n_1,n_2)p(n_1,n_2 ) ] + d[n_1 + 1-r_1(n_1 + 1,n_2)]p(n_1 + 1,n_2)\nonumber\\ & + g[1-r_1(n_1,n_2 - 1 ) ] p(n_1,n_2 - 1 ) -d[n_2-r_2(n_1,n_2)]p(n_1,n_2)\nonumber \\ & -g[1-r_1(n_1,n_2 ) ] p(n_1,n_2 ) + d[n_2 + 1-r_2(n_1,n_2 + 1)]p(n_1,n_2 + 1)\;. \nonumber \ ] ] the terms with coefficient @xmath8 in ( [ meubinf ] ) represent generation of @xmath65 when the switch is not in switch state i. the terms with coefficient @xmath7 in ( [ meubinf ] ) represent reduction of @xmath65 with rate @xmath66 when the switch is not state @xmath67 and reduction with rate @xmath68 when the switch is in state @xmath67 ( this is because the bound protein does not degrade ) . the stationary state of ( [ meubinf ] ) obeys detailed balance with respect to generation and degradation of @xmath4 and @xmath5 individually , thus @xmath69}{[n_1-r_1(n_1,n_2 ) ] } \\ & = & p(n_1 - 1,n_2)\frac{g}{d}\frac{[1+k(n_1 - 1)]}{[1+k(n_1+n_2 - 1 ) ] } \frac{1}{n_1 } \frac{[1+k(n_1+n_2)]}{[1+k(n_1+n_2 - 1)]}\nonumber \label{pubinf}\end{aligned}\ ] ] with a similar equation holding for detailed balance in @xmath5 . then these equations may be iterated to obtain @xmath70 the constant @xmath71 will be determined by normalisation of the sum of probabilities to one . the expression ( [ pubinf ] ) is a single - peaked distribution which is symmetric in @xmath4 and @xmath5 . to see this one can identify the stationary points of @xmath17 through the conditions @xmath72 and @xmath73 which yield @xmath74 the solution of these equations is @xmath75 where , when @xmath76 is large , @xmath77n - 2 \frac{g}{d } = 0\;. \label{nubinf}\ ] ] thus , in the @xmath78 limit the system has reached a symmetric state with the probability distribution peaked at @xmath75 . in this section we develop a mean field theory in which some correlations in the numbers of proteins are ignored . we start from the exact master equations ( [ master0][master2 ] ) for the evolution of probabilities @xmath11 . the zeroth moments of @xmath65 are the probabilities @xmath79 i.e. @xmath80 we now define the first moments of @xmath81 of @xmath65 as follows @xmath82 and the second moments @xmath83 summing ( [ master0 ] ) gives @xmath84 \label{r0t}\\ \frac{\partial \overline{n_1 ^ 0}}{\partial t } & = & gr_0 -d\overline{n_1 ^ 0 } -b[\overline{(n_1n_1)^0}+\overline{(n_1n_2)^0 } ] + u[\overline{(n_1)^1}+\overline{(n_1)^2}+r_1]\;. \label{n10t}\end{aligned}\ ] ] note that the physical meaning of e.g. @xmath81 is the probability of being in switch state @xmath6 ( @xmath79 ) multiplied by the mean number of type @xmath67 , _ given _ that the switch is in state @xmath6 . similarly , summing ( [ master1 ] ) gives @xmath85 we now invoke symmetry between switch state 1 and 2 : @xmath86 then the exact steady - state versions of equations ( [ r0t][n21 t ] ) read @xmath87 = gr_0 -d\overline{n_1 ^ 0 } + u [ \overline{(n_1)^1}+ \overline{(n_1)^2}+r_1 ] \label{n10ss}\\ b\overline{(n_1n_1)^0 } = -gr_1 + ( d+u)\overline{n_1 ^ 1 } + b\overline{(n_1)^0 } \label{n11ss}\\ b\overline{(n_1n_2)^0 } = ( d+u)\overline{n_2 ^ 1}\;. \label{n12ss}\end{aligned}\ ] ] note that if we sum ( [ n10ss][n12ss ] ) we obtain the exact relation @xmath88 which simply gives the overall birth / death balance for @xmath4 . also note that ( [ n11ss],[n12ss ] ) give exact relations between the second moments and first moments . however to actually evaluate these moments one would have to consider equations for higher moments , leading to a hierarchy of equations . we now make a mean - field approximation that expresses second moments in terms of first moments : @xmath89 note that a symmetry condition from ( [ symm ] ) has been explicitly used in the last equation of ( [ n1n2mf ] ) . the first relation ( [ n1n1mf ] ) comes from the assumption that @xmath4 has a poisson distribution when the switch is in the 0 state . this means that the second moment is equal to the square of the mean plus the mean itself . however , there is an important factor @xmath46 , which comes from the fact that @xmath90 is the mean square value of @xmath4 _ given _ that the switch is in the 0 state and @xmath91 is the mean value of @xmath4 _ given _ that the switch is in the 0 state . the poisson approximation is in fact exact in the limit where @xmath10,@xmath9 tend to zero ( see section [ sec : ub0 ] ) . the second relation ( [ n1n2mf ] ) is a simple factorization scheme which ignores correlations between the values of @xmath4 and @xmath5 when the switch state is 0 . using this approximation scheme ( [ n11ss ] ) becomes @xmath92 \label{n11mf}\ ] ] and ( [ n12ss ] ) becomes @xmath93 using expressions ( [ n11mf ] ) and ( [ n12mf ] ) in ( [ n10ss ] ) , combined with the previous approximations ( [ n1n1mf ] ) and ( [ n1n2mf ] ) yields the following quadratic equation for @xmath91 : @xmath94 \left(\frac{\overline{n_1 ^ 0}}{r_0}\right ) -g=0\;. \label{n10mf}\ ] ] one must take the positive root of this quadratic which yields @xmath95\;.\ ] ] then using ( [ rss ] ) one obtains @xmath96^{-1}\\ \overline{n_1 ^ 0 } & = & \left(\frac{\overline{n_1 ^ 0}}{r_0}\right)\left [ 1 + \frac{2b}{u}\frac{\overline{n_1 ^ 0}}{r_0}\right]^{-1}\;.\end{aligned}\ ] ] one may check the limits of section 2 from the quadratic ( [ n10mf ] ) . in the limit @xmath9,@xmath10 @xmath97 with @xmath98 one obtains @xmath99 and @xmath100^{-1}$ ] in agreement with section 2.3 where it is shown that @xmath4 follows a poisson distribution with mean @xmath101 when the switch is in state @xmath55 or @xmath102 . in the limit @xmath9,@xmath10 @xmath103 with @xmath98 fixed the quadratic ( [ n10mf ] ) reduces to @xmath104 this quadratic for @xmath105 is the same as the quadratic ( [ nubinf ] ) for the value of @xmath106 that maximises @xmath17 in the exact solution of section [ sec : ubinfty ] . the mean field theory we have developed can be compared with the simulations by studying the zeroth and first order moments of the probabilty distributions , i.e. , @xmath107 . ( the probabilities @xmath42 and @xmath47 may automatically be obtained from @xmath46 , since @xmath108 . ) different values of the mentioned quantities are given in table @xmath13 , where the parameters of the models have different values . the reference for this table is the set of e. coli values , and only the parameters that are changed are written . for the simulations performed , @xmath46 is always in good agreement with the mean field theory approximation , and so are @xmath42 and @xmath47 . also , @xmath50 is in quite good agreement , too . .results from different quantities from simulations and mean field theory approach . the values of @xmath46 and @xmath50 predicted by the mean field theory are always in good agreement with the simulations . @xmath109 and @xmath110 predictions are quite far from the simulation results , except on the limit @xmath111 , where our approximation is exact . [ cols="<,<,<,<,<",options="header " , ] @xmath109 and @xmath110 are different in the mean field theory , which is an improvement over simpler approximations where they have the same value . however , the values of @xmath109 and @xmath110 are rather different from the simulations , and this comes from ignoring higher correlations of the numbers of proteins and the state of the switch . only when @xmath111 are the values in close agreement with the simulation values as expected in this limit where the mean field theory is exact . mean field theory can , in principle , be improved by considering higher order moments and correlations . however , the algebra soon gets quite complicated . in this section we develop a perturbative approach that allows the steady state probabilities @xmath11 to be computed as a power series in the unbinding rate @xmath10 . we begin by considering the formal solution of the master equation system ( [ master0][master2 ] ) . to transform the system of equations into a system of partial differential equations , we take the generating function of the different probability distributions : @xmath112 where @xmath113 . in this way we obtain a system of linear partial differential equations with non - constant coefficents : @xmath114 \frac{\partial k_0}{\partial z_1 } + \left[d-(d+b)z_2\right]\frac{\partial k_0}{\partial z_2}\nonumber & & \\ \hspace{7cm}+uz_1k_1+uz_2k_2 & = & 0\label{k0ss}\\ \left[g(z_1 - 1)-u\right]k_1+d(1-z_1)\frac{\partial k_1}{\partial z_1 } + d(1-z_2)\frac{\partial k_1}{\partial z_2 } + b\frac{\partial k_0}{\partial z_1 } & = & 0\label{k1ss}\\ \left[g(z_2 - 1)-u\right]k_2+d(1-z_1)\frac{\partial k_2}{\partial z_1}+d(1-z_2 ) \frac{\partial k_2}{\partial z_2}+b\frac{\partial k_0}{\partial z_2 } & = & 0 \label{k2ss}\end{aligned}\ ] ] where the right - hand side terms have been set to @xmath115 , since the stationary probabilities @xmath11 are the main quantities to be determined in this paper . the second and the third equation of the system work in a completely analogous way , because of the symmetry of species @xmath13 and @xmath14 , so it will be enough to deal with the first two equations . we also note the symmetries @xmath116 and @xmath117 . in appendix @xmath118 we give a formal solution to the system ( [ k0ss][k2ss ] ) . however , in practice it is not clear how to actually compute e.g. probability distributions from this solution . in order to do this we develop instead a perturbative approach . in this section we develop a perturbative approach to the problem of finding the exact stationary state . to do so we require a suitable small parameter of the model which we choose to be @xmath10 , the unbinding parameter . in the @xmath119 , the exact solution is simple : if , for example , one protein of type @xmath13 is bound , the proteins of this kind will obey the usual poisson distribution regulated by death and birth terms , while the number of proteins of type @xmath14 will just decay to @xmath115 . this limit is the starting point for a perturbative solution , wherein the probability distribution will be expanded in a power series of @xmath10 : @xmath120 owing to the symmetry of the system @xmath121 we need only consider @xmath122 and @xmath21 . writing out the expansion explicitly we have @xmath123 note that the constant term @xmath124 since @xmath125 in the limit of no unbinding . this approach also makes sense when the typical e. coli values for the parameters are considered @xcite : @xmath126 @xmath10 is , along with @xmath7 , the smallest of the parameters . an expansion in @xmath127 could also be developed , but we find the expansion in @xmath10 more convenient . in the zeroth order of the @xmath10 expansion of the stationary master equation ( [ master2 ] ) ( with l.h.s set to zero ) we find @xmath128+d[(n_1 + 1)p_1 ^ 0(n_1 + 1,n_2)+(n_2 + 1)p_1 ^ 0(n_1,n_2 + 1)\nonumber\\ \fl & & -(n_1+n_2)p_1 ^ 0(n_1,n_2)]\;.\end{aligned}\ ] ] we define the generating function @xmath129 which obeys @xmath130 the solution of which is independent of @xmath131 : @xmath132 where @xmath133 is a constant to be determined . analogously @xmath134 . applying the normalization condition @xmath135 and the symmetry consideration @xmath136 leads to @xmath137 expanding as a power series in @xmath138 yields @xmath139 this is a poisson distribution for @xmath4 with mean @xmath101 , with @xmath5 fixed to be zero . the normalisation factor @xmath140 is so that @xmath141 is normalised to unity . we substitute the expansion ( [ ppert ] ) into the stationary master system ( [ master0][master2 ] ) with time derivatives set equal to zero . arranging orders of @xmath10 the equations may be written as @xmath142 for @xmath143 where the action of the linear operators @xmath144 is @xmath145\nonumber \\ & + d\left[(n_1 + 1)p_0^{(n)}(n_1 + 1,n_2)+(n_2 + 1)p_0^{(n)}(n_1,n_2 + 1)\right . \nonumber \\ & \left . \\ \fl \mathcal{l}_1 p_1^{(n)}(n_1,n_2 ) = & g[p_1^{(n)}(n_1 - 1,n_2)-p_1^{(n)}(n_1,n_2 ) ] \nonumber \\ & + d\left[(n_1 + 1)p_1^{(n)}(n_1 + 1,n_2)+(n_2 + 1)p_1^{(n)}(n_1,n_2 + 1)\right . \nonumber \\ & \left.-(n_1+n_2)p_1^{(n)}(n_1,n_2)\right ] \label{lp1}\end{aligned}\ ] ] and the inhomogenous terms are @xmath146 as noted above , we can first determine the zeroth order @xmath147 and @xmath148 as functions of the parameters of the model . then , owing to the form of the equation ( [ f0 ] ) , @xmath149 is determined . this allows us to solve for @xmath150 which in turn determines @xmath151 and allows us to solve for @xmath152 . continuing in this fashion the rest of the probabilities will be found following the structure : @xmath153 in general , @xmath154 will be found before @xmath155 . having laid out the general perturbation scheme and established the zeroth order ( @xmath156 ) solution , we now outline how equations ( [ lpf ] ) can be solved . we note that for each switch state @xmath6 the same linear operator @xmath144 appears at all orders @xmath157 . this means that the homogeneous parts of the equations are independent of the order ( with only the inhomogeneous term on the right hand side varying between orders ) and that they only have to be solved once . let us define a green function @xmath159 for the operator @xmath158 through @xmath160 so that the solution of ( [ lpf ] ) may be written @xmath161 we define a generating function @xmath162 the equation for which is obtained by summing ( [ q ] ) @xmath163 with @xmath164 . in order to solve ( [ k0eq ] ) we use the method of characteristics ( see e.g. @xcite ) . the characteristic equations are @xmath165 where @xmath166 is a time - like parameter and ( [ z1s],[z2s ] ) define characteristic curves along which the partial differential equation ( [ k0eq ] ) reduces to the ordinary differential equation ( [ k0s ] ) . solving ( [ z1s],[z2s ] ) yields the curves @xmath167 where @xmath168 and @xmath118,@xmath169 are two constants to be fixed . equation ( [ k0s ] ) can be rewritten as an ordinary differential equation in @xmath170 as @xmath171 = \frac{1}{d+b } v^{\frac{2gb}{(d+b)^2}-1 } { \rm e}^{-\frac{g(a+b)}{(d+b)}v } \left(\frac{d}{d+b } + a v\right)^{n_1 ^ 0 } \left(\frac{d}{d+b } + b v\right)^{n_2 ^ 0}\;. \label{k0de}\ ] ] to integrate ( [ k0de ] ) we choose an end - point of the integration as @xmath172 and set this to correspond to an arbitrary point in the @xmath173@xmath131 plane . this fixes the two constants @xmath118,@xmath169 as @xmath174 thus we obtain @xmath175 we now expand as a power series in @xmath173 , @xmath131 to obtain @xmath176 from ( [ k0def ] ) @xmath177 where @xmath178 is defined as the integral @xmath179 and @xmath180 and @xmath181 are the usual euler gamma function and confluent hypergeometric function respectively . if we define a green function for the operator @xmath182 through @xmath183 this equation only has a solution when @xmath184 . to see this we note that summing the left hand side of ( [ q1 ] ) over all @xmath4,@xmath5 yields zero i.e. the operator conserves probability . therefore one can not solve ( [ q1 ] ) or indeed ( [ lpf ] ) for an arbitrary right hand side ; one requires that the sum of the right hand side of ( [ lpf ] ) over all @xmath4,@xmath5 yields zero . however , since the null space of the operator @xmath182 is concentrated on @xmath185 ( i.e. the stationary state in ( [ p10 ] ) is proportional to @xmath186 ) we can find a solution of ( [ q1 ] ) for @xmath187 . it is simplest to proceed by considering the solution @xmath21 for an arbitrary right hand side @xmath188 @xmath189 that satisfies @xmath190 we define generating functions @xmath191 then summing ( [ lph ] ) yields @xmath192 in order to solve ( [ k1eq ] ) we again use the method of characteristics . the characteristic equations are this time @xmath193 solving the first two equations yields @xmath194 where now @xmath195 and @xmath118 , @xmath169 are constants to be fixed . the final equation becomes @xmath196 = \frac{h(z_1,z_2)}{d v } { \rm e}^{-a\frac{g}{d}v } \label{k1s2}\ ] ] we choose the integration to be from @xmath197 to @xmath172 where @xmath197 corresponds to @xmath198 and @xmath172 corresponds to an arbitrary point in the @xmath173@xmath131 plane which implies @xmath199 and @xmath200 . we then obtain the solution of ( [ k1s2 ] ) @xmath201 expanding as a power series in @xmath173,@xmath131 implies @xmath202 as discussed above , for @xmath203 we can define the green function ( [ q1 ] ) by means of which the solution of ( [ lph ] ) may be written @xmath204 then we can read off the green function from ( [ p1h ] ) as @xmath205 for @xmath185 the solution ( [ p1h ] ) reads @xmath206 some care is required with the @xmath207 term of the sum in ( [ p1hn20 ] ) since the integral @xmath208 does not converge . however the property ( [ hcon ] ) of the function @xmath209 implies that the coefficient of the offending integral is zero . to see this one can write the @xmath207 term of ( [ p1hn20 ] ) as @xmath210 in the final equality , the binomial expansion of @xmath211 has been used with the term @xmath212 not present since its coefficient vanishes due to ( [ hcon ] ) . all the integrals in ( [ rn1term ] ) then converge . the first - order contribution to the stationary probability @xmath213 is given by : @xmath214 with @xmath215\;.\ ] ] although @xmath176 is expressed in terms of known integrals in ( [ q0 ] ) , it is advisable to go back to the explicit expression of the integrals ( [ idef ] ) to evaluate the sums appearing in ( [ p01 ] ) more easily . in that way , @xmath216 can be written as : @xmath217 where the label _ symm _ refers to the fact that there will be another term equal to the written one , apart from a switch in the variables @xmath218 and @xmath219 . in [ appb ] it is shown how the expression may be simplified to the result @xmath220+symm\;. \label{p01res}\end{aligned}\ ] ] once we have @xmath150 , we can plug this result into the @xmath152 equation which becomes for @xmath203 @xmath221 where @xmath222 . we consider separately the two terms of @xmath151 . the first is @xmath223 since the calculation with the green function ( see section [ sec : gl1 ] ) is only for @xmath224 this term does not contribute and the only contribution comes from the second term in @xmath151 involving @xmath150 . the resulting expression , obtained by substituting @xmath225 in ( [ p1h ] ) with @xmath150 given by ( [ p01res ] ) , is @xmath226+symm(n_1 ^ 0 + 1,n_2 ^ 0)\bigg\}\times \nonumber \\ & { { n_2 ^ 0 \choose n_2 } } \sum_{r=0}^{n_1 } \left(\frac{g}{d}\right)^{r } \frac{1}{r ! } { { n_1 ^ 0 \choose n_1-r}}i(n_1+n_2-r , n_1 ^ 0+n_2 ^ 0-n_1-n_2 + 2r+1 ; g / d ) \nonumber\end{aligned}\ ] ] where the symmetric term in this case corresponds to the term inside the curly brackets , exchanging the places of @xmath227 and @xmath219 coming from the symmetric term in @xmath228 . for @xmath185 we obtain the result shown in ( [ p1hn20 ] ) and can not be simplified further . equations ( [ p1hn20 ] ) , ( [ p01res ] ) and ( [ p11res ] ) are the main results of this section and give closed form expressions for the first - order contributions to the stationary proabilities . the normalization constant @xmath229 appearing in ( [ p1hn20 ] ) is obtained from the condition : @xmath230=0\;.\ ] ] once the values of the probabilities @xmath231 have been obtained numerically , they are multiplied by the unbinding parameter @xmath10 and added to the zeroth order probabilities . in that way , the probability distributions @xmath232 can be computed and plotted up to the first order of the expansion . figure [ fig : finalpics ] shows the probability distributions for different values . the agreement is visually good in the first two examples : typical e. coli values ( [ ecv ] ) and another interesting case , with smaller @xmath8 . the order of magnitude is well reproduced in almost every point of the probability distribution . however , the approximation does not work accurately if the value of the unbinding parameter @xmath10 is of the order or bigger than the other parameters . although the first two examples of figure [ fig : finalpics ] seem visually in good agreement with the simulations , the best way to check this is to plot different slices of the probability distribution , that is the probability distribution of @xmath5 where @xmath4 is held constant ( figure [ fig : axes1 ] ) . we choose the values of @xmath4 to correspond to the slices with large probability mass , i.e. @xmath233 for e. coli values . figure [ fig : axes1 ] ) shows that along the slice with greatest probability mass ( @xmath234 ) the analytical and simulation plots show good agreement . for @xmath235 , there is reasonable agreement for the e. coli values ( figure [ fig : axes1 ] b ) ) , whereas for larger @xmath10 the agreement is not so good ( figure [ fig : axes1 ] d ) ) . since the proposed method is general , calculations can be performed up to any necessary order to get better results . for example , in the case of e. coli in axes @xmath236 , @xmath237 is enough to compute the second order to have very good results ( figure [ fig : axes2 ] ) . in general , the method can be iterated as many times as required . ) . the probability distributions have the same shape and the order of magnitude is well reproduced in almost every point . b ) @xmath238 the order of magnitude is again well reproduced . in this case , the two peaks get closer to the origin as the ratio g / d is smaller . c ) e. coli values with @xmath18 . the approximation at first order is no longer accurate as the value of @xmath10 is no longer small compared to the rest of the parameters of the model.,title="fig : " ] * a * ) e. coli values . ) . the probability distributions have the same shape and the order of magnitude is well reproduced in almost every point . b ) @xmath238 the order of magnitude is again well reproduced . in this case , the two peaks get closer to the origin as the ratio g / d is smaller . c ) e. coli values with @xmath18 . the approximation at first order is no longer accurate as the value of @xmath10 is no longer small compared to the rest of the parameters of the model.,title="fig : " ] * b * ) @xmath238 . ) . the probability distributions have the same shape and the order of magnitude is well reproduced in almost every point . b ) @xmath238 the order of magnitude is again well reproduced . in this case , the two peaks get closer to the origin as the ratio g / d is smaller . c ) e. coli values with @xmath18 . the approximation at first order is no longer accurate as the value of @xmath10 is no longer small compared to the rest of the parameters of the model.,title="fig : " ] * c * ) e. coli values with @xmath18 . where @xmath239 is fixed and chosen to correspond to slices with largest probability mass . the agreement in slice @xmath234 is good for both figures a ) and c ) . there is some quantitative difference in b ) , which corresponds to the e. coli values in slices with less probability mass . this is improved with second order calculations , as can be seen in figure [ fig : axes2 ] . in d ) the difference is clear , since the first order approximation is no longer accurate , as discussed in figure [ fig : finalpics ] . the error in the values is negligible and , in all the cases , is smaller than the size of the used symbols.,title="fig : " ] * a * ) e. coli values . slice @xmath234 . where @xmath239 is fixed and chosen to correspond to slices with largest probability mass . the agreement in slice @xmath234 is good for both figures a ) and c ) . there is some quantitative difference in b ) , which corresponds to the e. coli values in slices with less probability mass . this is improved with second order calculations , as can be seen in figure [ fig : axes2 ] . in d ) the difference is clear , since the first order approximation is no longer accurate , as discussed in figure [ fig : finalpics ] . the error in the values is negligible and , in all the cases , is smaller than the size of the used symbols.,title="fig : " ] * b * ) e. coli values . slices @xmath240 . where @xmath239 is fixed and chosen to correspond to slices with largest probability mass . the agreement in slice @xmath234 is good for both figures a ) and c ) . there is some quantitative difference in b ) , which corresponds to the e. coli values in slices with less probability mass . this is improved with second order calculations , as can be seen in figure [ fig : axes2 ] . in d ) the difference is clear , since the first order approximation is no longer accurate , as discussed in figure [ fig : finalpics ] . the error in the values is negligible and , in all the cases , is smaller than the size of the used symbols.,title="fig : " ] * c * ) @xmath238 . slice @xmath234 . where @xmath239 is fixed and chosen to correspond to slices with largest probability mass . the agreement in slice @xmath234 is good for both figures a ) and c ) . there is some quantitative difference in b ) , which corresponds to the e. coli values in slices with less probability mass . this is improved with second order calculations , as can be seen in figure [ fig : axes2 ] . in d ) the difference is clear , since the first order approximation is no longer accurate , as discussed in figure [ fig : finalpics ] . the error in the values is negligible and , in all the cases , is smaller than the size of the used symbols.,title="fig : " ] * d * ) e. coli values with @xmath18 . slice @xmath237 . and @xmath241 from second order analytical calculations and simulations , for e. coli values . the second order is clearly enough to get accurate results . ] the exclusive genetic switch presented in this paper represents a minimal model of two populations that compete in an indirect way , that is , through a common promoter site that when occupied by a member of one population stops the production of the other . it is a non - equilibrium system because the microscopic processes are irreversible and detailed balance does not hold . the non equilibrium stationary state exhibits interesting properties of bistability i.e. generically at any given time the system is dominated by one of the populations . we have developed two main analytical approaches to study the stationary state : a mean field theory and an exact perturbative approach . in addition we have presented some exactly solvable limits . we also have studied the system and checked the analytical results by using monte carlo simulations . these simulations show that the symmetry of the system is always broken , that is , the system is always functioning as a switch , with two opposite states in which one population is much more abundant than the other . this holds generically except in the limit @xmath78 where the state is symmetric . as the relevant parameters for the switch become smaller , the probability of finding bound proteins decreases , but the distributions always remain asymmetric , showing no phase transition between asymmetric and symmetric states the mean field theory of section [ sec : mft ] is based on exact moment equations which are factorised at the level of second moments . the approximation scheme is constructed to be exact in the limiting parameter case of vanishing binding / unbinding rates @xmath9 , @xmath10 with the binding constant @xmath242 held constant the theory shows good agreement with some quantities obtained from the simulation , but there are some other quantities whose agreement varies and depends on the value of the parameters . the mean - field theory could be improved upon by systematically considering higher order moments and correlations . as a first attempt , to our knowledge , at an exact solution of this nonequilibrium system we have developed an exact perturbative approach which consists of an expansion in the unbinding rate @xmath10 . we computed the expansion to first order and this has allowed us to obtain the whole probability distribution for the typical e. coli and other values , in good agreement with the simulations . higher orders can be obtained systematically by iterating the proposed method . the analytical expressions at first order ( [ p01res],[p11res ] ) already illustrate the complexity of the nonequilibrium stationary state . in principle , the green functions that have been calculated in section 4 allow the expansion to be carried out to arbitrary order although analytical expressions for the higher order terms in the expansions might be long or difficult to simplify . however , they still can be computed numerically by programming the proposed operations and , in principle , the method can be iterated as many times as necessary to get more accurate results . in this paper we have not attempted to study the dynamics , but it would be of interest to do so . in particular the dependence of the flip time or first passage time ( the time for the system to change from being dominated by one population to the other ) on the parameter values is of interest . it may be possible to extend our analytical solution to consider such flip times and to provide estimate of first passage times between the two bistable states . it might be that the green functions we have computed in section 4 contain some useful dynamical information . the techniques that we have developed should be applicable to the understanding of other related systems and properties of genetic switches in general . the factorisation of the moment equation hierarchy is straightforward to implement to obtain a mean field theory ; the exact perturbative approach is more involved but can be used in general for problems with similar probability distributions , as long as the green functions are analytically solvable . thus the techniques represent standard procedures to analyze this class of systems . juan venegas - ortiz would like to acknowledge the award of a college studentship from the university of edinburgh , and would like to thank francisco cordobs - aguilar for useful discussions about numerical implementation of the method . 99 u. alon _ an introduction to systems biology _ chapman & hall / crc , london ( 2006 ) p.b . warren and p.r . ten wolde , ( 2005 ) , _ j. phys . b _ * 109 * , 6812 p. visco , r. j. allen and m. r. evans ( 2009 ) , _ phys . rev . e _ * 79 * , 031923 m. ptashne _ a genetic switch : phage @xmath243 and higher organisms , 2nd ed . _ cell press and blackwell scientific , cambridge , ma , ( 1992 ) j. l. cherry and f. d. adler ( 2000 ) , _ j. theor . biol . _ * 203 * , 117 - 123 p.b . warren and p.r . ten wolde , ( 2004 ) , _ phys . lett . _ * 92 * , 128101 b. barzel and o. biham , ( 2008 ) , _ phys . e _ * 78 * , 041919 p. visco , r. j. allen and m. r. evans ( 2008 ) , _ phys . rev . lett . _ * 101 * , 118104 a. lipshtat , a. loinger , n. q. balaban and o. biham ( 2006 ) , _ phys . rev . * 96 * , 188101 t. s. gardner , c. r. cantor and j. j. collins ( 2000 ) , _ nature _ * 403 * , 339 - 342 h. h. mcadams and a. arkin ( 1997 ) , _ proc . * 94 * , 814 - 819 a. arkin , j. ross and h. h. mcadams ( 1998 ) , _ genetics _ * 149 * , 1633 a. becskei and l. serrano ( 2000 ) , _ nature _ * 405 * , 590 - 593 t. b. kepler and t. c elston , ( 2001 ) , _ biophys . j _ * 81 * , 3116 r. j. allen , p. b. warren , and p. r. ten wolde ( 2005 ) _ phys lett . _ * 94 * , 018104 a. loinger , a. lipshtat , n. q. balaban and o. biham ( 2007 ) , _ phys . e _ * 75 * , 021904 d. mukamel in _ soft and fragile matter : nonequilibrium dynamics , metastability and flow _ iop publishing , bristol ( 2000 ) m. r. evans , d. p. foster , c. godreche , and d. mukamel , ( 1995 ) phys . 74 , 208211 c. godreche , j - m luck , m. r. evans , d. mukamel , e. r. speer and s. sandow , ( 1995 ) , journal of physics a : mathematical and general , * 28 * , 6039 v. shahrezaei and p. s. swain , ( 2008 ) _ pnas _ * 105 * 17256 p - z shi and h qian , ( 2011 ) , _ j. chem . phys . _ * 134 * , 065104 r. karmakar ( 2010 ) , _ phys . rev . e _ * 81 * , 021905 j. e. m. hornos , d. schultz , g. c. p. innocentini , a. m. walczak , j. wang , j. n. onuchic and p. g. wolynes ( 2005 ) , _ phys . e _ * 72 * , 051907 j. ohkubo ( 2010 ) , _ arxiv _ * [ physics.bio-ph ] * , 1009.3085v1 d. t. gillespie ( 1977 ) , _ j. phys . chem . _ * 81 * , 2340 - 2361 d. schultz , j. n. onuchic and p. g. wolynes ( 2007 ) , _ j. chem . * 126 * , 245102 e.zauderer _ partial differential equations of applied mathematics _ john wiley & sons , new york ( 1989 ) according to @xcite , there is a way to solve , or at least simplify , a system of linear first - order partial differential equations . first of all , considering a three dimensional space with a vector @xmath244 for the three probabilites , the system has to be written as:@xmath245where @xmath118,@xmath169 and @xmath246 are @xmath247 matrices , and @xmath248 and @xmath249 are column matrices . the requirement to solve the system is that @xmath250 or @xmath251 . in this case the expression of the matrices is : @xmath252 & 0\\0 & 0 & -[g(z_2 - 1)-u]\\ \end{array } \right)\nonumber \\ \nonumber \\ d(z_1,z_2 ) = \left ( \begin{array}{c } 0\\ 0\\0 \end{array } \right)\end{aligned}\ ] ] this is not the most general system that could be written with this notation , since the matrix @xmath248 is zero , and @xmath118 and @xmath169 only depend on @xmath173 , @xmath131 , respectively . the requirement to solve the system is that @xmath250 or @xmath251 , which is fulfilled almost in every point . multiplying by the matrix @xmath253 , the matrix @xmath118 is transformed into : @xmath254}{d(1-z_2)(d-(d+b)z_2 ) } & 0 & \frac{(1-z_1)}{1-z_2}\\ \end{array } \right ) \\ \end{eqalign}\ ] ] whose eigenvalues , written as columns in a matrix @xmath255 are : @xmath256 this hyperbolic system can be transformed with elementary matrix operations : @xmath257 the method then states that the components of @xmath170 obey the following system of equations : @xmath258 where @xmath259 are the components of the matrix @xmath260 once we have performed the transformation with @xmath255 : @xmath261 . as can be seen from the last equation , we have uncoupled the derivative terms and , even if this system can not be solved analytically , it is easier to deal with it computationally . the problem itself can be written as : @xmath262v_0 + \left[\frac{uz_1(z_1 - 1)}{d(z_1-z_2)(z_2 - 1)}\right]v_1\nonumber\\ & & + \left[\frac{-dg(z_1 - 1)^2+b[d-(g+u)z_1+gz_1z_2]}{d(z_2 - 1)[d(z_1 - 1)+bz_1]}\right]v_2\nonumber \\on & & \quad \frac{dz_1}{dz_2}=\frac{1-z_1}{1-z_2}\nonumber\\ \frac{dv_1}{dz_2 } & = & \left[\frac{uz_2}{d(z_2-z_1)}\right]v_0+\left[\frac{(z_1 - 1)[-gz_1+(g+u)z_2]}{d(z_2-z_1)(z_2 - 1)}\right]v_1\nonumber\\ & & + \left[\frac{dg(z_2 - 1)^2+b[-d+(g+u)z_2-gz_1z_2]}{d(z_2 - 1)[d(z_1 - 1)+bz_1]}\right]v_2\nonumber \\ on & & \quad \frac{dz_1}{dz_2}=\frac{1-z_1}{1-z_2}\nonumber \\ \frac{dv_2}{dz_2 } & = & \left[\frac{uz_1}{d(z_2-z_1)}+\frac{u}{d-(d+b)z_2}\right]v_0+\left[\frac{uz_1[d(z_1 - 1)+bz_1]}{d(z_1-z_2)(d-(b+d)z_2)}\right]v_1 \nonumber \\ & + & \left[\frac{-u+g(z_1+z_2 - 2)}{d(z_2 - 1)[d(z_2 - 1)+bz_2]}\right]v_2 \nonumber \\ on & & \quad \frac{dz_1}{dz_2}=\frac{d-(d+b)z_1}{d-(d+b)z_2 } \label{exactsol}\end{aligned}\ ] ] although this method in principle solves exactly the system of partial differential equations it appears a formidable task to actually integrate equations ( [ exactsol ] ) . in this appendix we give the detailed derivation of ( [ p01res ] ) . we begin from ( [ p01int ] ) @xmath263 where the label _ symm _ refers to the fact that there will be another term equal to the written one , apart from a switch in the variables @xmath218 and @xmath219 . separating the parts of the expression that can be summed , the following simplification can be obtained by changing the order of the sums appropriately : @xmath266 where @xmath267 . note that this sum could be simplified further , but that the simplification will not allow us to perform the integration over @xmath170 in closed form . in this and following equation , we will try to obtain the simplest expressions globally , knowing that simplifying one part can lead to further complications in another .

a computer simulation has to be fast to be helpful , if it is employed to study the behavior of a multicomponent dynamic system . this paper discusses modeling concepts and algorithmic techniques useful for creating such fast simulations . concrete examples of simulations that range from econometric modeling to communications to material science are used to illustrate these techniques and concepts . the algorithmic and modeling methods discussed include event - driven processing , `` anticipating '' data structures , and `` lazy '' evaluation , poisson dispenser , parallel processing by cautious advancements and by synchronous relaxations . the paper gives examples of how these techniques and models are employed in assessing efficiency of capacity management methods in wireless and wired networks , in studies of magnetization , crystalline structure , and sediment formation in material science , in studies of competition in economics .

cs0405077

dynamic systems with many interacting components are encountered in areas ranging from econometric modeling to communications to material science . individual components and their local interactions may or may not be complex , but their mere multiplicity often creates a complexity which needs to be addressed . as an example , consider the flexchan feature , introduced in new releases of the tdma wireless system . according to the flexchan method , each base station that serves a cell in the service area , periodically checks interference level on different channels and dynamically rearranges the channels in the order of the increase of the expected interference . a new service request ( a new call or a handoff request from a mobile with the call in progress ) is allocated according to an accepted strategy taking into account availability of free channels and the segregation order . for instance , a `` greedy '' strategy places the call on the unoccupied channel with the least expected interference . all the channel capacity of the system is potentially available to each cell in flexchan in contrast with fixed allocation schemes , that are largely in use today . the latter prespecifies a partition of capacity among the cells in anticipation of the traffic . under the flexchan , cells themselves negotiate the capacity , dynamically forming an allocation pattern in response to the actual traffic . thus , the flexchan eliminates the manual planning , reduces the operating cost and presumably increases capacity and qos . does it ? will the channel segregation adapt to the traffic , or maybe it will instead oscillate somehow , with base stations `` fighting '' each other for the capacity ? if it does adapt , how long would the adaptation process take depending on algorithm parameters such as the frequency of checking the interference by base stations ? short of actual system deployment , only simulations can answer these questions . to be convincing , the simulation should be dynamic with base stations asynchronously working out their flexchan routine while users are moving in the service area . many base stations should be represented in order to demonstrate the algorithm viability ; it is possible to imagine how the algorithm would work for just a few base stations , but break down for , say , a hundred base stations . it becomes obvious that a crucial element of this simulation is the computing efficiency of the simulation algorithm . this paper reviews algorithmic and programming techniques which were used to answer the posed questions by simulation@xcite . the program was written so that , say , for about 100 base stations and 1000 active mobiles , simulating one hour of operation requires only several hours of processing on a single pc or workstation . similar algorithmic concepts and techniques for improving efficiency of _ discrete event _ simulations recur in diverse applications such as : - an econometric model of telephone companies , like at&t and mci , fighting among themselves for the residential telephone market quotas . to attract the customers , the players introduce various payment plans , like `` friends and family '' or `` volume discount . '' a question may be : which policy / discount works better given the same cost for the player . the customers behave randomly , their response is staggered , but they tend to behave in accordance with their individual economic interests . it was possible to simulate such systems with more than 100,000 customers with individualized connections and `` habits '' in calling each other during several simulated years with a single run taking a few minutes on a pc . - a dynamic routing in a wired network , like the long distance at&t network , where the use of an `` anticipating '' data structure allows one to cut processing time of simulating one operating hour from a hundred hours to under a ten hours and where further speed improvement using parallel processing shrinks the processing time to a few minutes . - multiparticle studies in computational physics , and material science , such as a model of magnetization , a model of particle deposition , a study of impurity - perturbed crystals . here `` lazy '' evaluation , poisson dispensing , and parallel processing lead to several order of magnitude speed improvements . some simulations previously thought impossible , as they would take years to complete under old techniques , with the new programming techniques move to the category of possible ones , those that can be completed in several hours . this paper is organized as follows . sections [ s : edr ] , [ s : lazy ] , and [ s : poisson ] present techniques and concepts for simulating multicomponent systems on a uniprocessor . the advantages and difficulties of event - driven processing are discussed in section [ s : edr ] , the balance between `` lazy '' and anticipatory methods of evaluation is explained in section [ s : lazy ] , and a general method of event - driven simulation which avoids , and which is faster than the event list method is introduced in section [ s : poisson ] . sections [ s : sequent ] and [ s : relax ] present technique for simulating efficiently large multicomponent systems on a multiprocessor . a `` cautious '' technique which avoids speculative computations is discussed in section [ s : sequent ] . if the latter technique is not feasible , one may resort to the synchronous relaxation method , presented in section [ s : relax ] . the lessons learned in the course of the practical application of the discussed methods are discussed in conclusion . a time - driven description of a dynamic system involves the global clock . the time - driven computer model keeps in memory the global state of the system and modifies the entire state as the global time advances in finite increments . although time - driven models are intuitive and convenient to think of , one should try to avoid them in simulations because of their computational inefficiency . converting the model from the time - driven to an event - driven form , whenever such conversion is possible , is a single best improvement to a computer simulation . that is a classical textbook recommendation . in a general case , it is not known , though , how to do the conversion . as an example of the conversion , consider simulation of a collection of chaotically colliding billiard balls . despite its toyish appearance , the `` billiards '' simulation technically is non - trivial and has serious applications . for simplicity consider billiards in one dimension . we may imagine a gutter bounded from both ends , which contains @xmath0 absolutely hard elastic balls of equal mass and size . the width of the gutter is just enough to let the balls move without friction in one dimension . the gutter is placed horizontally which excludes the effect of gravitation on the ball movements . the gutter filled with @xmath1 balls is shown at the bottom of figure [ gutter ] which also presents the initial time - space trajectories of the balls for times @xmath2 . the trajectories are indicated by dashed lines , they are initiated at @xmath3 by the shown positions and velocities . to simulate the system evolution in a time - driven way , we integrate equations of motion along small time segments of duration @xmath4 . thus , starting with time @xmath5 we advance the state to time @xmath4 , then to @xmath6 , then to @xmath7 , and so on . if a ball experiences no collision during the advancement interval , then , because there is no friction , we have @xmath8 where @xmath9 is the one - dimensional position coordinate and @xmath10 is the velocity of the ball ( center ) at time @xmath11 . at each step the time - driven method monitors distances between components that might come in contact . ball - ball collisions are detected by monitoring distances @xmath12 between centers of balls @xmath13 and @xmath14 . such a collision occurs during time interval @xmath15 if @xmath12 becomes smaller that ball diameter @xmath16 at time @xmath17 , while @xmath18 at time @xmath11 . once detected , the collision is processed by exchanging velocities of the balls at time @xmath19 : @xmath20 it follows , that for a single ball the algorithm detects and processes at most one collision during a @xmath4 step . with a large @xmath4 , the algorithm may fail to represent interdependent collisions that follow over a single @xmath4 step . hence , @xmath4 determines the accuracy of the simulation ; smaller @xmath4 higher the accuracy . a high accuracy is required in many applications of the billiards simulation , for instance , for generating a pattern of impurity perturbed crystal . a material scientist might be interested to know the geometry of displacements of particles in a crystal formed of identical particles with a tiny fraction of inserted isolated particles of a different kind , the impurity . the mechanism of generating a displacement pattern in two dimensions , see figures [ hex ] and [ mosaic ] , can be modeled by billiards as follow . we introduce an impurity _ larger _ particle into a perfect hexagonal crystal of circular particles of common diameter packed inside a bounded geometric shape , say , inside a hexagon . to fit the larger diameter circle we loosen the assembly by simultaneously decreasing the diameters of all the circles . the decrease of the particles matches the excess size of the impurity so that we then able to substitute a single regular particle with the larger impurity particle without creating particle overlaps . the impurity particle is in the center in figures [ hex ] and [ mosaic ] . we then randomly assign a velocity to each particle and let them all uniformly increase in diameter , so that the ratios of the diameters of the particles remain constant . `` swelling '' particles chaotically move in different directions colliding with each other thereby negotiating the limited free space inside the bounded region until the final `` jammed '' state delivers the sought pattern . figure [ hex ] represents vectors of displacement of the particles from their respective original positions in the unperturbed crystal and figure [ mosaic ] represents the particles themselves in the central square fragment of the pattern in figure [ hex ] . the pattern is robust . it repeats itself under different initial conditions within a range of parameters such as a ratio of the larger particle to the regular ones and a rate of particle expansion . at the final phase of the expansion , when the particles are almost `` jammed '' a quick sequence of interdependent collisions has a high chance to occur within the array of 11,000 particle . the accuracy required for capturing this interdependence is high and the @xmath4 required for it is small . should the @xmath4 be not small enough , the simulation fails to produce the correct pattern . unfortunately , the accuracy does not agree with the efficiency of computations . in a typical @xmath4 step , if the @xmath4 is small , the state of most particles is advanced according to with no collision . such is the case for the @xmath4 step shown in figure [ gutter ] . the no - collision advancements waste computing power : generating the pattern similar to the one shown in figures [ hex ] and [ mosaic ] would take more than a year on a personal computer . that is an estimate though , because the time - driven method was never tried successfully in this example . instead , the pattern was generated using an event - driven method and the computations for a pattern like the one in figures [ hex ] and [ mosaic ] were typically completed in under 10 hours of processing on a pc . the event - driven method assumes a discrete event model of the system , wherein the system and its components change their states instantaneously at discrete times ; those changes are called the _ events_. the state remains constant on the intervals between the events . the trajectory of the system is a directed acyclic _ event dependency graph_. the nodes of it are the events , and the links represent cause - effect relations in pairs of events . this graph can be also viewed as a data - flow diagram . to each node - event corresponds the event _ descriptor _ , which consists of the time of the event and of the specification of the state change represented by the event . if events @xmath21 are all the immediate causes of event @xmath22 , then the descriptor of event @xmath22 is a function of the descriptors of @xmath21 . generating the descriptor of event @xmath22 from the descriptors of @xmath22 s causes is _ processing _ event @xmath22 . figure [ gutter ] represents not only the 4-ball system trajectory but can be also viewed as the event dependency graph corresponding to this trajectory . the events @xmath23 , @xmath24 , are ball - ball and ball - wall collisions . the arrows from an event to an event along a ball trajectory can be seen as cause - effect links on the event dependency graph or as data - dependency links of the data - flow diagram . an event descriptor here consists of : the collision time , the positions and velocities of the involved balls immediately after the collision . one can check , for instance , that ( the descriptor of ) event @xmath25 , a collision of balls 3 and 4 , is a function of ( the descriptors of ) event @xmath26 , a collision of 2 and 3 , and @xmath27 , a bounce of 4 against the right end of the gutter . namely , the time component of @xmath25 descriptor , @xmath28 , is computed as @xmath29 where @xmath30 and @xmath31 are position and velocity of ball @xmath13 , @xmath32 , at time @xmath11 , @xmath33 is the time of event @xmath23 as indicated on the time axis in figure [ gutter ] , and @xmath16 is the ball diameter . as claimed , the right hand - side in is a function of event @xmath27 descriptor , including @xmath34 , @xmath35 , and ( negative ) @xmath36 , and event @xmath26 descriptor , including @xmath37 , @xmath38 , and ( positive ) @xmath39 . producing the simulated history of a system under study is the goal of a simulation . the goal is achieved by processing events along the history . but how do we know which events to process ? neither the event descriptors nor the topology of the event dependency graph is known in advance . the event - driven simulation must do both construct the event dependency graph and process the events on it . the former activity , is also referred to as event _ scheduling_. the task of designing an efficient scheduling usually constitutes the main difficulty of recasting a time - driven model into an event - driven form . a well known mechanism of scheduling an event - driven simulation on a uniprocessor is a _ queue of events_. with the queue one is able to insert future events in the schedule , and to retrieve the next event for processing . a popular data structure @xcite for the event queue is _ heap _ with which either operation is performed in order of @xmath40 computing operations where @xmath0 is the number of events in the queue . by contrast , a straightforward scheduling method would need order of @xmath0 operations either for retrieving or for inserting or for both . if @xmath41 , say , then slow down may be substantial . it should be stressed though , that handling event queue efficiently does not cover the entire task of efficient event - driven simulation @xcite . the simulationist also has to impose appropriate modeling assumptions . for example , in the simulation of billiards , were the balls not absolutely hard , the collisions would not be instantaneous . instead of a simple velocity exchange in , this would force a ( slow ) time - driven evaluation of the collisions . equally important for the speed of computation is the assumption of the motion without friction . with friction , the ball motions would not be integrable as simply as in . the wireless simulations discussed in section [ s : intro ] are naturally of a discrete event type , events being call arrival , termination , handoff , interference measurement , adjustment of the channel segregation order . among few exceptions is the users mobility . the users may move along curvy trajectories with variable speeds . it is decided , however , that the simulated users move with constant velocities along straight line segments . as in the billiards model , this assumption leads to an event - driven processing . validity of the simulation results is not seriously affected because a curvy motion with variable velocity can be approximated with a sequence of the motions of the considered type . selecting a model agreeable with the event - driven mechanism and using an efficient event queue may still be not enough for efficiency of an event - driven simulation . the following sections will discuss more techniques and approaches toward the goal of a fast multicomponent simulation . a simulation may generate queries concerning the system state . the anticipatory method can be defined as constructing answers for questions which have not yet been asked . it can be contrasted with a `` lazy '' evaluation method according to which generation of the answer is delayed until the latest possible moment when the query is issued . the anticipatory evaluation tends to be event - driven , while the lazy evaluation time - driven . let us discuss the two approaches in the following simple example . in the wireless simulation , discussed in section [ s : intro ] , it is necessary to output , as a function of time , the number of mobile users that request but do not obtain a connection . together with other information , this quantity is to be displayed on the pc screen as the simulation progresses . thus , given a time increment @xmath4 , for each time instance @xmath42 , the simulation is supposed to report the number of @xmath43 users among all the users present at time @xmath11 . the lazy method is straightforward : we scan the list of users and depending on whether a user is @xmath43 or non-@xmath43 at time @xmath11 , increment the corresponding quantity @xmath44 by one or not . before each scanning instance @xmath11 , we reset counter @xmath44 to @xmath5 . this is obviously a time - driven method . the anticipatory method is driven by events . it does not reset or recompute @xmath43 at each reporting time instant . instead , it maintains the correct value of @xmath43 continuously by updating it appropriately at every event which may change @xmath43 . such events are : call arrivals to the system , call releases from the system , and users that obtain a connection after a period of retrials . for example , when a user arrives to the system , we would increment @xmath43 by one if the user does not immediately get connected . note that both solutions `` deliver '' the same value of counter @xmath43 . this facilitates the development of the simulation program . in the initial phase , one may program a simpler lazy solution . then an anticipatory solution should be programmed . both solutions should then be tried in representative examples , where first we compare values of the results , and then , after the values are found identical , we compare the running times . an example of the anticipatory data structure is space _ sectorization_. billiards simulation in dimensions two and higher share this method with wireless mobile simulations . the query , in the case of wireless simulations , may be to locate all the mobiles that are close to a given point @xmath45 on the plane . similar queries exist in the billiards simulation . let us consider the wireless simulation case . there exists a straightforward method to locate all mobiles which are within a given distance @xmath46 from the point @xmath45 . in this method , we would find the position @xmath47 of each mobile , compute the distance @xmath48 from @xmath47 to @xmath45 , and then select those @xmath49 for which @xmath50 . in practice , @xmath46 is small and there are just a few such @xmath49 , but to find them we would have to scan all the mobiles . a structure that helps reduce the computation cost is partitioning the simulation area into smaller sectors . the geometry of a sector may be different , a usual choice is a square . given such a square @xmath51 , there is a set of all the mobiles @xmath52 that are inside it . knowing positions of all the mobiles , we can , of course , find the set @xmath52 using a total scan of all mobiles when a query is offered . in the anticipatory method , we maintain the sets @xmath52 continuously independently of queries . this entails updating two such sets each time a mobile crosses a demarcation line between the squares ; the crossing becomes an extra event to process . having invested in this anticipatory structure , we now find the mobiles @xmath46-close to a point @xmath45 in a different manner . namely , we first determine the square @xmath53 in which @xmath45 is located , then scan nearby squares @xmath51 that intersect with any circle of radius @xmath46 with center anywhere in @xmath53 , and then check only the mobiles that belong to sets @xmath52 . the same two query evaluation approaches , anticipatory and lazy , can be considered with respect to web search engines . for many query types , when a search request is submitted on a web search service , a lazy _ on line _ evaluation of the request is non - feasible . too few users would be willing to wait for the server to scan the world wide web . the global scanning is being performed but _ off line_. mechanical and manual methods are used to create structures which contain answers to various anticipated questions and mechanisms exist to quickly retrieve these answers concentrated at a few data base sites of the search server . the `` economics '' of a web server query evaluation and that in a simulation are different , though . in the former , high computational and manual labor costs are justifiable by the need of short response time . in the latter , an anticipatory method can be only justified if total computing effort when using the method decreases . with a lazy evaluation , while there may be high cost of retrieving the answer when needed , no resource is spent for anticipating the question . with an anticipatory evaluation , the cost of retrieving the answer may be smaller , but possibly a high investment may be needed for anticipating the question . anticipatory mechanisms should be advanced only as far as savings on the retrieving end of the balance are higher than spendings on the investing end . the situation may be illustrated with the following example @xcite of simulating a circuit - switched wired network , like the at&t telephone network . in the model , the network consists of @xmath0 nodes that represent the switches and @xmath54 links that connect the nodes . a link consists of a fixed number ( possibly zero ) of _ trunks_. the model assumes , that if a call is to be carried along some path in the network , then one trunk from each link on the path must be allocated for the exclusive use of the call . the model also assumes that calls are randomly generated between node pairs ( @xmath55 ) and that a _ routing policy _ decides whether to block ( reject ) such a call or carry it on some path between @xmath56 and @xmath57 . the simulation is used to assess the quality of different routing policies in terms of the blocking produced for given traffic loads . consider the _ least busy alternative _ ( lba ) policy and the _ aggregated least busy alternative _ ( alba ) policy . both policies allocate a trunk on the link between nodes @xmath56 and @xmath57 if an idle trunk is available . if not , both offer the call to a two - link path that uses an additional node @xmath58 . the intermediate node @xmath58 is called the _ via_. in lba policy , the overflowing call is offered to the two - link path that has the most idle capacity . specifically , path @xmath59 is used if @xmath60 maximizes @xmath61 where @xmath62 is the number of idle trunks currently on link @xmath63 . if no idle capacity is available - expression is zero for all @xmath58 - then the call is blocked . the alba policy is a coarsening of lba , which is less costly to implement . the overflow calls are offered to a two - link path with a minimum _ load class_. load classes are defined by a fixed number of prespecified capacity boundaries . for instance , three load classes can be defined with boundaries , where say @xmath6480% and @xmath6590% . if on a given link the proportion of occupied trunks is @xmath66 , then the link belongs to the load class @xmath13 with @xmath67 . thus , class 1 contains the lightly loaded links , class 2 contains the moderately loaded links , and class 3 contains the heavily loaded links . the load class of a two - link pass is defined to be the smaller of the load classes of the two links . an overflowing call between @xmath56 and @xmath57 uses a two - link path whose load class is the minimum among all two - link paths between this node pair . if no two - link path has sufficient idle capacity , then the call is blocked . the query subject to discussion here is finding the via @xmath58 when no idle capacity exists on the direct path @xmath68 . the `` lazy '' method operates exactly as the definition says : when the query is issued , it scans all possible vias @xmath58 , for each @xmath58 computes the idle capacity and chooses the @xmath58 which yields the maximum to in the case of lba policy , or it chooses a via with the minimum of the load class in the case of alba policy . an anticipatory data structure may be suggested for the case of lba policy which for each node pair ( @xmath55 ) consists of a list of vias @xmath58 sorted decreasingly according to . whenever a least busy via is needed , the first via in the list is chosen which greatly decreases the computations on the retrieving end of the balance . on the investing end of the balance , each time a trunk is allocated or freed on any link @xmath69 , the @xmath70 sorted lists for all links ( @xmath55 ) which can use either @xmath71 or @xmath72 as a via have to be updated . as a result , there is no overall saving if anticipatory approach is used for the lba simulation . however , for the alba simulation , the anticipatory approach yields a significant improvement comparing with the lazy approach . the capacity of a link is usually measured in hundreds if not thousands of trunks while there are typically only a few , say three load classes . thus , when a link trunk occupancy changes , the link rarely changes its load class . if the load class does not change , no expensive update of sorted class - via lists is needed . in the alba case , both lazy and anticipatory methods were tried . while a useful lazy simulation run of the network took more than 100 hours , the corresponding anticipatory simulation run took 3 to 10 hours on the same uniprocessor . we now examine event - driven billiards simulations with respect to the topic of anticipatory vs. lazy evaluation . the query concerning a ball location is better accommodated by the anticipatory sectoring as discussed above . another basic query in the billiards simulation is : which collision has to be processed next ? in the anticipatory method of d. rapaport @xcite , the data structure , which accommodates the answer , includes for each ball a set of all future events which can not be easily excluded . ( future collisions with distant balls in dimensions two and higher can be excluded using the sectoring method . ) the minimum time event in this list is the best candidate event to next occur with the given ball . the minimum time event among all the best candidates is the query answer . this event is to be processed next . maintaining such lists is an easy task for the gutter billiards , as there are only two candidates for a collision with any ball @xmath13 , its left neighbor @xmath73 and its right neighbor @xmath14 . ( the neighbor may be a gutter end for balls 1 and @xmath0 . ) in higher dimensions , the task of maintaining the lists becomes more involved , since many balls can not be excluded beforehand as candidates for future collisions of a ball , even if using the sectoring method . these anticipatory lists may include large and variable number of candidates and they have to be examined and updated for each event processed . on the positive side , method @xcite handles easily the collision _ preemption _ which occurs as follows . suppose the algorithm schedules a collision of two balls , say @xmath74 and @xmath75 , for some future time @xmath76 and this collision becomes the best candidate event to occur for each ball @xmath74 and @xmath75 . however , before the processing reaches time @xmath76 , a third ball , say @xmath77 , intercepts ball @xmath74 with a collision scheduled to occur at time @xmath78 so that @xmath79 . now , of course , the collision of @xmath74 with @xmath77 becomes the best candidate for @xmath74 . what should become the new best candidate for ball @xmath75 ? the answer is ready , anticipated in the list of ball @xmath75 : the previously rated next - best candidate event for ball @xmath75 is upgraded in status to become the best candidate . this anticipatory event - driven algorithm can be contrasted with the method @xcite , which also presents an event - driven , but `` lazy '' simulation algorithm for billiards . here only one future event is kept in each individual ball list . maintaining this structure is easy : when a better candidate emerges , it replaces the old one , which is simply discarded . however , we should reexamine for the `` lazy '' method the described above collision preemption situation with balls @xmath74,@xmath75 , and @xmath77 . what will be the new candidate event for ball @xmath75 ? the solution @xcite introduces a new event type on a ball trajectory , an `` advancement '' event . when processing such an event , the ball is advanced to the new position without changing its velocity . thus , when the collision between @xmath74 and @xmath75 , which was scheduled for time @xmath76 , is preempted by an earlier collision between @xmath74 and @xmath77 , the new candidate event for ball @xmath75 is such an advancement to the former collision site with ball @xmath74 . the time of the advancement event is @xmath76 , which is the time of the previously scheduled and then discarded collision of @xmath74 and @xmath75 . this is similar to the time - driven @xmath4 advancement in ; the lazy method @xcite can be viewed as a fall back to a time - driven mechanism . however , the lazy method proved to be not slower than the anticipatory method , while the overall algorithm is simpler . consider in more detail mentioned in section [ s : intro ] econometric model @xcite of two competing telephone providers , name them company 1 and company 2 ( the model generalizes easily to @xmath80 providers ) . the model also includes @xmath0 telephone customers . at any time instant @xmath11 each customer subscribes either to company 1 or to company 2 . let @xmath81 denote the subscription status of customer @xmath13 , @xmath82 , at time @xmath11 , so @xmath83 or @xmath84 . potential bill amounts @xmath85 can be computed for each customer @xmath13 if being a subscriber to company @xmath86 , @xmath87 . that is , the service would cost @xmath88 dollars per month to customer @xmath13 if @xmath89 . for example , to compute @xmath88 under the mci s `` friends and family '' plan , we add up the minutes - per - month customer @xmath13 calls all his / her calling parties who subscribe to the same provider ( here the mci ) , and multiply this by the plan discount price . then we add the minutes - per - month customer @xmath13 calls the calling parties who subscribe to the other provider , multiplied by the larger regular price . according to this model , a tendency to switch the provider originates in comparing alternative bills : if @xmath90 then customer @xmath13 wants to be served by provider 1 , and if @xmath91 , then by provider 2 . a customer who subscribes to one provider but wants to be served by the other one is subject to a `` pull '' to the opposite provider . the intensity of the pull toward provider 1 for customer @xmath13 who is currently with provider 2 in case @xmath90 is expressed by the rate @xmath92 where @xmath93 is an explicitly defined monotonically increasing function . if such customer @xmath13 is with provider 2 at time @xmath11 , then during a small time interval @xmath94 , customer @xmath13 tosses a coin and switches to provider 1 with probability @xmath95 . if the switch attempt is not successful , the switch is similarly reattempted during the next interval @xmath96 , then next interval and so on . a similarly defined `` pull '' toward company 2 affects a customer @xmath13 who is currently served by but is not happy with company 1 . the switch attempts are statistically independent of each other for different customers within the same @xmath4 interval and they do not depend on the system state before time @xmath11 . the model assumes an individual calling pattern for each customer @xmath13 ; the pattern is defined by specifying minutes - per - month @xmath97 customer @xmath13 calls each `` friend '' @xmath51 of his / her . the calling habits of the customers are stationary , that is , the calling volume matrix @xmath98 is independent of time . the parameters of the providers plans , e.g. , prices , are also time - independent . ( these assumptions can be relaxed . ) despite the stationarity , @xmath88 and the strength @xmath99 of the pull may change with time , @xmath100 , @xmath101 . this is so , because in plans like `` friends and family '' when your call - destination `` friend '' switches the allegiance , it affects your bill . you become more susceptible to switching to the same provider if your `` friend '' has done so , and that is what plans like `` friends and family '' count upon . an obvious method of simulating the outlined model is time - driven , it proceeds exactly as the model states . time is increased in small @xmath4 steps . at each step each customer who is not satisfied with the provider randomly attempts to switch . then based on the new assignments @xmath102 , new @xmath88 and new rates @xmath99 are computed to be valid for the next @xmath4 step . now we describe an alternative event - driven method of simulating this model . the method is based on the observation that the sequence of times of switching allegiance for each customer forms a poisson process with the rate @xmath99 which varies in time , @xmath103 . note that a stationary poisson process ( that with a constant arrival rate @xmath104 ) can be simulated as a sequence of arrival times @xmath105 with independent distributed exponentially with mean @xmath106 interarrival times . that is , given arrival @xmath107 , to sample @xmath108 one draws an independent sample @xmath109 of a random value uniformly distributed in @xmath110 , and then one computes @xmath111 . the poisson property is preserved when several arrival processes are aggregated into a single arrival process . say , we have @xmath0 poisson arrival processes and @xmath13th process has arrival rate @xmath112 , @xmath113 . the aggregate process is defined as the one that has an arrival when any component process does . so defined , the aggregate is also a poisson process . the arrival rate of the aggregate is @xmath114 . one way of arranging an event - driven simulation of the competition between the telephone providers is as follows . a next anticipated switch event is associated with each of the @xmath0 customers . the event time is generated by sampling the exponential distribution with the current rate of `` pull '' discussed above . the case of rate 0 is reserved for customers who are satisfied with the provider . such a customer does not wish to switch and the next switch time is set to infinity . the event queue is arranged as usual . processing each customer switch modifies some rates @xmath99 . this , in turn , modifies the time remaining to switch in the events scheduled for the affected customers ; sometimes it may reduce the rate to 0 which would postpone the switch event to the infinite future . having the poisson property preserved under the aggregation , the event - driven simulation of this model can be arranged in a different way without presampling and then updating future switch events and without the event queue . in the alternative method , these are replaced with the following _ poisson dispenser _ mechanism . a single poisson arrival stream with rate @xmath115 is generated and then is being `` dispensed '' among the component poisson streams in accordance with their partial arrival rates @xmath116 , larger the @xmath116 more probable is to delegate an arrival of the aggregate process to customer @xmath13 s process . rate @xmath116 of a component poisson process may experience changes within the time interval between the consecutive arrivals of the component process . by contrast , the aggregate rate @xmath115 remains constant between arrivals of its process . hence the interarrivals of the aggregate are distributed exponentially ; once next event is scheduled its time never gets changed or postponed . this simplifies and speeds up the computing . the dispenser procedure consists of the cyclic repetition of the two steps specified below ; the execution begins with @xmath117 and current time @xmath118 . + + + + + this procedure is formulated above for a general multicomponent system . in the considered example , a telephone customer is a component , and the state change of component @xmath13 is switch @xmath119 of the telephone provider for customer @xmath13 , where map @xmath120 turns 1 into 2 and 2 into 1 as required to effect the provider switch . since after the switch the customer is `` happy '' with the new provider , the new arrival rate in step 2 is zero ; it may become positive again later as a result of `` friends '' switching . also , values @xmath121 , are indeed a probability distribution , as @xmath122 and @xmath123 . in a straightforward method the selection in step 2 is implemented as follows . first , we draw an independent sample @xmath109 of a random value uniformly distributed in @xmath110 . then we scan monotonically non - decreasing sequence @xmath124 , @xmath125 , ... @xmath126 , ... @xmath127 , fitting value @xmath128 between consecutive terms @xmath129 . this is possible , since @xmath130 . the found @xmath13 is index of the customer to whom the arrival is delegated . unfortunately , an order of @xmath0 computations is required for the scanning and this is slow for a large @xmath0 . better methods exist . figure [ tree ] exemplifies the binary tree method in which the component @xmath13 is found in @xmath131 steps ( if @xmath0 is a whole degree of 2 ; otherwise @xmath132 steps ) . this method also starts with an independent sample @xmath109 of a random value uniformly distributed in @xmath110 . then , instead of a linear scan , a binary tree is traced to fit this value . the tracing is started from the root which is entered with @xmath133 . we then steer our way down the tree in @xmath131 steps . at step @xmath134 , having value @xmath135 , @xmath136 , we enter node with inscription @xmath137 on it . if @xmath135 happens to be smaller than @xmath138 , then we go to the left branch @xmath138 producing @xmath139 . if @xmath140 , then we go to the right branch @xmath141 producing @xmath142 . to maintain current the weights on the tree , each time a rate @xmath99 changes , the contents of @xmath143 nodes is updated . the update begins with the bottom node @xmath13 , continues up , and terminates at the root . a useful simulation run is the switching history of several thousand customers during a few simulated years . a time - driven version takes several computing hours to complete the run . the corresponding event - driven version with a poisson dispenser implemented using the binary tree , takes several seconds . the latter version allows one to easily simulate markets of a much larger size while keeping the running time bearable . for example the behavior of a 100,000-customer market during several simulated years can be simulated in less than 2 minutes . another application for the dispenser technique , is an _ ising spin model _ @xcite in computational physics . as a computational mechanism , the ising spin simulation is very similar to the econometric model of competition between the telephone providers which has been just discussed . @xmath0 ferromagnetic particles are in place of @xmath0 telephone customers . an external magnetic field pulling the particles so that they would align in the field s direction plays the role of the economic incentive for customers to `` align '' their allegiance to the provider which gives best saving . and the additional local magnetic field around a particle when its neighbors have already aligned is analogous to the additional incentive for a customer to switch the provider when the customer s calling parties have done so . specifically , the magnetic state called _ spin _ of each particle @xmath13 , takes on two values , @xmath144 or @xmath145 . depending on the current spins @xmath146 of the near - neighbors @xmath51 of particle @xmath13 , pulls to flip @xmath83 to @xmath147 and to flip @xmath148 to 1 are defined . the dependence involves external field direction and intensity . ( the form of this dependence is not essential for the present discussion . ) similarly to the model discussed before , the flip mechanism is probabilistic . simulation of the dynamics of ising spins can thus be arranged using the efficient binary tree poisson dispenser mechanism discussed above . in this method flipping one spin takes @xmath149 computations . one feature in the ising model is different from the telephone competition model though . the ising particles are arranged in a regular fashion . for example , they are placed in vertices of a two - dimensional square lattice where each particle has four near - neighbors : at the north , east , south , and west . by contrast , the calling volume matrix @xmath97 which defines calling parties of a customer is not assumed to be regular . because of the regularity in the ising model , for any particle , there is only a small number of possible combinations of neighboring particle states . let us count these combinations , say , for a planar square grid arrangement of the particles . each of the north , east , south , and west spin can take on 2 values , which yields 16 combinations . the particle itself can be in the two spin states . hence there are at most 32 different combinations each of which may define a different pull - to - flip rate . therefore there are at most 32 different rates and this finite set is the same for all @xmath0 particles . ( examining the form in which the rate depends on the combination , this set further reduces to only 10 different rates @xcite . ) thus , each of the @xmath0 leaves in the dispenser tree contains one of a small number of possible values . the finiteness allows one @xcite to improve the poisson dispenser algorithm so that delegating one arrival of the aggregate process to a component process takes a constant amount of computations , instead of @xmath149 computations . sequential random update is a general mechanism for modeling evolution of a multicomponent dynamic system . in this model , we consider a system with @xmath0 components , the state of the system is composed of states @xmath102 of its components , the system state changes at discrete instances @xmath150 according to the following cyclic procedure . + + + + + both the ising spin model and the model of competition between telephone providers discussed in section [ s : poisson ] fit the sequential random update scheme , if two transformations are made : 1 ) uniformizing the event arrival rates among the system components , 2 ) abandoning the continuous time and retaining only the update counter @xmath49 . the latter transformation is obvious . let us explain the former one . we choose an upper bound @xmath151 on the event arrival rates @xmath116 among the components . this can be easily done in the ising model : @xmath151 is the largest among the finite number of possible rates @xmath99 , see the discussion in section [ ss : ising ] . in the provider competition model , the most `` unhappy '' customer yields the upper bound on the switch rate . we attribute the same rate @xmath151 to all components so that the components are selected with equal probability as required in step 1 above . suppose a component @xmath13 at step 1 is selected . to compensate for the smaller rate with which the component @xmath13 is being updated , we make an additional coin toss and choose to update state @xmath102 in step 2 with probability @xmath152 . with the complimentary probability @xmath153 the state does not change . the no - change of state @xmath102 does not violate the format of the procedure ; it is a special case of the update when @xmath154 . the model of circuit - switched wired network discussed in section [ s : lazy ] almost fits this scheme , if we take @xmath155 links between the switches as components and equate the number of occupied trunks on a link with the state of the component . if different node pairs @xmath156 have different call arrival rates , we uniformize the rates as in the previous two models . the feature that does not fit the scheme is that when a call request arrives between a node pair @xmath157 and is placed via node @xmath58 instead of the direct link , the state of components @xmath158 and @xmath159 is updated instead of the state of the original component @xmath157 . other instances of the sequential random update include cellular arrays and neural nets . we will now discuss an example @xcite of a _ ballistic particle deposition_. the deposition model is aimed at studying morphology of amorphous layers growing on planar substrates , the subject of interest to material scientists . in the model , spheres of equal diameters 1 are falling vertically down toward the flat horizontal surface , see an example in figure [ 100disk ] . the deposition process is arranged as follows . particles are deposited one at a time . to deposit a particle , first its center @xmath160 coordinate is sampled randomly and uniformly over the adsorbing range . the range is segment ( 0,10 ) in figure [ 100disk ] . different particles have their @xmath160 generated independently of each other . the initial height of the center , is chosen sufficiently high above the surface . then the particle is falling vertically down until a contact occurs . for the particles that begin the process , the contact is likely to be with the adsorbing surface . later in the process , a falling particle is more likely to contact a stationary one , deposited before . in either case , the first contact instantly stops the incoming particle thus defining its @xmath161 coordinate . the ballistic deposition scheme is an instance of the random sequential update . we split the adsorbing range into sectors of equal measure ( congruent segments in 2d or planar figures of equal area in 3d ) and declare sectors to be the components of the system . the particle order index @xmath49 becomes the update order index . the state @xmath162 of component @xmath13 consists of the coordinates of those among the first @xmath49 particles , that have been deposited over sector @xmath13 . that is , say in a two dimensional case , the @xmath160 of those particles centers must be in segment @xmath13 . a particle from sector @xmath13 but located close to the sector s boundary may attach itself to a particle which was earlier deposited in sector @xmath51 , @xmath163 . that is how components / sectors @xmath51 may get involved in the state update of component @xmath13 in step [ stateupd ] of the sequential update scheme . whereas figure [ 100disk ] represents a small - scale `` educational '' example of deposition in two dimensions , interesting simulations are in three dimensions with two dimensional adsorbing range of a size larger than say 1000@xmath1641000 . many millions if not billions of particles are supposed to be processed . figure [ cumden ] presents particle density as a function of both time and height @xmath161 in a three dimensional deposition of 100 million particles . running such an experiment on a workstation would take about a week of execution time . can parallel processing be employed to speed up the deposition and other sequential update schemes ? one idea of making the update scheme parallel may be to have a parallel computer dedicate @xmath0 processing elements ( pes ) to the @xmath0 components of the system so that @xmath165 would host component @xmath13 , @xmath166 . the components would be updated concurrently without an organizing order . @xmath165 would repeatedly update the state of component @xmath13 , asynchronously obtaining from other pes the current values of states of those components @xmath51 which are required for computing the new value of state of component @xmath13 . this proposal can be criticized from several viewpoints . the most basic objection is that it violates the standard model of reproducible computer execution . this entails various shortcomings . the generated trajectory of the system may be different from that generated sequentially . say in the deposition example , the deposit structures generated in parallel may be statistically different from those generated sequentially . moreover , the state change trajectories resulted in different runs of the same system with the same initial states , in general , will be different . this is very inconvenient as it makes both studying the obtained structures and debugging the simulation program very difficult . another proposal is to use @xmath167 pes . a single `` master '' pe would be dedicated to dispensing the updates among the components . the @xmath0 `` slave '' pes would host the @xmath0 components . a `` slave '' would be responsible for updating the state of the hosted component . without further elaboration of the `` master - slaves '' scheme , we note that it can be organized so that the computations are reproducible and generate trajectories identical to those in the sequential procedure . moreover , for a small number of pes , the procedure may even be efficient . however , it does not scale for a large number of pes . the sequential dispensing performed by the `` master '' becomes a bottleneck for a large @xmath0 . in the example of ballistic deposition , we now describe another method of running the sequential random update scheme in parallel . unlike the `` chaotic '' and the `` master - slaves '' methods , this method possesses both desirable properties : it generates a reproducible , correct simulated trajectory , and its good performance scales to a large size systems and large number @xmath0 of pes . the first step of the method is a reformulation of the sequential random update scheme in continuous time . in the old formulation the components are updated at discrete instances @xmath168 . in the new formulation the components are updated at instances @xmath169 of the continuous time , the instances constitute a poisson process . an arbitrary positive @xmath170 is chosen and fixed . the rate of the poisson process is @xmath171 . it follows , that each of the @xmath0 component processes also has arrivals that form a poisson process . the component processes will be mutually independent . the rate of each component process will be @xmath170 . of course , in the ising spins simulation model and in the provider competition model we have the poisson processes to start with . we can reuse them with their original rates for the purpose of rendering the models in parallel . however , in the deposition model , the poisson process is an additional structure , introduced only for the purpose of running the model in parallel . in the poisson dispenser method discussed in section [ s : poisson ] , we aggregated arrivals of individual components into a single stream which was to be sampled by the sequential computer . here we do the opposite , namely , disaggregate the summary poisson arrival stream into the independent streams of the components and let each component @xmath13 stream be sampled by a separate @xmath165 which would be also responsible for maintaining state @xmath102 of component @xmath13 . let @xmath172 denote the poisson clock maintained by @xmath165 . that is , changes of state @xmath102 occur at time instances @xmath173 . in the beginning of simulation each @xmath165 sets its poisson clock @xmath172 to 0 and has @xmath102 assume its initial value . then each @xmath165 asynchronously from other pes executes the following procedure , which can be called `` cautious advancements . '' + + + + + the pes may execute this procedure with no other synchronization than that in the wait statement in step [ wait ] , which is supposed to assure the `` cautious advancement '' of local times @xmath174 . because of the asynchrony , it might be not obvious that the procedure is correctly defined , let alone works correctly . for example , if concurrently with @xmath165 updating @xmath102 and using state values @xmath175 , @xmath176 is changing @xmath175 , is the state update in step [ pstateupd ] well defined ? it was shown @xcite that the cautious advancement scheme is correctly defined , works and is correct . specifically , with probability 1 , during an update of state @xmath102 , other states @xmath175 , the values of which are used in the @xmath102 update , are not themselves being changed . cautious advancement is deadlock - free : no pe waits forever in step [ wait ] . the sequence of updates obtained by merging the sequences of component updates generated by each pe is statistically identical to the sequence generated by the sequential procedure . reproducibility also takes place : if the cautious advancement procedure is executed twice with the same initial settings , including the same seeds of pseudorandom number generators , used for sampling poisson arrivals , then the two resulting sequences of state updates will be identical with probability 1 . this is despite that execution timings in different runs may be different . to get insight into the behavior of the cautious advancements , consider a simple example presented in figure [ tlines ] . time lines of 10 components of a simulated model are depicted . for concreteness , we assume that the model represents the deposition of unit diameter circular particles over the segment of length 10 as in figure [ 100disk ] . in the latter case , the substrate segment ( 0,10 ) is divided into 10 smaller segments : [ 0,1),[1,2), ... segment [ @xmath177 ) is component @xmath13 and it is hosted by @xmath165 . dots on the time lines mark the poisson arrivals . at each arrival to segment [ @xmath177 ) , a particle with coordinate @xmath160 , @xmath178 , is deposited . given that both the particle diameter and component segment length are unity , to know the landing height of the particle , there is no need to know positions of particles previously deposited over segments that are more than one component - segment away . however , a caution is exercised with respect to the immediate left and right neighbors of segment [ @xmath177 ) . @xmath165 only deposits a particle at time @xmath174 if its two neighbors advanced their simulated times to reach or exceed @xmath174 . the immediate left neighbor is segment [ @xmath179 ) , unless @xmath180 . the immediate right neighbor is segment [ @xmath181 ) , unless @xmath182 . points 0 and 10 representing the same point , the immediate left neighbor of component 1 is component @xmath0 , whose immediate right neighbor is component 1 . to be able to relate the arrival times for neighbors 1 and @xmath0 , the time line of component 1 is drawn twice . although no additional synchronization is necessary for correctness and efficiency of the general cautious advancement procedure , it would help in understanding the procedure of deposition , if we assume that it is executed in lockstep . that is , no pe executes step 2 , before step 1 is completed by all the pes that are non - waiting at step [ wait ] of a preceding cycle . then , in step [ wait ] , all pes check the simulated times achieved by their neighbors . as a result , the set of all pes splits into those able to proceed further , and the rest which must wait . the non - waiting pes begin step [ pstateupd ] only after all pes have finished the checking in step [ wait ] . finally , the new cycle by the non - waiting pes begins not earlier , than all the non - waiting pes have updated their state , i.e. , deposited a particle . the lockstep execution enables us to say at which cycle each state update occurs , that is , each particle gets deposited . in the situation of figure [ tlines ] , @xmath183 , @xmath184 , and @xmath185 are lucky to process an event at cycle 1 , while the rest of pes are waiting . values @xmath186 , and @xmath187 get advanced to the second arrival time and as a result @xmath188 , @xmath189 , @xmath190 , and @xmath191 can process their events at cycle 2 . values @xmath192 , and @xmath193 get advanced to the second arrival time and as a result @xmath184 is lucky again and processes its second event . in figure [ tlines ] , the cycle - per - event assignment is followed up to cycle 10 . as an exercise , the reader may continue the task for the following cycles . the computational efficiency is determined by the fraction of non - waiting pes at each cycle . this fraction is close to 25% on average in the example shown . looking at the picture it seems likely that the fraction remains bounded from below and separated from zero when the size of the system gets larger and number of pes increases proportionally . mathematical studies @xcite of this assertion confirm it in a general case . this means scalability of the parallel simulation performed by the cautious advancement mechanism . using additional methods @xcite , in particular , allowing one pe to hold a larger sector or segment , the efficiency can be substantially increased , e.g. , from 25% to 60% and higher . the mentioned above 100 million particle deposition experiment was run on a maspar mp-1216 computer with 16,384 pes . the run took 620 seconds ( instead of a week on a workstation ) . it becomes clear from the deposition example , that the efficiency of the cautious advancement parallel method for sequential random update depends on the topology of the connections among the components . the sparse fixed connections and a large diameter of the connection graph increase the efficiency . a small - diameter graph with all - to - all connections or close to such makes the pes to be too cautious ; too few of them would be non - waiting . in the worst case only one pe dares to advance the local time during a cycle while all the other pes are cautiously waiting . that is the case in the circuit - switched network simulation , where node pairs @xmath156 are close to each other in the sense of network connectivity . even if we somehow resolved the difficulty that this model does not completely fit the random sequential update model as discussed in the beginning of the section , its parallel execution by the described method would not be efficient . however , the actual event dependency along the executional path is rather sparse which presents an opportunity for parallelism . unfortunately , this parallelism is not extracted by the cautious advancement method , because the method requires a variable event dependency graph to be upper bounded by the fixed component connectivity graph . ising model and the phone provider competition model have a sparse component connectivity , but they still may fail to produce an efficient simulation using the described technique . this is because , for example in the ising model , among the parameters the spin flip rate depends on is the temperature @xcite , and a low temperature causes the ratio between a particle update rate and its upper bound to be large . even though a high enough fraction of pes do not wait with processing their events , most of the processed events are trivial time advancements without a flip . this ruins the efficiency of the parallel execution in low temperature regimes . a successful use of the cautious advancement technique and its further developments for an ising model at a non - very - low temperature has been demonstrated @xcite with the model running on 400 pes of a parallel supercomputer t3e and yielding a speedup of 260 . another example is a wireless simulation @xcite where the calls are dynamically arriving at random positions in the service area according to a fixed distribution and the users are not moving during the calls . with such assumptions it is possible to arrange a variant of efficient cautious advancement parallel processing . next section discusses an alternative method of parallel execution . it aims to extract parallelism as it emerges during the execution rather than relying on a worst case estimate given by the connectivity graph . also it needs no uniformization of the event arrival rates . as in section [ s : sequent ] , we consider a discrete event simulation of a dynamic system with @xmath0 components . the simulation is to be performed on a parallel computer with @xmath0 processing elements . as before , the procedure is to give each pe a component to host , and have the pe produce the state change history of that component . the method to do so will be different from the one discussed in section [ s : sequent ] . unlike the previous method in which the event processing is final , the present method involves speculative computations wherein an event can be processed and then later rejected . the procedure was called _ synchronous relaxation_@xcite . in this procedure , each pe keeps track of the simulated time before which no event is to be rejected in the course of further processing ; this quantity is called _ committed time . _ the pes increase the committed time in lockstep , its value is common to all pes . each step consists of several iterations . at each iteration , while the committed time value does not change , each pe produces a speculative state change trajectory of the component it hosts beyond the committed time . the pe extends the trajectory until its local time reaches the committed time plus @xmath4 , where @xmath4 is the step size of committed time increases . unlike the @xmath4 of a time - driven simulation discussed in section [ s : edr ] , here @xmath4 does not define the accuracy of simulation and may be not small . since components are , in general , connected , to produce a correct trajectory of its component , a pe needs to know the correct histories of other components . but they are not known , because other pes are in the same quandary . the mechanism of generating the correct trajectories is by iterations . during the first iteration , each pe makes the simplest assumption about the trajectories of the other components , for example , that the other trajectories are empty of events , i.e. , states do not change . this will enable the pe to produce the hosted component trajectory . after every pe generates the speculative trajectory for additional @xmath4 units , they compare the trajectories . this comparison phase is started only after all pes have generated the trajectories . as a rule , a pe will detect inconsistencies between the assumed and actually generated trajectories of other components . if so , pes perform more iterations . during subsequent iterations , if a pe needs to know the trajectory of a component hosted by another pe , it uses the trajectory generated in the last iteration . the goal of producing correct trajectories at a step is achieved if no pe detects any inconsistencies between the assumed and actually generated trajectories of other pes . once this happens , all pes increase committed time by @xmath4 and continue . the synchronous relaxation parallel algorithm @xcite , used on a maspar computer with 16,384 pes , cuts the running time of simulating a circuit - switched wired network to a few minutes ( from several hours in the best sequential implementation on a fast workstation ) . naturally , the efficiency of the synchronous relaxation method relies on the convergence to be achieved at each step in a small number of iterations . to assess the number of iterations we examine the event dependency graph . for the event chains like the one in figure [ chain ] there will be many iterations . figure [ chain ] depicts an artificially difficult , worst case example . such event dependency graph may correspond to a single indivisible object which moves in space visiting the areas hosted by different pes . it is not feasible to make an efficient parallel simulation in such a special example . in figure [ levels ] , on the other hand , an `` average '' example is presented . it is obtained by randomly `` sprinkling '' the events - circles and possible event dependency links on the time - space diagram , without a particular application in mind . a good upper bound on the number of iterations can be supplied by counting _ levels_. the levels can be identified without knowing the system partitioning into components hence no such partitioning is shown in figure [ levels ] . level 0 consists of already processed events that are positioned below the @xmath4 strip . level 1 consists of those events at or above the lower boundary of the strip , which are immediate effects of only level 0 events . by induction , level @xmath86 consists of the events at or above the lower boundary of the strip , whose immediate causes are events at levels @xmath194 . in addition , to qualify for being on level @xmath86 , the event must have al least one event on level @xmath195 among its immediate causes . before the step begins , all level 0 events are correct . after all the pes generate their trajectories at iteration 1 , all level 1 events at least will be among the correctly settled events . it can be seen by induction that after iteration @xmath86 , all events on level @xmath86 or lower are correct . thus , the number of levels for those events of the event dependency graph that fit within the considered @xmath4-strip is an upper bound on the number of iterations needed for correctly determining all events for this strip . the actual number of iterations can be smaller than this upper bound for two reasons : 1 ) initial guesses of events are correct , 2 ) the event dependency subgraph hosted by a pe contains a complete set of cause - effects for several levels without need to know events in the neighboring pes . situation 1 is not always negligibly rare : in the applications in which there are only two choices for an event , reasonable initial guessing might save iterations . an extreme case of situation 2 is completely independent subsystems hosted by different pes , or , for that matter , just a single pe which hosts the entire system . in these conditions , all events are determined correctly at the first iteration . the question remains : how many event levels fits in the @xmath4-strip on an `` average '' ? a conjecture can be proposed which says , that , in a `` generic '' example , if @xmath0 tends to infinity , the `` average '' number of levels increases not faster than @xmath40 . this has been established in several applications , for example , in the simulation of circuit - switched networks @xcite . one may notice a similarity of the synchronous relaxation algorithm and the time warp algorithm @xcite . indeed , both algorithms use speculative event processing . the time warp procedure can be qualified as an _ asynchronous _ relaxation . instead of frequent synchronization , the original tw procedure allows each pe to proceed at its own pace , without explicitly synchronizing with other pes . as a result of tighter synchronization , the synchronous relaxation performs better than the tw in a worst case . the tw is known to sometimes enter undesirable modes like rollback avalanche or cascading , which might slow down unduly even well parallelizable simulations . unlike the synchronous relaxation , no mathematical guarantee of scalability of the tw algorithm to a large @xmath0 has been offered . whenever there is a choice between a method with speculative computations and a method without , if both methods should deliver a scalable parallel simulation , the non - speculative method should be taken because speculative computations always involve a heavy computing overhead . sometimes for the same simulation model in some regimes one can do well without speculative computations , while in the other regimes one can not . such is the ising model example . the uniformization entails a heavy overhead only for a low temperature and then synchronous relaxation is warranted . for higher temperatures , the method discussed in section [ s : sequent ] provides a reasonable alternative . there are many aspects in computer simulations , such as visualization , user interface , convenience and efficiency of programming etc . the aspect which comes first in simulating large dynamic systems is that of computing efficiency . a lesson learned from experiences in such tasks is that computing efficiency is determined by the properties of the underlying computational technique whereas the choice of the best technique is not defined by the modeling area . the same algorithmic idea may work well across diverse applications and modeling areas . another lesson is that no single `` silver bullet '' technique for efficient simulation has been offered thus far and that a concrete simulation model may need a combination of available techniques to work fast . sometimes one has to modify the model to fit a good technique . yet in other cases very substantial improvements in computing speed are achieved if a basic technique is modified or augmented to fit the application , rather than being used in a fixed `` prepackaged '' form . s. eick , a.g . greenberg , b.d . lubachevsky , and a. weiss , synchronous relaxation for parallel simulation with applications to circuit switched networks , " _ acm trans . on modeling and computer simulation * 3 * _ , no . 4 , pp . 287314 , 1993 . greenberg , b.d . lubachevsky , d.m . nicol , and p.e . wright , efficient massively parallel simulation of dynamic channel assignment schemes for wireless cellular communications , " _ proc . 8th workshop on parallel and distributed simulation , pads94 _ , edinburgh , scotland , uk , pp.187 - 194 . linhart , b.d . lubachevsky , r. radner , and m.j . meurer , ` friends and family ' and related pricing strategies , " proc . 2nd russian - swedish control conference , aug . 1995 , st.petersburg state technical university , russia , pp . 192 - 196 .

we study the cluster mass function in mixed dark matter ( mdm ) models , using two cobe normalized simulations with and , and and , both with 2 massive s ( mdm1 and mdm2 , respectively ) . for the sake of comparison , we also simulate a cdm model with spectral index ( tcdm ) , also cobe normalized . we argue that , in our non hydro simulations , where cdm particles describe both actual cdm and baryons , the galaxy distribution traces cdm particles . therefore , we use them to define clusters and their velocities to work out cluster masses . as cdm particles are more clustered than hdm and therefore have , in average , greater velocities , this leads to significant differences from ps predictions . such predictions agree with simulations if both hdm and cdm are used to define clusters . if this criterion is adopted , however , we see that : ( i ) mdm corresponds values slightly but systematically greater than cdm ; ( ii ) such exhibit a scale dependence : on scales , we find or 1.8 for cdm or mdm , respectively ; at greater scales the required decreases and a substantial cluster excess is found , at the large mass end ( ) . clusters defined through cdm in mdm models , instead , are less numerous than ps estimates , by a factor , at the low mass end ; the factor becomes.8 , depending on the mix , on intermediate mass scales ( ) and almost vanishes on the high mass end . therefore : ( i ) mdm models expected to overproduce clusters over intermediate scales are viable ; ( ii ) the greater reduction factor at small scales agrees with the observational data dependence on the cluster mass ( which , however , may be partially due to sample incompleteness ) ; ( iii ) the higher spectral normalization is felt at large scales , where mdm models produce more objects ( hence , large clusters ) than cdm . mdm1 even exceeds donahue et al . ( 1998 ) findings , while mdm2 is consistent with them . simulations are performed using a parallel algorithm worked out from couchman ap3 m serial code , but allowing for different particle masses and used with variable time steps . this allowed to simulate a cubic box with side of 360mpc , reaching a plummer resolution of 40.6kpc , and using particles . 0.2 truecm pacs : 95.35 ; 98.80 ; 98.65.cw # 1#2#3#4#1 * # 2 * , # 3 ( # 4 ) &

astro-ph9905281

cosmological n body simulations have become the basic tool to study the non linear stages of structure formation and evolution in the universe . in the least few years , the use of parallel computers has allowed simulations with an increasingly wide dynamical range and it is now possible to simulate boxes , with side up to hundreds mpc s , with distance and mass resolutions well below the galaxy cluster scale . in this paper we shall report the results of three n body simulations , in a box of side 360@xmath15mpc ( @xmath18 is the hubble parameters in units of 100 km@xmath19 ) , aimed to study the properties of galaxy clusters . in particular , we shall concentrate on mixed models , which , up to now , were sometimes disregarded in large parallel simulations . two of the models tested are therefore mixed models , both with 2 massive neutrino ( @xmath4 ) species ; @xmath4 masses are @xmath20ev and 1.63@xmath21ev , to yield @xmath0 and 0.14 , respectively . ( more details on the models are given below ; see , in particular , table i. ) models were selected , first of all , to approach observational data . in this work , however , we wish to quantify a peculiar behaviour of the cluster mass function , that we expect to occur for mixed models . within this context , the third model , a tilted cdm , was mostly selected for the sake of comparison . clusters of galaxies are the largest bound systems in the universe and are also fairly rare objects . they have been systematically studied in the recent years , using both optical and x ray data . in principle , this allows to use them to provide stringent constraints to cosmological models . the behaviour of their mass function , in particular , is a critical test . mass function data are given and/or discussed in several recent papers ( @xcite ) . for any cosmological model , the press @xmath22 schechter ( ps ) approach provides a semi analytical estimate of such mass function and model simulations already showed a basic agreement with such estimates . some discrepancies were however outlined at the low and high mass ends , which will be also confirmed by our analysis . there is however a well defined point that we wish to explore in this work . we shall show that the presence of a hot component may substantially widen the gap between expected and observable mass functions . the qualitative reasons of this effect can be explained soon , but the extent and the mass dependence of the gap can only be numerically explored . as is known , fluctuations in the different components of mixed models evolve differently . on the scales where hot fluctuations were initially erased , @xmath4 s can be trapped again later on , as they become slower , by fluctuations which survived in cdm ( and baryons ) . eventually , in most reasonable models , at @xmath23 , the transfer functions of the three massive components are again almost equal . at the non linear level , instead , @xmath4 s can be distributed differently from cdm and baryons , even at @xmath23 . several simulations ( see , e.g. , @xcite ) showed their final distribution , which follows the shape of non linear structures , but with a tendency to stay away from structure knots , rather inhabiting peripheral regions . such behaviour can be more or less pronounced , according to the scale considered . in the real world , baryons are the main mass tracer . galaxies emit because of their presence and diffuse gas emits radiation because of the electrons accompanying them . locating baryons in a simulation with no hydrodynamics may be hard . e.g. , in a recent work , valdarnini , ghizzardi @xmath22 bonometto ( 1999 ) have shown that using cdm instead of baryons , to study the global cluster structure , may be misleading and surely changes the scores of various cosmological models , when compared with x ray data . the question exists also for simulations involving no hot component , but the presence of a hot component might complicate the answer . in principle , however , the outcome would be simple . when initial conditions for the simulation are set at @xmath24 , we use separate transfer functions for hdm , cdm and baryons . the latter two components , however , have almost identical transfer functions , as baryon infall in potential wells , set by cdm at the end of recombination , is already complete . at @xmath24 , instead , hdm is still differently distributed . moreover , @xmath4 s keep average velocities @xmath25 , of thermal origin . simulations are then performed with two kinds of particles , conventionally called cdm and hdm . cdm particles , however , account both for baryons and actual cdm . hdm particles , instead , account for @xmath4 behaviour and , since the beginning of the simulation , have quite a different treatment , according to their peculiar properties , including their large average velocity @xmath26 . the obvious conclusion is that , in a dissipationless simulation , we should seek baryons where we put them , @xmath27 with cdm _ particles_. previous authors , however , distinguished between galaxy and gas setting , even in dissipationless simulations . e.g. , kofman et al . ( 1996 ) , who performed a pm simulation of mdm with 2 @xmath4 s of mass @xmath28ev and @xmath29 , in a box of @xmath30mpc ( @xmath31 ) , using @xmath32 particles and a force resolution @xmath33kpc ( hence a greater mass resolution and a slightly worse force resolution in a much smaller box , if compared with simulations described in this work ) , stated that distribution and motion of galaxies are primarily signatures of cdm particles . however , to try to spot x ray emitting gas , they set it in hydrostatic equilibrium , assuming a spherically symmetric static gravitational potential given by a king like profile , while the temperature profile was worked out using data on the x ray surface brightness of the abell cluster a2256 , which is a rich , spherically symmetric , almost relaxed cluster , quite similar to coma . this kind of prescriptions are however almost useless if simulations are compared with galaxy data ; cdm and galaxy distributions can be assumed to be similar on the scales resolved in this work . therefore : ( i ) even in mixed models , to compare simulations with data , we define clusters using cdm particles only . ( ii ) in order to evaluate cluster masses to fit data , we estimate them through the velocity distribution of cdm particles only . observational estimates , of course , are performed along a complex pattern . we shall be using recent results provided by girardi et al . ( 1998 ) , obtained using enacs data , and previous results of the same group ( biviano et al . 1993 ) . in principle we should work out a mock catalog from our simulation and use it , so to reproduce all biases that data may have and use all prescriptions followed to try to overcome them . we plan to do so in a forthcoming work . meanwhile , however , it is clear that cluster masses are to be determined through velocity distributions . the procedure to work out such _ virial _ masses , aiming to approach what is done in observations , will be better described below . week gravitational lensing , instead , depends on the whole mass , carried either by cdm or hdm particles , through the potential it produces . we shall make no comparison here with masses obtained through lensing data . just notice that , according to our arguments , in mixed models , they may be grater than _ virial _ masses by a factor which exceeds unity by more than @xmath34 . if we associate these points to the fact that cdm and hdm are distributed differently and expected to have different velocity distributions , it is clear that significant discrepancies from ps expectations can be present in mixed models . let us examine which could be their trend . first of all , cdm and hdm distributions are more and more alike as we go to greater scales . henceforth , above a suitable scale , the peculiarity of mixed models might become negligible . it is however hard to give an analytical estimate of such scale , if it exists . then , ps mass functions are worked out from transfer functions . if they are equal for the three components at low @xmath35 , we can expect that ps predictions hold if clusters are defined using both cold and hot particles . this expectation is however to be numerically verified . assuming that this is the case , the smoother distribution expected for @xmath4 s , in respect to cdm , may lead to a decrease of the amplitude of the actual cluster mass function . in particular , we shall see some examples where @xmath4 clouds around smaller cdm condensations create bridges among them . in similar cases , clusters defined using all particles turn into multiple systems of smaller objects , when only cdm particles are considered . but , quite in general , hdm is mostly periferal and helps to extend the volume where cdm particles are attributed to the cluster . forgetting hdm particles , therefore , leads to a decrease of the very cdm particles belonging to most clusters . this causes a decrease of cluster masses which may exceed the decrease due to neglecting hdm . its extent , however , is to be numerically tested . however , besides using cdm particles only , we must recall that real mass measurements are not based on counts of particle _ numbers _ ; obviously , this can be done only in simulations . optical and @xmath36ray data , instead , are essentially obtained through velocities . if hdm particles tend to stay farther from the main condensations , preferring regions where the potential well is shallower , we can expect that cdm particles , in average , take an extra share of the overall kinetic energy , that will depend on detailed @xmath4 parameters . if their kinetic energy is used to test the cluster mass , we can expect a systematic overestimate . once again , a quantitative evaluation of the effect can hardly be obtained without numerical means . these are the main physical points that we wish to explore through our simulations . as we shall see , such effects do exist and are quantitatively relevant . exploring the world of mixed models can be particularly expensive , from the numerical point of view . in a given standard cdm ( scdm ) simulation , a change of normalization can be interpreted as a change of the time when @xmath23 ; furthermore , scdm can be turned into @xmath37cdm ( see , e.g. , @xcite ) by rescaling the box and the shape parameter @xmath38 , which , in the transfer function of @xmath37cdm models , always appears in the combination @xmath39 . ( see section 3 for a discussion on the meaning of @xmath40 . ) such multiple interpretations of a single simulation do not exist for mixed models . even apart of the @xmath37cdm variant , a change of normalization should be accompanied by a shift of the velocities of hot particles at the simulation start . besides the restriction `` one model one simulation '' , mixed models need at least 3 particles instead of 1 , to obtain the same level of resolution obtainable with pure cdm . in fact cdm and hdm components need different particles and the latter ones must be double in number , to account for thermal velocities , without introducing a spurious non zero linear momentum density . further families of models which can be simulated using one kind of particles only are @xmath41cdm and models with total density parameter @xmath42 ( ocdm ) . the importance of @xmath41cdm models has greatly increased after an improved analyses of snia data ( @xcite ) indicated an accelerating cosmic expansion . ocdm models , instead , seem favoured by the presence of high redshift large scale matter condensations ( @xcite ) and by the statistics of arclets ( @xcite ) . the latter conclusions , however , might have to be partially reconsidered on the light of some results of this analysis . this work was made possible by our implementation of a parallel n body code , based upon the serial public ap3 m code of couchman ( 1991 ) , extended in order to treat variable mass particle sets and used varying the time steps , when needed . our simulation , dealing with a 360@xmath15mpc cubic box , needed either 180@xmath43 ( for pure cdm ) or @xmath44 ( for mdm ) particles . the force resolution is set by a plummer equivalent smoothing parameter @xmath45kpc . our comoving force and mass resolutions approach the limits of the computational resources of the machine we used ( the hp exemplar spp2000 x class processor of the cilea consortium at segrate such resolutions are close to the ones of the four simulations for pure cdm with different initial conditions described by colberg et al . ( 1997 ) , thomas et al . ( 1997 ) , or cole et al . see also gross et al . more details on the simulations are given on section 3 . the plan of the paper is as follows . in the next section we discuss first the linear features of the models treated . in section 3 we give all technical details concerning our simulations . their outputs are then used to show the evolution of the spectra ; the final spectrum , at @xmath23 , will then be compared with apm data and current approximated expressions , finding discrepancies which characterize mixed models ( section 4 ) . in section 5 , five different algorithms , used to select clusters , are briefly discussed . an extended comparison among their outputs is outside the scope of this work ; our basic issues are contained in section 6 , where the effects of the presence of a hot component , on the cluster mass function , will be suitably summarized . section 7 is devoted to a final discussion . quite in general , to define a model we require : background metric , substance mix and primeval spectrum . the background metric is fixed by the hubble parameter @xmath46 and the overall density parameter @xmath47 ( @xmath48 : present average density , @xmath49 : critical density ) . substance is fixed by partial density parameters . e.g. , @xmath50 or @xmath51 , which are the ratios between baryon or cold dark matter ( cdm ) and critical densities . hot dark matter ( hdm ) is fixed by two of input data : besides of @xmath52 , we must give its number of spin degrees of freedom @xmath53 . in this paper , instead , @xmath54 . further quantities to be specified are the present cbr temperature and the effective number of massless@xmath4 spin degrees of freedom ( @xmath55 ) . finally , early deviations from homogeneity are parametrized by the amplitude @xmath56 and the spectral index @xmath57 of the initial fluctuation spectrum @xmath58 ( here : @xmath59 is the comoving horizon distance ; @xmath60 is the wave number related to the comoving length scale @xmath61 and mass scale @xmath62 ) . in this section , we report predictions on large scale structure ( lss ) obtainable from the linear theory , for the models we simulated . in all our models @xmath63 and the hubble parameter @xmath31 . the other model parameters are shown in table i , together with the above mentioned predictions , worked out from transfer functions . transfer functions are obtained by solving numerically a large set of coupled differential equations . the set is particularly wide for mixed models ( see , e.g. , @xcite ) . recent work @xcite has shown that a wide parameter space exists , for which mixed models are consistent with lss features that we can estimate within linear theory . assuming the presence of scalar fluctuation modes only , in @xmath63 models , the angular fluctuation spectrum of cbr , for small @xmath64 values , can be approximated with its sachs & wolfe expression ( see , @xmath65 , @xcite ) : @xmath66 ( @xmath67 are bessel fuctions ) . as @xmath68 ( see , @xmath65 , @xcite ) the spectral amplitude @xmath56 is obtainable from data for @xmath69 . the models we simulate have different @xmath57 values : for the tilted cdm ( tcdm ) we took @xmath70 while , for mdm1 and mdm2 , @xmath1 and @xmath71 , respectively . among physical unknowns , there is the contribution of primeval gravitational waves to the observed @xmath69 value . as a working solution , in order to leave room for such signals , without spoiling the fit with cobe data , we decided to work out @xmath56 values from lower 3@xmath72 limits on @xmath69 . to our knowledge , simulations of mixed models with @xmath73 were never performed before , for any box size and dynamical range , except by lucchin et al . ( 1996 ) , who however considered a model with @xmath74 and a single @xmath4 family , normalized to the central cobe value , and therefore significantly violating some linear constraints ( e.g. , @xmath75 , see below ) . using the transfer function @xmath76 , we can evaluate the mass variance over the comoving scale @xmath61 : @xmath77 here , for a top hat window function , @xmath78 . for @xmath79 and 25 mpc , we evaluate @xmath80 and @xmath81 , which allow to work out @xmath82 the connection between this general definition and some current approximated expressions , like @xmath83 , can be found in @xcite . its values , as well as the values of @xmath80 , are reported in table i. peacock & dodds ( 1994 ) , using apm data , and borgani et al . ( 1994 ) using abell / aco samples , obtained the ( 2@xmath72 ) intervals 0.190.27 and 0.180.25 for @xmath38 . according to the ps approach , the expected cumulative number density of clusters is related to @xmath80 and @xmath38 and reads : @xmath84 \exp(-u^2/2 ) ~. \eqno ( 2.6)\ ] ] here @xmath85 is defined so that the mass variance ( expressed in function of mass scale , instead of length scale ) @xmath86 ; @xmath6 values from 1.4 to 1.8 ( @xcite ) were considered . making use of eq . ( 2.5 ) with a top hat window function , the transfer function can be used to compute @xmath87 for @xmath88mpc and @xmath89 . values of @xmath75 , obtained within the ps approach , are given in table i , for different models . an observational range is @xmath90 ( @xcite ) , but this result is still subject to a number of uncertainties and values within a factor 2 from upper and lower limits can not be safely rejected . however , even with such extra freedom , our mdm1 model would appear out of the observational range . we must however bear in mind that ps estimates are based on the transfer function at @xmath23 , which are almost identical for all dark matter and baryon components . as we shall see , non linear effects lead to greater condensations in cdm ( and baryons ) than in hdm , at all @xmath35 s . testing the consequences of such effects is one of the aims of this work . taking them into account , for reasons widely discussed below , @xmath75 may become substantially smaller and even mdm1 will be found compatible with data . further severe tests for tilted or mixed models concern the formation of structures at high @xmath35 . among them , one of the most stringent is the amount of gas expected in damped lyman@xmath91 systems , which can be expressed through the parameter @xmath92 . here @xmath91 is an efficiency parameter ( @xmath93 ) . let then @xmath94 be the mass variance , worked out from the transfer function @xmath95 , for the mass scale @xmath14 at redshift @xmath35 . using its value , @xmath96~. \eqno ( 2.7)\ ] ] is easily obtained . accordingly , we evaluate @xmath97 , taking @xmath98 and @xmath99 in @xmath100 , @xmath101 and a top hat window function ( see , however , @xcite ) . data provided by storrie lombardi et al . ( 1995 ) give @xmath102 , while @xmath103 values expected for the different models are given in table i. expected bulk velocities were also evaluated for our models and found consistent with potent reconstructions of velocity fields . altogether , therefore , our models can be said to predict linear features consistent with most current linear observational constraints . the simulations start from the redshift @xmath104 . the particle sampling of the density field is obtained applying the zeldovich approximation ( @xcite ) starting from a regular grid . for mdm models , the prescriptions of jing @xmath22 fang ( 1994 ) or borgani et al . ( 1997 ) were followed ( see also the seminal paper by @xcite ) we adopt the same random phases in all simulated models . particular care is taken to treat the hdm component , sampled by couples of particles initially set in identical positions , with opposite thermal velocities , in order not to have a spurious linear momentum . henceforth , the simulation will contain a double number of hdm particles compared to cdm . in the definition of initial mode amplitudes , the same random numbers were adopted for cold and hot components . this allows a fair fit of numerical and analitical amplitude growings , at the initial stages . as already mentioned , our simulations study a periodic cubic box of side @xmath105mpc . cdm+baryons are represented by @xmath106 particles , whose individual mass is @xmath107 for tcdm , @xmath108 for mdm1 and @xmath109 for mdm2 ; the mass of the @xmath110 hdm particles is @xmath111 for mdm1 and @xmath112 for mdm2 . outputs of the simulations are preserved at 20 intermediate redshifts @xmath113 set at constant time intervals @xmath114yr . the comoving force resolution is given by the softening length , @xmath115kpc . the force @xmath116 is evaluated considering each particle as a smoothed distribution of mass , with shape @xmath117 for @xmath118 ( this is the so called s2 shape , @xcite ) . it behaves as @xmath119 when @xmath120 . since the softening of the force is usually referred to a plummer shape , @xmath121 , we used a least @xmath122 test to estabilish the best approximation between the forces generated by the two different shapes . the minimum @xmath122 occurs when @xmath123 . in our case , this corresponds to a plummer equivalent softening @xmath124kpc ; we will use this latter value of the softening as our nominal force resolution . for the sake of comparison , the numerical simulations of colberg et al . ( 1997 ) , thomas et al . ( 1998 ) have @xmath125 or @xmath126 particles in a 239.5@xmath15mpc cubic box with @xmath127kpc , and cole et al . ( 1997 ) have a cubic box of side @xmath128mpc , a softening @xmath129kpc with @xmath130 particles . gross et al . ( 1998 ) using a parallelized standard pm code simulated a mixed model and various cdm variants in a box containing @xmath131 particles and @xmath132 grid points . they performed simulations both at high and low resolution in boxes @xmath133 and @xmath134mpc wide which correspond grid sizes of @xmath135 and @xmath136kpc . the claimed softening is @xmath137 times the grid size . the program used to compute the evolution of particles , under the action of gravitational forces , is a parallel code developed starting from the public ap3 m serial code of couchman ( 1991 ) . such code divides the interparticle forces in long range and short range ones . the long range forces are accounted for by the pm ( _ particle mesh _ ) part , which solves the equations of gravitation in the fourier space , where they are essentially algebric , making use of fft to transfer results from coordinate to momentum space and _ viceversa_. the resolution of pm calculations is set by the number of grid cells , that we fixed at @xmath138 . further resolution , in our case down to 0.08@xmath139 , is attained by direct summation between neighbour particles in the pp ( _ particle particle _ ) part . however , where the particle density attains or exceeds @xmath140 times the mean value , the evaluation of the short range forces is performed by repeating the above scheme ( refinement ) , therefore furtherly subdividing the forces in long and short range ones , and treating them with a pm and pp calculation respectively . the boundary conditions for the refinements are therefore not periodic . if the refinement volumes still contain high density regions , the refinement process can continue to deeper levels . the parallelization of the code was made taking into account the technical characteristics of the hp exemplar spp2000 x class processor of the cilea consortium ( segrate milan ) we employed . the machine architecture is shared memory . this architecture allows to parallelize the calculation without explicitly managing the data distribution . the spp2000 of cilea is composed by two hypernodes , each with 16 hp pa risc 8000 processors and a total memory of 4@xmath21gbyte . the code worked on a hypernode of the machine , using 8 of its processors . this is due to system specific requirements . the memory resources we used were up to @xmath141 and @xmath142gbyte ( out of @xmath143gbyte of shared memory available ) for tcdm and mdm , respectively . approximately half required memory is spent to store particle coordinates , momenta and masses . the rest of ram is then needed for all other calculation purposes ( density field storage , fft overhead storage , etc . ) . the main effort to parallelize the program was concentrated on pp , both at level 0 and refinements . once the pm part is executed , pp calculations at level 0 are shared among available processors , as the linked list mechanism of the code automatically divides consequently the different volumes where the pp calculation is being performed . when the threshold for refinement is attained , the related operations are executed by a single processor . further refinements are assigned to another processor , etc . ; this allows to execute , with the setup we used , up to 8 refinements simultaneously . as is usual , let us define the speed up @xmath144 as the ratio of the cpu time used for serial execution versus the solar time spent for parallel execution on @xmath145 processors . our program speed up is @xmath146 for the pp part at the first steps , but , as is expected , @xmath147 slows down to @xmath148 at the last steps , due to unbalancing . as far as refinements are concerned , we reach a value @xmath149 at the last steps . the pm part , instead , has no substantial advantage from using system parallel fft . altogether , the speedup remains roughly constant along the simulations , because of the increasing computational weight of the short range calculus , as clustering develops and , in average , never overtakes a value @xmath150 . the number of steps made was different for the 3 simulations . 1000 equal @xmath145time steps were used for tcdm ( the time parameter @xmath151 ; at these late redshifts , the expansion is nearly matter dominated , also in the presence of the hdm component , and equal @xmath145 intervals yield equal time intervals ) . mdm1 , instead , was run splitting each of the first 100 above steps into two equal parts . hence it used 1100 time steps . the time step choice for mdm2 was more complex . with reference to the 1000 equal time steps ( as used for tcdm ) , we give a succession of number pairs ; the former one indicates the step numbers out of 1000 , the latter one the step numbers they have became ( in order to fulfill the criteria outlined herebelow ) : [ 2,10 ] [ 23,92 ] [ 50,100 ] [ 125,125 ] [ 800,400 ] . such step choices were dictated by two criteria : ( i ) energy conservation . according to layzer irvine equations ( see , e.g. , @xcite ) , it had an overall violation @xmath152 for tcdm and mdm2 and @xmath153 for mdm1 . ( ii ) cole et al . ( 1997 ) requirements , that the rms displacement of particles in a step is @xmath154 and the fastest particle has a displacement @xmath155 were never violated . in fact , for mdm1 , the ( ii ) criterion would have been violated in the first 100 of 1000 equal time steps , because of the residual thermal velocity of @xmath4 s . the problem was even more serious for mdm2 . this led to the subdivision of the initial part of the runs . however , we also performed the initial 10@xmath156 of mdm1 in 100 time steps , thus violating the cole criterion . final particle positions are displaced up to 1.6@xmath157 , although the average displacement can hardly be distinguished from computational noise . ( similar displacements are found when the same program is run on two different machines . ) accordingly , no discrepancy is present in spectra . we should therefore conclude that the ( ii ) criterion is perhaps too restrictive . in fig . [ fig1 ] we show 3 slices ( one for each simulation ) , @xmath158mpc@xmath43 wide , of the simulation volume at @xmath23 , projected along their shortest side . approximately half of cdm particles are showed . an eye inspection of the plots shows two clear features : ( i ) the distributions of matter , in the two slices , have similar shapes , as is to be expected , as the two simulations are started from the same random numbers . ( ii ) mdm1,2 show more pronounced features on a greater scale , in respect to tcdm . this should be ascribed to the lack of power on intermediate scales ( 100500@xmath15mpc ) , which is one of the characteristics ( and problems ) of tilted models ; the shape of the spectra of the two models at @xmath23 can be seen in fig . [ fig3 ] herebelow . in this section we describe the evolution of the simulation spectra for matter distributions . in fig . [ fig2 ] the evolution of the spectrum of the 3 models , starting from initial conditions and up to @xmath23 is shown . two sets of 3 plots refer to the two components ( cold and hot ) of mdm1 and mdm2 and to their overall spectrum . the last plot refers to tcdm . the spectral points are worked out from particle coordinates , performing the following operations : on 180 grid points ( @xmath159 ) we construct a matter density field @xmath160 using the cic ( cloud in cells ) scheme ( here @xmath159 is a vector with components @xmath161 indicating the discrete coordinate of each cell ) . let then be @xmath162 , and let @xmath163 be its fourier transform . here @xmath164 values with components multiple of @xmath165 , up to 180@xmath166 , are considered . the moduli of @xmath164 span a set of discrete values , to be averaged to obtain @xmath167 ( averaging is over directions and on a modulus interval of width @xmath168 ) . the plots show that particle distributions are able to reproduce the initial power spectrum only above a scale @xmath169mpc . this is a standard feature for simulations , when initial conditions are set using a grid . white ( 1996 ) suggested an alternative approach based on a glass ( for a recent discussion of this point and a comparison between initial conditions set on a grid or on a glass , see @xcite ) . at variance with pm , p3 m simulations explore scales well below the resolution set by the initial conditions , by exploiting the particle particle part of the code . this is a standard and welcome feature of such simulations and there has been a wide debate on the reliability of p3 m outputs down to such scales ( see , e.g. , splinter et al . 1998 and references therein ) . here we keep to the standard approach , but one should bear in mind that some reserves on this point were risen in the literature . spectra are plotted at @xmath170 , at @xmath171 , @xmath172 and at @xmath173 . the two intermediate redshifts correspond to fractions 0.1 and 0.3 of the total time of non linear evolution @xmath174 . a further comment concerns the evolution of the spectrum of the hot component in mdm . its high@xmath175 rise at @xmath176 is a well known feature , appearing in mixed simulations , and originates from shot noise , as hot particles cover distances greater than cell sizes , because of thermal motions . several attempts are reported , in the literature , to reduce such effect ( an obvious way amounts to increasing the total number of particles used to simulate the hot component ) and to evaluate how far it can affect final results . in this work no advanced of sophisticated treatment is applied to this feature and we follow the standard pattern discussed by klypin et al . elsewhere , we plan to discuss such effect in more detail , stressing a peculiar feature , which is to be overcame and however risks to cause unphysical differences among simulations of the same model with different numbers of hot particles , unless suitable cautions are taken . in fact , pairs or sets of hot particles , initially in the same position , once reaching a distance @xmath177 , feel a mutual gravitational attraction , whose potential energy will cause a decrease of their thermal kinetic energy . such braking effect is due to representing @xmath4 s through galactic mass particles . in our simulation , we introduced an @xmath178@xmath179 option , switching off the interaction within each hot particle pair until they reach a mutual distance @xmath180 , coinciding with the average initial distance between cold and hot particles . at @xmath181 the whole option is switched off and all gravitational interactions are resumed . as is also shown by the appearence of shot noise at such time , however , the fraction of particles still in binding danger is then negligible . in fig . [ fig3 ] we compare the output spectra at @xmath23 with the linear spectrum at @xmath23 , the spectrum at @xmath23 corrected for non linearity according to peacock @xmath22 dodds ( 1996 ) and apm reconstructed spectral points ( @xcite ) . at variance with fig . [ fig2 ] , power spectra here are corrected for the effect of cic convolution , by dividing them by a squared top hat window function @xmath182 , where @xmath183 is 0.85 times the particle spacing width . as is shown in more detail by fig . [ fig4 ] , in the case of tcdm , simulation outputs almost overlap the the spectrum corrected for non linearity . for mdm , instead , simulation outputs systematically exceed the theoretical curve , although by a small amount . one of the main aims of the simulations is that of obtaining a large set of model clusters for each cosmological model , at various redshifts . this will enable us to study cluster evolution and to create mock cluster catalogs . it is not clear , instead , how far the global cluster morphology can be understood within the frame of non dissipative simulations ( see , e.g. , @xcite ) . here we shall report some basic results and general properties of the clusters selected in the simulation outputs . in particular , we shall give their cumulative mass function @xmath184 . the clusters we consider here were found using a spherical overdensity ( so ) algorithm , yielding the cluster locations , the radii @xmath185 inside which a density contrast @xmath186 is attained and the total mass @xmath14 of particles within @xmath185 . for mixed models , the procedure will be applied both to the whole set of cold and hot particles and to cold particles only , although only the latter output bears a direct physical significance . let us now describe our so procedure implementation . as a first step , candidate clusters are located using a standard fof algorithm , with linking length @xmath187 ( here @xmath188 is the average particle particle separation ) , giving as outputs groups with more than @xmath189 particles . we then perform the following operations : ( i ) we find the center of mass @xmath190 of each group and ( ii ) we determine the radius @xmath191 , inside which the density contrast is @xmath192 ( all particles are to be included , not only those initially found by fof ) . in general , the new center of mass is not @xmath190 . the operations ( i ) and ( ii ) , define a new particle group , on which the very operations ( i ) and ( ii ) can be repeated . the procedure is iterated until we converge onto a stable particle set . if , at some stage , the group comprises less than @xmath189 particles , we discard it . the final @xmath191 is @xmath185 . in the actual implementation of so , it may happen that the same particle is a potential member of two particle groups . in this case the procedure assigns it to the more massive one . as a matter of fact , this has the consequence that , sometimes , more massive groups swallow smaller ones . a consequence of this choice is a slight decrease of the total number of clusters , over all mass scales , and this is confirmed by some preliminary comparisons described below . in fact , before using this procedure on our simulations , we performed a number of checks on test simulations of the same cosmological models , with a smaller mass resolution ( @xmath193 64 ^ 3 $ ] particles , in the same volume ) . clusters found by our so procedure were compared with those found by other standard public cluster identification algorithms , namely : fof ( @xcite , http://www-hpcc.astro.washington.edu/tools/fof ) itself , hop ( @xcite ; http://www.sns.ias.edu/~ eisenste / hop / hop.html ) and denmax / skid ( http://www-hpcc.astro.washington.edu/tools/skid ) . we confirm a good agreement between the cumulative cluster mass functions @xmath184 obtained using group masses from fof ( @xmath194 ) and hop ( suitably tuning its parameters ) , and that most massive hop clusters show a systematic offset towards greater masses ( @xcite ) . the mass function found by denmax ( again suitably tuned ) finds a slightly smaller @xmath184 , while the so mass function mostly lays slightly below all of them . discrepancies are however small , in spite of being enhanced by the reduced mass resolution in the test simulations . the results of a detailed comparison among different cluster finders , applied to higher resolution simulations , will be presented elsewhere , stressing also morphological differences . the so procedure , however , is the one which provides clusters which appear to satisfy a precise virialization requirement ( a given overdensity in a sphere ) . in this work so was started setting @xmath194 , and used @xmath195 ( when using cdm particles only ) or 75 ( when also hdm particles are taken ) . in fig . [ fig5 ] we show the location and the mass of clusters in the same simulation slices shown in fig . [ fig1 ] . circles are centered on cluster centers of mass and their radii are proportional to cluster masses obtained summing all particle masses within @xmath185 ( the radius scale is set on graphic criteria and circles mostly exceed the projected physical volumes of clusters ) . two features can be easily seen even by eye : ( i ) the location of clusters is similar in the three simulations , as is to be expected , as they are started with the same random numbers . ( ii ) in average , mdm1 clusters are more massive than tcdm , but not so numerous . qualitative differences between mdm2 and tcdm are not so striking . in fig . [ fig6 ] we report the cumulative cluster mass functions , for tcdm and for all particles of mdm1 and mdm2 ( open circles ) , compared with ps mass function curves , as given by eq . ( 2.6 ) , taking five equally spaced @xmath196 values between 1.4 and 1.8 ( @xmath197 corresponds to an isolated spherical protocluster collapse ) . these plots show a substantial agreement between theoretical expectations and simulation outputs , confirming previous results by @xcite . some discrepancies , that will be discussed soon , can be fairly easily understood and are really minor effects , when one considers how complex is the problem that the theoretical ps expression tries to face . discrepancies can be expressed through the dependence of @xmath6 on scales and models . for all models @xmath6 decreases when passing from smaller to greater scales . however , while for mdm1 it starts from values exceeding 1.8 , at scales @xmath198 , for mdm2 , 1.8 is never exceeded and , for tcdm , the start occurs for @xmath199 . from @xmath200 to @xmath201 , @xmath6 approximately decreases by 0.1 ( some irregularity seems however present in mdm2 ) . at even greater scales a further excess is met , leading to values @xmath6 up to @xmath202 , for tcdm . although the final @xmath6 is smaller for mdm1 and mdm2 , the excess of large mass clusters met in mdm1 is greater . similar features were already outlined by previous analyses of pure cdm simulations ( see , e.g. , @xcite , for low masses , @xcite , for the large mass end ) . our analysis shows that they are present also in mixed simulations ; furthermore , we find a decrease of cluster numbers , compared to ps expectations , for mixed models . the scale dependence is thought to arise because of deviations from spherical isolated growth . the smaller number of clusters we find in mixed simulations , instead , can be attributed to a slower gravitational growth , caused by the enhanced thermal velocities in one of the matter components . however , when mixed model predictions are to be compared with optical observations , the above analysis is unsuitable , as already argued in the introduction . with the resolution allowed by our simulations , hydrodynamical effects are expected to be negligible . baryons emitting light are therefore to be sought where they are initially set , when cdm _ particles _ were distributed at @xmath24 , using the ( already coincident ) transfer functions of cdm and baryons and granting them an overall density parameter @xmath203 . only hydrodynamics could separate cdm from baryons . let us remind that , instead , hdm particles were initially distributed according to a ( still ) different transfer function , their evolution permanently resented initial thermal velocities , which baryons certainly do not have , and they have a @xmath23 spatial distribution different from cdm particles , namely in those sites where non linearity forced cdm to evolve so rapidly that hdm could not overcome its pace and reach it . in the introduction we reported on attempts by previous authors to locate baryons in non hydro mixed simulations . suitable prescriptions were suggested to disentangle gas from cdm , in simulations with higher resolution than ours ; however , even with such greater resolution , no doubt was cast on the fact that galaxies are cdm tracers . at @xmath173 , discrepancies among transfer functions , set by linear evolution , are marginal . at the linear level , therefore , the distributions of baryons and dark matter components are already alike , down to galactic scales . non linear evolution , instead , is faster and the final distribution of cold and hot particles , involved in non linear processes , keep different . in fig . [ fig7 ] , [ fig8 ] and [ fig9 ] , we show several model clusters , at different mass scales and with different morphologies , for tcdm , mdm1 and mdm2 , respectively . in fig . [ fig8 ] and [ fig9 ] , where 2dimensional projections of 4 clusters from mdm1 and mdm2 simulations are shown , such differences are evident . in fact , an eye inspection of cdm and hdm distributions shows that dense knots are mostly populated by cdm particles , while the hdm particle distribution is significantly smoother . e.g. , the second cluster in fig . [ fig8 ] has a triple structure in cdm , which is almost absent in hdm . a similar situation holds for the third one , whose double structure is erased at the hdm particle level . the last two clusters are lighter ones , but it is still visible how central knots are essentially made by cdm particles . similar features are visible ( and perhaps stronger , probably because hdm particles are even lighter ) for mdm2 . multiple features and knots , present in cdm distributions , are attenuated or vanish in hdm . similar properties can be observed in broedbeck et al . ( 1998 ) visualization work , based on mdm pm simulations performed in a smaller volume ( a 50@xmath204mpc side cube ) , but with a force resolution similar to ours and a better mass resolution . accordingly , their cluster sample is 40 times less numerous than ours . it should be mentioned , however , that multiple clusters , such as those shown in the second and third panels of fig . [ fig8 ] , are , in a sense , survivors . in most cases , cdm bridges are not adequate to unify the smaller condensations . then we identify several smaller clusters instead of a big single one . but , more often , some of the smaller condensations pass below the lower mass threshold and do not contribute to the cluster mass function . these effects are unique to identifying clusters with cdm particles only in mixed simulations ; if we use all particles , we find a consistent extra number of multiple clusters like the ones shown in fig . [ fig8 ] . we conclude that most features expected in mixed model simulations are indeed present . let us now quantify them . in order to pass from cluster masses as sums of particle masses , to cluster masses comparable with optical data , two operations are to be performed . we first apply the so algorithm to cold particles only . once cold particles within @xmath185 are selected ( let @xmath14 be their total mass ) , we evaluate their rms velocity @xmath205 and use the relation @xmath206 to work out the virial mass @xmath207 . here @xmath208 is the gravitational constant ; the factor 2/3 implies that particle distribution is assumed to be approximately isothermal ( peebles 1993 , @xcite ) . in fig . [ fig10 ] we compare the behaviours of @xmath184 obtained for mdm1 and mdm2 in three different ways : ( i ) using all particles to define clusters and summing their masses ; ( ii ) using only cdm particles to define clusters and summing their masses ; ( iii ) using cdm particles to define clusters and eq . ( 6.1 ) to work out @xmath207 , as cluster mass , from cdm particle velocities . for the sake of comparison , in fig . [ fig11 ] , mass functions , obtained according to the ( ii ) and ( iii ) criteria , are shown also for tcdm . the @xmath184 behaviour , built to give the best approach to cluster mass estimate from data , is the ( iii ) one . in general , passing from ( i ) to ( iii ) , the decrease of the cumulative mass function , at the low mass end , approaches half order of magnitude . the difference becomes gradually smaller for greater and greater mass scales . the behaviour , however , depends on the matter mix and it may be premature to give an expression for cluster abundance reduction , on various mass scales . moreover , the method used to select particles belonging to a cluster , which has already an impact in defining masses according to the ( ii ) criterion , bears an even greater effect on the ( iii ) criterion : here , adding or subtracting a few high speed particles may significantly change @xmath205 ; the operational assumption of spherical symmetry in the so and similar procedures , in principle , might cause significant @xmath207 shifts , for intrinsically non spherical clusters . ( let us mention that procedures allowing non spherical clusters may have even more serious difficulties to select really bound particles ) . however , approximately , on mass scales @xmath12@xmath13 , the reduction factor from ( i ) to ( iii ) is 0.60.8 . cen ( 1997 ) applied eq . ( 6.1 ) to cdm simulations , but then went further , mimicking a number of observational biases and , in particular , evaluated @xmath205 from all particles in observational _ cylinders _ of radii @xmath209 or 2@xmath15mpc and constant depth @xmath210mpc , centered on cluster images . here we compare @xmath207 with @xmath14 , directly estimated in spheres of radius @xmath185 ( instead of cylinders ) , aiming to disentangle observational biases on @xmath205 , arising from projection effects due to particles in cylinders ( interlopers ) , from possible real physical features . in principle , for pure cdm , @xmath207 and @xmath14 should coincide . on the contrary , cen ( 1997 ) found @xmath207 values slightly exceeding @xmath14 , at small mass scales . the excess disappeared at the top mass end , where he found @xmath207 values smaller than @xmath14 . this trend certainly includes projection effects . in fact , the expected number of interlopers is proportional to the volume @xmath211 , and is independent from the mass @xmath14 of the cluster . on the contrary , within a fixed @xmath212 , just for geometrical effects , the number of galaxies belonging to the cluster is expected to be @xmath213 . observational biases for top mass clusters should therefore be smaller . this point is apparently confirmed by fig . [ fig12 ] , where the straight line , in all panels , yields @xmath214 . the right panel shows what happens for a pure cdm model . at lower masses , the number of clusters is greater , and the range of @xmath215 ratios is wider . in spite of that , it seems clear that an underestimate of masses is present all over the scales , if eq . ( 6.1 ) is used . the best fit straight line through @xmath14,@xmath207 points has a slope @xmath216 . this value , safely below unity , confirms that the effect found by cen ( 1997 ) is not only due to projections . let us outline that the best fit line can be hardly found by eye , from fig . [ fig12 ] , where the distribution of points on vertical bars at fixed @xmath14 can not be spotted . this effect has also an impact on the cluster mass function for tcdm . in fig . [ fig11 ] we show @xmath184 for tcdm , obtained using either @xmath14 or @xmath207 as cluster masses . the latter curve is smaller by a factor @xmath217 . in our opinion , the latter value are a better approach to observational data . let us now compare the effect found for pure cdm with what happens in mixed models . in spite of the noise due to peculiar cluster features , the left panel of fig . [ fig12 ] ( mdm1 ) shows that the data trend is steeper than for cdm . at the low mass end , virial masses seem consistent with particle masses . in average , this no longer seems true in the top mass range , even though the noise is large . a numerical fit gives a slope @xmath218 . the central panel refers to mdm2 and its features are somehow intermediate . at low @xmath14 , @xmath207 values have a deficit which looks much alike the one in pure cdm . this deficit decreases for greater @xmath14 and , at the top mass end , there seems to be a fair coincidence between @xmath14 and @xmath207 . fit slope is @xmath219 . this is intermediate between cdm and mdm1 , but statistically distinct from both . the first and main conclusion from the above arguments is that virial mass estimates give values whose relation with masses obtained summing all the particles depends on the substance by which the real world is made . the trend shown in fig [ fig12 ] is confirmed if we focus at the top mass end ( see fig . [ fig13 ] , which is analogous to a figure given by @xcite ) . by comparing the trend of @xmath215 in mdm with cdm models , there seems to be some evidence that , in average , the amount of hdm trapped inside a cluster of mass @xmath14 increases with @xmath14 itself . more hdm means a deeper potential well and also more cdm can be trapped . within this interpretation the difference between mdm1 and mdm2 is to be ascribed to the different amounts of hdm . we can conclude that : ( i ) the cumulative cluster abundance , that we expect to measure on optical data , in the presence of a hot component , is significantly smaller than standard ps estimates . the reduction amounts to a factor @xmath100.4 at the low mass end and reduces to @xmath110.8 , depending on the mix , on mass scales @xmath12@xmath13 , where cosmological models are usually compared with cluster observations . ( ii ) we should not expect that this effect is covered by projection effects . the discrepancy between @xmath14 and @xmath207 , found in cen ( 1997 ) is confirmed also excluding them . notice also that the very cluster mass function for cdm has a significant shift , when using virial masses . ( iii ) a contribute to the discrepancy between @xmath14 and @xmath207 , which is found also in the cdm case , is likely to arise from the assumptions that the matter distribution , inside clusters , is approximately isothermal . we plan to deepen this point in a forthcoming analysis , studying cluster profiles and other features of matter distributions in clusters . in fig . [ fig10 ] we compare simulation mass functions with data . open circles are biviano et al . ( 1993 ) data points , filled circles are girardi et al . ( 1998 ) data points . the latter analysis uses a wider sample and is corrected for several biases . if the same corrections are performed on the older data set , however , there is an excellent fit between the two observational outputs . the biases considered were : ( i ) subclustering , which might vary individual cluster masses by a large factor , but is found not to have significant effects on the overall function , which is quite alike both if a @xmath220 of substructured clusters is included or excluded . ( ii ) interlopers , whose exclusion causes an average reduction of cluster masses by a factor accounting for half of the discrepancy . ( iii ) boundary effects , which account for the rest of the shift in cluster masses . interlopers are clearly absent in simulation analysis . however , the assumption of isothermal particle distribution in clusters is admittedly just a basic approximation . the analysis performed on data , taking into account individual galaxy points as well as boundary effects , should be fully repeated on mock catalogs built from simulations , to improve the comparison level . this might even cause mass corrections similar to those due to boundary effects , which cover half of the shift between the two sets of data points . accordingly , in spite of the fact that the two data sets meet mdm1 and mdm2 mass functions , respectively , we should refrain from stating that this shows that mdm2 is a better data fit than mdm1 . tcdm , instead , is farther from data ( see fig.[fig11 ] ) . let us also draw the attention on the reduction of @xmath184 at the low mass end , which is mostly due to using cdm particles only , to define clusters . our simulations will allow an analysis of clustering evolution with @xmath35 , in mixed models . in fig . [ fig14 ] we show a preliminary but significant result , by comparing the cluster mass functions at @xmath23 and @xmath221 . as expected , the degree of evolution of the three models is similar and an excess of clusters at the top mass end is however present . as can be seen from fig . [ fig14 ] , in mdm1 there are 2 clusters with mass @xmath222 . in mdm2 the 3 largest clusters have a nearly identical mass @xmath223 . in tcdm , instead , there is 1 cluster of @xmath224 ; all other clusters are less massive . this behaviour is coherent with the spectral features of the three models , but is also linked to the higher @xmath80 value required in mixed models , to fit the observed cluster abundance at @xmath23 , over scales @xmath12@xmath13 . if we scale donahue et al . ( 1998 ) findings to the simulation volume , we expect there @xmath225 clusters of mass @xmath226 . this allows to suggest that , thanks to the different spectral shape and to the required higher normalization , mixed models , with @xmath63 , may not be in conflict with donahue et al . ( 1998 ) data analysis . in this paper we have reported the basic outputs of two large n body simulations , for mixed models and compared them to a tilted cdm simulation . all models are consistent with linear constraints . the numerical code was built by implementing couchman ap3 m program for a parallel shared memory machine . this allowed to attain a significant dynamical range with reasonable computational times ( @xmath227 week ) . we first reconstructed linear and non linear spectra , from particle distributions . at scales exceeding @xmath228mpc , non linear effects are small and reconstructed spectra fit linear expectations , even at @xmath173 . spectra are plotted against reconstructed apm 3dimensional data ; the best fit , at large scales , is given by mdm1 , which nicely passes through observational bars . it has been known since several years that mixed models provide a good fit of several features of lss ( ) . recent work ( @xcite ) has shown that the fit of galaxy spectrum data , about the @xmath229mpc peak , improves if a stronger bending of the spectrum , arising from lighter hdm particles , is partially compensated by a blue spectral index . both such ingredients rise the cmb angular spectrum @xmath230 , which can therefore fit both cobe and more recent experiments ( @xcite ) . e.g. , fig . 3 shows that mdm2 , which has a smaller @xmath52 and @xmath57 just above 1 , is not such a good fit to apm data as mdm1 . its score , however , is better than tcdm . ( moreover , mdm2 is still a good fit to @xmath230 at large @xmath64 values , while here tcdm fails . ) as far as the non linear behaviour is concerned , we found excellent fits with peacock @xmath22 dodds ( 1996 ) expression . mixed models , however , give non linear features slightly , but systematically , exceeding their expression . our simulations allowed us to follow the evolution of cdm and hdm components , while clusters form and in the field between different clusters . one might expect that the fraction of @xmath4 s , captured in a cluster , increases with the deepening of its potential well . this is neither what we see in the simulations , nor was it seen in previous works ( see , e.g. , klypin et al . 1997 , and references therein ; let us also draw the reader s attention on the visualization work by broedbeck et al . rather , @xmath4 s tend to remain on the outskirts of clusters , whose knots , in average , are partially @xmath4emptied . the point is that gravitational growth , in highly non linear sites , can be so fast that hdm fluctuations can hardly cover the gap they still had at @xmath24 , in respect to cdm , and that they almost cover in the linear growth . the main concern of this work , however , was the cluster mass distribution . the expectations outlined in the introduction section were confirmed by our numerical analysis . in particular we found that the actual number of clusters _ produced _ by a mixed model is significantly smaller than the amount predicted by ps estimates . on a scale @xmath231 , as the one usually adopted to test models using their transfer function , the reduction factor is @xmath2320.8 . simulations were actually compared with optical observational data ( biviano et al . 1993 , girardi et al . the latter data are obtained from a much wider sample , but most discrepancies arise just from the treatment of observational biases . the mass function obtained from simulations directly excludes some of such biases , e.g. , the presence of interlopers . some other corrections might however be needed before the comparison is safe , which might cause a further decrease of estimated cluster masses . accordingly , although mdm2 seems quite consistent with observations and provides an excellent fit of data at various mass scales , also mdm1 can not be safely excluded . ps estimates are to be decreased by a suitable ( smaller ) factor , also for cdm , in accordance with previous findings of cen ( 1997 ) . in this work we did not pursue a comparison of simulations with x ray data on cluster abundance ( see , e.g. eke et al . they are known to be in fair agreement with optical data for @xmath233@xmath234 ( see , e.g. , fig . 9a in @xcite ) . at lower masses various problems are to be solved , both concerning the completeness of real cluster samples and the dynamics of gas in simulated clusters . one of our findings is the significant reduction of optical cluster masses , when external layers , connected by hdm bridges , are no longer included in the mass computation . in the analysis of real cluster samples , galaxies belonging to such layers would appear as interlopers and an eye analysis of fig . 7 of @xcite seems to confirm that including ( at least some of ) the interlopers might rise the cluster mass function , approaching ps estimates . this point was not deepened in this analysis , as the natural approach is through the use of mock catalogs , as we plan to do in a forthcoming work . cen ( 1997 ) analysis of the relation between virial mass estimates @xmath207 and real masses @xmath14 is widened here to mixed models . we find that the @xmath235 ratios depend on the cluster mass scale . in general , the relation between cluster virial mass estimates and real cluster masses , is likely to depend on dark matter nature . the hypothesis of isothermal distribution of particles , inside clusters , is also found to yield masses ( slightly ) smaller than real , in pure cdm simulations . in mixed simulations the discrepancy between @xmath207 and cold particle masses is smaller than in pure cdm . on the contrary , taking into account also hdm particles , the gap is substantially greater and shows a significant scale dependence , although the detailed trend depends on the substance of the model . let us finally outline that , at variance with what tends to happen for galactic mass systems , the abundance of early clusters , in mixed models , tends to exceed cdm . in our volume of 360@xmath15mpc , 2 clusters with mass @xmath236 ( with @xmath237 ) were found in the mdm1 simulation . also mdm2 provides a cluster abundance compatible with donahue et al . ( 1998 ) observational findings . such numbers , however , are too small , both on the observational and on the simulation sides , to provide real confirms of models . we thank the consortium cilea for allowing us to run a former trial simulation free of charge and a particular thank is due to giampaolo bottoni of cilea , for his expert technical assistance . elena donghia , who gave us useful suggestions for the parallelization of the pp part of the code , is also to be gratefully thanked . hugh couchman is to be thanked for making public his ap3 m serial code and for some private communications . anatoly klypin is also to be thanked for accurate discussions of several points of this work . we would finally thank stefano borgani for early discussions on the topics of this work . colberg j.m . , white s.d.m . , jenkins a.r . , pearce f.r . , frenk c.s . , thomas p.a . , hutchings r.m . , couchman h.m.p . , peacock j.a . , efsthatiou g.p . , @xmath22 nelson a.h . , 1997 , the virgo consortium : the evolution @xmath22 formation of galaxy clusters , in large scale structure : proc . of the ringberg workshop sept . 1996 , ed . d.hamilton , preprint astro - ph/970286 postman m. , cluster as tracers of the large scale structure , in evolution of large - scale structure : from recombination to garching : proc . mpa / eso cosmology conference , garching , germany , august 1998 , preprint astro ph/9810088 storrie lambardi l.j . , mcmahon r.g . , irwin m.j . , @xmath22 hazard c. , 1995 , high redshift lyman limit @xmath22 damped lyman - alpha absorbers , in : proc . eso workshop on qso a.l . , preprint astro ph/9503089 l|c|c|c + & & & + @xmath52 & 0.26 & 0.14 & - + @xmath238ev & 3.022 & 1.627 & - + @xmath239 & 6.8 & 9 & 6 + @xmath57 & 1.2 & 1.05 & 0.8 + @xmath240k & 12.1 & 13 & 17.4 + @xmath80 & 0.75 & 0.62 & 0.61 + @xmath38 & 0.18 & 0.23 & 0.32 + @xmath75 ( ps ; @xmath241 ) & 14 . & 5.2 & 5.7 + @xmath75 ( sim ) & 10 . & 4.7 & 6.0 + @xmath103 & 1.3 & 1.2 & 1.3 + + +

by making use of the green function concept of quantization of the electromagnetic field in kramers kronig consistent media , a rigorous quantum mechanical derivation of the rate of intermolecular energy transfer in the presence of arbitrarily shaped , dispersing , and absorbing material bodies is given . applications to bulk material , multi - slab planar structures , and microspheres are studied . it is shown that when the two molecules are near a planar interface , then surface - guided waves can strongly affect the energy transfer and essentially modify both the ( frster ) short - range dependence of the transfer rate and the long - range dependence , which are typically observed in free space . in particular , enhancement ( inhibition ) of energy transfer can be accompanied by inhibition ( enhancement ) of donor decay . results for four- and five - layered planar structures are given and compared with experimental results . finally , the energy transfer between two molecules located at diametrically opposite positions outside a microsphere is briefly discussed .

quant-ph0107150

intermolecular energy transfer as a fundamental process in many biochemical and solid - state systems has been of increasing interest @xcite . it is often distinguished between two cases , namely ( radiationless ) short - range transfer ( also called frster transfer @xcite ) and ( radiative ) long - range transfer . in the former the distance @xmath2 between donor and acceptor is small compared with the electronic energy transfer wavelength @xmath3 , . the free - space transfer rate behaves as @xmath0 , which can be explained by the instantaneous ( longitudinal ) coulomb interaction between the two molecules . in the latter the intermolecular distance substantially exceeds the transition wavelength , . the observed @xmath1 dependence of the transfer rate can be regarded as being the result of emission and reabsorption of real ( transverse ) photons . it is worth noting that in a rigorous approach to the problem ( e.g. , within the framework of the multipolar formalism of qed @xcite ) the @xmath0 and @xmath1 distance dependences are limiting cases of a unified theory @xcite . when the two molecules are near material bodies , then the electromagnetic field felt by them can be quite different from that in free space and the intermolecular energy transfer can change accordingly . the effect has attracted attention , because it offers the possibility of controlling the energy transfer , with regard to potential applications , e.g. , in high - efficiency light - harvesting systems , optical networks , and quantum computing . enhanced energy transfer between molecules randomly distributed within a single glycerol droplet ( of about 10@xmath4 m diameter ) @xcite and within a polymer fabry - prot microcavity @xcite has been observed . using monomolecular layers of donor and acceptor molecules ( separated by distances of @xmath5 nm ) in planar microstructures , the dependence of short - range energy transfer on the local photon mode density has been demonstrated @xcite . calculations of the energy transfer rate have been performed in order to include the effect of bulk material @xcite , microspheres @xcite , and planar microcavities @xcite . the quantum theory given in ref . @xcite is based on a microscopic model that allows for both dispersing and absorbing bulk material . in refs . @xcite the classical field generated by a donor dipole and felt by an acceptor dipole in the presence of a microsphere is substituted into the free - space fermi s golden rule expression . a strictly quantum mechanical treatment that starts from a mode decomposition of the electromagnetic field according to the helmholtz equation of the macroscopic maxwell equations is given in refs . unfortunately , the microscopic theory developed for bulk material @xcite becomes quite cumbersome when boundaries are present , and studies based on the standard mode expansion @xcite can not incorporate material absorption . in the present paper we give a rigorous derivation of the rate of intermolecular energy transfer in the presence of arbitrarily shaped , dispersing , and absorbing material bodies , starting from the quantized version of the macroscopic electromagnetic field . the quantization is based on the introduction of langevin noise current and charge densities into the classical maxwell equations , which can then be transferred to quantum theory , with the electromagnetic - field operators being expressed in terms of a continuous set of fundamental bosonic fields via the classical green tensor ( see @xcite and references therein ) . in particular , we show that the minimal - coupling scheme and the multipolar - coupling scheme yield exactly the same form of the rate formula . it is worth noting that the formalism includes material absorption and dispersion in a consistent way , without restriction to a particular frequency domain , and applies to an arbitrary ( inhomogeneous ) medium configuration . here , we apply the theory to bulk material , multi - slab planar structures , and microspheres , with special emphasis on media of drude lorentz type . in particular , we show that the energy transfer can be strongly modified , if the two molecules are sufficiently near an interface and surface - guided waves at the energy transfer wavelength exist . four- and five - layered planar structures are studied , and the results are compared with recent measurements @xcite . finally , the effect of surface - guided waves and whispering - gallery waves in the case of the molecules being near a microsphere is briefly discussed . the paper is organized as follows . in section [ rate ] the basic - theoretical concept of electromagnetic - field quantization is outlined and the energy transfer rate is derived . section [ disc ] is devoted to applications , with special emphasis on multi - slab planar structures , and concluding remarks are made in section [ conl ] . some deepening calculations are given in the appendix . let us consider an ensemble of point charges , interacting with the quantized electromagnetic field in the presence of absorbing media . the minimal - coupling hamiltonian in coulomb gauge reads @xcite @xmath6&&\hspace{4ex } + \sum_\alpha { 1\over 2m_\alpha } \left [ \hat{\bf p}_\alpha - q_\alpha \hat{\bf a}(\hat{\bf r}_\alpha ) \right]^2 % } \nonumber\\[.5ex]&&\hspace{4ex } + \,{\textstyle\frac{1}{2 } } \int\ ! { \rm d}^3{\bf r}\ , \hat{\rho}({\bf r } ) \hat{\phi}({\bf r } ) + \int\ ! { \rm d}^3{\bf r}\ , \hat{\rho}({\bf r } ) \hat{\varphi}({\bf r } ) , \end{aligned}\ ] ] where @xmath7 is the position operator and @xmath8 is the canonical momentum operator of the @xmath9th ( nonrelativistic ) particle of charge @xmath10 and mass @xmath11 . the first term of the hamiltonian is the energy of the medium - assisted electromagnetic field , expressed in terms of bosonic vector fields @xmath12 with commutation relations @xmath13 = \delta_{kk ' } \delta({\bf r}\!-\!{\bf r}')\delta(\omega\!-\!\omega'),\ ] ] @xmath14 = 0\,.\ ] ] the second term is the kinetic energy of the charged particles , and the third term is their coulomb energy , where the corresponding scalar potential @xmath15 is given by @xmath16 with @xmath17 being the charge density of the particles , and @xmath18 the vacuum dielectric permittivity . the last term is the coulomb energy of interaction of the particles with the medium . the scalar potential @xmath19 and the vector potential @xmath20 of the medium - assisted electromagnetic field are given by @xmath21 @xmath22 where @xmath23 with @xmath24 and @xmath25 being the transverse and longitudinal dyadic @xmath26-functions , respectively , and @xmath27 here , @xmath28 is the classical green tensor , which obeys the inhomogeneous , partial differential equation @xmath29 \bbox{g}({\bf r},{\bf r}',\omega ) = -\,\bbox{\delta}({\bf r}-{\bf r}')\ ] ] together with the boundary condition at infinity [ @xmath30 is the dyadic @xmath26-function ] , with @xmath31 @xmath32 @xmath33 @xmath34 @xmath35 being the complex , space- and frequency - dependent permittivity . let us consider the case where the particles are constituents of neutral molecules ( at positions @xmath36 ) that are well separated from each other . the hamiltonian ( [ n001 ] ) can then be decomposed into an unperturbed part @xmath37 and an interaction part @xmath38 as follows @xmath39 @xmath40 @xmath41 here , @xmath42 is the hamiltonian of the @xmath43th molecule , @xmath44 is the coulomb interaction energy between the @xmath43th and the @xmath45th molecule , and @xmath46&&\hspace{4ex } + \sum_{\alpha_m } \left(\frac{q_{\alpha_m}^2}{2m_{\alpha_m}}\right ) \hat{\bf a}^2(\hat{\bf r}_{\alpha_m } ) + \int\ ! { \rm d}^3{\bf r}\ , \hat{\rho}_m({\bf r } ) \hat{\varphi}({\bf r } ) \end{aligned}\ ] ] is the interaction energy between the @xmath43th molecule [ charge density @xmath47 and the medium - assisted electromagnetic field . in what follows we shall restrict our attention to the ( electric-)dipole approximation , so that eq . ( [ n017 ] ) simplifies to @xmath48 where @xmath49 is the dipole operator of the @xmath43th molecule . disregarding the @xmath50 term in eq . ( [ n018 ] ) , which does not give rise to off - diagonal molecular matrix elements , making use of eqs . ( [ n009])([n011 ] ) , and applying the dipole approximation , @xmath51 takes the form of @xmath52 where @xmath53 \mbox{\boldmath $ \delta$}^\perp({\bf r}-{\bf r}_m ) } \nonumber\\&&\hspace{15ex } + \hat{\bf d}_m { } \mbox{\boldmath $ \delta$}^\parallel({\bf r}-{\bf r}_m ) . \end{aligned}\ ] ] let us consider the resonant energy transfer between two molecules @xmath54 and @xmath55 at positions @xmath56 and @xmath57 . the initial ( final ) state @xmath58 ( @xmath59 ) describes the excited molecule @xmath54 ( @xmath55 ) , the molecule @xmath55 ( @xmath54 ) being in the ground state , and the medium - assisted field in vacuum , @xmath60 @xmath61 ( cf . @xcite ) . note that imposing this initial condition requires that the time of state preparation is sufficiently short compared with the time of energy transfer . using the born expansion @xcite up to the second order perturbation theory , the ( total ) rate of energy transfer can be given by @xmath62 where @xmath63 is the occupation probability of the state @xmath58 , and @xmath64 with @xmath65 applying the decomposition ( [ n015 ] ) , we may write @xmath66&&\hspace{4ex } = \langle a , b'| \hat{v}_{ab}|a',b\rangle + \langle a , b'|\hat{\cal t}|a',b\rangle , \end{aligned}\ ] ] where @xmath67 } \nonumber\\[.5ex]&&\hspace{4ex}\times \left[e_i-\hat{h}_0 + is\right]^{-1 } \left[\hat{h}_{a\,\rm int } + \hat{h}_{b\,\rm int}\right]|a',b\rangle . \end{aligned}\ ] ] let us first consider the coulomb term @xmath68 , \end{aligned}\ ] ] where @xmath69 in order to calculate @xmath70 , we make use of eqs . ( [ n019 ] ) and ( [ n020 ] ) , perform the summation and integrations over the possible intermediate states @xmath71 and @xmath72 . after some calculation we derive , on applying eq . ( [ n007 ] ) and the relationship @xcite , @xmath73&&\hspace{2ex } = \int { \rm d}^3{\bf s}\ , \frac{\omega^2}{c^2}\ , \varepsilon_{\rm i}({\bf s},\omega ) g_{km}({\bf r},{\bf s},\omega ) g^\ast_{lm}({\bf r'},{\bf s},\omega ) , \end{aligned}\ ] ] @xmath74&&\hspace{1ex } \times \left\ { \frac{\left[{\bf d}_{b'b } { } \bbox{\delta}_b({\bf r}',-\omega ) { } { \rm i m } \,\bbox{g } ( { \bf r}',{\bf r},\omega ) { } \bbox{\delta}_a({\bf r},-\omega ) { } { \bf d}_{aa'}\right ] } { -\hbar \omega_{a'a}-\hbar \omega + is}\right . \nonumber\\[.5ex]&&\hspace{2ex } \left . + \,\frac{\left[{\bf d}_{b'b } { } \bbox{\delta}_b({\bf r}',\omega ) { } { \rm i m } \,\bbox{g } ( { \bf r}',{\bf r},\omega ) { } \bbox{\delta}_a({\bf r},\omega ) { } { \bf d}_{aa'}\right ] } { \hbar \omega_{a'a}-\hbar \omega + is}\right\ } , \end{aligned}\ ] ] where @xmath75 and @xmath76 [ note that ] . recalling that @xmath77 @xmath32 @xmath78 , we may rewrite eq . ( [ n039 ] ) as @xmath79&&\hspace{2ex } \times \left\ { \frac{\left[{\bf d}_{b'b } { } \bbox{\delta}_b({\bf r}',\omega ) { } { \rm i m } \,\bbox{g } ( { \bf r}',{\bf r},\omega ) { } \bbox{\delta}_a({\bf r},\omega ) { } { \bf d}_{aa'}\right ] } { \hbar \omega_{a'a}-\hbar \omega + is\,\,\mbox{sign}\,(\omega ) } \right\ } \end{aligned}\ ] ] or , equivalently , @xmath80&&\hspace{1ex}\times\ , \frac{1}{2i}\,\left\ { \left [ \frac{\left[{\bf d}_{b'b } { } \bbox{\delta}_b({\bf r}',\omega ) { } \bbox{g } ( { \bf r}',{\bf r},\omega ) { } \bbox{\delta}_a({\bf r},\omega ) { } \bbox{\bf d}_{aa'}\right ] } { \hbar \omega_{a'a}-\hbar \omega + is\,\,\mbox{sign}\,(\omega)}\right]\right . \nonumber\\&&\hspace{2ex } \left . - \left [ \frac{\left[{\bf d}_{bb ' } { } \bbox{\delta}_b({\bf r}',\omega ) { } \bbox{g } ( { \bf r}',{\bf r},\omega ) { } \bbox{\delta}_a({\bf r},\omega ) { } { \bf d}_{a'a}\right ] } { \hbar \omega_{a'a}-\hbar \omega -is\,\ , \mbox{sign}\,(\omega)}\right]^*\right\}. \end{aligned}\ ] ] the @xmath81-integral in eq . ( [ n041 ] ) may now be evaluated by means of contour - integral techniques , by taking into account that the green tensor is a holomorphic function of @xmath81 in the upper complex half - plane , which asymptotically behaves as @xcite @xmath82 we therefore close the path of integration by an infinitely large semicircle in the upper complex half - plane , , and subsequently subtract the semicircle integral . it is easily seen that only the terms in @xmath83 and @xmath84 [ eq . ( [ n029 ] ) ] which are proportional to @xmath81 contribute to the integral over the semicircle , @xmath85.\ ] ] it is further seen that only the first term in the curly bracket contributes to the integral over the closed path . we thus arrive at @xmath86 } \nonumber\\[.5ex]&&\hspace{6ex } -\,\frac{\omega_{a'a}^2}{\varepsilon_0 c^2 } \,\left[{\bf d}_{b'b } { } \bbox{g } ( { \bf r}_b,{\bf r}_a,\omega_{a'a } ) { } { \bf d}_{aa'}\right ] . \end{aligned}\ ] ] substitution of the expressions ( [ n038 ] ) and ( [ n045 ] ) into eq . ( [ n037 ] ) yields the transition amplitude @xmath87 . \end{aligned}\ ] ] note that the first term in eq . ( [ n045 ] ) and the coulomb term ( [ n038 ] ) exactly cancel out . we eventually combine eqs . ( [ n023 ] ) and ( [ n046 ] ) and find that the rate of energy transfer between the chosen states @xmath88 and @xmath89 reads as ( ) @xmath90&&\hspace{4ex}\times\ , \left| { \bf d}_{b'b } { } \bbox{g } ( { \bf r}_b,{\bf r}_a,\omega_{a'a } ) { } { \bf d}_{aa'}\right|^2 \delta(\omega_{a'a}-\omega_{b'b } ) . \end{aligned}\ ] ] it can be proved ( appendix [ multipol ] ) that the use of the multipolar hamiltonian @xcite instead of the minimal - coupling hamiltonian ( [ n001 ] ) exactly leads to the same expression of the energy transfer rate . let us now consider the total energy transfer rate according to eq . ( [ n023a ] ) , by taking into account the vibronic structure of the molecular energy levels . restricting our attention to oppenheimer approximation and neglecting the weak dependence of the electronic transition - dipole matrix element on the nuclear coordinates ( see , e.g. , @xcite ) , we may factorize the dipole transition matrix elements according to @xmath91 where @xmath92 is the purely electronic transition - dipole matrix element of the transition between the lower and the upper electronic state of the molecule @xmath93 , and @xmath94 are the overlap integrals between the vibrational quantum states in the two electronic states of the respective molecule . note that the vibrational overlap integrals take account of both displaced and distorted energy surfaces . combining eqs . ( [ n023a ] ) and ( [ n047 ] ) yields @xmath95&&\hspace{4ex}\times \left| { \bf d}^*_b { } \bbox{g } ( { \bf r}_b,{\bf r}_a,\omega_{a'a } ) { } { \bf d}_a\right|^2 \delta(\omega_{a'a}-\omega_{b'b } ) , \end{aligned}\ ] ] which can be rewritten as @xmath96 where @xmath97 and @xmath98 and @xmath99 respectively , are proportional to the ( single - photon ) emission spectrum of molecule @xmath54 and the ( single - photon ) absorption spectrum of molecule @xmath55 in _ free space _ each @xcite . thus , the rate of energy transfer is proportional to the overlap of the two spectra weighted by the square of the absolute value of the actual green tensor . it is worth mentioning that eqs . ( [ n047])([n047e ] ) apply to the resonant energy transfer between two molecules in the presence of an arbitrary configuration of dispersing and absorbing macroscopic bodies . all the relevant parameters of the bodies are contained in the green tensor . note that the emission ( absorption ) spectrum observed in this case is not proportional to @xmath100 $ ] in general , as it can be seen from a comparison of eq . ( [ n047d ] ) with eq . ( [ b8 ] ) . in particular when the green tensor slowly varies with frequency on a scale given by the ( relevant ) vibrational frequencies of the molecules , then @xmath101 is also a slowly varying function of frequency and can ( approximately ) be taken at the electronic energy transfer frequency and put in front of the integral in eq . ( [ n048 ] ) , thus @xmath102 where @xmath103 in this case , the influence of matter environment on the ( total ) energy transfer rate is fully contained in @xmath104 . clearly , when the two molecules are near a resonator - like equipment , so that the molecule can `` feel '' sharply - peaked field resonances , then @xmath101 can not be assumed to be a slowly varying function of frequency in general ( see section [ sphere ] ) . it may be interesting to compare the rate of energy transfer with the donor decay rate . straightforward generalization of the well - known formula for a two - level transition yields , on applying the born oppenheimer approximation , @xmath105 where @xmath106,\ ] ] and @xmath107 is given by eq . ( [ n047d ] ) . whereas the decay rate is determined by the imaginary part of the green tensor ( taken at equal positions ) , the transfer rate is determined by the full green tensor ( taken at different positions ) . thus , decay rate and transfer rate can quite differently respond to a change of the environment . let us first consider the case when the two molecules are embedded in bulk material of arbitrary complex permittivity @xmath108 . using the well - known expression of the bulk - material green tensor @xmath109 ( see , e.g.,@xcite ) , application of eq . ( [ n047c ] ) yields @xmath110 where @xmath111 } \nonumber\\[.5ex]&&\hspace{1ex}\times \left [ - \left ( { \bf d}^*_b { } { \bf d}_a - 3 \frac{{\bf d}^*_b { } { \bf r}}{r}\ , \frac{{\bf d}_a { } { \bf r}}{r } \right ) \left ( { 1\over q^3(\omega)r^3}- { i\over q^2(\omega)r^2 } \right ) \right . \nonumber\\[.5ex]&&\hspace{7ex } \left . + \left ( { \bf d}^*_b { } { \bf d}_a - \frac{{\bf d}^*_b { } { \bf r}}{r}\ , \frac{{\bf d}_a { } { \bf r}}{r } \right ) { 1\over q(\omega)r } \right ] \end{aligned}\ ] ] with @xmath112 the energy transfer rate is then obtained according to eq . ( [ n048 ] ) . obviously , the green tensor of bulk material can be regarded as being a slowly varying function of frequency , so that the approximation ( [ n047f ] ) applies . from eqs . ( [ n057 ] ) and ( [ n051 ] ) it is seen that the energy transfer rate includes both the small - distance case ( frster transfer ) , with the rate being proportional to @xmath0 , and the large - distance ( radiative ) case , where the rate becomes proportional to @xmath1 . note that the exponential @xmath113|^2 $ ] @xmath32 @xmath114 $ ] , which typically arises from material absorption , drastically diminishes the large - distance energy transfer [ @xmath115 @xmath32 @xmath116 @xmath32 @xmath117 . in eq . ( [ n057 ] ) local - field corrections are ignored . they may be taken into account by applying , e.g. , the scheme used in ref . @xcite for correcting the rate of spontaneous decay . it is worth noting that the above given result , which is based on the quantization of the macroscopic maxwell field for given complex permittivity , exactly corresponds to the result obtained in ref . @xcite within the framework of a fully microscopic approach on the basis of some model medium coupled to the radiation field and a heat bath . already from the study of the spontaneous decay of an excited atom near an interface @xcite it is clear that in the case of inhomogeneous media ( of complicated atomic structure ) a microscopic approach would be rather involved and closed solutions would hardly be found . let us consider a planar multi - slab structure and assume that the two molecules are in the same slab . the relevant green tensor ( for the energy transfer between the two molecules relevant ) of an inhomogeneous system of this type can always be written in the form of @xmath118 where @xmath119 is the green tensor according to eq . ( [ n051 ] ) , with @xmath108 being the permittivity of the slab in which the two molecules are located , and the reflection term @xmath120 insures the correct boundary conditions at the surfaces of discontinuity . clearly , a decomposition of the type of eq . ( [ n059 ] ) is also valid for other than planar systems , provided that the two molecules are located in a region of space - independent permittivity . to be more specific , let the @xmath121-direction be the direction of variation of the permittivity of the multi - slab system and assume that @xmath56 and @xmath57 are in the @xmath122th slab of thickness @xmath123 ( fig . [ mlayer ] ) . the reflection term in eq . ( [ n059 ] ) can then be given by @xcite ( see also ref . @xcite ) @xmath124&&\hspace{4ex } = { i \over 4\pi } \int_0^\infty { { \rm d } k_\| k_\| \over 2\beta_j } \,e^{i\beta_j d_j } \tilde{\bbox{g}}{^{\rm refl}}({\bf r}_b,{\bf r}_a,\omega , k_\|)\end{aligned}\ ] ] [ @xmath125 @xmath32 @xmath126 ; @xmath127 @xmath32 @xmath128 . choosing the coordinate system such that @xmath129 @xmath32 @xmath130 , the nonvanishing components of @xmath131 read @xmath132 \nonumber\\[.5ex]&&\hspace{11ex } + \,c^s_+\left[j_0(k_\|r_x ) + \!(-)\ , j_2(k_\|r_x)\right ] , \\[.5ex ] \label{ml3 } & & \tilde{g}^{\rm refl}_{xz(zx ) } = -(+)\ , 2i \ , { \beta_jk_\|\over k_j^2 } \ , s^p_{+(- ) } j_1(k_\|r_x ) , \\[.5ex ] \label{ml3a } & & \tilde{g}^{\rm refl}_{zz } = 2 \ , { k_\|^2\over k_j^2 } \,c^p_+ j_0(k_\|r_x)\end{aligned}\ ] ] [ @xmath133 - bessel function ] , where @xmath134&&\hspace{10ex } + \,(-)\ , 2 r^q_+ r^q_- \cos(\beta_j r_z ) e^{i\beta_j d_j } \bigr ] d_q^{-1 } , \\[.5ex ] \label{ml5 } & & s^q_{+(- ) } = \bigl [ r^q_-e^{i\beta_j(z_a+z_b - d_j ) } - r^q_+e^{-i\beta_j(z_a+z_b - d_j ) } \nonumber\\[.5ex]&&\hspace{10ex } + \,(-)\ , 2i r^q_+ r^q_- \sin(\beta_j r_z ) e^{i\beta_j d_j } \bigr ] d_q^{-1 } , \\[.5ex ] \label{ml6 } & & d_q = 1-r^q_+ r^q_- e^{2i\beta_j d_j}. % , \qquad q = p , s\ , \end{aligned}\ ] ] here , @xmath135 @xmath32 @xmath136 means tm(te ) polarized waves , and @xmath137 are the total reflection coefficients at the upper ( lower ) stack of slabs [ ] of the waves in the @xmath122th slab ( for details , see ref . @xcite ) . note that when @xmath56 and @xmath57 are in the top ( bottom ) slab , then eqs . ( [ ml1])([ml6 ] ) ( formally ) apply provided that and are set . if the frequencies of the vibronic transitions that are involved in the energy transfer are sufficiently far from a medium resonance , so that material absorption ( in the @xmath122th slab ) may be disregarded , then the permittivity may be considered as being real and positive . in this case , it may be useful to decompose the integral in eq . ( [ ml1 ] ) into two parts , @xmath138 @xmath139&&\hspace{2ex } = { i \over 4\pi } \int_0^{\sqrt{\varepsilon_j}\omega / c } { { \rm d } k_\| k_\| \over 2\beta_j } \,e^{i|\beta_j| d_j } \tilde{\bbox{g}}^{\rm refl}({\bf r}_b,{\bf r}_a,\omega , k_\|),\end{aligned}\ ] ] @xmath140&&\hspace{2ex } = { i \over 4\pi } \int_{\sqrt{\varepsilon_j}\omega / c}^\infty { { \rm d } k_\| k_\| \over 2\beta_j } \,e^{-|\beta_j| d_j } \tilde{\bbox{g}}^{\rm refl}({\bf r}_b,{\bf r}_a,\omega , k_\|).\end{aligned}\ ] ] obviously , @xmath141 results from waves that have a propagating component in the @xmath121-direction , whereas the waves that contribute to @xmath142 are purely evanescent in the @xmath121-direction . let the two molecules be embedded in a half - space medium ( medium @xmath143 ) and assume that in the relevant frequency interval the permittivity of the medium @xmath144 can be regarded as being real and positive . when the molecules are near the interface between the two half - space media such that , it can be proved that eqs . ( [ ml1])([ml3a ] ) reduce to ( ) @xmath145&&\hspace{4ex } \simeq \,{1 \over 4\pi k_1 ^ 2}\ , { \varepsilon_2\!-\!\varepsilon_1 \over \varepsilon_2 \!+\!\varepsilon_1}\ , { ( z_a\!+\!z_b)^2 -\!(+)\,2r_x^2 \over [ ( z_a\!+\!z_b)^2+r_x^2]^{5/2}}\,,\end{aligned}\ ] ] @xmath146&&\hspace{4ex } \simeq\ , + ( -)\,{1 \over 4\pi k_1 ^ 2}\ , { \varepsilon_2\!-\!\varepsilon_1 \over \varepsilon_2 \!+\!\varepsilon_1}\ , { 3(z_a\!+\!z_b)r_x \over [ ( z_a\!+\!z_b)^2+r_x^2]^{5/2}}\,,\end{aligned}\ ] ] @xmath147&&\hspace{15ex}\times \left\ { { 2(z_a\!+\!z_b)^2-r_x^2 \over k_1 ^ 2[(z_a\!+\!z_b)^2+r_x^2]^2 } \!+ \!1 \right\}\end{aligned}\ ] ] [ @xmath148 , complex permittivity of medium @xmath149 . note that for @xmath56 @xmath32 @xmath150 , eqs . ( [ int6])([int5 ] ) just give the green tensor whose imaginary part determines the influence of the interface on the rate of spontaneous decay of a single molecule @xcite . ( for some special cases , see also ref . @xcite . ) under the assumptions made , the main contribution to @xmath151 comes from @xmath142 . hence surface - guided waves ( including decaying waves ) play an important role and can noticeably influence the resonant energy transfer . in particular when mediu is a metal or a dielectric with , ( and typically @xmath152 @xmath153 @xmath154 ) , then a strong effect is observed for , which is nothing but the condition for best excitation of surface - guided waves @xcite . in the numerical calculation of @xmath101 [ eq . ( [ n047c ] ) ] , which contains the relevant information about the influence of the interface on the rate of energy transfer [ see eqs . ( [ n048])([n047e ] ) ] , we have assumed that the two molecules are situated in vacuum [ ] above a half - space medium of drude lorentz type and restricted our attention to a single - resonance medium , @xmath155 here , @xmath156 corresponds to the coupling constant , and @xmath157 and @xmath158 are respectively the medium oscillation frequency and the linewidth . recall that the drude lorentz model covers both metallic ( ) and dielectric ( ) matter and features a band gap between @xmath157 and . we have performed the calculations using the exact green tensor [ eqs . ( [ n059])([ml6 ] ) ] . comparing the results with those obtained by using the approximately valid green tensor [ eq . ( [ n059 ] ) together with eqs . ( [ int6])([int5 ] ) ] , we have found good agreement . the behavior of @xmath101 is illustrated in fig . [ into ] . it is seen that outside the band gap where the modification of @xmath101 due to the presence of the interface is small even for small distances of the molecules from the interface . since in this frequency domain @xmath101 may be regarded as being slowly varying on a frequency scale defined by the vibrational frequencies of the molecules , eq . ( [ n047f ] ) applies . thus , the energy transfer rate is simply proportional to @xmath104 . inside the band gap , however , the interface can significantly affect @xmath101 if , according to eqs . ( [ int6])([int5 ] ) , ( in fig . [ into ] ) , that is to say , if the energy transfer transition under consideration is tuned to a surface - guided wave . note that a negative real part of the medium permittivity can easily be realized by metals . careful inspection of the contributions @xmath159 and @xmath151 to @xmath160 reveals that the enhancement of @xmath101 results from @xmath151 , whereas the reduction reflects some destructive interference of @xmath159 and @xmath151 . another interesting feature is that the reduction of @xmath101 can go hand in hand with an enhancement of the corresponding quantity @xmath161 [ eq . ( [ n067a ] ) ] for the donor decay rate @xmath162 [ eq . ( [ n067 ] ) ] ( see the inset in fig . [ into ] ) . further , fig . [ into ] reveals that with increasing material absorption ( i.e. , with increasing value of @xmath158 ) @xmath101 varies less rapidly inside the band - gap region , and enhancement and reduction are thus less pronounced . clearly , the strong influence on @xmath101 of the interface which is observed for small material absorption must not necessarily lead to a correspondingly strong change of the energy transfer rate , because of the integration in eq . ( [ n048 ] ) . nevertheless , the results show the possibility of controlling the resonant energy transfer by surface - guided waves . figure [ intr ] illustrates the dependence of @xmath101 on the intermolecular distance for the case when @xmath81 corresponds to a surface - guided wave frequency and a noticeable change of @xmath101 is observed ( in the figure ) . it is seen that the @xmath163 dependence , which is typical of the frster transfer in free space , is observed for much shorter intermolecular distances . the relative minima of @xmath101 below the free - space level , which are observed for somewhat larger intermolecular distances , again result from destructive interference between @xmath159 and @xmath151 . eventually , the large - distance reduction of @xmath101 below the free - space level results from material absorption . as already mentioned , the behavior of @xmath101 in fig . [ intr ] is dominated by surface - guided waves that decay exponentially along the @xmath164-directions . with increasing material absorption the penetration depths decrease , so that on average @xmath101 becomes closer to the free - space level . the possibility of controlling the ultrashort - range energy transfer by varying the distance of the molecule from the surface is illustrated in the inset . in fig . [ intz ] the dependence of @xmath101 ( again for ) on the molecule surface distance is plotted , and the contributions to @xmath101 from ordinary waves having a propagating component in @xmath121-direction ( @xmath141 ) and surface - guided waves ( @xmath142 ) are shown . it is clearly seen that when the two molecules are very near the surface , then energy transfer between them is mediated by surface - guided waves , whereas for larger distances ordinary waves play the dominant role . note that the oscillatory behavior is typical of the latter case . clearly , for very large distances ( @xmath165 @xmath166 @xmath167 ) the free - space behavior is observed . recently , experiments have been carried out to study the transfer of excitation energy between dye molecules confined within planar optical microcavities @xcite . in the experiments , donors ( eu@xmath168 complex ) and acceptors ( 1,1-dioctadecyl-3,3,3,3-tetramethylindodicarbocyanine ) embedded within a transparent material ( 22-tricosenoic acid ) bounded by no ( weak - cavity structure ) , one ( half - cavity structure ) , or two ( full - cavity structure ) silver mirrors are considered . to compare the experimental results with the theoretical ones , we have modeled the half - cavity structure by a planar four - layered system and the full - cavity structure by a five - layered system . the former consists of vacuum , dielectric matter ( 22-tricosenoic acid , @xmath169 @xcite , thickness @xmath170 ) , metal ( silver , @xcite , thickness @xmath171 nm ) , and vacuum , and the latter consists of vacuum , metal ( silver , thickness @xmath172 nm ) , dielectric matter ( the same as above , thickness @xmath170 ) , metal ( silver , thickness @xmath171 nm ) , and vacuum . in each system , the donor is situated in the middle of the dielectric layer , while the position of the acceptor is shifted towards the silver mirror of @xmath171 nm thickness . the green tensors of the two systems can be calculated according to eqs ( [ n059])([ml6 ] ) . assigning to silver a drude lorentz - type permittivity @xcite , it can be proven that in the relevant frequency interval ( of overlapping donor emission and acceptor absorption spectra ) @xmath101 [ eq . ( [ n047c ] ) ] and @xmath161 [ eq . ( [ n067a ] ) ] sufficiently slowly vary with @xmath81 , so that [ cf . ( [ n047f ] ) ] and , similarly , . thus , @xmath104 and @xmath173 can be viewed as measures of the energy transfer rate and the donor decay rate , respectively . figure [ barnes1 ] shows the dependence on @xmath170 of @xmath173 and @xmath104 ( averaged over the dipole orientations ) . from fig . [ barnes1](a ) it is seen that at ( i.e. , for ) the ratio of the donor decay rates for the five- and four - layered systems is , which ( within the measurement accuracy ) is in sufficiently good agreement with experimental result ( see fig . 2d in ref . note that in the vicinity of the ratio of the two rates sensitively responds to a change of @xmath174 . comparing @xmath173 [ fig . [ barnes1](a ) ] with @xmath104 [ fig . [ barnes1](b ) ] , we see that for the four - layered system and @xmath174 @xmath175 @xmath176 ( i.e. , for ) both @xmath173 and @xmath104 decrease with increasing @xmath170 and an approximately valid linear relation between the energy transfer rate and the donor decay rate can be established in agreement with experimental results in ref . @xcite . from the data reported in ref . @xcite it could be expected that the linear relation between the two rates is generally valid . this is of course not the case . since the energy transfer rate is determined by the full ( two - point ) green tensor , whereas the donor decay rate is only determined by the imaginary part of the ( one - point ) green tensor , the two rates can behave quite differently , as it is demonstrated in fig . [ barnes1 ] . in particular , the increase of the donor decay rate at the cavity resonances can be accompanied with a decrease of the energy transfer rate , because of destructive interferences . in the experiments in ref . @xcite , the measurements are performed on an ensemble of donors and acceptors whose distance is fixed in the @xmath121-direction but variable in the @xmath177-direction . the question thus arises of whether the measured data refer to a single nearest - neighboring donor acceptor pair or not . in fig . [ barnes2 ] we have plotted the dependence on @xmath170 of @xmath104 ( averaged over the dipole orientations ) for the five - layered system and various values of @xmath178 , with @xmath179 being fixed . we see that the rates of energy transfer between molecules whose distances are larger than that of nearest - neighboring molecules can be quite comparable with those of the latter . moreover there are also cases where the energy transfer rate increases with the donor acceptor distance . the experimentally determined energy transfer rates are thus averaged rates , which not necessarily show the characteristic features of single - pair transfer rates . averaging in fig . [ barnes2 ] @xmath104 over all values of @xmath178 , the resulting curve is expected to be substantially flatter than the solid - line curve , particularly when @xmath170 sweeps through @xmath3 . an analysis of the contributions of @xmath141 [ eq . ( [ ml8 ] ) ] and @xmath142 [ eq . ( [ ml9 ] ) ] to @xmath151 [ eq . ( [ ml7 ] ) ] reveals that for cavity lengths of ( i.e. , for ) evanescent waves dominate the influence of the cavity system on both the rate of intermolecular energy transfer and the donor decay rate and lead to a strong increase of them . whereas for cavities lengths of ( i.e. , for ) evanescent waves only weakly affect the donor decay rate , they can strongly affect the intermolecular energy transfer up to cavity lengths of a few micrometers . note that the resonance lengths seen in fig . [ barnes1 ] originate from propagating waves . microspheres have been of increasing interest , because of the whispering - gallery ( wg ) and surface - guided ( sg ) waves , which may be employed , e.g. , for reducing the thresholds of nonlinear optical processes @xcite . intermolecular energy transfer in the presence of microspheres has been considered for molecules near a small metallic spheroid ( spheroid s linear extension @xmath180 ) in the nonretardation limit , for molecules embedded within a dielectric microsphere @xcite , and for the case where one molecule is inside a dielectric microsphere and the other outside it @xcite . here we restrict our attention to the influence of wg and sg waves on the energy transfer between two molecules outside a microsphere , taking fully into account retardation effects . let @xmath144 and @xmath148 be respectively the permittivities outside and inside the sphere . if the transition dipole moments are parallel to each other and tangentially oriented with respect to the sphere , the relevant ( spherical - coordinate ) components of @xmath151 are ( , ) @xmath181&&\hspace{2ex } = { ik_1 \over 4\pi } \sum_{l=1}^\infty { ( 2l+1)\over l(l+1 ) } \biggl\ { { \cal b}_l^m h_l^{(1)}(k_1r_a ) h_l^{(1)}(k_1r_b ) \nonumber\\[.5ex]&&\hspace{4ex}\times\ , \bigl [ l(l+1)p_l(\cos\theta_b ) - \cos\theta_b p'_l(\cos\theta_b ) \bigr ] \nonumber\\[.5ex]&&\hspace{8ex } + \ , { \cal b}_l^n { [ k_1r_a\,h_l^{(1)}(k_1r_a)]'\over k_1r_a } \nonumber\\[.5ex]&&\hspace{10ex}\times\ , { [ k_1r_b\,h_l^{(1)}(k_1r_b)]'\over k_1r_b } p_l'(\cos\theta_b ) \biggr\ } \end{aligned}\ ] ] ( for the green tensor of a sphere , see , e.g. , @xcite ) , and for radially oriented dipoles the relevant components are ( , ) @xmath182&&\hspace{4ex}\times\ , { \cal b}_l^n h_l^{(1)}(k_1r_a ) h_l^{(1)}(k_1r_b ) p_l'(\cos\theta_b)\,,\end{aligned}\ ] ] where @xmath183&&\hspace{4ex } = - \frac { \bigl [ a_2j_l(a_2)\bigr ] ' j_l(a_1 ) - \bigl [ a_1j_l(a_1)\bigr ] ' j_l(a_2 ) } { \bigl [ a_2j_l(a_2)\bigr ] ' h_l^{(1)}(a_1 ) - j_l(a_2 ) \bigl[a_1 h_l^{(1)}(a_1)\bigr ] ' } \,,\end{aligned}\ ] ] @xmath184&&\hspace{0ex } = - \frac { \varepsilon_1(\omega ) j_l(a_2 ) \bigl[a_1 j_l(a_1)\bigr ] ' - \varepsilon_2(\omega)j_l(a_1 ) \bigl[a_2 j_l(a_2)\bigr ] ' } { \varepsilon_1(\omega ) j_l(a_2 ) \bigl [ a_1 h_l^{(1)}(a_1)\bigr ] ' -\varepsilon_2(\omega ) \bigl[a_2 j_l(a_2)\bigr ] ' h_l^{(1)}(a_1 ) } \nonumber\\ & & \end{aligned}\ ] ] [ ; @xmath185 , microsphere radius ; @xmath186 , spherical bessel function ; @xmath187 , spherical hankel function ; @xmath188 , associated legendre function ] . in fig . [ sph ] the dependence on frequency of @xmath101 is illustrated for the case where vacuum is outside the sphere and the two molecules are placed at diametrically opposite positions , with the transition dipole moments being radially oriented . it is clearly seen that the energy transfer can greatly be facilitated at the positions of the sphere - assisted field resonances , the enhancement of @xmath101 at the positions of sg resonances ( inside the band gap ) being larger than those at the positions of wg resonances ( outside the band gap ) . maximum values of @xmath101 are observed where the sg resonances overlap . the energy transfer rate for tangentially oriented dipoles ( not shown ) is in general smaller than that for radially oriented dipoles . note that when @xmath101 is sharply peaked at the sphere - assisted field resonances , such that it is not slowly varying in the frequency interval where the ( free - space ) donor emission and acceptor absorption spectra overlap , then it can not be taken at the electronic energy transfer frequency and put in front of the integral in eq . ( [ n048 ] ) . in this case , the change of the energy transfer rate will be less pronounced than it might be expected from the frequency response of the electronic part , because of the frequency integration . we have given a rigorous , strictly quantum mechanical derivation of the rate of intermolecular energy transfer in the presence of dispersing and absorbing material bodies of arbitrary shapes , showing that both the minimal - coupling scheme and the multipolar coupling scheme lead to rate formulas of exactly the same form . the dependence on the material bodies of the energy transfer rate is fully expressed in terms of the green tensor of the macroscopic maxwell equations for the medium - assisted electromagnetic field . in the macroscopic approach , the dispersing and absorbing material bodies are described , from the very beginning , in terms of a spatially varying permittivity , which is a complex function of frequency . the macroscopic approach has similar to classical optics the benefit of being universally valid , without the need of involved _ ab initio _ microscopic calculations . in so far as such calculations for simple model systems have been performed , the results agree with those obtained from the microscopic approach . clearly , macroscopic electrodynamics is valid only to some approximately fixed length scale which exceeds the average interatomic distance in the material bodies . whereas the donor spontaneous decay rate is determined by the imaginary part of the green tensor in the coincidence limit , the donor acceptor energy transfer rate depends on the full two - point green tensor . hence , the decay rate and the energy transfer rate can be affected by the presence of material bodies quite differently . our calculations for planar multilayer structures have shown that enhancement ( inhibition ) of spontaneous decay and inhibition ( enhancement ) of energy transfer can appear simultaneously . they have further shown that surface - guided waves can strongly affect the energy transfer , thus being very suitable for controlling it . in free space it is often distinguished between two limiting cases , namely the short - distance nonradiative ( frster ) energy transfer and the long - distance radiative energy transfer . the former is characterized by the @xmath0 distance dependence of the transfer rate , and the latter by the @xmath1 dependence . in particular , in the short - distance limit the energy transfer rate rapidly decreases with increasing distance between the molecules . this must not necessarily be the case in the presence of material bodies , because of the possibly drastic change of the dependence on the distance of the energy transfer rate . so , our calculations for planar multilayer structures have shown that the energy transfer rate can also increase with the distance . we thank s. scheel and a. tip for discussions . is grateful to the alexander von humboldt stiftung for financial support . this work was supported by the deutsche forschungsgemeinschaft . the multipolar - coupling hamiltonian can be obtained from the minimal - coupling hamiltonian by means of the power zienau transformation @xcite , @xmath189 where @xmath190 \,,\end{aligned}\ ] ] with @xmath191&&\hspace{10ex } \times \int_0 ^ 1 { \rm d}\lambda \ , \delta\!\left[{\bf r}\!-\!{\bf r}_m\!-\!\lambda \left(\hat{\bf r}_{\alpha_m } - { \bf r}_m\right)\right]\end{aligned}\ ] ] being the polarization associated with the @xmath43th molecule . using @xmath192 from eq . ( [ n001 ] ) , we derive ( see , for details , @xcite ) @xmath193&&\hspace{2ex } + \sum_m \sum_{\alpha_m}\frac{1}{2m_{\alpha_m } } \bigg\ { \hat{{\bf p}}_{\alpha_m } \nonumber\\[.5ex]&&\hspace{3ex } + q_{\alpha_m}\int_0 ^ 1\!{\rm d}\lambda\,\lambda \left ( \hat{\bf r}_{\alpha_m}\!-\!{\bf r}_m \right ) \times \hat{\bf b}\left [ { \bf r}_m\!+\!\lambda \left ( \hat{\bf r}_{\alpha_m } \!-\!{\bf r}_m \right ) \right ] \bigg\}^2 \nonumber\\[.5ex]&&\hspace{2ex } + \,\sum_{m } \int\ ! { \rm d}^3{\bf r } \ , \left[\frac{1}{2\varepsilon_0 } \hat{\bf p}_m({\bf r } ) \hat{\bf p}_{m}({\bf r } ) \right ] \nonumber\\[.5ex]&&\hspace{2ex } -\,\sum_m \int\ ! { \rm d}^3{\bf r}\ , \left[\hat{{\bf p}}_m({\bf r } ) \hat{\bf e}({\bf r } ) \right]\ , , \end{aligned}\ ] ] where [ with @xmath20 from eq . ( [ n010 ] ) ] , and @xmath194 and neutral molecules with non - overlapping charge distributions are again assumed . note that in the multipolar - coupling scheme the operator of the electric field strength is defined according to @xmath195 - \bbox{\nabla } \hat{\varphi}({\bf r } ) - \bbox{\nabla } \hat{\phi}({\bf r})\ , , \end{aligned}\ ] ] which implies the following relation between @xmath196 and @xmath197 : @xmath198 hence , @xmath199 has the meaning of the displacement field with respect to the molecular polarization . from eq . ( [ n062 ] ) it is seen that the molecules now interact only via the medium - assisted electromagnetic field . in particular , in the ( electric-)dipole approximation eq . ( [ n062 ] ) simplifies to @xmath200 where @xmath201 with @xmath202 is the unperturbed hamiltonian of the medium - assisted electromagnetic field and the molecules , and @xmath203 is the interaction energy between them . comparing the multipolar - coupling energy given by eq . ( [ n066 ] ) with the minimal - coupling energy @xmath204 given by eq . ( [ n015 ] ) together with eqs . ( [ n017a])([n020 ] ) , we see that the two energies ( formally ) become equal to each other , if we remove in the latter the coulomb term and replace @xmath205/\hbar\omega$ ] with @xmath206 . having these changes in mind , we now follow step by step the derivation of eq . ( [ n047 ] ) in section [ two molecules ] . starting from the corresponding eigenstates of the unperturbed multipolar - coupling hamiltonian ( instead of those of the unperturbed minimal - coupling hamiltonian ) , it is not difficult to see that the result is again eq . ( [ n047 ] ) . it should be pointed out that the above mentioned difference between @xmath196 and @xmath197 [ eq . ( [ n062b ] ) ] does not affect the energy transfer rate . in the electric - dipole approximation and the rotating - wave approximation , the hamiltonian for a single molecule ( at position @xmath56 ) that ( with regard to the vibronic transitions @xmath207 @xmath208 @xmath209 ) resonantly interacts with the medium - assisted electromagnetic field reads , by appropriately specifying eqs . ( [ n014a])([n020 ] ) , @xcite @xmath210 @xmath211&&\hspace{4ex } + \sum_a \hbar\omega_a |a\rangle\langle a| + \sum_{a ' } \hbar\omega_{a ' } |a'\rangle\langle a'|,\end{aligned}\ ] ] @xmath212,\ ] ] where @xmath213 is the positive - frequency part of @xmath196 defined by eq . ( [ n006 ] ) , and the vibronic transition - dipole matrix elements @xmath214 of the vibronic transitions are given , in the born oppenheimer approximation , by eq . ( [ n034 ] ) . let us assume that the molecule is initially ( at time @xmath215 @xmath32 @xmath130 ) prepared in a statistical mixture of vibrational states in the upper electronic state and the medium - assisted electromagnetic field is in the vacuum state , i.e. , @xmath216 the time - dependent spectrum of light observed at position @xmath217 ( in free space ) by means of a spectral apparatus of sufficiently small passband width can be given by ( see , e.g. , @xcite ) @xmath218 , \end{aligned}\ ] ] where @xmath219 and @xmath220 are respectively the setting frequency and the operating time of the spectral apparatus . in order to calculate the electric - field correlation function associated with the light emitted by the molecule during the spontaneous decay of the upper electronic state , we may restrict our attention to the perturbative expansion of the time evolution operator up to the first order in @xmath204 @xcite , @xmath221 we make use of eqs . ( [ b1b ] ) , ( [ n006 ] ) [ together with eq . ( [ n007 ] ) ] , ( [ b6 ] ) , and ( [ b7 ] ) , apply eq . ( [ b5 ] ) , and derive after some calculation , on recalling the relation ( [ n030 ] ) , ( see also @xcite ) @xmath222&&\hspace{2ex } = 2\pi \sum_{a , a ' } p_{a ' } |v_{a'a}|^2 |{\bf f}({\bf r},{\bf r}_a , \omega_{a'a})|^2 \delta(\omega_s - \omega_{a'a } ) , \end{aligned}\ ] ] where @xmath223&&\hspace{2ex } = { 1 \over \pi\varepsilon_0 } \int_0^\infty { \rm d}\omega \ , { \omega^2 \over c^2}\ , { \rm im}\ , \bbox{g } ( { \bf r},{\bf r}_a,\omega ) { } { \bf d}_a \zeta(\omega_{a'a}\!-\!\omega ) \nonumber\\[.5ex]&&\hspace{15ex } \simeq\ , - { i \omega_{a'a}^2 \over \varepsilon_0 c^2}\ , \bbox{g}({\bf r},{\bf r}_a,\omega_{a'a}){\bf d}_a \end{aligned}\ ] ] [ @xmath224 @xmath32 @xmath225 @xmath34 @xmath226 ; @xmath227 , principal value ] . in the derivation of eq . ( [ b8 ] ) , retardation has been ignored and the relation @xmath228&&\hspace{4ex } = \lim _ { t\rightarrow\infty } { \sin^2(\omega t/2 ) \over t(\omega /2)^2 } = 2\pi\delta(\omega ) \end{aligned}\ ] ] has been used . j. s. avery , proc . london * 88 * , 1 ( 1966 ) ; l. gomberoff and e. a. power , _ ibid . _ * 88 * , 281 ( 1966 ) ; e. a. power and t. thirunamachandran , phys . a * 28 * , 2671 ( 1983 ) ; d. l. andrews and b. s. sherborne , j. chem . phys . * 86 * , 4011 ( 1987 ) ; d. l. andrews , chem . phys . * 135 * , 195 ( 1989 ) ; d. l. andrews and g. juzelinas , j. chem . phys . * 96 * , 6606 ( 1992 ) ; d. p. craig and t. thirunamachandran , chem . phys . * 167 * , 229 ( 1992 ) . ho trung dung , l. knll , and d .- welsch , phys . a * 57 * , 3931 ( 1998 ) ; s. scheel , l. knll , and d .- welsch , _ ibid . _ * 58 * , 700 ( 1998 ) ; a. tip , l. knll , s. scheel , and d .- welsch , _ ibid . _ * 63 * , 043806 ( 2001 ) .

in this paper , we study optimization of the first eigenvalue of in a bounded domain under several constraints for the function . we consider this problem in various boundary conditions and various topologies of domains . as a result , we numerically observe several common criteria for for optimizing eigenvalues in terms of corresponding eigenfunctions , which are independent of topology of domains and boundary conditions . geometric characterizations of optimizers are also numerically observed . * keywords : * eigenvalue problem , topology optimization , dependence of optimizers on topology . * ams subject classifications :* 35q93 , 49j20 , 65n25 , 74g15

1408.5370

in this paper , we consider the following eigenvalue problem @xmath3 in a bounded domain @xmath4 in @xmath5 with suitable boundary conditions , where @xmath6 and @xmath7 denotes the lebesgue measure of @xmath4 in @xmath5 . let @xmath8 , then ( [ k ] ) immediately implies @xmath9 under these settings , we consider the following problem : [ prob - rho - abst ] find @xmath10 which attains the supremum or the infimum of the first eigenvalue @xmath11 on @xmath12 under suitable boundary condition . if it exists , characterize the shape of the domain @xmath13 . we shall call the smallest positive eigenvalue the first eigenvalue , since we consider this problem under dirichlet , neumann or mixed boundary condition . our motivation for the above problem is as follows . assume that two different materials are in a given domain @xmath4 with fixed volume ratio . how do we arrange such materials to optimize the heat conductivity of @xmath4 ? since the long - time behaviour of heat transfer of the heat equation is controlled by the first eigenvalue of @xmath14 , we may consider problem [ prob - rho - abst ] as one of toy models of this problem . in this paper , we numerically study the eigenvalue optimization for ( [ ev - prob - rho ] ) and we observe the following results for problem [ prob - rho - abst ] : 1 . the element @xmath10 optimizing @xmath11 can be characterized by inequalities with respect to @xmath15 , where @xmath16 is the eigenfunction associated with @xmath17 of ( [ ev - prob - rho ] ) . as a consequence , the domain @xmath13 is given by the super- or the sub - level set of @xmath18 . this characterization is independent of topology and geometry of @xmath4 and boundary conditions on @xmath19 . 2 . if @xmath4 is star - shaped and symmetric in a certain direction , then @xmath20 has the same symmetry . optimized region @xmath20 depends continuously on a parameter of the boundary condition if the robin boundary condition is imposed . the precise statements are described in section [ section - numerical ] ( observation [ observation - criteria - heat ] - [ observation - dependence - heat ] ) with various numerical results . this paper is organized as follows . in section [ section - setting ] , we provide more precise setting of our problems . a numerical method for finding optimizers we apply here , _ the level set approach _ , is also derived here . in section [ section - prior ] , numerical and mathematical known results for a well - considered problem are discussed . section [ section - numerical ] is where our main discussion is developed . we show several numerical observations about eigenvalue optimization criteria , geometry of the level set @xmath13 for the optimizer @xmath21 and continuous dependence of @xmath20 on boundary conditions . here we provide the precise setting of problem [ prob - rho - abst ] . [ prob - rho ] for @xmath22 define @xmath23 find an element @xmath10 which attains @xmath24 if it exists , characterize @xmath21 and the shape of domain @xmath13 . we also consider the same question for @xmath25 . it is mathematically known that the first eigenvalue of the linear operator @xmath26 is characterized by the rayleigh quotient ( [ ev - rho ] ) ( see e.g. @xcite ) . here , we also consider the minimization of @xmath11 as a comparison with the maximization of it . we consider the problem in the cases of various type of domains like ones with piecewise smooth boundary , non - convex or non - simply connected ones , as well as various boundary conditions ( dirichlet , neumann or mixed boundary condition ) . a well - known mathematical theory ( see e.g. @xcite ) tells us that the linear operator @xmath27 associated with ( [ ev - prob - rho ] ) possesses discrete eigenvalues @xmath28 in the case of the homogeneous dirichlet boundary value problem , @xmath29 is not an eigenvalue and hence the infimum in ( [ ev - rho ] ) is attained in @xmath30 being not identically zero . thus , in the case of dirichlet boundary value problems , we call @xmath30 admissible if and only if @xmath31 in an appropriate function space . on the other hand , in the case of the homogeneous neumann boundary value problem , @xmath29 is admitted as an eigenvalue with a constant function as the eigenfunction . @xmath29 is thus the smallest eigenvalue . according to the general theory of eigenfunctions , eigenfunctions associated with @xmath11 have to be orthogonal to constant functions in the inner product on @xmath32 . we thus call @xmath30 admissible , if and only if @xmath30 satisfies @xmath33 . our problem is one of typical problems called _ topology optimization_. one of well - known methods for determining topology of the optimal object is the level set approach , originally developed in osher and sethian @xcite . this method provides an efficient way of describing time - evolving curves and surfaces which may undergo topological changes . osher and santosa @xcite improve the method so that it can be applied to optimization problems with one or more constraints like volume constraint . we follow their method as a numerical approach for determining optimizers in our problems . here we briefly review implementations of the level set approach discussed in @xcite . we describe a subset @xmath34 as the super - level set of a function @xmath35 , i.e. @xmath36 then , the function @xmath2 in problem [ prob - rho ] can be represented as a function of @xmath37 , and hence @xmath38 in problem [ prob - rho ] can be also regarded as a function of @xmath37 since @xmath38 originally depends on @xmath2 . we may thus rewrite our optimization problem in the minimization problem of the following energy functional @xmath39 : @xmath40 where @xmath41 for minimizing eigenvalues and @xmath42 for maximizing eigenvalues , @xmath43 is the lagrange multiplier and @xmath44 . here , we remark that the equation @xmath45 corresponds to the volume constraint ( [ constraint ] ) . the first variation of @xmath39 and @xmath43 can be calculated by ordinary methods . our optimization problem is then reduced to solving the gradient flow associated with @xmath39 , namely , the flow which decreases the energy @xmath39 along solutions . more precisely , our problem is reduced to the problem to solve the following evolution equation : @xmath46 in a certain function space , where @xmath47 is given by @xmath48 and @xmath49 is the associated eigenfunction of @xmath50 . here we omit the detail how to obtain this system , since it is completely done by following the arguments in @xcite . since the differentiability of the solution @xmath37 of ( [ sdf ] ) may be lost during evolution of ( [ sdf ] ) , we may not construct numerical solution of ( [ sdf ] ) . to avoid this difficulty , instead of ( [ sdf ] ) , we solve @xmath51 with sufficiently small @xmath52 . then ( [ viscos ] ) is a semi - linear parabolic evolution equation , and hence the well - known explicit or implicit scheme enables us to solve this equation numerically keeping smoothness of @xmath37 during evolutions . this technique is well - known as _ viscosity vanishing method_. as for hamilton - jacobi type equations like ( [ sdf ] ) , it is known that solutions of ( [ viscos ] ) approach to those of ( [ sdf ] ) as @xmath53 in a suitable sense ( see e.g. @xcite ) . as for eigenvalue optimization problems , there are a lot of earlier works for @xmath54 this is a generalized eigenvalue problem of the laplacian , which is , for example , well - considered for studying frequency of drums with spatially inhomogeneous density on @xmath4 . osher - sethian @xcite and osher - santosa @xcite considered ( [ ev - prob - sigma ] ) with the level set approach ( see section [ section - levelset ] ) , which has been well treated during a variety of improvements . in @xcite and many related works , the following problem is considered , which is an alternative one of problem [ prob - rho ] . [ prob - sigma ] for @xmath55 define @xmath56 find an element @xmath57 which attains @xmath58 if it exists , characterize @xmath59 and the shape of domain @xmath60 . for problem [ prob - sigma ] , there are several mathematical results which determine optimizers of @xmath61 as well as geometries of domain @xmath60 . in @xcite , krein mathematically considered the largest and the smallest @xmath62-th natural frequency of strings with fixed endpoints on the interval @xmath63 . in words of problem [ prob - sigma ] , he considered @xmath64 , @xmath65 , on @xmath66 with the homogeneous dirichlet boundary condition . he completely solved this problem by detecting the optimizer @xmath59 . a couple of decades later , cox and mclaughlin extended krein s arguments in arbitrary dimensions @xcite . they obtained the following optimization criteria for eigenvalues in terms of corresponding eigenfunctions . [ preceding - sigma ] consider problem [ prob - sigma ] with the homogeneous dirichlet boundary condition . if @xmath67 is the minimizer of @xmath68 in @xmath12 , then there is a positive constant @xmath69 such that , for the corresponding eigenfunction @xmath70 , the following inequalities hold : @xmath71 similarly if @xmath72 is the maximizer of @xmath68 in @xmath12 , then there is a positive constant @xmath73 such that the following inequalities hold : @xmath74 moreover , they obtained results for geometric properties of @xmath20 using symmetry arguments derived from the maximum principle . [ preceding - geom ] consider problem [ prob - sigma ] with the homogeneous dirichlet boundary condition . assume that @xmath75 is the minimizer of @xmath68 in @xmath12 and @xmath76 . if @xmath4 is convex and symmetric in @xmath77 orthogonal directions , then @xmath78 is also convex and symmetric in @xmath77 directions . moreover , @xmath78 is star - shaped with respect to the center of symmetry . similarly , if @xmath79 is the maximizer of @xmath68 in @xmath12 and @xmath80 . then , under the same assumption as minimizers , the same statements hold for @xmath81 , where @xmath82 is the complement of @xmath83 . theorem [ preceding - geom ] claims that geometric properties of @xmath78 and @xmath84 are also characterized by corresponding eigenfunctions , since @xmath78 and @xmath84 are determined by the super- or the sub - level sets of corresponding eigenfunctions . on the other hand , these researches focus on problem [ prob - sigma ] only with the homogeneous dirichlet boundary condition . furthermore @xcite , @xcite and @xcite are not concerned with problem [ prob - sigma ] with any other boundary condition , nor are they concerned with problem [ prob - rho ] . and we know no paper about them ( even numerically ) in the last few decades . lou and yanagida @xcite discuss an indefinite linear eigenvalue problem related to biological invasion of species . the formation of their problem is similar to problem [ prob - sigma ] with the homogeneous neumann condition , but the assumption on the weight @xmath85 is different . in @xcite , @xmath86 is assumed to be bounded with a fixed negative total weight . they consider only one - dimensional problems and prove that the global minimizer @xmath59 of the principal eigenvalue @xmath68 ( i.e. the first eigenvalue in our arguments ) should be the specific two - valued function , which is often called bang - bang " type . such bang - bang type weight @xmath59 is characterized by the super - level set of the eigenfunction associated with @xmath87 , which is similar to theorem [ preceding - sigma ] . in @xcite , chamillo , greiser , imai , kurata and onishi discuss geometries of the minimizing configuration @xmath20 for the first eigenvalue problem @xmath88 under the volume constraint ( [ constraint ] ) . they prove that @xmath20 does not inherit symmetry of @xmath4 for appropriate values of @xmath89 and @xmath90 when @xmath4 is either an annulus or a dumbbell . problem [ prob - rho ] is considered by cox and lipton @xcite , conca , laurain and mahadevan @xcite , conca , mahadevan and sanz @xcite and some other related works . in @xcite , cox and lipton study an shape optimization problem for the heat equation with the dirichlet boundary condition , which corresponds to problem [ prob - rho ] . they discuss the existence of optimal eigenvalues and optimizers . moreover , they obtain the exact optimal configuration in the one - dimensional problem . on the other hand , in conca , laurain and mahadevan @xcite and conca , mahadevan and sanz @xcite , they discuss the asymptotic expansion of the first eigenvalue @xmath91 and the asymptotic behavior of the minimized eigenvalue . they also discuss a conjecture of the optimal configuration when @xmath4 is a ball . the conjecture claims that the optimal configuration @xmath20 for the minimization problem of the first dirichlet eigenvalue is a concentric ball . however , the conjecture is disproved in @xcite . at least in two and three dimensional cases , the optimal configuration is a union of concentric annuli and a disk under a certain condition of @xmath89 and @xmath90 . on the other hand , since all the discussions therein are based on the asymptotic expansion of eigenpairs , hence the characterization of optimal configurations remains as an open problem for large @xmath89 . we also note that the attainability of the supremum or the infimum of the first positive eigenvalue in problem [ prob - rho ] still remains open . in fact , the subset @xmath12 is not weak@xmath92-closed in @xmath93 ( e.g. @xcite ) . the supremum and the infimum of the first eigenvalue are thus attained in the weak@xmath92-closure @xmath94 of @xmath12 . in @xcite , they prove that , in problem [ prob - rho ] , @xmath95 and the similar result for supremum also holds . but there is still a possibility that @xmath96 . we finally remark that the optimizer @xmath59 in problem [ prob - sigma ] can be chosen as an element of @xmath12 under sufficient smoothness assumptions on @xmath19 and eigenfunctions . this fact is the consequence of @xcite and @xcite . so far problem [ prob - rho ] is well - considered neither mathematically nor numerically , but the same level set approach as in the case of problem [ prob - sigma ] can be applied . recall that optimizers of eigenvalues are determined by stationary solutions of ( [ sdf ] ) . if @xmath37 is a stationary solution of ( [ sdf ] ) , then @xmath37 should satisfy @xmath97 on @xmath98 . moreover if @xmath37 satisfies @xmath99 on @xmath100 , then it should associate @xmath10 , which is given by @xmath101 if and only if @xmath102 , and the eigenfunction @xmath16 of ( [ ev - prob - rho ] ) with @xmath103 such that @xmath104 on @xmath100 , since @xmath105 is equal to the constant @xmath43 on @xmath100 , where @xmath106 is a constant . hence one can guess that optimizers are characterized by @xmath107 . first we study the eigenvalue problem in @xmath108 shown in figures [ fig - rho1dim-1 ] and [ fig - rho1dim-2 ] . as shown in figure [ fig - rho1dim-1 ] , the differential @xmath109 of the eigenfunction @xmath30 has discontinuities on @xmath100 and there is little hope of the correspondence between @xmath110 for optimizers and @xmath111 . next we consider the correspondence between @xmath110 for optimizers and @xmath112 instead of @xmath111 . the graph of @xmath112 and @xmath110 for optimizers are shown in figure [ fig - rho1dim-2 ] . both in the case of dirichlet and neumann boundary conditions , we may guess a correspondence between @xmath110 for optimizers and level sets of @xmath112 as similar to theorem [ preceding - sigma ] . next , we consider two - dimensional problems . in the following numerical experiments , we fix @xmath113 and the constant of volume constraint @xmath114 in ( [ constraint ] ) unless otherwise noted and consider the following @xmath4 : @xmath115{$ ( 0,3)\times ( 0,3)\setminus \{((0,1]\times ( 0,1 ] ) \cup ( ( 0,1]\times [ 2,3 ) ) \cup ( [ 2,3)\times ( 0,1 ] ) \cup ( [ 2,3)\times [ 2,3))\ } $ } \subset \mathbb{r}^2,\\ & \quad \quad \text{(cross , e.g.\ figure \ref{fig - rhod - cross})}\\ & \omega_3 : = \ { x^2+y^2 < 1\}\subset \mathbb{r}^2,\quad \text{(disk)}\\ & \omega_4 : = \{x^2+y^2 < 1\ } \setminus [ \{(x-0.5)^2+y^2 < 0.04\}\cup \{(x+0.5)^2+y^2 < 0.04\}]\subset \mathbb{r}^2,\\ & \quad \quad \quad \text{(disk with two holes , e.g.\ figure \ref{fig - rhod - nonstar})}\\ & \omega_5 : = ( -\pi , \pi)\times ( -2\pi , 2\pi)\subset \mathbb{r}^2,\quad \text{(rectangle)}\\ & \omega_6 : = \scalebox{0.8}[1.0]{$ ( 0,5)\times ( 0,3)\setminus \{((0,2]\times ( 0,1 ] ) \cup ( ( 0,2]\times [ 2,3 ) ) \cup ( [ 3,5)\times ( 0,1 ] ) \cup ( [ 3,5)\times [ 2,3))\ } $ } \subset \mathbb{r}^2,\\ & \quad \quad \quad \text{(rectangular cross , e.g.\ figure \ref{fig - rhon - cross})}\\ & \omega_7 : = \{x^2+y^2/4 < 1\}\subset \mathbb{r}^2\quad \text{(ellipse)}\quad \text { and } \\ & \omega_8 : = \scalebox{0.9}[1.0]{$ \ { x^2+y^2/4 < 1\ } \setminus [ \{(x-0.5)^2+y^2/16 < 0.04\}\cup \{(x+0.5)^2+y^2/16 < 0.04\ } ] $ } \subset \mathbb{r}^2.\\ & \quad \quad \quad \text{(ellipse with two holes , e.g.\ figure \ref{fig - rhon - nonstar})}\end{aligned}\ ] ] in two - dimensional problems , we apply the freefem++ library @xcite to all the computations . focus on figure [ fig - rhod - square ] , [ fig - rhod - cross ] , [ fig - rhod - disk ] , [ fig - rhod - nonstar ] , [ fig - rhon - rect ] , [ fig - rhon - cross ] , [ fig - rhon - ellipse ] and [ fig - rhon - nonstar ] . in these cases , @xmath110 for optimizers corresponds to the super- or the sub - level set of @xmath18 regardless of topologies of @xmath4 and boundary conditions as in one - dimensional problems , where @xmath16 is the associated eigenfunction of @xmath17 . more precisely , if @xmath116 minimizes or maximizes @xmath38 and if @xmath38 is simple ( see table [ table - simple - rho ] ) , then the corresponding eigenfunction @xmath70 satisfies @xmath117 inequalities ( [ inequality - min - rho ] ) and ( [ inequality - max - rho ] ) can be considered as analogue of those in theorem [ preceding - sigma ] . if ( [ inequality - min - rho ] ) and ( [ inequality - max - rho ] ) hold , then the simpleness of @xmath17 holds and vice versa ( table [ table - simple - rho ] ) . indeed , if @xmath38 is not simple , then the inequality ( [ inequality - max - rho ] ) does not hold , which will be seen from figure [ fig - rhon - square ] and table [ table - simple - rho ] . the maximum principle guarantees the simpleness of the smallest eigenvalue for ( uniformly ) elliptic operators . in the case of ( [ ev - prob - rho ] ) with the homogeneous dirichlet boundary condition , the smallest eigenvalue @xmath38 is positive . the smallest eigenvalue @xmath11 is then always simple for any @xmath22 in the case of the homogenous dirichlet boundary condition . however , under the homogeneous neumann boundary condition , @xmath11 is the second smallest eigenvalue , since the smallest eigenvalue is @xmath29 , and the simpleness of @xmath11 is not guaranteed . assume that @xmath118 attains the maximum of @xmath38 on @xmath12 and that @xmath119 has multiplicity two as shown in figure [ fig - rhon - square ] . then the energy functional in ( [ lagrange ] ) with multiplicity two eigenvalue @xmath120 can be also written by @xmath121 corresponding @xmath122 in ( [ v0 ] ) for obtaining the steepest descent flow is @xmath123 since @xmath124 holds for optimizers , then one can see that the optimizer should satisfy @xmath125 on @xmath100 , where @xmath126 . by the definition of @xmath110 by @xmath37 , @xmath127 holds on @xmath100 and hence the above equality on @xmath100 is equivalent to @xmath128 moreover , the constraints @xmath129 and @xmath130 must be kept during evolution of @xmath37 by ( [ sdf ] ) . optimizers @xmath37 then have to satisfy @xmath131 for the first variation @xmath132 of @xmath37 . in particular , it follows that @xmath133 should be satisfied by calculations discussed in @xcite . in particular , positive constants @xmath134 and @xmath135 must be identical and hence @xmath136 , @xmath137 . since @xmath136 and @xmath138 are arbitrary , we may choose @xmath139 . if we normalize eigenfunctions @xmath70 and @xmath140 so that @xmath141 ( @xmath142 ) , then the following inequality will be the maximization criterion for @xmath11 with multiplicity two : @xmath143 here we chose @xmath144 in ( [ inequality - max - rho - s ] ) , which is natural because @xmath2 and @xmath145 have the same order in ( [ inequality - min - rho ] ) and ( [ inequality - max - rho ] ) . we can see that ( [ inequality - max - rho-2 ] ) is actually satisfied ( see figure [ fig - rhon - square ] ) . we conclude our numerical observations for optimization criteria in problem [ prob - rho ] . [ observation - criteria - heat ] consider problem [ prob - rho ] with bounded a domain @xmath1 . if @xmath146 is the minimizer of @xmath38 in @xmath12 , then the eigenfunction @xmath70 associated with @xmath38 satisfies ( [ inequality - min - rho ] ) . similarly , if @xmath147 is the maximizer of @xmath68 in @xmath12 and if @xmath119 is simple , then the eigenfunction @xmath70 associated with @xmath119 satisfies ( [ inequality - max - rho ] ) . if @xmath147 is the maximizer of @xmath38 in @xmath12 and if @xmath119 has multiplicity two , then the eigenfunction @xmath70 associated with @xmath119 satisfies ( [ inequality - max - rho-2 ] ) . these optimization criteria will be generalized to higher dimensional problems , in particular , the case that @xmath11 has multiplicity @xmath148 in the similar manner . next , we consider geometry of the domain @xmath149 defined by the optimizer @xmath21 . inequalities ( [ inequality - min - rho ] ) , ( [ inequality - max - rho ] ) and ( [ inequality - max - rho-2 ] ) imply that various properties of @xmath20 for optimizers come from corresponding eigenfunctions . it is natural to expect that some geometric properties inherit from @xmath4 . here , we focus on connectivity , convexity , star - shapedness and symmetry . in figure [ fig - rhod - square ] ( square with the homogeneous dirichlet boundary ) , neither @xmath20 nor @xmath150 are even connected , even if @xmath4 is convex . in the case of the homogeneous neumann boundary value problem , either @xmath20 or @xmath150 is convex if @xmath4 is convex , according to our computation results . on the other hand , in the case of the non - simply connected domain @xmath151 ( figure [ fig - rhon - nonstar ] ) , @xmath20 maximizing @xmath11 is not connected , although @xmath151 is connected . however , by the inequality ( [ inequality - max - rho ] ) , one can easily confirm that we can choose @xmath152 so that @xmath20 maximizing @xmath11 is connected . as a consequence , there is generally no topological correspondence between @xmath4 and @xmath20 ( or @xmath150 ) which are independent of boundary conditions on @xmath19 or @xmath90 . next we focus on symmetry of @xmath20 . inequalities ( [ inequality - min - rho ] ) and ( [ inequality - max - rho ] ) imply that symmetry of @xmath20 comes from that of @xmath153 given by corresponding eigenfunction . if @xmath153 is symmetric in a certain axis direction or in rotation , then , thanks to the original equation @xmath154 , @xmath30 and @xmath145 will be also symmetric . finally symmetry of @xmath2 will hold from symmetry of @xmath153 . the key consideration is that whether the symmetry of @xmath15 associated with the optimizer @xmath21 inherits from @xmath4 . our numerical simulations argue that symmetry of @xmath15 inherits from @xmath4 ( figure [ fig - rhod - square ] , [ fig - rhod - cross ] , [ fig - rhod - disk ] , [ fig - rhod - nonstar ] , [ fig - rhon - rect ] , [ fig - rhon - cross ] , [ fig - rhon - ellipse ] and [ fig - rhon - nonstar ] ) . since @xmath16 is the solution of ( [ ev - prob - rho ] ) , symmetry of @xmath15 leads to that of @xmath16 and @xmath21 . one then observes the following . [ observation - symmetry - heat ] for problem [ prob - rho ] , let @xmath21 be the optimizer of @xmath11 , @xmath16 be the associated eigenfunctions of @xmath17 and @xmath13 . if @xmath4 is star - shaped and symmetric in a certain direction or in rotation , then @xmath20 and @xmath150 are also symmetric in the direction regardless of boundary conditions on @xmath19 . our examples also show that there is a possibility that both @xmath20 and @xmath150 are symmetric even if @xmath4 is not star - shaped ( see figures [ fig - rhod - nonstar ] and [ fig - rhon - nonstar ] ) . however , in the dirichlet boundary case , there is also a possibility that symmetry breaking of @xmath20 occurs in the case of @xmath4 being either an annulus or a dumbbell , which are not star - shaped . this is indeed the case of the eigenvalue problem @xmath88 as mentioned in section [ section - prior ] and the same phenomenon may occur in problem [ prob - rho ] . on the minimization problem for the disk with the homogeneous dirichlet boundary condition , we also mention that the optimal configuration @xmath20 is a union of concentric annuli and a disk ( see figure [ fig - rhod - disk ] ) . this observation implies that the result of @xcite mentioned in section [ section - prior ] may also hold if c is not close to 1 . finally we consider problem [ prob - rho ] with the mixed boundary conditions to discuss continuous dependence of optimizers on boundary conditions . let @xmath155 and the boundary condition on @xmath156 be @xmath157 where @xmath158 is a non - negative constant . in general , as in the case of the homogeneous dirichlet and the homogeneous neumann boundary conditions , the unique existence of the boundary value problem of elliptic equation @xmath159 is well - known under suitable assumptions , where @xmath39 is an elliptic operator , say , @xmath160 for a given bounded function @xmath161 and @xmath162 is a piecewise continuous function on @xmath19 . moreover , the unique solution @xmath30 depends continuously on @xmath162 in @xmath163 ( see e.g. @xcite ) . in previous subsections we observed that optimizers are dominated by eigenfunctions associated with corresponding eigenvalues . it is then natural to consider that optimizers also depend continuously on boundary conditions as well as eigenfunctions as solutions of elliptic equations . [ observation - dependence - heat ] consider problem [ prob - rho ] with the boundary condition ( [ robin - bc ] ) . then the region @xmath20 given by the optimizer @xmath21 of @xmath38 depends continuously on @xmath158 . sample numerical results are shown in figure [ fig - rho - cont - min ] and [ fig - rho - cont - max ] . although these are just simple examples , combining this observation with continuous dependence of eigenfunctions on boundary conditions , we may well expect that one can mathematically prove the continuous dependence of optimizers on boundary conditions in a suitable topology . as a comparison with problem [ prob - rho ] , we consider the optimization of the first eigenvalue for ( [ ev - prob - sigma ] ) . there are many studies on this type of problems , for example @xcite , while almost all such considerations are only with the homogeneous dirichlet boundary condition on rectangular domain . in this section we set @xmath164 for @xmath86 under consideration and fix @xmath165 . calculations similar to those in subsection [ section - prob - rho ] with @xmath166 in ( [ v0 ] ) yield numerical results listed in figures [ fig - sigmad - square ] , [ fig - sigmad - cross ] , [ fig - sigmad - disk ] , [ fig - sigmad - nonstar ] , [ fig - sigman - rect ] , [ fig - sigman - cross ] , [ fig - sigman - ellipse ] , [ fig - sigman - nonstar ] and [ fig - sigman - square ] . these results imply if @xmath55 minimizes or maximizes @xmath68 and if @xmath68 is simple ( see table [ table - simple - sigma ] ) then the corresponding eigenfunction @xmath70 associated with @xmath68 satisfies @xmath167 examples of mixed boundary value problem are also shown in figures [ fig - sigma - cont - min ] and [ fig - sigma - cont - max ] , which give us the same observation as in observation [ observation - dependence - heat ] . note that inequalities ( [ inequality - min - sigma ] ) and ( [ inequality - max - sigma ] ) are exactly the same correspondences as the optimization criteria for the homogeneous dirichlet boundary value problems : theorem [ preceding - sigma ] . one of key considerations of these observations is that @xmath68 is assumed to be simple . table [ table - simple - sigma ] shows the ratio between @xmath61 and the second eigenvalue @xmath168 . by discussions similar to subsection [ section - prob - rho ] , we obtain the following eigenvalue optimization criteria in problem [ prob - sigma ] including eigenvalues with the multiplicity two . [ observation - criteria - sigma ] consider problem [ prob - sigma ] in bounded domain @xmath1 . if @xmath67 is the minimizer of @xmath68 in @xmath12 , then the eigenfunction @xmath70 associated with @xmath68 satisfies ( [ inequality - min - sigma ] ) . similarly , if @xmath72 is the maximizer of @xmath68 in @xmath12 and if @xmath169 is simple , then the eigenfunction @xmath70 associated with @xmath169 satisfies ( [ inequality - max - sigma ] ) . if @xmath72 is the maximizer of @xmath68 in @xmath12 and if @xmath169 has multiplicity two , then eigenfunctions @xmath70 and @xmath140 associated with @xmath169 satisfies @xmath170 under the normalization @xmath171 ( @xmath172 , @xmath173 ) , which can be confirmed in figure [ fig - rhon - square ] . next we consider geometry of @xmath20 . in the case of the homogeneous dirichlet boundary condition , there is a mathematical result which describes the geometric property of @xmath174 : theorem [ preceding - geom ] . in the case of the homogeneous neumann boundary value problems in problem [ prob - sigma ] , unlike dirichlet boundary value problems , associated eigenfunctions do not have identical sign in @xmath4 . we thus consider connected components of off - zero - level set of eigenfunctions , which is called _ nodal domains_. it is well - known that the eigenfunction associated with @xmath61 has exactly two nodal domains in the case of the homogeneous neumann boundary value problems ( see e.g. @xcite ) . assume that @xmath30 is the eigenfunction associated with @xmath61 under the homogeneous neumann boundary condition . then the function @xmath175 and @xmath176 obtained by restricting @xmath30 on each nodal domain , say , @xmath177 and @xmath178 , are eigenfunctions of the same equation with an identical sign in @xmath179 , respectively , with @xmath180 in this case , the same discussion as in the case of homogeneous dirichlet boundary condition can be applied to analyzing properties of @xmath181 ( see e.g. @xcite ) including our criteria ( [ inequality - min - sigma ] ) and ( [ inequality - max - sigma ] ) . on the other hand , in figure [ fig - sigman - rect ] , [ fig - sigman - cross ] , [ fig - sigman - ellipse ] and [ fig - sigman - nonstar ] , @xmath4 is symmetric with respect to the set @xmath182 . in figure [ fig - sigman - square ] , @xmath177 is rotationally symmetric with respect to the origin , namely , @xmath183 . in such cases , we can see that @xmath20 is also symmetric with respect to the null set of @xmath30 ( or @xmath184 ) . consequently we observe the following : [ observation - geometry - sigma ] consider problem [ prob - sigma ] with the homogeneous neumann boundary condition . assume that @xmath75 is the minimizer of @xmath68 in @xmath12 and @xmath76 . if each nodal domain @xmath179 ( @xmath172 , @xmath173 ) is convex and symmetric in @xmath77 orthogonal directions , then @xmath185 also has the same properties . similarly , assume that @xmath79 is the maximizer of @xmath68 in @xmath12 and @xmath80 . then , under the same assumption as minimizers , the same statements hold for the set @xmath186 . finally , if @xmath4 is symmetric with respect to the null set of eigenfunctions @xmath187 ( or @xmath188 in the case that @xmath87 is not simple ) , then @xmath20 and @xmath150 are also symmetric . these observations are numerically confirmed in figure [ fig - sigman - rect ] , [ fig - sigman - cross ] , [ fig - sigman - ellipse ] , [ fig - sigman - nonstar ] and [ fig - sigman - square ] . we also discuss the geometry of @xmath20 in problem [ prob - sigma ] in the case that @xmath4 is not even star - shaped . our numerical results partially answer the inheritance problem in this case , as shown in figure [ fig - sigmad - nonstar ] , [ nonstar ] and [ fig - sigman - nonstar ] . theorem [ preceding - geom ] refers to the star - shapedness of @xmath20 only in the case that @xmath4 is convex . our numerical results newly suggest if @xmath4 is star - shaped then @xmath20 is also star - shaped ( see figure [ fig - sigmad - cross ] ) . however , if @xmath4 is not star - shaped , neither @xmath20 nor @xmath150 are necessarily star - shaped . in figure [ nonstar ] , minimization in problem [ prob - sigma ] for @xmath151 with the homogeneous dirichlet boundary condition is considered . only difference between two figures is the ratio of the volume constraint in ( [ constraint ] ) . one of them is @xmath189 and the other is @xmath190 . in the case of @xmath189 , @xmath191 for the minimizer @xmath192 of @xmath61 is star - shaped . on the other hand , in the case of @xmath193 , neither @xmath78 nor @xmath194 is star - shaped . as a conclusion , if @xmath4 is not star - shaped , in general , neither the optimized domain @xmath20 nor its complement @xmath150 is start - shaped . as for symmetry , if @xmath4 is star - shaped and symmetric in a certain direction , then both @xmath20 and @xmath150 have the same symmetry both in the dirichlet and the neumann boundary problems . summarizing our arguments we obtain the following observation : [ observation - symmetry - sigma ] for problem [ prob - sigma ] , let @xmath59 be the optimizer of @xmath61 , @xmath16 be the associated eigenfunction of @xmath87 and @xmath60 . if @xmath4 is not star - shaped , the star - shapedness of @xmath20 and @xmath150 generally depends on @xmath4 and @xmath90 . if @xmath4 is star - shaped and is symmetric with respect to a certain direction or rotationally symmetric , then @xmath20 and @xmath150 also have the same symmetry . we finally remark that , even if @xmath4 is symmetric and non - star - shaped , there are cases that @xmath20 and @xmath150 have symmetry which @xmath4 possesses , as are shown in figure [ fig - sigmad - nonstar ] and [ fig - sigman - nonstar ] . throughout numerical studies in this paper , we numerically solved ( [ viscos ] ) with @xmath195 via the following implicit scheme @xmath196 where @xmath197 is an arbitrary element of a finite element subspace of @xmath198 with suitable boundary conditions . fix the initial level set function by @xmath199 . then we successively solve ( [ scheme ] ) so that all @xmath200 keep the volume constraint @xmath201 by following the discussion in @xcite . here @xmath113 ( problem [ prob - rho ] ) , @xmath202 ( problem [ prob - sigma ] ) and @xmath203 are fixed in all cases , and the step size @xmath204 is defined by @xmath205 in each step . examples are shown in figure [ fig - rhod - conv]-[fig - sigman - conv ] . although the initial shape of @xmath206 or @xmath207 matters , we may observe that the convergence rate is independent of geometry of @xmath4 in each problem with each boundary condition . in this paper , we have studied the eigenvalue optimization of spatially inhomogeneous diffusion operator @xmath208 with a given constraint , which is motivated by the control of heat conductivity of spatially inhomogeneous media . we applied the level set approach to characterize optimizers for problem [ prob - rho ] . collecting our numerical observations , one knows the following : the region @xmath209 determined by the eigenvalue optimizer @xmath21 is characterized by the super- or the sub - level set of @xmath18 even if @xmath11 has multiplicity greater than two , where @xmath16 is the eigenfunction associated with @xmath17 . this characterization is independent of topologies of @xmath4 and boundary conditions on @xmath19 . moreover , if @xmath4 is star - shaped and has symmetry in a certain direction , @xmath20 also possesses the same symmetry . one of key considerations in our numerical studies here is that eigenvalue optimizers can be characterized by associating eigenfunctions including symmetry . once such characterizations are mathematically confirmed , various properties of optimizers , such as symmetry and continuous dependence on boundary conditions in a suitable topology , will follow from those of corresponding eigenfunctions , as is true in the case of problem [ prob - sigma ] . this research was partially supported by jst , crest : a mathematical challenge to a new phase of material science , based on discrete geometric analysis . km was partially supported by coop with math program , a commissioned project by mext . hn was partially supported by grants - in - aid for scientific research ( c ) ( no . 26400067 ) . we thank prof . hideyuki azegami for providing us with very meaningful advice for our computational study . we also thank prof . motoko kotani for introducing us to this study and giving us a lot of advice for writing this paper . 10 c. bandle , _ isoperimetric inequalities and applications _ , monographs and studies in mathematics , vol . 7 , pitman ( advanced publishing program ) , boston , mass . , 1980 . s. chanillo , d. grieser , m. imai , k. kurata , and i. ohnishi , _ symmetry breaking and other phenomena in the optimization of eigenvalues for composite membranes _ , comm . * 214 * ( 2000 ) , 315337 . c. conca , a. laurain , and r. mahadevan , _ minimization of the ground state for two phase conductors in low contrast regime _ , siam j. appl * 72 * ( 2012 ) , 12381259 . c. conca , a. laurain , r. mahadevan , and d quintero , _ minimization of the ground state of the mixture of two conducting material in a small contrast regime _ , arxiv:1408.4981 ( 2014 ) . c. conca , r. mahadevan , and l. sanz , _ shape derivative for a two - phase eigenvalue problem and optimal configurations in a ball _ , canum 2008 , esaim proc . , vol . 27 , edp sci . , les ulis , 2009 , pp . 311321 . r. courant and d. hilbert , _ methods of mathematical physics . i and ii _ , interscience publishers , new york - london , 1962 . s. cox and r. lipton , _ extremal eigenvalue problems for two - phase conductors _ , arch . rational mech . 136 * ( 1996 ) , 101117 . s. j. cox and j. r. mclaughlin , _ extremal eigenvalue problems for composite membranes . * 22 * ( 1990 ) , 153167 . to3em , _ extremal eigenvalue problems for composite membranes , ii _ , appl . * 22 * ( 1990 ) , 169187 . m. g. crandall and p .- l . lions , _ viscosity solutions of hamilton - jacobi equations _ , trans . * 277 * ( 1983 ) , 142 . f. hecht , _ new development in freefem++ _ , j. numer . * 20 * ( 2012 ) , no . 3 - 4 , 251265 . mr 3043640 m. g. krein , _ on certain problems on the maximum and minimum of characteristic values and on the lyapunov zones of stability _ , ( 2 ) * 1 * ( 1955 ) , 163187 . y. lou and e. yanagida , _ minimization of the principal eigenvalue for an elliptic boundary value problem with indefinite weight , and applications to population dynamics _ , japan j. indust . appl . math . * 23 * ( 2006 ) , 275292 . s. osher and j. a. sethian , _ fronts propagating with curvature - dependent speed : algorithms based on hamilton - jacobi formulations _ , j. comput . phys . * 79 * ( 1988 ) , 1249 . s. j. osher and f. santosa , _ level set methods for optimization problems involving geometry and constraints . i. frequencies of a two - density inhomogeneous drum _ , j. comput . * 171 * ( 2001 ) , 272288 . n. s. trudinger , _ maximum principles for linear , non - uniformly elliptic operators with measurable coefficients _ , math . z. * 156 * ( 1977 ) , 291301 . in problem [ prob - rho ] ( problem [ prob - sigma ] ) with the homogeneous neumann boundary condition , the simpleness of the first eigenvalue @xmath11 ( @xmath61 ) is nontrivial . @xmath210 and @xmath211 ( @xmath192 and @xmath212 ) denote the minimizer and the maximizer of @xmath11 ( @xmath61 ) , respectively . as for the homogeneous neumann boundary value problems for @xmath213 or @xmath151 , @xmath214 ( @xmath215 ) holds for minimizations and maximizations , which implies that @xmath11 ( @xmath61 ) is simple in those cases . in the case of the maximization problem for @xmath216 , the ratio @xmath217 ( @xmath218 ) is close to @xmath219 and hence we may not consider that the first eigenvalue is simple . indeed , figure [ fig - rhon - square ] and figure [ fig - sigman - square ] imply that ( [ inequality - max - rho ] ) and ( [ inequality - max - sigma ] ) do not hold , respectively . on the other hand , @xmath220 and @xmath221 hold , which imply that the first eigenvalue for @xmath177 has multiplicity two . minimization in problem [ prob - sigma ] for the non - star - shaped region @xmath223 ( cf . figure [ fig - sigmad - nonstar ] ) with homogeneous dirichlet boundary condition is considered . figure ( a ) is the minimizer @xmath192 with the volume constraint ratio @xmath189 and ( b ) is @xmath192 with the volume constraint ratio @xmath193 . the super - level set @xmath224 is star - shaped in case of ( a ) , which is not the case of ( b ) . [ nonstar ] for problem [ prob - rho ] , figures of eigenfunction stand for @xmath225 and for problem [ prob - sigma ] , figures of eigenfunction stand for @xmath226 . each figure shows optimizer of the first eigenvalue and associated eigenfunction . the red region in figures of optimizer stands for @xmath227 or @xmath228 . in the following figures , we calculate the optimal configuration for @xmath113 and @xmath203 unless otherwise noted . figure ( a ) shows the maximizer @xmath118 of @xmath11 . figure ( b ) and ( c ) show corresponding @xmath235 and @xmath236 , where @xmath237 and @xmath238 are associated eigenfunctions of @xmath119 and @xmath239 ( actually equal to @xmath119 ) , respectively , after the normalization so that @xmath240 holds . figure ( d ) shows @xmath241 after normalizations . [ fig - rhon - square ] figure ( a ) shows the maximizer @xmath212 of @xmath61 . figure ( b ) and ( c ) show corresponding @xmath242 and @xmath243 , where @xmath237 and @xmath238 are associated eigenfunctions of @xmath169 and @xmath244 ( actually equal to @xmath169 ) , respectively , after the normalization so that @xmath245 holds . figure ( d ) shows @xmath246 after normalizations . [ fig - sigman - square ] cccccc ( a ) & on @xmath177,title="fig:",width=85 ] & on @xmath177,title="fig:",width=85 ] & on @xmath177,title="fig:",width=85 ] & on @xmath177,title="fig:",width=85 ] & on @xmath177,title="fig:",width=85 ] ( b ) & on @xmath177,title="fig:",width=85 ] & on @xmath177,title="fig:",width=85 ] & on @xmath177,title="fig:",width=85 ] & on @xmath177,title="fig:",width=85 ] & on @xmath177,title="fig:",width=85 ] & @xmath247 & @xmath248 & @xmath249 & @xmath250 & @xmath251 \(a ) shows the minimizer @xmath230 and ( b ) shows @xmath252 for the eigenfunction @xmath253 of @xmath254 with various @xmath255 the minimization criterion ( [ inequality - min - rho ] ) is actually satisfied in each case . the larger @xmath255 becomes , the closer @xmath230 is to the minimizer with homogeneous dirichlet boundary condition ( figure [ fig - rhod - square ] , a ) . [ fig - rho - cont - min ] cccccc ( a ) & on @xmath177,title="fig:",width=85 ] & on @xmath177,title="fig:",width=85 ] & on @xmath177,title="fig:",width=85 ] & on @xmath177,title="fig:",width=85 ] & on @xmath177,title="fig:",width=85 ] ( b ) & on @xmath177,title="fig:",width=85 ] & on @xmath177,title="fig:",width=85 ] & on @xmath177,title="fig:",width=85 ] & on @xmath177,title="fig:",width=85 ] & on @xmath177,title="fig:",width=85 ] & @xmath247 & @xmath248 & @xmath249 & @xmath250 & @xmath251 \(a ) shows the maximizer @xmath118 and ( b ) shows @xmath256 for the eigenfunction @xmath257 of @xmath258 with various @xmath255 the maximization criterion ( [ inequality - max - rho ] ) is actually satisfied in each case . the larger @xmath255 becomes , the closer @xmath118 is to the maximizer with homogeneous dirichlet boundary condition ( figure [ fig - rhod - square ] , c ) . [ fig - rho - cont - max ] cccccc ( a ) & on @xmath177,title="fig:",width=85 ] & on @xmath177,title="fig:",width=85 ] & on @xmath177,title="fig:",width=85 ] & on @xmath177,title="fig:",width=85 ] & on @xmath177,title="fig:",width=85 ] ( b ) & on @xmath177,title="fig:",width=85 ] & on @xmath177,title="fig:",width=85 ] & on @xmath177,title="fig:",width=85 ] & on @xmath177,title="fig:",width=85 ] & on @xmath177,title="fig:",width=85 ] & @xmath247 & @xmath248 & @xmath249 & @xmath250 & @xmath251 \(a ) shows the minimizer @xmath192 and ( b ) shows @xmath259 for the eigenfunction @xmath253 of @xmath232 with various @xmath255 the minimization criterion ( [ inequality - min - sigma ] ) is actually satisfied in each case . the larger @xmath255 becomes , the closer @xmath192 is to the minimizer with homogeneous dirichlet boundary condition ( figure [ fig - sigmad - square ] a ) . [ fig - sigma - cont - min ] cccccc ( a ) & on @xmath177,title="fig:",width=85 ] & on @xmath177,title="fig:",width=85 ] & on @xmath177,title="fig:",width=85 ] & on @xmath177,title="fig:",width=85 ] & on @xmath177,title="fig:",width=85 ] ( b ) & on @xmath177,title="fig:",width=85 ] & on @xmath177,title="fig:",width=85 ] & on @xmath177,title="fig:",width=85 ] & on @xmath177,title="fig:",width=85 ] & on @xmath177,title="fig:",width=85 ] & @xmath247 & @xmath248 & @xmath249 & @xmath250 & @xmath251 \(a ) shows the minimizer @xmath212 and ( b ) shows @xmath260 for the eigenfunction @xmath253 of @xmath169 with various @xmath255 the maximization criterion ( [ inequality - max - sigma ] ) is actually satisfied in each case . the larger @xmath255 becomes , the closer @xmath212 is to the maximizer with homogeneous dirichlet boundary condition ( figure [ fig - sigmad - square ] , c ) . [ fig - sigma - cont - max ] we calculate the convergence to the optimizer . each figure shows the density function @xmath91 and @xmath86 after @xmath261 steps as we solve ( [ viscos ] ) towards the optimizer in problem [ prob - rho ] and problem [ prob - sigma ] , respectively . cccccc ( a ) & with dirichlet boundary , title="fig:",width=85 ] & with dirichlet boundary , title="fig:",width=85 ] & with dirichlet boundary , title="fig:",width=85 ] & with dirichlet boundary , title="fig:",width=85 ] & with dirichlet boundary , title="fig:",width=85 ] & @xmath262 & @xmath263 & @xmath264 & @xmath265 ( b ) & with dirichlet boundary , title="fig:",width=85 ] & with dirichlet boundary , title="fig:",width=85 ] & with dirichlet boundary , title="fig:",width=85 ] & with dirichlet boundary , title="fig:",width=85 ] & with dirichlet boundary , title="fig:",width=85 ] & @xmath262 & @xmath266 & @xmath267 & @xmath268 in all cases @xmath113 and @xmath203 are fixed . the rightmost graph is the evolution of @xmath11 as we solve ( [ viscos ] ) . ( a ) : maximization of @xmath11 on @xmath177 . ( b ) : maximization of @xmath11 on @xmath223.[fig - rhod - conv ] cccccc ( a ) & with neumann boundary , title="fig:",width=85 ] & with neumann boundary , title="fig:",width=85 ] & with neumann boundary , title="fig:",width=85 ] & with neumann boundary , title="fig:",width=85 ] & with neumann boundary , title="fig:",width=85 ] & @xmath262 & @xmath269 & @xmath270 & @xmath266 ( b ) & with neumann boundary , title="fig:",width=85 ] & with neumann boundary , title="fig:",width=85 ] & with neumann boundary , title="fig:",width=85 ] & with neumann boundary , title="fig:",width=85 ] & with neumann boundary , title="fig:",width=85 ] & @xmath262 & @xmath266 & @xmath271 & @xmath263 ( c ) & with neumann boundary , title="fig:",width=85 ] & with neumann boundary , title="fig:",width=85 ] & with neumann boundary , title="fig:",width=85 ] & with neumann boundary , title="fig:",width=85 ] & with neumann boundary , title="fig:",width=85 ] & @xmath262 & @xmath266 & @xmath272 & @xmath263 in all cases @xmath113 and @xmath203 are fixed . the rightmost graph is the evolution of @xmath11 as we solve ( [ viscos ] ) . ( a ) : maximization of @xmath11 on @xmath229 . ( b ) : maximization of @xmath11 on @xmath151 . ( c ) : maximization of @xmath11 on @xmath177 . in this case the optimized eigenvalue @xmath119 has multiplicity two ( cf . figure [ fig - rhon - square ] and table [ table - simple - rho ] ) and hence the function @xmath273 in the level set evolution ( [ sdf ] ) is set @xmath274 after the normalization @xmath275 . [ fig - rhon - conv ] cccccc ( a ) & with dirichlet boundary , title="fig:",width=85 ] & with dirichlet boundary , title="fig:",width=85 ] & with dirichlet boundary , title="fig:",width=85 ] & with dirichlet boundary , title="fig:",width=85 ] & with dirichlet boundary , title="fig:",width=85 ] & @xmath262 & @xmath271 & @xmath276 & @xmath267 ( b ) & with dirichlet boundary , title="fig:",width=85 ] & with dirichlet boundary , title="fig:",width=85 ] & with dirichlet boundary , title="fig:",width=85 ] & with dirichlet boundary , title="fig:",width=85 ] & with dirichlet boundary , title="fig:",width=85 ] & @xmath262 & @xmath271 & @xmath277 & @xmath267 ( c ) & with dirichlet boundary , title="fig:",width=85 ] & with dirichlet boundary , title="fig:",width=85 ] & with dirichlet boundary , title="fig:",width=85 ] & with dirichlet boundary , title="fig:",width=85 ] & with dirichlet boundary , title="fig:",width=85 ] & @xmath262 & @xmath266 & @xmath263 & @xmath265 in all cases @xmath202 and @xmath203 are fixed . the rightmost graph is the evolution of @xmath61 as we solve ( [ viscos ] ) . ( a ) : minimization of @xmath61 on @xmath177 . ( b ) : minimization of @xmath61 on @xmath229 . ( c ) : minimization of @xmath61 on @xmath223 . cccccc ( a ) & with neumann boundary , title="fig:",width=85 ] & with neumann boundary , title="fig:",width=85 ] & with neumann boundary , title="fig:",width=85 ] & with neumann boundary , title="fig:",width=85 ] & with neumann boundary , title="fig:",width=85 ] & @xmath262 & @xmath270 & @xmath278 & @xmath263 ( b ) & with neumann boundary , title="fig:",width=85 ] & with neumann boundary , title="fig:",width=85 ] & with neumann boundary , title="fig:",width=85 ] & with neumann boundary , title="fig:",width=85 ] & with neumann boundary , title="fig:",width=85 ] & @xmath262 & @xmath266 & @xmath271 & @xmath263 ( c ) & with neumann boundary , title="fig:",width=85 ] & with neumann boundary , title="fig:",width=85 ] & with neumann boundary , title="fig:",width=85 ] & with neumann boundary , title="fig:",width=85 ] & with neumann boundary , title="fig:",width=85 ] & @xmath262 & @xmath279 & @xmath280 & @xmath281 in all cases @xmath202 and @xmath203 are fixed . the rightmost graph is the evolution of @xmath61 as we solve ( [ viscos ] ) . ( a ) : minimization of @xmath61 on @xmath229 . ( b ) : minimization of @xmath61 on @xmath223 . ( c ) : maximization of @xmath61 on @xmath151 . in this case the optimized eigenvalue @xmath169 has multiplicity two ( cf . figure [ fig - sigman - square ] and table [ table - simple - sigma ] ) and hence the function @xmath273 in the level set evolution ( [ sdf ] ) is set @xmath282 after the normalization @xmath283 .

helium ion microcopy ( him ) based on gas field ion sources ( gfis ) represents a new ultra high resolution microscopy and nano fabrication technique . it is an enabling technology that not only provides imagery of conducting as well as uncoated insulating nano structures but also allows to create these features . the latter can be achieved using resists or material removal due to sputtering . the close to free form sculpting of structures over several length scales has been made possible by the extension of the method to other gases such as neon . a brief introduction of the underlying physics as well as a broad review of the applicability of the method is presented in this review .

1311.1711

high resolution imaging , in particular in biology and materials science , is mostly done using scanning electron microscopy ( sem ) . the ease of use and the widespread availability has made this the number one method for imaging in the aforementioned fields . structuring and manipulation of nano sized features is traditionally the domain of focused ion beams . here , mainly liquid metal ion sources ( lmis ) using gallium are used however , other techniques such as various types of gfis,@xcite alloy lmis,@xcite magneto optical trap sources ( motis)@xcite and multicusp plasma sources@xcite are runners up . good reviews discussing these two techniques can be found in refs . [ ] and [ ] for sem and fib , respectively . helium ion microscopy presents a technique that unites many of the advantages of sem and fib in a single tool . more importantly , it also overcomes some of the deficiencies of sem and fib . in particular , the possibility to image biological and in general insulating samples without coating at highest resolution is important for many scientific questions currently discussed in materials science and biology . another important charged particle beam technique namely transmission electron microscopy ( tem)depends on very thin samples free of defects from the preparation . the unique nano sized beam of the him makes it possible to not only mill and pattern smallest features but also do this with minimal damage to the crystal lattice . in the following we will give an outline of the working principle and signal generation in helium ion microscopy , followed by two sections highlighting specific imaging applications and examples of materials modification . the initial idea of a scanning ion microscope has been put forward by knoll and ruska already in 1932.@xcite the working principle of a helium ion microscope can be divided into three different stages . 1 . helium ionization and acceleration 2 . beam formation and control 3 . sample interaction in this introduction we will only briefly touch points 1 and 2 . instead , we will focus on the physics that is important to understand the application of the technique to imaging and nanofabrication . the first is made possible by using a newly developed gas field ion source ( gfis ) . gfis have been investigated for a long time@xcite for their use in microscopes.@xcite the idea is based on the initial design of a field ion microscope by e. mller.@xcite however , only recently a stable ion source with a high brightness and small virtual source size has been realized by ward , notte and economou for use in a microscope.@xcite it is based on an emitter whose apex has been shaped into a three sided pyramid ( see fig . [ fig : him - source ] ) . work is done currently to understand and optimize the supertip formation process in order to maximise the achievable current.@xcite using scanning field ion microscopy ( sfim ) the apex of the tip can be monitored and shaped using high fields that can ultimately remove weakly bound atoms from the apex . in this way the configuration of the tip apex can be controlled at the atomic level . although monomers are possible , trimers are more stable . figure [ fig : him - source](b ) shows a sfim image of the tip . the trimer in the center and the edge of the first two shelves below the trimer are visible . the combination of a pyramidally shaped tip and the carefully shaped apex concentrates the electric field , so that subsequent helium ionization predominantly happens at the top most atoms of the pyramid . using apertures in the beam path allows to select current originating from ionization events on a single apex atom . typical source parameters are listed in table [ tab : column - param ] . .typical helium ion source parameters@xcite [ cols="<,^ , < " , ] the second point is technologically demanding and requires a high degree of knowledge on the design and implementation of the involved electrostatic lenses , quadrupoles , octopoles , etc . for more details the reader is refereed to numerous monographs available on charged particle optics such as [ ] . the critical source parameters important for the ion optical performance of the column are given in table [ tab : column - param ] . the energy spread of 1ev is an upper bound . earlier measurements indicate that the values could in fact be lower by a factor of two to four.@xcite one of the important consequences of the parameters listed in table [ tab : column - param ] is the image side beam convergence angle @xmath0 . typical values for @xmath0 are well below 1mrad . this small beam divergence results in a large depth of field @xmath1 here , @xmath2 denotes the minimum feature that can be resolved in the image . assuming identical resolutions the him will have a superior depth of field as compared to low voltage sem with typical @xmath0 values of 8mrad.@xcite once the focused ion beam strikes the sample , the interaction of the accelerated particles with the substrate atoms and electrons will give rise to a large number of different signals . we will cover the available signals in the following section . before we do so , we will briefly discuss the processes that occur during ion / sample interaction and that are important for the achievable resolution in charged particle beam imaging . besides the small beam diameter , the shape and size of the actual interaction volume between the impinging particle and the sample plays an important role to reach ultimate resolution . figure [ fig : charged - part - beams ] compares monte carlo ( mc ) simulation results for different charged particle beams . from the figure it is evident that the interaction volume relevant for secondary electron ( se ) generation of the focused he beam is smaller than for the other two.@xcite for the case of a ga beam the large cross section of ga with in this case si leads to substantial scattering in the near surface region relevant for the se signal generation . for a low energy electron beam needed to simultaneously optimize resolution and surface sensitivity in sem electron electron scattering in the sample widens the beam dramatically in the first few nanometers deteriorating the achievable resolution . scattering also occurs in him . in the case of a 30kev he beam scattering occurs with the nuclei of the sample atoms . however , due to the low mass of helium , scattering is not very efficient in the first few nanometers of sample material . this results in minimal beam divergence inside the sample . consequently the collected se originate from a cylinder with a minimal volume . the somewhat lower energy of the se in him@xcite and the lower characteristic escape depth for se in him for light elements@xcite enhances this difference between sem and him even further . next we will discuss available particles and correoponding signals . we will in particular highlight their benefits and drawbacks when used for imaging and what kind of physical quantities can be accessed using them . the sequence in which they are discussed is determined by their abundance in the tool . this also corresponds to the ease of use and how widespread the technique is available in the current instrument base . an initial overview of some of the different contrast mechanisms is given by bell.@xcite secondary electron ( se ) emission generated by ions can be split into two parts . in the logical order we will first discuss se generation followed by the transport of electrons through the material . the latter is important as the generated electrons still need to reach and subsequently leave the surface into the vacuum so they can be detected . as has been proposed by bethe@xcite the rate of secondary electron generation @xmath3 ( electrons per incoming ion ) is proportional to the stopping power of the incident particle @xmath4 in ev / . @xmath5 here , @xmath6 denotes a scaling constant . it is assumed that at least in the relevant near surface region , atomic collisions do not play an important role and @xmath4 depends only on the electronic stopping power . the generated se1 are mostly excited by the incoming ions via kinetic emission ( ke).@xcite two types of secondary electrons of type 2 ( se2 ) exist in ion beam imaging . type 2 electrons can be generated either by recoiling target atoms or from other ses of type 1 . the second type of se2 generation ( se generated by se ) does not play an important role in him . this becomes clear when looking at the maximum energy of the se , which is taken to be equal to the energy loss of the impinging he for a head on collision@xcite @xmath7 here , @xmath8 is the electron mass , @xmath9 the ion velocity , and @xmath10 denotes the fermi velocity . the cross section for such a collision is highest if the ion velocity@xmath11 m/s for a primary energy of 35kev matches the fermi velocity of the electrons in the material . for gold and 35kev he this yields a maximum se energy of 45ev . however , this energy is approximately equal to or even below the effective energy needed for se generation by electrons in many materials.@xcite thus the size of the electron electron collision cascade is restricted . however , kinetic excitation of electrons is also possible by recoil atoms , provided they are fast enough so that their speed still matches the fermi velocity of the target material . electrons produced by recoiling target atoms are usually called se2 . ramachandra et al.@xcite calculated the ratio between se2 and se1 as function of energy and material . the conclusion that can be drawn from their calculation is that for higher primary energies a smaller se2/se1 ratio can be achieved for most materials and consequently a higher resolution is possible . the other process for electron emission is potential emission ( pe ) via auger neutralization . however , only for very low energies below 5kev pe becomes dominant@xcite over ke . once secondary electrons have been generated they still need to escape from the solid . this process can be described as a diffusion process . the characteristic length scale of this diffusion process the effective diffusion length of secondary electrons @xmath12is of the order of 1 nm for nearly all materials.@xcite this leads to the fact that only the first few nanometers of the material add to the emitted ses . measurements of the effective se yield in him show variations between 1 for carbon and values as high as 8 for platinum.@xcite the number and energy distribution of these ion induced secondary electrons differs from what is found in a sem . a sharper maximum at lower energies is usually found@xcite in him . in fig . [ fig : se - yield](a ) a comparison of calculated se yields in sem , ga fib and him is presented . the calculations done by ohya et al . still overestimate peak position as well as peak width.@xcite these differences between actual measurement results of se yields in a him and calculations are attributed to se generation mechanisms not considered in the calculations . indications exist that bulk plasmon excitations can play an important role@xcite for se generation in him . an actual measurement of se yield as a function of se energy and the work function dependence of the distribution maximum is presented in fig . [ fig : se - yield](b ) . . ( b ) experimentally obtained se yield for three different metals . the dependence of the peak position on work function is shown as an inset . reproduced with permission from proc . spie , vol . 8036 , 80360o-180360o-10 ( @xcite ) . copyright ( 2011 ) spie . ] a consequence of the particular energy distribution of se in him and the small se generation volume is an enhanced surface sensitivity . this has been shown in a recent comparative study of him and sem imaging performance on carbon coated gold nano rods.@xcite only at very low acceleration voltages is sem able to visualize the thin carbon layer covering the gold particles . however , low voltage sem suffers from carbon deposition in the imaged area making detailed studies challenging . an example of the high sensitivity of him with respect to the material work function is the visualization of different half unit cell surface termination in ti@xmath13sic@xmath14.@xcite depending on the position of the cleavage plane the surface is either terminated by si ( half unit cell ) or ti ( full unit cell ) . the difference in the chemistry of the top surface layer of atoms results in different se yields for the two terminations . as a result they can be discriminated in him . a software package called _ ionise _ developed by p. rack and coworkers is available that allows the calculation of the expected se yield for a large number of materials.@xcite although good agreement has been achieved between simulation and experiment , only a limited number of analytical applications based on se yield are known.@xcite sample topography in him is made visible in a similar way as in sem . the dependence of the se yield on the specimen tilt with respect to the incoming beam can be described the following secant law @xmath15 however , experimental studies showed that the actually measured se yield at the detector deviates from the expected secant behaviour.@xcite the deviations lead to a more linear response curve , which in turn should reduce the strong edge effect known from sem . nevertheless , a very strong edge effect has been observed in thin layers.@xcite in summary , the achievable high resolution and surface sensitivity in him is based on the fact that the ses originate from a cylinder at the beam penetration point with a diameter of approximately 1nm given by the beam diameter and a length of less than 3nm determined by @xmath12 . this small probe volume helps to achieve very high image resolution in him . it should be noted that obtaining such high resolution images has become substantially easier since the introduction of uhv him.@xcite at small fields of views usually high fluences are reached as a consequence of the large pixel density . in an uhv him implantation and sputtering can still negatively affect the sample during imaging . carbon deposition on the other hand can be excluded.@xcite the removal of hydrocarbons from the sample chamber vacuum prevents the formation of carbon deposits in the imaged area . as a consequence some exceptional imaging results could be achieved ( see fig . [ fig : bio](b ) ) . what backscattered electrons are to the sem , backscattered helium ( bshe ) atoms and ions are to him . this rather bold statement is true in several ways , as will be highlighted in the next paragraphs . two different ways to utilize bshe are available in current him . first , and most commonly used , is a microchannel plate ( mcp ) detector to acquire qualitative element distribution maps . second , a silicon drift detector can be used to obtain spectroscopic information from microscopic areas . the latter allows quantitative element identification based on the same principles as rutherford backscattering spectroscopy ( rbs ) . while se images usually are rich in morphological contrast , bshe images obtained with the mcp are poor in topography and rich in elemental contrast . in contrast to the se images , the information in these images originates from the bulk of the sample . to measure them the mcp is inserted below the pole piece in the primary beam path . while a center hole allows the primary beam to reach the sample , this geometry maximizes the solid angle , and thus the collection efficiency of the detector . the obtained contrast can be understood by examining the rutherford scattering cross section @xmath16 where @xmath17 is the elementary charge , and @xmath18 and @xmath19 denote the atomic number of impinging and target particle , respectively . assuming that the target atom is at rest , @xmath20 is the energy of the impinging particle , @xmath21 is an arbitrary element of solid angle , and @xmath22 is the scattering angle . for fixed @xmath18helium for this review and a given energy , a dependence on @xmath23 leads to a strong contrast between different elements . a more detailed analysis of the scattering cross section shows that a dependence that is related to the structure of the periodic table of elements also exists . this is a result of the change in screening along the rows of the periodic table . in fig . [ fig : bshe - yield ] , experimentally obtained bshe yields for various elements are presented . . the row structure of the periodic table can clearly be seen on top of a general increase in bshe yield for heavy elements . data courtesy of carl zeiss ag . ] the dependence of the yield on the structure of the periodic table is clearly present . at this point it is important to realize that a high sensitivity is needed from the detector that is used . under standard imaging conditions in him , typically 500 or less ions are used per pixel . assuming a high backscatter yield of 20% and taking into account the detector solid angle , no more than 50 ions will reach the detector . a large fraction of these backscattered he particles are also neutral@xcite due a charge transfer process which occurs once they enter the sample . however , this would represent an ideal case for a heavy element bulk target . in practice these numbers can be substantially smaller and not a single ion should be lost in the detector.@xcite one should realize that contrary to the electrons discussed in the previous section , bshe represent a bulk signal . depending on the atomic number sampling depth , between a few tens of nanometers and a few hundred nanometers can be achieved . as can be seen from equ . ( [ equ : cross - section ] ) , the backscatter efficiency can be increased by lowering the acceleration voltage . as a side effect , this will also reduce the range of the helium and thus the sampling depth . in addition to the qualitative element distribution , also quantitative information on the elemental composition can be obtained . for a binary collision when momentum and energy are conserved the kinetic energy of the backscattered helium @xmath24 depends on the on the ratio of the masses of the impinging ( @xmath25 ) and target ( @xmath26 ) particle . measuring the energy loss at a fixed angle allows the mass of the collision partner , as well as its position relative to the sample surface , to be determined . at high kev or low mev energies this is known as rutherford backscatter spectroscopy ( rbs ) . this method is known to deliver high sensitivity and excellent depth - resolution . the nature of the involved square function makes the method very sensitive to differences between light elements . a silicon drift detector with a resolution of approximately 4kv can be used for this purpose . however , due to the relatively low primary energy of only 35kev , the measured peaks are not as well defined as in classical rbs . nevertheless , the nature of him allows such spectra from areas of only several m@xmath27 to be obtained.@xcite the technique has been successfully applied to measure zro@xmath14 layer thicknesses on si substrates with monolayer sensitivity@xcite and for thickness measurements of cobalt / nickel nano rings.@xcite the achievable resolution is not only limited by the detector . as a result of the already low primary energy , the backscattered particles have a high probability to undergo multiple scattering events . this occurs for the impinging , as well the backscattered helium particle . consequently , a statistical broadening of the exit energies of the bshe exists . it has been shown by simulations that even for an ideal combination , such as a 75 nm thick heavy hafnium oxide film on silicon , an intrinsic uncertainty of 14% exists for the depth resolution of hf . given the current detector resolution this value is further increased to 60%.@xcite compositional characterization will have even larger errors due to a severe peak overlap.@xcite figure [ fig : hfo ] presents an experimentally obtained spectrum from hafnium oxide . the simnra@xcite calculated fits to obtain elemental composition and layer thickness have been added for different real and ideal detector resolutions . an increase of detector resolution to 1kv would substantially improve the quality of the obtained spectra and consequently of the fit accuracy . these curves should be taken to be an indication of what could be possible and not an exact representation of the achievable resolution . one also needs to keep in mind that the majority of the backscattered particles are neutral,@xcite complicating any post scattering treatment . the yield of both signals secondary electrons as well as back scattered helium also depends on the crystalline orientation of the sample with respect to the beam . this allows for two additional contrast mechanisms in him . the well known channeling contrast@xcite can be exploited to obtain the crystal orientation of samples with a large lateral resolution.@xcite this technique makes use of calculations of the blocked area fraction , which yields results similar to stereographic projections of channeling minima or laue back reflection patterns . one of the surprising results is that despite the very small wavelength of he ions , scanning transmission ion microscopy ( stim ) is possible . both dark field and bright field images could be recorded in which thickness fringes and line defects could be identified.@xcite generally speaking , ionoluminescence ( il ) is a phenomenon of light emission due to the optical transitions of an electronic system which has been excited by ion irradiation . three main stages can be distinguished in the luminescence process:@xcite 1 . energy absorption and excitation of the electronic system ; 2 . system relaxation and energy transfer to the emission centers ; 3 . transition of the system into the ground state by photon emission . during sample bombardment with he@xmath28 ions , light can be obtained from the excited backscattered neutral he,@xcite excited sputtered atoms and molecular complexes,@xcite or from the sample material itself.@xcite for the case where emission originates from excited species which have left the sample , the spectrum consists of discrete doppler broadened lines . these lines corresponds to the optical transitions within atomic ( or molecular ) orbitals . this light is usually observed at a distance up to a few millimeters from the target surface.@xcite in the case of organic samples , ionoluminescence reveals the electronic structure of the organic compounds.@xcite for the investigation of material properties we are mainly interested in the luminescence from the sample itself . the physics of the emission processes is usually described by considering the sample s electronic band structure , or by using configuration coordinate diagrams.@xcite according to its origin , there are two general types of luminescence : _ extrinsic _ and _ intrinsic_.@xcite in the case of extrinsic luminescence , the light generation is related to the presence of impurity atoms or ions ( so called _ activators _ ) in the sample material . depending on their electronic structure , activators can be divided in the following groups : 1 . transition metal ions with d@xmath29 electronic configuration ( e.g. ti@xmath30 , cr@xmath30 , mn@xmath31 ) ; 2 . ions with s@xmath32-configuration ( e.g. tl@xmath28 , pb@xmath31 , sb@xmath30 ) ; 3 . earth elements ( ree@xmath33 ) ; 4 . actinides ( e.g. uo@xmath34 , cm@xmath30 ) . sometimes the presence of a certain impurity ( _ sensitizer _ ) is needed for the luminescence of an activator ( for example ce@xmath30 for activation of tb@xmath30 ions @xcite ) . as opposed to sensitizers , _ quenchers _ suppress light emission from an activator . for example , fe@xmath31 ions act as quenchers for emission from mn@xmath31 in apatite.@xcite at high activator concentrations self quenching may occur due to resonant absorption processes . intrinsic luminescence on the other hand , is emission from the sample material itself . two cases can be distinguished . first , optical transitions from delocalized states , or in other words the recombination of free electrons from the conduction band with holes in the valence band . this can include direct and indirect transitions . the radiative recombination of free excitons also falls in this category . second , optical transitions from localized states can also be attributed to intrinsic luminescence . this includes the following processes : recombination of excitons trapped at defect sites ( so called self trapped excitons@xcite ) , emission from excited defects known as color centers ( e.g. nitrogen vacancy centers in diamond@xcite ) , and transitions of charge carriers from delocalized into localized states . the shape and width of the emission peaks and bands strongly depends on electron phonon interactions and thermal effects . a strong electron phonon interaction leads to a stokes shift and peak broadening . as a result , it is desirable to perform ionoluminescence measurements at cryogenic temperatures , which has not been done to date in him . since ionoluminescence is in many aspects similar to the cathodoluminescence ( cl ) phenomenon often observed in scanning electron microscopy ( sem ) , databases of cl studies can be employed for the interpretation of il spectra . ionoluminescence studies are significantly complicated by the fact that an ion beam not only induces light emission , but also directly influences the optical properties of the target due to defect creation . ion irradiation can lead to target coloration ( e.g. in alkali halides@xcite ) and enhanced emission , but also quench the luminescence ( e.g. semiconductor materials@xcite ) . however , the use of him to observe il phenomena allows these processes to be followed in situ with a high lateral resolution . in this section we will try to give an overview of applications of helium ion microscopy that make use of the special imaging capabilities of the him . although , the high resolution is the most prominent fact that allows for very accurate critical dimension measurements,@xcite many successful applications of him make use of other distinct characteristics of him . this overview is by no means complete but will attempt to highlight interesting and eventually surprising imaging applications . the use of electrons for charge neutralization enables him to obtain high resolution images of insulating , and in particular uncoated , biological samples . in fig . [ fig : bio](a ) a high resolution him image of a butterfly wing is presented.@xcite the black ground scales of papilio ulysses and other butterflies are imaged without any prior coating , which allows the smallest features to be identified . the large depth of field in him also makes it ideal for creating anaglyphs using different angles for imaging . high precision measurements of otherwise not accessible feature heights are possible this way.@xcite however , only very few groups have demonstrated the applicability of him for imaging of cells.@xcite this is astonishing since very high resolution should be possible . figure [ fig : bio](b ) demonstrates the achievable resolution on such biological , soft and insulating samples . the imaged protein crystal ( flat bovine liver catalyse ) exhibits a simple rectangular unit cell with a lattice spacing of 8.8 nm @xmath35 6.7 nm . these values are in excellent agreement with the expected values from literature.@xcite the use of bshe for imaging allows amongst other things subsurface processes such as the formation of buried contacts to be visualized . this overcomes an existing limitation in how we currently try to follow subsurface diffusion processes . helium ion microscopy offers a unique , destruction free method to reveal the in plane shape of a diffusion front . in fig . [ fig : buried ] results of a him study on the subsurface formation of pd interconnects are presented.@xcite covered pd@xmath14si interconnects . ( b ) simultaneously recorded bshe image showing the pd deposits ( bright , top and bottom ) and the buried pd@xmath14si layer . the shape of the two pd diffusion fronts is clearly discernible in between the two pd contacts . ] while a large degree of surface detail is present in the se image presented in fig . [ fig : buried](a ) , no direct indication of the buried pd@xmath14si interconnect is visible . however , from the simultaneously recorded bshe image shown in fig . [ fig : buried](b ) the morphology of the pd diffusion front can be seen . the benefit of this method over cross section approaches is evident . besides the obvious ease with which the shape of curved interfaces can be resolved , sample preparation is also substantially easier . knowledge of the non straight nature of the diffusion front is important to understand the reason for large variations in device performance . the shape of the subsurface diffusion front is not accessible using cross section techniques . because the penetration depth of the ions can be varied by using different primary energies , it is also possible to get an estimate of the thickness of the cover layer.@xcite there are several reviews@xcite introducing various applications of il for material characterization that also reveal possible complications of the technique . here , we will focus on il imaging with light ions and results obtained with him . the possibility to do il imaging using a him was demonstrated by boden et al.@xcite the authors tested a wide range of samples such as quantum dots ( qd ) , semiconductors , rare earth doped nanocrystals , and ce@xmath30 doped garnet . both bulk samples and nanocrystals doped with ree were found to exhibit il . this is a result of the optical transitions of the @xmath36 electrons in the dopant . however , the il signal was found to decay with prolonged beam exposure . this decay has been attributed to defect accumulation which hinders electron transport to the actual emitting rare earth atoms . several direct bandgap semiconductor materials were investigated in ref . [ ] , but no il was detected . however , recently we managed to obtain clear ionoluminescence information from such a material . figure [ fig : fig_gan ] shows the il image obtained from a thin gan film on sapphire using a fluence of only @xmath37@xmath38 . beam energy is 35kev . fov : 45 ] the dark lines in the il image ( fig . [ fig : fig_gan ] ) are dislocations that are initially present in the film . these act as centers for non radiative recombination@xcite and appear dark . the black dot with the grey halo in the center of the image is the result of the long term irradiation of a single pixel prior to recording the image . due to the high dose applied there all il is quenched rapidly . this is the result of the creation of various types of defects that provide non radiative paths for the electron de excitation.@xcite in contrast to bulk semiconductors , the authors of ref . [ ] could record il images of agglomerates of semiconductor quantum dots . it is suggested , that crystal defects in qds are generated at a lower rate due to the small size of the particles ( @xmath395 nm ) . hard nuclear collisions occur deeper in the bulk , while in the first tens of nanometers electronic stopping is predominant . the observed emission corresponds to the expected band gap transition . nevertheless , the signal is quenched with increasing ion fluence . while qd aggregates were detected with relative ease , attempts to record il images with higher magnification showing single qds were not successful . alkali halides are known to exhibit intrinsic il as a result of defect production due to the ion beam radiation.@xcite figure[fig : fig_nacl ] demonstrates beam energy is 35kev . fov : 25 ] simultaneously acquired images of residues from nacl solution on a metal substrate . the bright areas in the se image ( fig . [ fig : fig_nacl](a ) ) are the aluminum substrate . the darker areas that exhibit a rough appearance are nacl and residue from the drop cast solution used . the il image ( fig . [ fig : fig_nacl](b ) ) clearly shows the presence of nacl as bright ramified structures . ogawa et al.@xcite have reported il measurements on sio@xmath40 using him . interestingly they observe an increase of the il intensity with increasing fluence . the authors identify several differences between cl and il spectra from the same sample , but the nature of the il signal remains unclear . the use of il for analysis of biological samples is possible , but not very often used . the advantage of il imaging over conventional fluorescence imaging is an absence of diffraction limits that restrict the spatial resolution . in that respect , application of il in him for bio imaging looks promising . few groups have used proton beams with energies in the mev range to investigate biological samples such as cells@xcite and skin tissue.@xcite in another study several promising dyes have been identified and can now be used to stain cells specifically for il investigations.@xcite however , only limited work is done using il for imaging of biological samples . using a classical ion source , a resolution of 150 nm has been demonstrated for ionoluminescence images in stim mode.@xcite graphene , the prototypical thin layer that currently receives a lot of attention , has successfully been imaged by several groups.@xcite this was done for both suspended and free standing graphene . in fig . [ fig : graphene - bubble ] a sio@xmath14 supported graphene flake has been imaged using mild imaging conditions . . using a dose of 1.8@xmath41@xmath38 per image the formation of a he bubble can be observed . as can be seen for the bubble marked in the lower part of the images , holes and edges allow the he to escape and the bubble collapses . at a fluence of 2.0@xmath42@xmath38 the bubble is filled and it collapses two frames later after increasing the dose to 2.3@xmath42@xmath38 . with further increase of the dose the cycle starts again . fov is 20 m . ion doses in @xmath38 are indicated in the figures . ] the flakes can be imaged easily with substantial contrast , despite their ultra thin nature . the sequence of images presented in fig . [ fig : graphene - bubble ] shows the formation and collapse of helium filled bubbles . graphene has been shown to be impermeable for many gases including helium,@xcite hence the backscattered helium remains trapped below the graphene sheet . these bubbles fill with he until either the rim of the flake or a hole in the layer is reached . this will allow the he to escape and the bubble then collapses . this process can be observed for the central bubble between doses of 2.0@xmath42@xmath38 ( initial filling ) and 2.3@xmath42@xmath38 ( collapse and start of refilling ) . the filling of the bubbles and subsequent deformation of the graphene sheet results in the stretching of wrinkles and folds that are present in the sheet . assuming that the bubble has the shape of a spherical cap with a height of 100 nm and a diameter of 4 m we can calculate the pressure under the graphene sheet . using the ideal gas law and a backscatter yield of 0.006 , as obtained from srim@xcite , we arrive at a pressure on the order of 10 mbar . this value agrees reasonably well with the observed pressure range for gas bubbles under graphene on sio@xmath14 obtained elsewhere.@xcite first principles calculations indicate that it should be possible to reveal the graphene lattice using him due to minute changes in the electron emission across the lattice.@xcite however , the realization of such an image might be difficult due to the destructive nature of the helium beam . it is not clear if sufficient signal can be collected before the defects will alter the local atomic configuration.@xcite defects created in graphene due to the ion graphene interaction have been investigated theoretically.@xcite recent experimental investigations indicate that already at very low doses severe damage is done to the graphene sheet and many defects are created.@xcite safe imaging doses for suspended graphene seem to be as low as 10@xmath43 to 10@xmath44@xmath38.@xcite for supported graphene , as shown above , the critical dose is certainly lower due to the more destructive nature of recoils from the support material . an exceptional imaging result has been achieved for carbon nano membranes ( cnm).@xcite such free standing membranes , that are formed by cross linking self assembled monolayers and then removing the substrate , are difficult to image using conventional sem . however , in him these cnms are not only visible , but morphological details such as folds and nanometer sized holes created by slow highly charged ions , are also visible.@xcite given the nature of ultra thin films and the sampling volume of the available secondary imaging particles in him , the ses are the obvious choice for imaging such samples . an example of the successful application of him to soft layers is the visualization of the phase separation in a mixed poly(3-hexylthiophene)/[6,6]-phenyl - c@xmath45-butric acid methyl ester ( p3ht / pcbm).@xcite such blends are typically used in organic photovoltaic solar cells and represent an important materials class . although both molecules are essentially carbon , the different number of @xmath46 and @xmath47 bonds in the two polymers results in minute changes of the se yield.@xcite an other example is the clear identification of self assembled monolayers ( sam ) deposited onto sio@xmath14.@xcite here , differences in work function allow the identification of sam molecules and give rise to the observed contrast . a delicate method of visualizing non continuous ultra thin layers and their different properties utilizes bshe . for a substrate oriented in a channeling condition , a very low bshe signal is recorded due to the extended range of he along the low index channeling direction.@xcite however , at places where a thin surface layer with different atomic positions is present , scattering will occur and an increased bshe signal can be observed . although the relative mass of the elements in the adlayers and the underlying bulk are important , this effect works for any combination of bulk and adlayer elements.@xcite in particular it works for light elements on a heavy substrate.@xcite similar to sem , local electromagnetic fields in the sample will influence the yield , angular distribution and energy of the generated se . this can be utilized to image dopant distribution@xcite and electronic potential distribution . a similar application utilizes static capacitive@xcite contrast to reveal conducting features buried below an insulating cover.@xcite the availability of different gases for gfis such as neon@xcite and the combination with a classic liquid metal gallium focused ion beam ( fib)@xcite makes the technique interesting for various types of materials modifications . the fine beam produced by the gfis has the potential to engineer structures with a length scale that is well below what is currently possible with lmis based fib techniques . contrary to e beam writing , which can achieve similar critical lengths , the removal of material is also possible . writing structures into resists is usually done using gas injection systems ( gis ) in fib or , if higher resolutions are needed , by using e beam lithography . in particular the latter suffers from the so called proximity effect.@xcite the deposition parameters necessary to achieve a constant feature size will depend on the distance to the next feature in the proximity of the beam . this effect is a consequence of the relatively large lateral range of the electrons in the resist material . structures produced by low mass ion beams are less sensitive to this effect.@xcite the near complete absence of this effect is related to the fact that in the surface near region there is practically no scattering when using ion beams ( see also fig . [ fig : charged - part - beams ] ) . furthermore , very little backscattering occurs in the usually relatively light resist materials . in addition the very low energy se ( < 10ev)@xcite in him are localized close to the beam path , whereas the ses in e beam lithography have a higher energy ( > 100ev ) and therefore travel relatively long distances in the resist material.@xcite as a consequence resist modification by ions happens only in a very small volume along the beam path . an extensive overview comparing electrons and ions for their useful application in lithography is given by melngailis.@xcite with the availability of he gfis new standards have been set in terms of achievable line width.@xcite this has at least partially been made possible by the above discussed , near complete , absence of the proximity effect when using a highly focused he beam . however , to date resists initially developed for e beam lithography such as hydrogen sylsesquioxane ( hsq ) and polymethyl methacrylate resist ( pmma ) are used . new resist materials optimized for the large amount of energy deposited by the ion beam could potentially lead to even better results . several precursor materials have been explored including mecpptme@xmath13 , teos , and xef@xmath14.@xcite recently , also a tungsten based metal precursors has successfully been used.@xcite with this extremely flexible method a wide range of features can be directly sculpted including 3d structures . figure [ fig : metal - deposition ] shows examples of structures created using ion beam induced deposition ( ibid ) in the group of p. alkemade.@xcite demagnified copy of a 13th century painting of the chinese poet li bai . ( b ) stacked cross formed by exploiting shadowing of the precursor gas ( fov : 1 m ) . ( adapted with permission from scanning 34 , 90100 ( * ? ? ? copyright 2012 wiley periodicals , inc . ) ] however , a detailed understanding of the deposition process is needed to unleash the full potential of the method for resist patterning . a model has been proposed that identifies primary ions and secondary electrons as being responsible for the vertical growth of the structures . scattered ions and recoil atoms on the other hand are responsible for the lateral growth of the structures.@xcite simulation software ( _ envision _ ) based on this model is available and can be used to predict the metal deposition process in him.@xcite the program is based on available concepts and code of an electron beam induced simulation@xcite and the _ _ ionise__@xcite monte carlo program . experimental confirmation of results showed the applicability of the simulation method.@xcite different from what has been discussed in the previous section , here we will discuss structuring results that do rely on the sputtering of sample atoms . sputtering is a well known phenomenon that has been extensively studied.@xcite however , with only a few exceptions these works usually focus on ions of lower energies of only a few hundred ev to a few kev for the relevant gases helium and neon . for higher energy ranges the bombardment species is either not a noble gas or generally a heavier element . the energy range of interest for him is not so well investigated . surprisingly , many as yet unexplained phenomena , such as very high and fluence dependent sputter yields are observed.@xcite due to the sub nanometer size and the collimated nature of the him ion beam , it is possible to mill features with very small critical dimensions not possible with current fib technology . we start by discussing results obtained on the most simple milled structure a pore . such pores are successfully used for biomolecule detection . the achievable diameter can be as small as 4 nm , which is 60% smaller compared to other one step milling techniques.@xcite most importantly the quality of these pores for biomolecule detection is comparable to state of the art pores fabricated using tem@xcite or fib.@xcite however , these techniques create smallest pores by closing existing bigger pores in a controlled way . the method relies on the fluidization of the material under irradiation and subsequent deformation which is driven by the surface tension and diffusion of ad atoms created by an ion beam.@xcite although it has not been demonstrated to date , this should also be possible by careful implantation of he into the area surrounding the pore . in addition to the effects observed elsewhere,@xcite an additional volume increase and subsequent reduction of pore size can occur due to the formation of nano - sized he bubbles.@xcite substantial redistribution of unsputtered material has already been observed when milling bigger pores.@xcite as a result , comparable diameters should be achievable . recently a gfis using neon@xcite became available . its combination with a classical liquid metal gallium fib column allows for efficient milling of structures at several length scales.@xcite figure [ fig : triple - milling ] shows a structure milled into gold using all three beams . while initial cut outs are made using the lmis gallium source ( fig . [ fig : triple - milling](a ) ) finer details are cut using the gfis and neon ( fig . [ fig : triple - milling](b ) ) . smallest details with a length scale of the order of 10 nm are patterned using helium and the gfis ( fig . [ fig : triple - milling](c ) ) . the same gas source combination is also used for imaging of the structures . another exciting application of helium ion beams for direct write lithography is the successful preparation of devices based on graphene.@xcite the possibility to precisely cut graphene with nanometer precision has been demonstrated earlier by lemme et al.@xcite unfortunately , the unique properties of graphene are very sensitive to damage introduced by the impinging ions.@xcite consequently , only very small fluences of the order of @xmath48@xmath38 can be used to image graphene without damaging it to the extent that the unique properties are lost.@xcite unfortunately , this excludes high resolution imaging of supported as well as free standing graphene . helium implantation is unavoidable in him . it is therefore important to understand the associated defect structures , their evolution and response to external influences . it has to be made clear that the phenomena discussed in this section appear at doses higher than those normally used in him imaging . however , special applications ( bshe or il imaging ) may require relatively high doses . obviously , he implantation plays an important role when the him is used for materials modification , as discussed in the previous section . a dramatic example is shown in ref . [ ] in the context of the fabrication of domain wall magnetoresistance devices . although successful fabrication of domain wall magnetoresistance devices was demonstrated at lower doses , at higher doses the implanted he leads to a swelling and surface deformation that in turn results in the destruction of the patterned devices . initial studies that allow the comparison of experimental damage volumes with srim were made by livengood et al.@xcite one of the results in this publication is summarized in an overview over the different damage regimes that are to be expected for a wide range of he doses ( see fig . [ fig : damage - diagram ] ) . interestingly , this diagram is valid for both semiconductor materials ( silicon ) and metals ( copper ) . the validity of these damage regimes has been demonstrated by several groups . a nice visualization of the bubbles that form at depths comparable to the range of the ions in the material is shown by ref . the pressure inside these bubbles and blisters is high . for the rather soft and ductile metal gold , pressures of up to several gpa can be reached in the initial he nano bubbles.@xcite these bubbles form at open volume defects that are either present in the material or are created by the beam.@xcite the growth of these initial , nano sized , he bubbles is independent of the primary energy and leads to observable materials modification at fluences in the low 10@xmath49@xmath38 range.@xcite however , with increasing he dose these bubbles start to coalesce , provided they are formed deep enough in the sample to not immediately reach the sample surface . what follows is a rapid expansion of a blister on the surface . the pressure inside such a blister of a few hundred mpa@xcite is substantially lower than in the initial nano sized bubbles . the effective formation and rapid growth of he induced blisters is at least partly due to the fact that highly compressed he gas is an ideal stopping medium for he . this results in an amplification of the damage in the vicinity of the bubble and the observed accelerated growth . in the past half decade helium ion microscopy has proven to be an interesting alternative to its direct and well established competitors : scanning electron microscopy and gallium focused ion beam . however , the authors strongly believe , and hope to have demonstrated above , that him is more than a replacement for sem and fib . it excels over sem and fib in particular for * high resolution imaging of uncoated biological samples * imaging of insulating samples * a high surface sensitivity * imaging of ultra thin layers * materials modification with unprecedented resolution * direct write lithography * resist patterning at the same time new applications and techniques are constantly being developed and refined . in particular , analytical additions are needed and are currently being developed . while ionoluminescence has its special applications , additions such as secondary ion mass spectroscopy@xcite still need to prove their applicability in real materials science problems . however , although spectroscopy is currently still in its infancy , the extension of the technique to other gases such as neon , and potentially even heavier ones , guarantees an exciting future of the technique in particular for nano fabrication applications . 141ifxundefined [ 1 ] ifx#1 ifnum [ 1 ] # 1firstoftwo secondoftwo ifx [ 1 ] # 1firstoftwo secondoftwo `` `` # 1''''@noop [ 0]secondoftwosanitize@url [ 0 ] + 12$12 & 12#1212_12%12@startlink[1]@endlink[0]@bib@innerbibempty link:\doibase 10.1116/1.2101792 [ * * , ( ) ] link:\doibase 10.1016/j.ultramic.2004.11.020 [ * * , ( ) ] link:\doibase 10.1021/nl801472n [ * * , ( ) ] link:\doibase 10.1116/1.1526694 [ * * , ( ) ] @noop _ _ , ed . ( , , ) link:\doibase 10.1116/1.2955728 [ * * , ( ) ] link:\doibase 10.1007/bf01342199 [ * * , ( ) ] link:\doibase 10.1063/1.326679 [ * * , ( ) ] link:\doibase 10.1116/1.584328 [ * * , ( ) ] link:\doibase 10.1143/jjap.31.l291 [ * * , ( ) ] http://www.pubmedcentral.nih.gov/articlerender.fcgi?artid=432639&tool=pmcentrez&rendertype=abstract [ * * , ( ) ] link:\doibase 10.1116/1.568497 [ * * , ( ) ] link:\doibase 10.1016/0167 - 9317(85)90022-x [ * * , ( ) ] link:\doibase 10.1116/1.584014 [ * * , ( ) ] link:\doibase 10.1007/bf01329651 [ * * , ( ) ] link:\doibase 10.1016/0169 - 4332(95)00351 - 7 [ * * , ( ) ] link:\doibase 10.1116/1.2357967 [ * * , ( ) ] link:\doibase 10.1007/s00339 - 004 - 2869 - 6 [ * * , ( ) ] link:\doibase 10.1016/j.ultramic.2013.03.013 [ * * , ( ) ] link:\doibase 10.1016/j.ultramic.2012.07.026 [ * * , ( ) ] link:\doibase 10.1063/1.4726112 [ * * , ( ) ] link:\doibase 10.1063/1.3702209 [ * * , ( ) ] link:\doibase 10.1016/j.ultramic.2012.11.005 [ * * , ( ) ] link:\doibase 10.1007/978 - 1 - 4615 - 0765 - 9 [ _ _ ] ( , , ) link:\doibase 10.1007/978 - 3 - 540 - 85916 - 1 [ _ _ ] , , vol . ( , , ) link:\doibase 10.1016/0169 - 4332(93)90301-q [ * * , ( ) ] link:\doibase 10.1016/j.phpro.2008.07.088 [ * * , ( ) ] in link:\doibase 10.1063/1.2799423 [ _ _ ] , vol . , ( , ) pp . in link:\doibase 10.1117/12.886347 [ _ _ ] , vol . , ( , , ) pp . link:\doibase 10.1016/j.ultramic.2009.01.013 [ * * , ( ) ] link:\doibase 10.1017/s1431927609090138 [ * * , ( ) ] link:\doibase 10.1116/1.3258634 [ * * , ( ) ] link:\doibase 10.1103/physrev.59.913 [ * * , ( ) ] link:\doibase 10.1016/j.nimb.2007.12.058 [ * * , ( ) ] in link:\doibase 10.1007/978 - 1 - 4615 - 2840 - 1_4 [ _ _ ] , , ( , , ) pp . link:\doibase 10.1002/sia.2107 [ * * , ( ) ] link:\doibase 10.1103/physrevb.24.4412 [ * * , ( ) ] link:\doibase 10.1016/j.nimb.2008.11.003 [ * * , ( ) ] link:\doibase 10.1134/s1027451010050186 [ * * , ( ) ] link:\doibase 10.1103/physrevlett.84.378 [ * * , ( ) ] link:\doibase 10.1016/j.ultramic.2013.07.010 [ * * , ( ) ] , link:\doibase 10.1016/j.scriptamat.2011.03.013 [ * * , ( ) ] link:\doibase 10.1166/jnn.2012.4260 [ * * , ( ) ] link:\doibase 10.1088/0957 - 4484/24/33/335702 [ * * , ( ) ] link:\doibase 10.1017/s1431927611005514 [ * * , ( ) ] link:\doibase 10.3762/bjnano.3.57 [ * * , ( ) ] link:\doibase 10.1116/1.3502667 [ * * , ( ) ] link:\doibase 10.1116/1.3271254 [ * * , ( ) ] in @noop _ _ ( , , ) in link:\doibase 10.1063/1.59188 [ _ _ ] ( , ) pp . link:\doibase 10.1103/physreva.39.3836 [ * * , ( ) ] link:\doibase 10.1017/s1431927610093682 [ * * , ( ) ] in link:\doibase 10.1007/978 - 3 - 642 - 67112 - 8_5 [ _ _ ] ( , , ) chap . , pp . in link:\doibase 10.1007/978 - 3 - 540 - 87529 - 1_1 [ _ _ ] , , ( , , ) chap . , pp . link:\doibase 10.1016/0031 - 8914(73)90154 - 7 [ * * , ( ) ] link:\doibase 10.1103/physreva.12.876 [ * * , ( ) ] link:\doibase 10.1016/0029 - 554x(65)90154 - 0 [ * * , ( ) ] link:\doibase 10.1016/0029 - 554x(78)90916 - 3 [ * * , ( ) ] link:\doibase 10.1021/ac50009a001 [ * * , ( ) ] in link:\doibase 10.1007/978 - 1 - 4615 - 5361 - 8_5 [ _ _ ] , ( , ) pp . link:\doibase 10.1016/s0168 - 583x(97)00182 - 1 [ * * , ( ) ] in @noop _ _ , ( , , ) ed . , pp . link:\doibase 10.1149/1.2408364 [ * * , ( ) ] link:\doibase 10.1306/d4267ab9 - 2b26 - 11d7 - 8648000102c1865d [ * * , ( ) ] link:\doibase 10.1016/0022 - 3697(90)90144 - 5 [ * * , ( ) ] link:\doibase 10.1016/j.physrep.2013.02.001 [ * * , ( ) ] in link:\doibase 10.1007/bfb0104335 [ _ _ ] , , vol . , ( , , ) link:\doibase 10.1017/s1431927612013463 [ * * , ( ) ] link:\doibase 10.1088/0957 - 0233/22/2/024004 [ * * , ( ) ] link:\doibase 10.1002/sca.20267 [ * * , ( ) ] in http://www.microscopy.org/mandm/2010/jiang.pdf[__ ] ( , ) link:\doibase 10.1038/srep03514 [ * * , ( ) ] link:\doibase 10.1016/s0022 - 5320(67)80038 - 8 [ * * , ( ) ] link:\doibase 10.1016/j.microrel.2012.06.130 [ * * , ( ) ] link:\doibase 10.1016/j.surfcoat.2006.01.075 [ * * , ( ) ] link:\doibase 10.1016/j.nimb.2011.10.070 [ * * , ( ) ] link:\doibase 10.1143/jjap.37.l398 [ * * , ( ) ] link:\doibase 10.1063/1.1337646 [ * * , ( ) ] @noop ( ) link:\doibase 10.1103/physrevb.14.2583 [ * * , ( ) ] link:\doibase 10.1080/00337577908234484 [ * * , ( ) ] link:\doibase 10.1017/s1431927612005922 [ * * , ( ) ] link:\doibase 10.1016/j.nimb.2009.03.069 [ * * , ( ) ] link:\doibase 10.1016/s0168 - 583x(01)00465 - 7 [ * * , ( ) ] link:\doibase 10.1016/j.nimb.2012.12.052 [ * * , ( ) ] link:\doibase 10.1021/nn900744z [ * * , ( ) ] , in http://meetings.aps.org/link/baps.2010.mar.h21.7[__ ] , vol . ( , , ) link:\doibase 10.1021/nn401992q [ * * , ( ) ] link:\doibase 10.1021/nl801457b [ * * , ( ) ] http://www.worldcat.org/isbn/096542071 http://books.google.com/books?vid=isbn096542071 http://www.amazon.com/gp/search?keywords=096542071&index=books&linkcode=qs http://www.librarything.com/isbn/096542071 http://www.worldcat.org/oclc/191686602[__ ] ( , ) link:\doibase 10.1103/physrevlett.109.265505 [ * * , ( ) ] link:\doibase 10.1088/0957 - 4484/22/17/175306 [ * * , ( ) ] link:\doibase 10.1063/1.4726053 [ * * , ( ) ] link:\doibase 10.1103/physrevb.81.153401 [ * * , ( ) ] link:\doibase 10.1016/j.progsurf.2012.05.001 [ * * , ( ) ] link:\doibase 10.1063/1.4792511 [ * * , ( ) ] link:\doibase 10.1021/nl202269n [ * * , ( ) ] link:\doibase 10.1002/app.1990.070391003 [ * * , ( ) ] link:\doibase 10.3762/bjnano.3.58 [ * * , ( ) ] link:\doibase 10.1021/la204437r [ * * , ( ) ] link:\doibase 10.1016/j.susc.2012.02.007 [ * * , ( ) ] link:\doibase 10.1209/0295 - 5075/85/46001 [ * * , ( ) ] link:\doibase 10.1093/oxfordjournals.jmicro.a023780 [ * * , ( ) ] link:\doibase 10.1002/sca.20268 [ * * , ( ) ] link:\doibase 10.1016/j.nima.2010.12.220 [ * * , ( ) ] link:\doibase 10.1017/s1551929512000715 [ * * , ( ) ] link:\doibase 10.1088/0957 - 4484/24/17/175302 [ * * , ( ) ] link:\doibase 10.1116/1.568515 [ * * , ( ) ] link:\doibase 10.1002/sia.740010103 [ * * , ( ) ] link:\doibase 10.1016/0168 - 583x(93)90781-z [ * * , ( ) ] link:\doibase 10.1116/1.3250204 [ * * , ( ) ] in link:\doibase 10.1117/12.862438 [ _ _ ] , vol . , ( ) p. link:\doibase 10.1007/978 - 3 - 7091 - 0424 - 8 [ _ _ ] , edited by and ( , , ) pp . link:\doibase 10.1002/sca.21009 [ * * , ( ) ] link:\doibase 10.1088/0957 - 4484/21/45/455302 [ * * , ( ) ] link:\doibase 10.1017/s1551929511000307 [ * * , ( ) ] in link:\doibase 10.1007/978 - 3 - 7091 - 0424 - 8_11 [ _ _ ] , ( , , ) pp . link:\doibase 10.1116/1.3237095 [ * * , ( ) ] link:\doibase 10.1116/1.4800983 [ * * , ( ) ] link:\doibase 10.1116/1.3517536 [ * * , ( ) ] link:\doibase 10.1088/0957 - 4484/19/41/415704 [ * * , ( ) ] link:\doibase 10.1016/0168 - 583x(87)90004 - 8 [ * * , ( ) ] link:\doibase 10.1016/0370 - 1573(81)90106-x [ * * , ( ) ] and , eds . , link:\doibase 10.1007/978 - 3 - 540 - 44502 - 9 [ _ _ ] , , vol . ( , , ) link:\doibase 10.1088/0022 - 3727/46/19/195501 [ * * , ( ) ] link:\doibase 10.1088/0957 - 4484/22/28/285310 [ * * , ( ) ] link:\doibase 10.1063/1.1947391 [ * * , ( ) ] link:\doibase 10.1038/35084037 [ * * , ( ) ] link:\doibase 10.3762/bjnano.4.53 [ * * , ( ) ] link:\doibase 10.3762/bjnano.3.67 [ * * , ( ) ] link:\doibase 10.1002/sca.21003 [ * * , ( ) ] link:\doibase 10.1116/1.3511509 [ * * , ( ) ] link:\doibase 10.1016/j.mee.2010.11.041 [ * * , ( ) ] link:\doibase 10.1016/j.carbon.2009.12.057 [ * * , ( ) ] link:\doibase 10.1088/0957 - 4484/23/39/395302 [ * * , ( ) ] link:\doibase 10.1116/1.3237101 [ * * , ( ) ] link:\doibase 10.1016/0001 - 6160(87)90034 - 4 [ * * , ( ) ] link:\doibase 10.1103/physrevb.38.1087 [ * * , ( ) ] link:\doibase 10.1063/1.4739240 [ * * , ( ) ] link:\doibase 10.1002/sca.20262 [ * * , ( ) ] link:\doibase 10.1016/j.nima.2010.12.123 [ * * , ( ) ]

in center vortex theory , beyond the simplest picture of confinement several conceptual problems arise that are the subject of this paper . recall that confinement arises through configuration averaging of phase factors associated with the gauge center group , raised to powers depending on the total gauss link number of a vortex ensemble with a given wilson loop . the simplest approach to confinement counts this link number by counting the number of vortices , considered in d=3 as infinitely - long closed self - avoiding random walks of fixed step length , piercing any surface spanning the wilson loop . problems arise because a given vortex may pierce a given spanning surface several times without being linked or without contributing a non - trivial phase factor , or it may contribute a non - trivial phase factor appropriate to a smaller number of pierce points . we estimate the dilution factor , due to these inert or partially - inert vortices , that reduces the ratio of fundamental string tension to the areal density of vortices from the ratio given by elementary approaches and find . then we show how inert vortices resolve the problem that the link number of a given vortex - wilson loop configuration is the same for any spanning surface of whatever area , yet a unique area ( of a minimal surface ) appears in the area law . third , we discuss semi - quantitatively a configuration of two distinct wilson loops separated by a variable distance , and show how inert vortices govern the transition between two possible forms of the area law ( one at small loop separation , the other at large ) , and point out the different behaviors in and higher groups , notably . the result is a finite - range van der waals force between the two loops . finally , in a problem related to the double - loop problem , we argue that the analogs of inert vortices do not affect the fact that in the baryonic area law , the mesonic string tension appears .

hep-th0601090

to some extent , our understanding of area laws in confining gauge theories is based on intuition and plausibility . certainly , there can be no doubt that , in the fundamental representation of @xmath6 , the expectation value @xmath7 of the trace of a simple flat wilson loop @xmath8 involves the area @xmath9 of the flat surface spanning it and not the area of any other spanning surface : @xmath10;\ ; \langle w \rangle = \exp [ -k_fa].\ ] ] [ here @xmath1 is the fundamental string tension , and we use imaginary anti - hermitean gauge potentials , incorporating the gauge coupling @xmath11 in them . ] but , at least in the center vortex picture of confinement , it is not always easy to see how some of these plausible results follow from the basically simple mechanism of confinement , based on linkages of vortices with wilson loops . in this paper we discuss several confinement puzzles , all of them connected by the theme of inert vortices . by this we mean vortices that do not link to wilson loops in the usual way , but occupy space that could have been occupied by truly - linked vortices . for brevity we also use this term to refer to partially - inert vortices that are linked , but with a smaller link number than would naively be expected . we seek the effects of inert vortices only for large wilson loops , those whose length scales are all large compared to the fundamental gauge - theory length @xmath12 . as the loop scales approach @xmath12 the effects of inert vortices either disappear or are substantially modified . in the center vortex picture , the area law arises through group - center phase factors raised to powers depending on the gauss link number of the vortex condensate with the wilson loop . this link number can be calculated from the intersections of a vortex with any surface spanning the wilson loop . to characterize the condensate we will stick for simplicity to d=3 , although there is no real qualitative difference between three and four dimensions . in d=3 , vortices are closed stringlike tubes of chromomagnetic flux , a finite fraction of which have infinite length , and in d=4 they are closed 2-surfaces , whose description raises complications . in @xmath13 we model the vortices as closed self - avoiding infinite - length random walks on a cubical lattice of lattice length @xmath12 . this model is similar in spirit , if not in implementation , to the @xmath14 models for @xmath4 and @xmath5 center vortices given by others @xcite , for use in lattice computations . it is also similar to the usual identification of center vortices as infinitesimally - thick objects , called p - vortices @xcite , that live on a lattice dual to the lattice where wilson loops live . even for the archetypical example of a flat wilson loop for @xmath4 , inert vortices are important . in this simple case , there are many types of inert vortices , including those that pierce the surface twice within a few characteristic lengths @xmath12 , or that pierce it an odd number of times but have link number less than the number of pierce points . the net effect of inert vortices for a simple flat wilson loop is that the density of linkage differs from the density of piercing by a factor @xmath0 , which we call the dilution factor , lying between zero and one . this factor is a rough but useful estimate of the various ways in which vortices can be inert . in sec . [ wloops ] we estimate that @xmath0 lies in the range [email protected] . in the dilute gas approximation ( dga ) , the usual result for the @xmath4 fundamental string tension @xmath1 is @xmath16 ; dilution by inert vortices modifies this to @xmath17 [ for @xmath5 , the standard dga gives @xmath18 ; the diluted dga is @xmath19 , as for @xmath4 . it may be that the @xmath5 dilution factor @xmath20 is not precisely the same numerically as it is for @xmath4 , but our estimates are not accurate enough to see much of a difference . for general @xmath6 there are in principle as many dilution factors as vortex densities . ] it would , of course , be good to compare our estimates with lattice data . unfortunately , it has turned out @xcite to be rather difficult to calculate the density @xmath2 on the lattice , for a number of reasons . among these are the dependence on gauge of the center - vortex location procedures ; effects of gribov copies ; and finite - size effects . @xcite states that lattice artifacts are so important that these authors can not really find a reliable value for @xmath2 . however , @xcite claims a value of @xmath21 of 1.4 , which taken literally might indicate a dilution factor around 0.7 , at least if the dga is more or less correct . the best way to attack the numerical estimation of dilution might be to simulate directly a model of self - avoiding random walks , in the spirit of @xcite , rather than to work with qcd itself . a second major issue arises because link numbers can be calculated ( through stokes theorem ) by counting intersections of vortices with a surface spanning the wilson loop , but any spanning surface can be used for the link - number calculation , not just the surface that ultimately appears in the area law . so it is not clear what area is to be used in the area law , nor even why there is a unique area . we discuss these issues in sec . [ unlink ] , showing how unlinked vortices resolve the paradox of two or more possible areas for a simple wilson loop . the third issue , elaborated in sec . [ 2loops ] , is an interesting variant on the question of when vortices are linked or not . this issue was raised before at a qualitative level @xcite . we consider two identical wilson loops separated by a certain distance , and ask how the overall vev of these two loops depends on separation . for @xmath4 this is a problem somewhat like the corresponding soap - bubble problem , where there are ( at least ) two minimal soap films that can appear for two wire frames close to one another . we can think of each loop as a @xmath22 meson , and our results indicate a van der waals potential between the two loops , which breaks ( as also happens for soap bubbles ) at a critical separation between the loops . similar but more elaborate results hold for @xmath5 . finally , in sec . [ baryon ] we consider the baryonic area law for @xmath5 . this has been explored in center vortex theory @xcite , where it is shown that the area law comes from three surfaces with quark world lines and a central line as boundaries . to some extent the baryonic area law seems to have issues like those of the double wilson loop in sec . [ 2loops ] ; in particular , there might appear to be correlations like the van der waals potential of the two - loop problem that could modify the accepted baryonic area law . however , we show that analogs of the inert vortices in the two - loop problem , which are in fact not inert even though they are in the same geometry except for orientation effects , do not affect the fact that the linearly - rising potential for the three quarks has precisely the mesonic string tension for each of its three sheets . [ conc ] contains a summary . except for secs . [ principles ] , [ wloops ] we use the dga approximation to describe our results . but it turns out to be convenient , for these two other sections , to use a standard form @xcite of the area law that contains the dga as a limiting low - density case . the center - vortex picture for gauge group @xmath6 invokes a vortex condensate . if @xmath23 there are @xmath24 types of vortices labeled by an integer @xmath25 that gives the vortex magnetic flux in units of @xmath26 . a vortex labeled @xmath27 is the antivortex of the vortex labeled @xmath28 . these vortices are characterized by an areal density @xmath29 for each vortex type . by this areal density we mean ( in all dimensions , not just @xmath13 ) that the average number of @xmath27-vortices that pierce any flat surface of area @xmath9 is @xmath30 . there is not much theoretical insight into the values of these different densities for @xmath6 with @xmath31 . however , for @xmath4 and @xmath5 there is only one density , which we term @xmath2 , that sets the scale for the string tension ( in @xmath5 there are two types of vortices , but one is the antivortex of the other and they have the same densities ) . in this paper we will only consider gauge groups @xmath4 and @xmath5 . a condensate of vortices can form only if a finite fraction of them has essentially unbounded length ( or an unbounded number of steps in the random walks describing the vortices ) . only such vortices , long compared to any wilson loop scale , can contribute to the area law . these can be linked or not , depending on the circumstances we encounter . finite - length vortices are therefore inert , in our terminology , and will be accounted for by a renormalization factor that we will not attempt to calculate here . consider now a simple flat wilson loop and the flat surface spanning it . this surface is taken to lie in a plan of the lattice dual to the vortex lattice and is divided into squares of this dual lattice ; we call these @xmath12-squares . any such square is pierced by a single vortex with probability @xmath32 . this probability is related to the areal density of the vortex condensate by @xmath33 the probability that a square is unoccupied is @xmath34 . as on the lattice , @xmath32 can be extracted @xcite from the vev of a square wilson loop one lattice unit on a side . denote this vev as @xmath35 ; then for @xmath4 @xmath36.\ ] ] the only difference from the lattice definition is that in lattice computations the length scale for the wilson loop is not a physical quantity @xmath12 but a lattice spacing , and the lattice version of @xmath32 must be scaled via the renormalization group to find a physical probability , such as used in the present model [ eq . ( [ prob ] ) ] . note that @xmath32 is bounded by 1/2 , since @xmath35 lies between 0 and 1 . moreover , note that the probability @xmath32 as derived from eq . ( [ pdef ] ) is not subject to dilution , which applies only to large wilson loops . the assumption that vortices in different @xmath12-squares are statistically independent leads to the standard argument for center - vortex confinement , which ignores inert vortices . the confining area law ( discarding perimeter effects ) for a wilson loop follows from an ensemble average of center - group elements . each of these is of the form @xmath37 , where @xmath38 is an element of the center of the gauge group as specified by the properties of the @xmath39 vortex ( and the group representation of the loop itself ) . for example , for the fundamental wilson loop in @xmath4 the only non - trivial element of the center has @xmath40 . in the above product , @xmath41 is the gauss linking number of this vortex with the wilson loop . the gauss linking number , a topological quantity , can be written as an intersection number of the vortex and any surface spanning the wilson loop ; its ( integral ) value is independent of the choice of surface . in the @xmath4 case the necessary average is @xmath42\rangle.\ ] ] aside from the assumption of independent @xmath12-squares , the critical assumption for expressing confinement in the center - vortex picture is that @xmath32 is the probability that a vortex is actually linked once to a flat wilson loop . when an odd number of vortices is linked once , the wilson loop has value -1 and when an even number is linked , the value is + 1 . if the assumption is true , the area law follows from multiplying the probabilities @xmath43 of eq . ( [ pdef ] ) for all the @xmath12-squares of the spanning surface . another useful way of expressing this area law is to write out the combinatorics for vortex occupancy of @xmath44 sites of a surface spanning a wilson loop @xmath8 : @xmath45^{n_s}=[1 - 2p]^{a_s/\lambda^2}.\ ] ] here the number @xmath46 of sites on a given spanning surface @xmath47 is @xmath48 , where the surface has area @xmath49 , and each term represents the number of ways of arranging empty and once - filled @xmath12-squares . next we need to modify the area law for dilution , which arises from several factors : if a vortex penetrates an even number @xmath50 of times , there are @xmath51 sites that lead to unit phase factor in the wilson - loop vev , although these sites are occupied . in effect , the vortices filling such sites are inert ( although they may , strictly speaking , be linked to the wilson loop topologically ) . similarly , if a vortex penetrates an odd number ( greater than 1 ) of times it is linked and gives a non - trivial phase factor , but three or more sites are occupied , rather than the single site assumed when we related the string tension and the piercing probability as in eq . ( [ vev ] ) . we will argue in sec . [ alpha ] that the diluted form of the standard equation ( [ vev ] ) is @xmath52^{a_s/\lambda^2},\ ] ] yielding a string tension @xmath53 the dga approximation is the small-@xmath32 limit of either eq . ( [ vev ] ) ( undiluted ) or of eq . ( [ dilstring ] ) [ diluted ; see eq . ( [ kvrho ] ) ] . note that while the effect of dilution on the probability @xmath32 is simply to renormalize it , the effect of dilution on a dimensionless quantity such as @xmath21 can not be characterized as a renormalization of a dimensionful quantity such as the density @xmath2 or the @xmath12-square area @xmath54 : @xmath55 it seems very difficult to resolve dilution problems completely by analytic methods ; the best we can do is to give a semi - quantitative discussion of how these factors renormalize downward the linkage probability from @xmath32 . in this section , dealing with the gauge group @xmath4 , we distinguish between the previously - introduced probability @xmath32 that a vortex pierces a @xmath12-square , thus contributing unit link number , and the density of link number per @xmath12-square , which is what we really need . this density is reduced by inert - vortex effects . we attempt to capture , in some mean - field sense , these effects approximately by introducing a dilution factor @xmath0 that effectively reduces the pierce probability @xmath32 to @xmath56 . we begin by classifying various ways in which vortices can pierce a spanning surface yet not be linked ( in the sense that they are associated with a trivial phase factor ) , or are linked but are to be associated with a reduced density of linkage numbers . for brevity we refer to all these as inert vortices . it is useful to distinguish three types of inert vortices ; only types ii and iii need detailed discussion . type i vortices have finite length , and correspond in some sense to localized particles . the vortex condensate may have some of these , but they can not explain confinement , since for large wilson loops those that are linked contribute only to sub - area effects such as perimeter terms in the vev . presumably the effect of such vortices is essentially to renormalize the areal density of vortices of unbounded length . we will not discuss such finite - length vortices any further , so from now on we are only concerned with the issue of vortices much longer than any wilson loop scale , and the extent to which these are or are not inert . the distinction between the remaining two types of long vortices is this . type ii vortices exhibit what we will call local return , by which we mean that a vortex , however long , penetrating a localized flat surface has its highest probability of returning to that surface after only a few more steps of the random walk . this probability is not to be confused with the probability that a random walk will be near to where it has been after a large number @xmath57 of steps ; this probability decreases like @xmath58 , where @xmath59 is the dimension of space - time ( see the appendix ) . for the present section , dealing with a flat spanning surface , this type is important , but once the renormalization of @xmath32 has been made , they are not very important in the further applications of secs . [ unlink ] , [ 2loops ] . finally , a type iii vortex is one which , having penetrated a surface , penetrates it a second time with high probability after a large number of steps . this can only be true , in view of the above remarks , if the surface is of a special type , including those used in secs . [ unlink ] , [ 2loops ] , and [ baryon ] such as a closed surface or one that has curvature radii comparable to its length scales . an infinite - length vortex must penetrate a closed surface at least twice . a qualitative model of type iii effects might be to assume that after renormalization for the effects of types i and ii vortices , the remaining effects are seen for vortices composed of infinitely - long straight lines intersecting surfaces compounded of flat segments . the rationale for the straight - line vortices is that the local returns have been accounted for by renormalization for type ii vortices . in this model no vortex can return locally to a single surface element and so explicit type ii vortices are missing ; more than one flat surface segment must be involved , and these must form a non - planar surface . we will not use this model in the present paper . among the type ii vortices , the only ones to be considered in this section , there is essentially an infinite number of subtypes . several of them are shown in fig . [ fig1 ] . in these figures , the surface being pierced lies in a plane on a lattice ( shown ) dual to the vortex lattice ( not shown ) . ( a ) simple piercing by a vortex ( thick line ) . ( b ) an inert vortex , penetrating the surface with two more steps after the first piercing . ( c ) an inert vortex that has rendered the square labeled x inaccessible to another vortex . ( d ) a partially - inert vortex , using up three @xmath12-squares for a single linking.,height=384 ] the first , fig . [ fig1](a ) , shows the standard penetration assumed in the usual area - law formula of eq . ( [ vev ] ) : unit link number associated with a single piercing . [ fig1](b ) shows an inert vortex ( zero link number ) produced by two additional steps from the first piercing . [ fig1](c ) shows another inert vortex , using three additional steps , that makes the square labeled x inaccessible for piercing by another vortex , because of the mutual - avoidance requirement . [ fig1](d ) shows a vortex with four extra steps that is linked , but has pierced three @xmath12-squares , thereby again reducing the @xmath12-squares available for more vortices . note that in every case shown in the figure except for fig . [ fig1](a ) the link number is reduced relative to the piercing number . there are other ways for vortices to be inert , for example , a vortex may have a link number that is a multiple of @xmath60 for @xmath6 with @xmath61 . related effects take place in the baryonic area law for @xmath5 ( sec . [ baryon ] ) . the flat surface is divided , as before , into @xmath44 @xmath12-squares . of these , on the average @xmath63 are pierced once by a vortex , and of the pierced squares , a fraction @xmath0 on the average contribute to the area - law formula with a minus sign ( in @xmath4 ) . we use a simple statistical model , ignoring shape effects and certain correlations and assuming that dilution is statistically independent of piercing . it is easy to see that the coefficient of @xmath64^{k_3}\;\;[\sum k_i = n_s]\ ] ] in the formal expansion of @xmath65^{n_s}\ ] ] is the statistical weight for a configuration with @xmath66 pierced @xmath12-squares , of which @xmath67 are going to give a minus sign in the wilson loop vev . we then have @xmath68^{n_s}=[1 - 2\alpha p]^{n_s},\ ] ] showing that , as previously specified , @xmath0 simply renormalizes @xmath32 . for the dga formula for the wilson - loop vev we find @xmath69,\ ] ] in which the string tension is renormalized by the factor @xmath0 from its previous value . the dga is perhaps made more plausible by dilution . not all the effects of inert vortices can be captured by the simple dilution factor defined above . vortices are correlated with each other through self - avoidance , as in fig . [ fig1](c ) , and the specific geometry of the portion of a random vortex walk that penetrates the spanning surface more than once can matter . unfortunately , even simpler problems can not be solved analytically . for example , the statistics of co - existing but mutually- and self - avoiding monomers and dimers ( see @xcite for a mean - field approach and earlier references ) , has no exact solution ( except in the limit of close - packed dimers @xcite ) . this is because dimers and higher multimers do not obey simple ( _ e . g. _ , multinomial ) statistics . one can appreciate this from the observation that on an empty lattice of @xmath44 sites a dimer can be laid down in @xmath70 ways , but only @xmath71 self - avoiding dimers can be put down in total . ( for monomers , of course , these two numbers are the same , namely @xmath44 . ) in the inert - vortex problem there are , in principle , a huge number of multimers , of various shapes and sizes , that should be accounted for . rather than attempting some elaborate generalization of the monomer - dimer problem , we proceed as follows . the problem is to estimate the probability that a self - avoiding random walk of infinite length , having pierced the wilson surface ( which we call @xmath72 ) once , pierces it again one or more times . we need two different types of probabilities for this problem . the first is the standard probability density @xmath73 that a random walk on the @xmath13 lattice is at the lattice point @xmath74 after exactly @xmath60 steps . if the difficult restriction of self - avoidance is dropped , this is given by ( see , _ e. g. _ , @xcite ) @xmath75 [ \cos \theta_1+\cos \theta_2+\cos \theta_3]^n\exp [ i\vec{\theta}\cdot\vec{m}].\ ] ] this is normalized so that @xmath76 and the sum over all @xmath74 of this density yields unity . in the limit of large @xmath60 and components of @xmath74 ( which is an integer - valued vector ) @xmath73 has the usual gaussian form . this is not the probability of real interest , although it can be used in certain circumstances to find the probability that we need . we will call the one that we do need @xmath77 , the probability density that a random walk piercing at the origin re - pierces for the first time at @xmath74 using exactly @xmath60 steps , with the vector @xmath74 restricted to lie in the lattice plane that is nearest neighbor to the wilson surface . the probability density @xmath78 is simply @xmath73 with @xmath74 restricted to the wilson surface . our interest is in probabilities summed over the wilson surface , so we define @xmath79 by the standard rules of probability @xcite @xmath80 which says that the random walk , having pierced the surface for the first time after @xmath81 steps , may penetrate it many times again before ending on the surface after @xmath60 steps . [ note that it takes @xmath82 additional steps of the random walk , at minimum , to re - pierce the surface , given that these steps are counted as starting with the first step after the first piercing . ] if the total number of piercings is odd ( even ) the vortex is linked ( unlinked ) . the case of no further intersections after the first piercing is to be included ; the standard area - law model of sec . [ principles ] is equivalent to assuming that this probability is unity , and all other probabilities are zero . this is far from the case , as we will see . ( [ compprob ] ) is easily solved in terms of generating functions @xmath83 we have @xmath84 note that we may express the elementary solution for @xmath85 , which is @xmath86 in a suggestive way for the original probabilities : @xmath87 ^ 2 + \dots\ ] ] showing how the probability @xmath88 is compounded from probabilities of first return [ see , _ e. g. _ , fig . [ fig1](d ) ] . the final probability of interest is the probability that the random walk ever re - pierces the surface ; this is clearly given by the sum of all the @xmath89 , or by @xmath90 . equation ( [ pvsq ] ) also holds in certain other problems , notably the gambler s ruin problem ( described in the appendix ) , which asks for the probability that a @xmath91 random walk starting at the origin ever returns to it . in this problem backtracking ( re - tracing the last step ) is allowed , and ( as we review in the appendix ) it is straightforward to show that the corresponding probability of ever returning is exactly unity . it is also unity in two dimensions . however , our problem differs from the gambler s ruin problem in two essential ways : no backtracking is allowed , and ultimately the problem becomes three - dimensional , when the radius of gyration @xmath92 of the random walk becomes large compared to the size scale @xmath93 of the wilson loop . [ actually , the radius of gyration for self - avoiding walks has an exponent somewhat different from 1/2 , but we ignore that complication here . ] it has long been known ( as reviewed in @xcite ) that the probability of return to the origin in @xmath13 is finite , with a value of approximately 0.34 . furthermore , self - avoidance completely changes the problem ; for example , a self - avoiding random " walk in @xmath91 has probability zero of ever returning to any site it has reached . let @xmath94 be the probability that a vortex , having penetrated the wilson surface once , never penetrates it again . then , with @xmath95 as the probability that it ever penetrates the surface again , @xmath96 or one may write , as in eq . ( [ qeqn ] ) , a probability sum rule saying that the sum of probabilities of piercing exactly once , exactly twice , etc . , must be unity : @xmath97 ^ 2 + \dots \}.\ ] ] as before , the sum over @xmath98 begins with 2 for self - avoiding random walks . it is now necessary to estimate @xmath99 , or equivalently @xmath95 . unfortunately , this is not a straightforward matter when there is self - avoidance ( see the appendix for a brief review of some of the well - known analysis when this condition is not imposed ) . one way to proceed is simply to count the number of self - avoiding paths going from an original piercing to another piercing as a function of their step length , and look for ways of partially re - summing the results . our proposal , given below , is more accurate for random walks that do not backtrack than for true self - avoiding walks ; the difference is that a non - backtracking walk may violate the condition of self - avoidance by looping back on itself . we propose the following expression for non - backtracking walks , valid for @xmath100 : @xmath101.\ ] ] the corresponding generating function , approximately valid for self - avoiding walks , is @xmath102.\ ] ] the explanation of the terms is as follows . the factor 4/25 is the probability for fig . [ fig1](b ) , for @xmath103 . for a non - backtracking walk in @xmath13 there are 5 possible choices to add a new step to the random walk . the horizontal step in this figure has probability 4/5 ( note that we are summing over all possible sites ) , and the next , vertical , step has probability 1/5 . the appearance at larger @xmath81 of 3/5 in the formula expresses the probability that a non - backtracking random walk will take its next step horizontally ( with respect to the wilson surface ) , and 1/5 is the probability of a vertical step in one particular direction ( up or down ) . [ the restriction to even powers of 1/5 is easily understood by drawing a few figures , in the style of fig . [ fig1 ] . ] so a random walk once started in a horizontal plane has a three times larger probability of staying in that plane than of moving to a plane higher above the wilson surface . it is easy to check the combinatorics of the low - order powers of 1/5 in eq . ( [ smallqeqn ] ) , and also the highest powers , for example , the last term ( 4/25)@xmath104 . this is the probability that the random walk goes as high as possible , takes one horizontal step , and then returns straight down to the wilson surface . this can only happen for even @xmath81 , and one easily sees that the maximum attainable height is @xmath105 . we have compared the approximation of eq . ( [ smallqeqn ] ) with explicit counting of self - avoiding walks through @xmath106 and find that the difference between self - avoiding and non - backtracking is acceptably small . the difference between non - backtracking walks and self - avoiding walks first appears at @xmath107 , where there are 144 non - backtracking walks but only 128 self - avoiding walks . this is about an 11% error , but because the erroneous contribution to the @xmath107 walks is 16/3125 @xmath108 0.0052 , out of a total of about 0.533 , the error in the final result from @xmath107 terms is less than 1% . in any case , to deal with true self - avoidance rather than just non - backtracking raises the same issues as having a condensate of mutually - avoiding vortices , and we do not discuss that issue in any detail in this paper . we find an approximation to @xmath95 by summing over all @xmath81 in eq . ( [ smallqeqn ] ) , with the result @xmath109=\frac{8}{15}\simeq 0.533.\ ] ] at this point one might guess that the dilution factor @xmath0 should be equal to the probability @xmath99 of never returning . but this is not quite right ; we really need the probability of being linked or unlinked . to do this , one should separately find the probabilities that there is an even or odd number of piercings , as expressed by separating the even and odd powers of @xmath110 in eq . ( [ feller2 ] ) . we use @xmath111 for the probability that a vortex is linked , and @xmath112 for the probability that it is unlinked ; these are defined by @xmath113 now it seems that @xmath114 . if we to use our approximation @xmath115 we would find the link probability ( dilution factor ) to be about 0.65 , somewhat larger because we are counting three , five , @xmath116 piercings as well as a single piercing . ( [ linkunlink ] ) is not quite right either , because while @xmath111 expresses the probability of linkage , when ( say ) three sites are used for the link instead of one , there is dilution that must be accounted for . we account for this by calculating a sum of weighted probabilities , where the weight for the appearance of @xmath98 powers of @xmath110 in eq . ( [ linkunlink ] ) is the inverse of the number of sites occupied by the linked vortex , or @xmath117 . for example , when @xmath118 three @xmath12-squares are pierced , as shown in fig . [ fig1](d ) . this result for @xmath0 , when modified by the original piercing probability @xmath32 , should give something like the link number per unit area , in effect increasing the size of @xmath12-squares to account dilution . the weighted sum for @xmath111 gives another estimate for @xmath0 : @xmath119.\ ] ] for @xmath115 this gives @xmath120 . it is plausible that @xmath95 lies between the no - self - avoidance @xmath13 value of 0.34 and the @xmath91 value of 1 . in that case , the estimate of @xmath0 with no inverse - site weighting [ eq . ( [ linkunlink ] ] gives a finite value even in the @xmath91 limit , where we find @xmath121 , so the dilution factor is 1/2 . this is a singular limit , because the original probability @xmath99 used to construct @xmath122 vanishes , but this is cancelled by a singularity in the sum over @xmath123s . it is not surprising that the limiting probabilities are each 1/2 , since there is no way of distinguishing an even number of piercings from an odd number . the value of @xmath0 from this equation for the @xmath13 value @xmath124 is about 0.75 . ( [ newalpha ] ) , with inverse - site weighting , gives 0.75 for @xmath124 and 0 for @xmath125 . presumably this latter case is unrealistic , because at some point the radius of gyration of the random walk is large compared to the wilson loop scale @xmath93 and the problem really is three - dimensional ; in any case , the @xmath91 self - avoiding case is completely opposite to the no - self - avoidance case . to be more accurate in estimating @xmath0 , one would need to account for mutual avoidance effects such as shown in fig . [ fig1](c ) , and be more precise about weighting various random - walk configurations . we will not attempt that here , and close by saying that our estimates for the dilution factor are consistent with @xmath126 . in the following sections we look for further manifestations of inert vortices of type iii , going beyond the local effects that led to the dilution factor . but these local effects still occur , and so everywhere in our arguments the original probability @xmath32 should be replaced by the diluted probabilities @xmath127 now we return to the problem that the spanning surface used for counting link numbers as piercings is arbitrary , yet there surely is a unique area in the area law . the simplest possible case , where there is no doubt as to the answer , is that of a flat wilson loop , and one can get the right area law by using the flat spanning surface in stokes theorem . yet one should also get the right answer by using any other surface . how does this come about ? ( [ fig2 ] ) shows such a flat wilson loop , with two spanning surfaces . the first , labeled @xmath128 , is flat and is correct . the second is labeled @xmath47 . we choose orientations so that the combined surface @xmath129 is oriented . of course , this surface is closed . a flat wilson loop , and two spanning surfaces , labeled @xmath128 and @xmath47.,height=288 ] we must now improve upon the techniques outlined in sec . [ unlink ] , to account for inert vortices of type iii . the calculation of the area law based on the flat surface @xmath128 needs no change . but what if we instead wished to calculate the area law based on surface @xmath47 ? here there is extra dilution . the vortices linked to surface @xmath47 are still those linked to surface @xmath128 , which must pass through surface @xmath128 and surface @xmath47 once each , but there are also inert vortices , which pass through surface @xmath47 twice and surface @xmath128 not at all . [ we will not account for vortices linking three , five , @xmath116 times , most of which are included in the dilution factor . ] we denote the number of linked vortices by @xmath130 and the number of inert vortices by @xmath131 . the factor of 1/2 in the latter definition simply reflects the fact that every inert vortex pierces surface @xmath47 twice , so that @xmath132 is the total number of pierce points of inert vortices . it is not convenient to introduce this factor of 1/2 for linked vortices , because the diluted pierce probability @xmath133 , introduced above , is related to the number @xmath130 of vortices linked to surface @xmath128 by @xmath134 where @xmath135 is the area of surface @xmath128 . that is , @xmath130 is the number of pierce points of vortices on surface @xmath128 . the total number of vortex piercings of the combined surface @xmath129 , denoted @xmath136 , is @xmath137 with the two arising because linked vortices penetrate both surface @xmath47 and surface @xmath128 . by hypothesis of a uniform areal density of vortices , this total number of vortex pierce points on the combined surface @xmath129 is also given by @xmath138 combining these equations , one finds @xmath139 to calculate the area law in the dga , we can easily use the formulas of sec . [ principles ] for the flat @xmath128 surface , which has a total of @xmath140 @xmath12-squares : @xmath141^{n_{\sigma}}\rightarrow e^{-k_f a_{\sigma}},\ ] ] with @xmath142 as before . it is more interesting to calculate it from the point of view of the other spanning surface @xmath47 . here we must account for the diminished probability of occupation of this surface by linked vortices , since some of them , as counted by @xmath143 , are inert . so we change the probabilities by adding to @xmath144 , the probability of no occupation , the probability @xmath145 of occupation by an inert vortex of type iii . this gives @xmath146 similarly , the new link probability @xmath147 is @xmath148 using eq . ( [ sigma ] ) with the new probabilities for the surface @xmath47 one gets , going to the dga limit as before : @xmath149^{n_s}\rightarrow e^{-2(\tilde{p}/\lambda^2 ) a_s } = e^{-2(\hat{p}/\lambda^2 ) a_{\sigma}}.\ ] ] consider the compound wilson loop shown @xmath150 in fig . [ fig3 ] , composed of two identical but oppositely - oriented wilson loops . [ the orientation is irrelevant in @xmath4 . ] they are separated transversely by a distance @xmath151 . the relevant expectation value is @xmath152 two identical but oppositely - oriented rectangular wilson loops 1 and 2 , separated by a distance @xmath151 . ( a ) when @xmath151 is small compared to loop scales , the wilson - loop area , labeled 3 , connects the loops . ( b ) when @xmath151 is large , the wilson - loop areas are the disjoint areas , labeled 1 and 2 , of each loop.,height=288 ] when the distance @xmath151 between the loops is small compared to the loop , intuition @xcite suggests that the surface whose area should appear in the area law , as in eq . ( [ wloop ] ) , is the minimal surface 3 joining the two loops as shown in fig . [ fig3](a ) . when @xmath151 is large , intuition suggests that the situation in fig . [ fig3](b ) holds , where each loop is spanned by its own minimal surface with no connection to the other surface . we will show that intuition is indeed correct for the compound area law of center vortex theory , and give an approximate interpolation formula for intermediate values of @xmath151 . begin with the @xmath4 case , where the orientation of the wilson loop does not matter , and @xmath153 . as usual , we assume that the time extent @xmath154 , the spatial extent @xmath155 of the wilson loop , and their separation @xmath151 are all large compared to the qcd scale length @xmath12 and assume that @xmath156 . in this limit the spanning surface @xmath157 of fig . [ fig3](a ) is nearly flat , and has area @xmath158 . ( we ignore the contribution @xmath159 from the top and bottom . ) the wilson loop surfaces @xmath160 have areas @xmath161 . imagine now a configuration where all surfaces @xmath162 exist , so that there is a closed surface with two marked contours , the wilson loops 1 and 2 . this constitutes a minor generalization of the configuration already considered in sec . [ unlink ] . there are several ways that vortices can be linked or inert ( in the sense of sec . [ types ] ) , after the renormalization of type i and type ii vortices . use the notation @xmath163 for the total number of vortices penetrating surface @xmath164 . these obey @xmath165 each of these is subdivided as follows : the number of vortices piercing @xmath166 and @xmath157 an odd number of times is called @xmath167 , with analogous notation for @xmath168 ; the number piercing surface @xmath169 and @xmath170 each an odd number of times is @xmath171 ; and the number entering @xmath157 and then exiting the same surface is @xmath172 . as indicated in sec . [ probsurface ] , the probability that a vortex known to be at a point remote from a surface that then penetrates the surface ( once ) is proportional to the surface area , and is finite when the qcd scale length @xmath12 vanishes . if we require the number of vortices penetrating one surface , say @xmath169 , and then another surface , say @xmath170 , that probability is proportional to @xmath173 . so we now assume that @xmath174 these vortex - linking numbers are related to the total vortex - piercing numbers by @xmath175 eqs . ( [ 2loopn],[nassump],[sumrule ] ) are easily solved to yield @xmath176 we will now compute the expectation value @xmath177 of the product @xmath178 of the two wilson loops , from the point of view of the surfaces @xmath160 . only the number @xmath167 contributes non - trivially to an @xmath4 wilson loop . as in sec . [ unlink ] we introduce modified probabilities @xmath179 just as in calculating the wilson loop vev from eq . ( [ newsurf ] ) , we have in the dga : @xmath180=\exp -[\frac{2k_ftrz}{r+z}].\ ] ] this formula is only approximate , but it shows features that we believe are generally correct ( and one feature that is not correct ) . for example , the heavy - quark potential @xmath181 [ coefficient of @xmath182 in the exponent of eq . ( [ newsurf2 ] ) ] has the behavior @xmath183 in the limit @xmath184 , showing that the the area law of two @xmath4 wilson loops [ or , as in fig . [ fig3 ] , two oppositely - directed wilson loops ] disappears as the two loops approach each other and form an @xmath60-ality zero configuration . ( for @xmath60-ality zero loops there is a pseudo - area law , coming from the finite size of vortices @xcite , at distances @xmath185 , but that is irrelevant here . ) in the opposite limit of @xmath186 we find @xmath187 , so the vev is just the product of the separate vevs for two wilson loops . while this in itself is correct , the approach to this limit can not be , for it would yield a residual potential @xmath188 which vanishes only like @xmath189 . in actuality , at some point when @xmath190 the spanning surface switches on the length scale @xmath12 from @xmath157 to @xmath191 , much as the corresponding soap - bubble surface would for two physical wire loops 1 and 2 . we do not see this breaking because we have not included effects coming from a network of gluon world lines running from loop 1 to loop 1 , from loop 2 to loop 2 , and from loop 1 to loop 2 . the simplest step in the formation of this network has been discussed @xcite in connection with baryonic and mesonic hybrids , having extra gluons along with their quark content . there it is shown that a single gluon acts like a physical string which separates a wilson - loop spanning surface into two surfaces , and the string requires extra energy to be stretched if the string stretching leaves the minimal spanning surface . the fluctuations associated with this stretching should yield the lscher @xcite term , as well as leading to the transition from surface @xmath157 to @xmath191 as the gluons in the network recombine . but when this stretching is not too great the potential should behave roughly as given in eq . ( [ newsurf2 ] ) . in @xmath6 with @xmath61 the orientation of the wilson loops does matter . if the loops have opposite orientation , as shown in fig . [ fig3 ] , the problem is essentially the same as for @xmath4 . but one should also consider the problem of two loops of the same orientation . since for @xmath5 the treatment of this problem is quite similar to that for the baryon , which is a compound of three wilson loops , we defer further discussion to sec . [ baryon ] . it is by now well - established both in center vortex theory @xcite and on the lattice @xcite that , as shown in fig . [ fig4 ] , in @xmath5 the heavy - quark baryonic potential has three wilson - loop surfaces spanning a boundary consisting of the quark world lines , with the three lines coinciding along a central line ( the so - called y law ) . a baryonic wilson loop in @xmath5 is composed of three simple wilson loops sharing a common central line ( expanded in the figure ) . the central line is invisible to @xmath5 center vortices.,height=288 ] this is a three - fold compound wilson loop , sharing some of the features of the double loop discussed in sec . [ 2loops ] . in particular , a vortex may pierce two loops . such a vortex is not inert , as were the vortices piercing surfaces 1 and 2 of fig . [ fig3 ] and discussed in the previous section . instead , linkage with these analogs of inert vortices are important to establish that the string tension in the baryonic area law is precisely the mesonic string tension @xmath1 . the analysis of the dilution factor @xmath0 for @xmath5 is slightly more complicated , because a vortex piercing twice is not unlinked as it is in @xmath4 ; it is simply equivalent to an antivortex piercing once . given the complete symmetry between vortex and antivortex , this means that an antivortex piercing twice is equivalent to a vortex piercing once . there is a modification of the numerical value of the dilution factor , since it is now possible to dilute unit link number by occupying a minimum of two sites , rather than a minimum of three as for @xmath4 . however , the string tension is still lessened by a single dilution factor @xmath0 , just as for @xmath4 , and we can continue to use the notation developed for that case . we will not pursue the question of what the value of the dilution factor is for @xmath5 ; it must be quite similar to that for @xmath4 . the @xmath5 version of the standard area law for a single wilson loop , given in eq . ( [ vev ] ) for @xmath4 , is @xmath192\}^{a_s/\lambda^2}=[1-\frac{3\hat{p}}{2}]^{a_s/\lambda^2}.\ ] ] here we use @xmath193 as the probability ( accounting for dilution ) that a vortex of flux @xmath194 pierces the surface , with equal probability that the antivortex of flux @xmath195 pierces it ; this means that , as before , @xmath196 . this leads immediately to the dga @xmath5 string tension @xmath197 . to discuss the baryonic area law in similar terms to those used with the double wilson loop , introduce a surface made of flat pieces , labeled 4 and shown in fig . [ fig5 ] , that forms a closed surface when combined with the surfaces spanning wilson loops 1 and 3 . we introduce a new surface labeled 4 that creates a closed surface consisting of the two wilson - loop surfaces for quarks 1 and 3 plus a surface joining these.,height=288 ] in what follows we will be careful to distinguish vortices from antivortices , although at the end the symmetry between them makes this simply a distinction for convenience of exposition . as in sec . [ 2loops ] we introduce the number @xmath167 as the number of times that the same vortex ( not antivortex ) pierces both spanning surface 1 and surface 3 an odd number of times , and @xmath198 is the number of times a vortex pierces both surface 1 and 4 . we consider that the dilution factor @xmath0 has been applied , so we can take this odd number to be just one . the number @xmath199 is the number of vortex piercings on surface 1 , and it obeys @xmath200 the factor of 2 in the denominator arises because there are just as many antivortices piercing any given surface ; it is the same factor of two dividing the explicit probabilities in the area law of eq . ( [ su3law ] ) . any vortex that pierces surface 1 once must leave through either surface 3 or surface 4 . if it leaves through surface 4 it is associated with a phase factor @xmath201 $ ] ; if it leaves through surface 3 it is associated with a phase factor @xmath202 $ ] . for any such vortex configuration there is an equally probable antivortex configuration , with the phase factors interchanged . so the _ total _ number of piercing , vortices plus antivortices , is @xmath203 , and the number of antivortices piercing surface 3 is @xmath167 , etc . this , plus the relation of eq . ( [ baryonpierce ] ) , means that the standard calculation of the baryonic area law using surface 1 ( plus surfaces 2 and 3 ) , giving the y law , is exactly the same as accounting for the linkages of vortices and antivortices through surfaces 1 and 4 , plus linkages through surfaces 1 and 3 ( plus permutations for the other two quarks ) . there really are no unlinked type iii vortices in this problem , the way there were in secs . [ unlink ] , [ 2loops ] . the baryonic string tension is precisely the mesonic string tension , because when it is calculated from the three quark surfaces 1 , 2 , and 3 separately , without regard to the other two quarks , it is just the standard simple wilson loop calculation . but it is just the same when linkages to two quarks are considered . of course , no one doubted the equality of string tensions for baryons and mesons , but the point was to see how it worked out in the center vortex picture . we have discussed semi - quantitatively some of the effects of inert vortices that do not couple as effectively as they might to a wilson loop , and that change the probability of linkage by a dilution factor @xmath204 . in turn , this changes the @xmath4 dga string tension from @xmath16 to @xmath205 , where @xmath2 is the areal density of vortices piercing a large flat surface . our estimate , based on partial sums of non - backtracking walks , is @xmath206 . understanding inert vortices also leads to understanding of how it is that there is a unique area in the area law , even though any surface spanning a wilson loop is suitable for counting link numbers in the center vortex picture . we found a finite - range van der waals force , due to inert vortices , that links two mesonic wilson loops , in this case vortices that are simultaneously linked to both wilson loops . and finally we showed that the analogs of inert vortices in the two - loop problem do not interfere with the usual formulation of the area law for @xmath5 baryons , based on considering the three wilson loops as independent . there seems to be no possibility of a detailed analytic approach to these problems , which therefore are best studied further with lattice computations . one can , of course , create a center vortex condensate through simulation of the underlying non - abelian gauge theory , but it would also be very interesting and possibly simpler to study inert - vortex effects with simulations that begin with an _ a priori _ vortex model , similar in spirit to the approach of @xcite , rather than having to deal with all the complications of the full gauge theory . indeed , it seems that relating @xmath1 and @xmath2 on the lattice has numerous complications @xcite , and has not been attempted recently . but the simulation of two wilson loops seems approachable on the lattice . this appendix discusses two issues : 1 ) a brief review of well - known analysis for return probabilities for random walks with backtracking allowed ; 2 ) the probability for a vortex to pierce two separated wilson loops . the first serves for cautionary notes in the problem calculating the dilution factor of sec . [ wloops ] , and the second arises in finding the area law for the compound wilson loop of sec . [ 2loops ] . for the most part and for dimension @xmath207 , analytic results are only available for random walks without self - avoidance constraints , and we will only discuss those in the next subsection . the only problems that one has any real chance to analyze are for non - self - avoiding walks , and we will review a famous one , the gambler s ruin problem , here . the basic probability concepts are the same for self - avoiding walks , and so this review should be helpful to those unfamiliar with the underlying ideas . the cautionary note here is that while the @xmath91 gambler s ruin problem seems superficially quite similar to those of sec . [ wloops ] , in practice they are very different , both because the questions of sec . [ wloops ] have three - dimensional effects and because they deal with self - avoiding walks . we ask for the probability @xmath208 that an unbiased random walk starting from the origin on a lattice of points in @xmath91 will return to the point @xmath209 for the first time after @xmath98 steps . the case @xmath210 is the standard gambler s ruin problem . for an unbiased walk ( all step probabilities equal to @xmath211 in @xmath59 dimensions ) plya long ago proved that in the limit of infinite steps the probability of return to the origin is unity in @xmath212 but less than one in all other dimensions this is perhaps plausible from the fact that the probability of being at the origin after @xmath60 steps in @xmath59 dimensions is , for @xmath213 , @xmath214 and the sum over @xmath60 diverges at large @xmath60 in @xmath212 but not in higher dimensions . as in sec . [ wloops ] , the relation between the @xmath215 and the @xmath216 is @xmath217 note that in the sum on the right @xmath218 has @xmath219 set equal to zero , because we are compounding the probability of first return to @xmath219 , as given by @xmath220 , with the probability of being at this same point @xmath219 after more steps . it is straightforward to find @xmath215 and @xmath221 . the probability @xmath215 is the standard random - walk probability , given by @xcite @xmath225^n\exp [ im\theta ] .\ ] ] from this we find the generating function ( for @xmath209 ) @xmath226\}^m(1-s^2)^{-1/2},\ ] ] and the gambler s ruin ( @xmath210 ) generating function is @xmath227^{-1/2}.\ ] ] this yields for the gambler s ruin problem the well - known result @xmath228^{1/2}.\ ] ] one learns from this that the probabilities for first return after @xmath98= 2 , 4 , 6 , 8,@xmath116 steps are 1/2 , 1/8 , 1/16 , 5/128,@xmath116 independent of the total number of steps in the random walk ( if this number is larger than @xmath98 ) , and that the probability of ever returning is unity . this follows from eqs . ( [ app1 ] , [ genfunct ] ) , which shows that @xmath229 diverges in @xmath91 ( and also in @xmath230 ) but not in higher dimensions . for non - zero @xmath219 we have @xmath231 it then follows from eqs . ( [ meqn ] , [ generate ] , [ mgenfunct])that @xmath232\}^m.\ ] ] by expanding in @xmath233 , one sees that the probabilities vanish for @xmath234 , as expected , and by setting @xmath235 in @xmath222 one sees that the probability of ever reaching @xmath219 is unity . the probability @xmath216 peaks for @xmath236 , when @xmath219 is comparable to the vortex radius of gyration . now consider calculating the probability of first return anywhere on an infinite plane in @xmath13 . the probability @xmath73 of going from the origin to lattice point @xmath74 in @xmath60 steps on an infinite lattice is given in eq . ( [ 3dprob ] and the corresponding generating function is @xmath237 \{1-\frac{s}{3}[\cos \theta_1+\cos \theta_2+\cos \theta_3]\}^{-1}\exp [ i\vec{\theta}\cdot\vec{m}];\;p_3(0;\vec{m})=\delta_{\vec{m},\vec{0}}.\ ] ] take @xmath238 in eq . ( [ 3dgen ] ) and sum from @xmath239 to @xmath240 over @xmath241 . the result is , not unexpectedly , a simple variant on the @xmath91 gambler s ruin problem , and yields @xmath242^{-1/2}.\ ] ] using a standard expansion and eq . ( [ d1q ] ) we find the probabilities @xmath243 : @xmath244 where the @xmath245 are gegenbauer polynomials . [ [ compoundloops - probability - that - a - vortex - pierces - two - separated - wilson - loops - sec.2loops ] ] [ compoundloops ] probability that a vortex pierces two separated wilson loops [ sec . [ 2loops ] ] ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ the next problem comes up in the compound wilson loop estimates of sec . [ 2loops ] , where a vortex can pierce two separated wilson loops . in @xmath13 , consider the plane surface of dimensions @xmath246 centered on and perpendicular to the @xmath151-axis at a distance @xmath247 along this axis from the origin . we ask for the probabilities @xmath248 to end up on this surface after @xmath60 steps ; the probability @xmath249 of reaching this surface for the first time after @xmath60 steps ; and the probability of ever reaching it . also to be calculated are the corresponding generating functions @xmath250 and @xmath251 . we assume that all lengths @xmath252 are large in lattice units , so that the number of steps is also large . the first probability is a sum over the surface of the probability given in eq . ( [ 3dprob ] ) : @xmath253 the sum is easily done to yield @xmath254 [ \cos \theta_1+\cos \theta_2+\cos \theta_3]^n\exp [ im\theta_3]\frac{\sin [ ( l_1 + 1)\theta_1 /2]}{\sin [ \theta_1 /2]}\frac{\sin [ ( l_1 + 1)\theta_2 /2]}{\sin [ \theta_2 /2]}.\ ] ] in the large-@xmath60 limit @xmath255 , one finds an area factor emerging : @xmath256 [ \cos \theta_1+\cos \theta_2+\cos \theta_3]^n\exp [ im\theta_3]\ ] ] where @xmath257 is the area of the surface . this happens because the integrand is only appreciable when @xmath258 . the generating function for this probability , if needed , is constructed as in eq . ( [ 3dgen ] ) , including summing over @xmath259 . the probability @xmath249 is determined by an analog of eq . ( [ feller ] ) , including a sum over surface variables : @xmath260 where the sum over @xmath259 is delimited as in eq . ( [ p3llm ] ) . to see what this sum means , write the @xmath261-probability in fourier form @xmath262 with @xmath263 . we find , using eq . ( [ 3dprob ] ) , @xmath264}{\sin[(\theta_j+\alpha_j+1/2 ] } \int \frac{d^3\alpha}{(2\pi ) ^3}[(\sum \cos \alpha_i)/3]^{n - j}.\ ] ] we anticipate , and can confirm later , that in the limit @xmath265 the @xmath266 are of order @xmath267 and the @xmath268 are of order @xmath269 . it turns out that @xmath270 is large compared to @xmath81 ( which is of order @xmath271 ) and so we can drop the @xmath272 in the argument of the sine functions in eq . ( [ qfourier1 ] ) . the @xmath0 integral then factors out , and is given by the probability @xmath273 of returning to a given point after @xmath270 steps . the problem is now solved , in principle , by using generating functions , as in eq . ( [ mgenfunct ] ) . or one can study the sum of eq . ( [ qfourier1 ] ) directly , and find by a scaling argument that the maximum value of the @xmath261-probability on the right - hand side behaves like @xmath274 . note that when lattice lengths @xmath252 are converted to physical lengths by multiplying by @xmath12 this probability remains finite . 99 m. engelhardt and h. reinhardt , nucl . b * 585 * , 591 ( 2000 ) ; m. engelhardt , m. quandt , and h. reinhardt , nucl . b685 * , 227 ( 2004 ) ; m. quandt , h. reinhardt , and m. engelhardt , phys . d * 71 * , 054026 ( 2005 ) . j. greensite , prog . . phys . * 51 * , 1 ( 2003 ) . l. del debbio , m. faber , j. giedt , j. greensite , and s olejnk , phys . d * 58 * , 094501 ( 1998 ) ; r. bertle , m. faber , j. greensite , and s olejnk , jhep * 0010 * , 007 ( 2000 ) ; m. faber , j. greensite , and s olejnk , jhep * 0111 * , 053 ( 2001 ) . de forcrand and m. pepe , nucl . b * 598 * , 557 ( 2001 ) . k. langfeld , o. tennert , m. engelhardt , and h. reinhardt , phys . b * 452 * , 301 ( 1999 ) . j. m. cornwall , phys . d * 69 * , 065019 ( 2004 ) . j. m. cornwall , phys . d * 69 * , 065013 ( 2004 ) . a. m. nemirovsky and m. d. coutinho - filho , phys . a*39 * , 3120 ( 1989 ) . h. n. v. temperley and m. e. fisher , phil . mag . * 6 * , 1061 ( 1961 ) ; m. e. fisher , phys . rev . * 124 * , 1664 ( 1961 ) ; p. w. kastelyn , j. math . * 4 * , 287 ( 1961 ) . e. w. montroll , j. siam * 4 * , 241 ( 1956 ) . w. feller , _ an introduction to probability theory and its applications _ , wiley , new york , 1951 . j. m. cornwall , _ workshop on non - perturbative quantum chromodynamics _ k. a. milton and a. samuel ( birkhuser , boston , 1982 ) , p. 119 ; phys . rev.d*57 * , 7589 ( 1998 ) . j. m. cornwall , phys . rev . d * 71 * , 056002 ( 2005 ) . m. lscher , nucl . b180 * , 317 ( 1981 ) ; m. lscher , k. symanzik , and p. weisz , nucl . phys . * b173 * , 365 ( 1980 ) . t. t. takahashi , h. matsufuru , y. nemoto , and h. suganuma , phys . lett . * 86 * , 18 ( 2001 ) ; phys . rev . d*65 * , 114509 ( 2002 ) . c. alexandrou , p. de forcrand , and o. jahn , mucl . b ( proc . suppl . ) * 119 * , 667 ( 2003 ) .

we report results from a pilot program to obtain spectroscopy for objects detected in the _ galaxy evolution explorer _ ( _ galex _ ) medium imaging survey ( mis ) . our study examines the properties of galaxies detected by _ galex _ fainter than the sloan digital sky survey ( sdss ) spectroscopic survey . this is the first study to extend the techinques of to estimate stellar masses , star formation rates ( sfr ) and the b ( star formation history ) parameter for star - forming galaxies out to . we obtain redshifts for 50 _ galex _ mis sources reaching ( ab mag ) , having counterparts in the sdss data release 4 ( dr4 ) . of our sample , 43 are starforming galaxies with , 3 have emission line ratios indicative of agn with , and 4 objects with are qsos , 3 of which are not previously cataloged . we compare our sample to a much larger sample of,000 matched galex / sdss galaxies with sdss spectroscopy ; while our survey is shallow , the optical counterparts to our sources reach magnitudes fainter in sdss _ r _ than the sdss spectroscopic sample . we use emission line diagnostics for the galaxies to determine that the sample contains mostly star - forming galaxies . the galaxies in the sample populate the blue sequence in the vs color - magnitude diagram . the derived stellar masses of the galaxies range from 10 to 10 m and derived sfrs are between 10 and 10 m . our sample has sfrs , luminosities , and velocity dispersions that are similar to the samples of faint compact blue galaxies studied previously in the same redshift range by , , & . however , our sample is mag fainter in surface brightness than the compact blue galaxies . we find that the star - formation histories for a majority of the galaxies are consistent with a recent starburst within the last 100 myr .

0706.3424

following the successful launch and early operations of the _ galaxy evolution explorer _ satellite in 2003 , we decided to undertake an initial spectroscopic assay of sources detected by _ galex _ using the deimos multiobject spectrograph @xcite on the keck ii telescope . although _ galex _ data have been matched to relatively deep optical surveys @xcite our study is the first to extend the modeling technique of @xcite from @xmath14 to @xmath15 . the power of the _ galex _ medium imaging survey to select interesting objects is noteworthy . we recall that the mis observations are 1500 sec in duration , imaging fuv and nuv simultaneously ( @xcite & @xcite ) . the _ galex _ satellite has investigated the ultraviolet universe to @xmath16 with the goal of determining among other things , the star formation rate ( sfr ) of galaxies in the local universe , @xmath17 @xcite and the evolution of the global star formation density out to @xmath16 @xcite . the ultimate goal of such work along with other surveys and observations of galaxies at similar and higher redshifts is to constrain the baryonic physics of galaxy formation and evolution . questions still remain as to how the sfr depends on different factors such as environment , mass , morphology etc . @xcite . previous work deriving sfrs with a uv - selected sample was carried out by @xcite who derive sfrs both from uv luminosities at 2000 and h@xmath18 luminosities from optical spectra for a sample of galaxies at redshifts of @xmath19 detected in the uv by the balloon - borne telescope foca @xcite . estimates of local star formation rates before this were only possible through either optical emission line luminosities of the recombination lines of hydrogen , mostly h@xmath18 , and [ oii]@xmath203727 or the far infrared luminosity from 10 - 100@xmath21 m @xcite . in this paper we present the observations and results from the pilot program for a sample of objects detected by _ galex _ and sdss with spectra obtained from the deimos spectrograph at the keck ii telescope . the analysis of this sample is enhanced greatly by the techniques of salim et al.(2005 ) that have been expanded to include the derivation of galaxy physical parameters such as stellar mass , from _ galex _ + sdss photometry and redshifts . the main goal of this paper is to characterize the galaxies detected by _ galex _ that are fainter than sdss spectroscopic galaxy sample , in terms of mass , sfr , and uv - optical color . we are predominantly ( about two - thirds of the sample ) seeing the faint blue galaxy population @xcite and are also sensitive to qsos . while _ galex _ observations have been undertaken in many deep fields with much higher median redshifts @xcite we have the advantage in this sample of obtaining spectroscopy with keck / deimos , which gives wavelength coverage sufficient to derive some detailed physical parameters from the spectra . in 2 we describe the sample and the photometric and spectroscopic data . redshift and emission line flux measurements from the spectroscopy are presented in 3 . the distributions of uv color , uv - optical color , and the sdss _ r _ magnitude for the sample are described in 4 . the derived galaxy parameters from sed fits are described in 5 . we give a discussion and summary of our findings in 6 and 7 . throughout this paper we assume _ km s@xmath10 mpc@xmath10 , @xmath23=0.3 , and @xmath24=0.7 . _ galex _ is a nasa small explorer mission aimed to survey the uv emission from galactic and extragalactic sources from 700 km circular orbit @xcite . _ galex _ images the sky simultaneously in two bands , the far - uv ( fuv 1344 - 1786 ) and the near - uv ( nuv 1771 - 2831 ) . each _ galex _ circular field is 1.25 deg . in diameter . we use fuv and nuv magnitudes and magnitude errors derived in elliptical apertures . the photometry is taken from the _ galex _ internal release 1.1 ( ir1.1 ) . the _ galex _ mis photometry for tile number 10273 has a limiting magnitude , m@xmath25(ab)@xmath26 . we use optical photometry for our objects obtained from sdss data release 4 ( dr4 ) @xcite . the sdss photometric data are taken with the 2.5 m telescope at apache point observatory . imaging is obtained in _ bands @xcite . the imaging data are photometrically @xcite and astrometrically @xcite calibrated . an overview of the sdss data pipelines and products can be found in @xcite . our sample is derived from objects with detections in the _ galex _ medium imaging survey ( mis ) tile 10273 , centered at a right ascension / declination of 17@[email protected]@xmath29 @xmath30 57@[email protected]@xmath33 . the _ galex _ field being much larger than the deimos slit mask , we selected objects from the _ galex _ field to populate two deimos slit masks . the two deimos fields were initially chosen to maximize the number of mis objects with optical counterparts in the sdss classified as galaxies with @xmath34 photometry from an early version of the _ galex _ data reduction pipeline . the number of galaxies matching this criteria was much smaller than the number of available slits , and so to take full advantage of the deimos field of view , the slits were populated by any galaxy detected by both _ galex _ and sdss . in addition , the remainder of available slits in both fields was populated with blue stellar - like objects ( qsos and white dwarfs ) , main sequence outlier stars ( @xmath35 or @xmath36 ) , and bright alignment stars . the first deimos field contains 26 objects classified as galaxies in sdss , and 10 blue stellar objects and main sequence outlier stars . the second deimos field contains 19 objects classified as galaxies in sdss , and 9 blue stellar objects and main sequence outlier stars . the spectroscopic data were obtained at the keck ii telescope with the deimos multi - object spectrometer on october 1 , 2003 @xcite . spectra were obtained using two slit masks , for a total 64 spectra . the 830 grooves mm@xmath10 grating was used with a slit width of @xmath37 giving a resolution of @xmath38@xmath39fwhm . the signal - to - noise ratio obtained over the continuum of each spectrum varied between sources from 1 to @xmath40 , depending on the brightness of the source . the integration time for both slit masks was 30 minutes . flux calibration of the spectra was not performed as no flux standards were observed . the spectra cover a wavelength range of @xmath41 . figure 1 shows a panel of several one dimensional spectra for star - forming galaxies in the sample at a range of redshifts , plotted in the rest wavelength , and figure 2 shows the spectra for objects with measured @xmath3 . in all , we measured 50 redshifts , 45 for sources classified as galaxies in sdss and 5 for sources classified as stars in sdss . of those five , three are qso s at @xmath3 and the other two are star - forming galaxies . the remaining 14 of the sources classified as stars by the sdss have stellar spectra . the redshifts for objects with detected emission lines were measured by eye . for objects with @xmath42 , multiple emission lines were detected including either h@xmath18 and/or h@xmath43 and [ oii]@xmath203727 , [ oiii]@xmath204959 , [ oiii]@xmath205007 , [ nii]@xmath206584,[sii]@xmath206717 , and [ sii]@xmath206731 . for all objects where the doublet [ oii]@xmath203727 was detected , h@xmath18 was redshifted beyond the wavelength range of the detector . for one source , at z=1.028 , we detected two features that we assigned as [ oii]@xmath203727 and mgii@xmath202800 emission . for the other three sources with @xmath3 we find only 1 feature that we identify as mgii@xmath202800 , based on the width of the line and the absence of other features in the spectrum . figure 3 shows the redshift distribution of the sample . the mean redshift of the sample is @xmath44 . two of the sources in the deimos sample have corresponding sdss spectra . the redshifts obtained for these two objects agree with the sdss redshifts . one of the objects with sdss spectra is the qso at @xmath45 and the other is a galaxy at @xmath46 . both the deimos spectra and the sdss spectra for the latter object have significant detections of only h@xmath18 and [ nii]6584 . we measure spectral emission line fluxes and equivalent widths for [ oii]@xmath203727 , h@xmath43 , [ oiii]@xmath204959 , [ oiii]@xmath205007 , h@xmath18 , [ nii]@xmath206584 , [ sii]@xmath206717 , and [ sii]@xmath206731 . fluxes , equivalent widths and errors for the emission lines were measured in idl using the mpfit function . the continuum of each spectrum was first fit by a polynomial , then the emission lines were simultaneously fit with gaussians . table 1 lists the objects , their equivalent widths and flux ratios of [ oiii]@xmath205007/h@xmath43 and [ nii]@xmath206584/h@xmath18 for objects with a 3@xmath47 measurement of at least one of the above lines . we did not perform a reddening correction to the fluxes , since the standard procedure requires either detections of both h@xmath18 and h@xmath43 or a measurement of the radio continuum @xcite . the effect of this on the flux ratios of [ oiii]@xmath205007/h@xmath43 and [ nii]@xmath206584/h@xmath18 is thought to be negligible due to the small wavelength separation of the emission lines . in the deimos sample we only have detections of both h@xmath18 and h@xmath43 for 18 objects , all at @xmath48 , and chose for consistency not to perform corrections on any of the spectra . we note that the foreground galactic extinction , e(b - v ) , calculated from the dust maps of @xcite is @xmath49 magnitudes for the field . fourteen objects have measured fluxes of h@xmath43 , [ oiii]@xmath205007 , h@xmath18 , and [ nii]@xmath206584 . figure 4 shows the emission line diagnostic of @xcite used to distinguish star - forming galaxies and type ii agn . the solid curve is the boundary between star - forming galaxies and agn determined through modeling of starburst galaxy spectra by @xcite . a star forming galaxy with only @xmath40% of the optical emission line flux due to an agn would lie above the star - forming / agn boundary @xcite . the dashed curve is the more stringent demarcation used by @xcite to distinguish between agn and star - forming galaxies in sdss . the shaded histogram shows the distribution of emission line ratios for a sample of 51,000 sdss / galex objects ( see 4 ) . the curve of @xcite distinguishes between the two different sources of emission line flux in this large sample with the agn occupying the parameter space to the right of the distribution of star - forming galaxies . in the deimos sample only three galaxies of the fourteen have emission line flux ratios above this curve , implying that some portion of the flux is due to an agn , though it is still within the errors that the emission line flux for two of these objects results entirely from star - forming regions inside the host galaxies . objects can also be agn if @xmath50 [ oiii]/h@xmath51 or @xmath50 [ nii]/h@xmath52 . another 13 objects also have only detections of h@xmath43 and [ oiii]@xmath205007 . of the thirteen , one has a flux ratio of @xmath50 [ oiii]/h@xmath51 , indicative of an agn . @xcite find a similar ratio of galaxies classified as star - forming / agn from examination of emission lines in their sample of uv selected galaxies . the deimos sample is @xmath53 magnitudes deeper in _ r _ than the sdss spectroscopic sample . figure 5 shows a histogram of the sdss _ r _ magnitudes for the sample plotted with the sdss _ r _ magnitude for _ galex _ mis sources in ir1.1 with matches in the sdss dr2 spectroscopic sample . the ir1.1/dr2 sample consists of @xmath54 galaxies at redshifts @xmath55 with derived star - formation histories ( sfhs ) by sed fitting ( see 5 ) . the ir1.1/dr2 histogram is normalized to the size of the deimos galaxy sample . the ir1.1/dr2 sample is mostly bounded in _ r _ by the sdss spectroscopic survey limits , @xmath56 @xcite . using the redshifts determined for the deimos sample we calculate the absolute _ r _ magnitude , m@xmath57 , of galaxies in the sample , k - corrected to the mean redshift of the ir1.1/dr2 sample , @xmath58 . k - corrections for all bandpasses were calculated using the publicly available code of @xcite version 4_1_4 . in figure 6 we construct a color magnitude diagram ( cmd ) of @xmath59 versus @xmath60 . we overplot our sample onto the ir1.1/dr2 sample , plotted as a shaded contour plot . in the figure the diamond symbols correspond to deimos objects with spectroscopically measured redshifts at @xmath61 , and the crosses correspond to objects with spectroscopically measured redshifts at @xmath62 . the contours enclose 40% and 80% of the objects in the ir1.1/dr2 sample . the cmd of the ir1.1/dr2 sample clearly shows the bimodality of galaxies seen by _ galex _ in the nearby universe , @xmath61 , with distribution peaks at blue @xmath59 colors of @xmath63 and at red @xmath5 colors of @xmath64 @xcite . we henceforth call these the blue and red sequences . the galaxies in the blue sequence are generally late - type in morphology and have spectra with emission line - ratios indicative of star - formation @xcite , while the galaxies in the red sequence typically have absorption line spectra and lower star - formation rates than blue sequence galaxies . the majority of the objects in the deimos sample lie in the region of the diagram occupied by blue star - forming galaxies in the ir1.1/dr2 sample , mostly along the blue edge of the distribution , with only one having a @xmath5 color greater than 4 . the spectrum for this object , at @xmath65 , is shown in figure 1 . the spectrum shows absorption features : nad , and mgib , but also shows weak emission features of h@xmath18 , [ nii]6584 , and [ sii]6717,6731 . in the bottom - left panel of figure 7 we again plot the color magnitude diagram for the sample , highlighting the objects with @xmath66 detections ( filled circles ) . the other 3 panels of this figure show the uv color distribution for the 24 galaxies in the deimos sample with @xmath66 detections . plotted for reference are galaxies from the ir1.1/dr2 sample with @xmath66 detections shown as the shaded countour plots or shaded histogram in the respective panels . besides the 3 galaxies in the deimos sample with @xmath67 the uv color for this sample shows a similar distribution with the ir1.1/dr2 sample . in the fuv cmd the objects with m@xmath68 show a range of uv color not seen in the local blue sequence galaxies of similar optical luminosity . both figures 6 and 7 show that this sample probes the type of galaxies that one obtains when looking at objects detected by _ several magnitudes deeper than the sdss spectroscopic limit . we derive the following galaxy parameters according to the approach of @xcite : the v - band dust attenuations , a@xmath69 in magnitudes , stellar metallicity z , the current star formation rate , sfr , averaged over the past 100 myr in m@xmath9 yr@xmath10 , the present - day stellar mass , m@xmath70 , of the galaxy in m@xmath9 , the fraction of stellar mass formed in bursts over the last 100 myr , f@xmath71 , and the @xcite b parameter , defined as the ratio of the current sfr to the past time - averaged sfr ( averaged over the estimated age , not hubble time ) . the galaxy parameters are derived from model libraries of galaxies at redshifts between 0.1 and 1.6 at redshift increments of 0.1 for the deimos sample , and at redshifts of 0.05,0.1,0.15,0.2 , and 0.25 for the ir1.1/dr2 sample . each library consists of up to @xmath410@xmath72 models . each model is parameterized according to galaxy age , optical depth , star formation history ( sfh ) , and metallicity . the sfh of each model is parameterized according to @xcite , with an underlying , continuous , exponentially declining sfr upon which bursts of star formation , random in time and amplitude , are superimposed . dust attenuation in each model is parameterized using the prescription of @xcite using an effective @xmath73-band optical depth @xmath74@xmath69 and absorption curve , @xmath75 resulting from both giant molecular clouds and the diffuse ism , with the fraction @xmath21 of @xmath74@xmath69 contributed only by the diffuse ism . the @xmath73-band optical depth from giant molecular clouds is taken to only affect stars younger than 10 myr . a description of the prior distributions of the model parameters is discussed in @xcite . spectral energy distributions ( seds ) are created for each galaxy in the library using the population synthesis code of @xcite . the model seds are convolved with the _ galex _ and sdss filter response curves . statistical estimates of physical galaxy parameters are derived by comparing the observed 7 band galex@xmath30sdss fluxes of each galaxy to all the convolved model seds in the nearest redshift library . probability density functions ( pdfs ) for each physical parameter are created by assigning weights to the parameters . the @xmath76 goodness of fit of the model determines the weight ( @xmath77 $ ] ) that is assigned to the parameters of that model . the median ( or most typical parameter value ) of the pdf is taken as the estimate of the galaxy parameter . we perform sed fits for galaxies not identified as qsos . in table 2 we list the galaxies , their redshifts , @xmath78 non k - corrected colors , and their derived parameters . the parameters derived from the sed fits are not used unless the reduced @xmath76 fit of at least one model is below 10 . only two of the 46 galaxies do not meet this criterion . figures 8 and 9 show the derived parameters for the deimos sample of galaxies overplotted on the ir1.1/dr2 sample plotted as shaded contour plots . the contours are labeled and encompass 52% , 84% , and 97% of the data . the diamonds as before are objects with @xmath61 , and the crosses are objects with @xmath62 . the objects span the range of derived parameters of the ir1.1/dr2 sample for z , and a@xmath69 . the derived metallicities , @xmath79 $ ] , for the sample are not very well constrained . the typical error , @xmath80 of the difference between the 97.5 percentile and the 2.5 percentile of the pdf , of the derived metallicities are @xmath8143% z@xmath9 . the derived m@xmath70 range from 10@xmath7 to 10@xmath8 m@xmath9 with an average error of @xmath82dex , but the majority of the m@xmath70 are at or below 10@xmath83m@xmath9 , about an order of magnitude below the mean of the ir1.1/dr2 sample . the derived sfrs lie between @xmath84 and @xmath85 m@xmath86 , with five galaxies having derived sfrs greater than 10 m@xmath86 . the average error on the derived sfrs is typically @xmath87dex . comparing the sfrs between galaxies with similar @xmath5 color or m@xmath88 , the galaxies at @xmath62 have sfrs about an order of magnitude greater than the galaxies at @xmath61 . this is mostly a selection effect since the more distant galaxies are also more luminous on average . a comparison of these sed derived sfrs with h@xmath18 derived sfrs can not be performed since the spectra are not flux calibrated and many lack h@xmath18 . the range of sfrs we find for this sample is similar to the range of sfrs for local uv galaxies found by @xcite derived both from h@xmath18 luminosity and from the uv luminosity , but is an order of magnitude larger than the sfrs found by @xcite in their sample of goods - north treasury keck redshift survey galaxies at @xmath89 when compared to galaxies in the deimos sample with similar redshifts . we find that the galaxies are largely starbursts with @xmath90 . this is shown in figure 10 . the distribution of @xmath91 for objects in the deimos sample with sed fits are plotted with the distribution for the ir1.1/dr2 sample normalized to the number of objects in the deimos sample with sed fits . the deimos sample shows a highly peaked value of @xmath91 compared to the ir1.1/dr2 sample which has two small peaks near @xmath92 and @xmath93 , showing that most of these galaxies currently have less star - formation now than in the past . the value @xmath94 that describes most of the galaxies in the deimos sample indicates that while these galaxies have less star formation than in the past , they are currently or have recently gone through a burst of star - formation . two of the galaxies have @xmath95 indicating that these galaxies are going through major starbursts in their histories . from a statistical standpoint , the sed fits reveal with a 95% reliability that at least three galaxies have not had a burst of starformation in the last 100 myr . half of the galaxies in the sample could have formed as much as @xmath4 10% of their stellar mass in a burst within the last 100 myr , 10 could have formed up to @xmath96 and five of the less massive systems could have had up to @xmath97 of their stellar mass form in bursts within the last 100myr . in the plot of @xmath98 vs @xmath6 our sample lies along a `` blue sequence '' of star forming galaxies and they have among the bluest colors for star forming galaxies in the local universe found by _ galex _ and sdss . while all the galaxies show blue @xmath99 color this is not a homogenous sample . fifteen of the galaxies show disk structure in the sdss images . the remaining two - thirds of the sample fit into the heterogeneous class of faint blue galaxies at intermediate redshifts previously studied by @xcite and @xcite in the hubble deep field and adjoining fields , and by @xcite and @xcite in the canada - france redshift survey . the faint blue subset of the deimos sample has similar luminosities , sfrs and optical colors to the compact galaxies found in the hubble deep field by @xcite , though none are `` compact '' ( optical half - light radii @xmath100.5 . ) this is a selection effect in our sample . galaxies with half - light radii this small are classified as stars in sdss and would not have been selected for spectroscopy in our sample . in figure 11 we plot , as filled diamonds , the rest frame absolute b magnitude versus the b magnitude surface brightness and sfr versus velocity dispersion . the absolute b magnitudes are calculated from the observed bandpasses using the k - correction code of @xcite , and the surface brightness are calculated with the half - light sdss r petrosian radii . the velocity dispersions are derived from the measured linewidths of [ oiii]5007 for objects with @xmath101 detections of the emission line . for comparison , we also plot , as stars and squares , the faint blue galaxy samples of @xcite and @xcite making the necessary corrections to our adopted cosmology . while our sample of galaxies have luminosities , sfrs , and velocity dispersions comparable with the previous samples , the size selection effect separates our sample from the previous samples in surface brightness . @xcite and @xcite propose that the subset of the faint blue population of galaxies with compact geometry ( half light radii @xmath100.5 ) and narrow emission lines ( @xmath102 ) will fade to become dwarf spheroidals by z=0 , while @xcite claims that the most luminous of these are too massive to become dwarf spheroidals , and instead will become the bulges of spiral galaxies . the most massive intermediate redshift galaxies ( @xmath103 ) in our sample could likely follow this latter evolution path . indeed , we see that the galaxies in our sample that show extended structure at low redshifts ( @xmath48 ) are among the the most massive of the sample . the conclusions of @xcite and @xcite rest partially on the assumption that galactic winds from the last starburst event will remove the remaining gas from these systems and halt star formation causing them to fade several magnitudes by z=0 . through modeling the uv and optical broadband colors we find that about half the galaxies in our present sample could have formed at most @xmath104 of their stellar mass in bursts within the past 100 myr , and only fifteen ( all having @xmath105 ) could have formed over 20% of their stellar mass within the last 100 myr . if the last starburst event removes all remaining gas from these systems , as @xcite propose , then why did the previous starforming events in these galaxies that produced the majority of the stars in these systems not halt star formation ? our derived starformation histories of these galaxies argues that the current star formation events will not halt starfromation in every galaxy . but what percentage of these will continue to experience further bursts of star formation is unknown . however , even in the low mass systems of the present sample , residual star formation or another burst of star formation is ultimately unlikely to change the evolutionary outcome proposed by @xcite and @xcite for the most compact low mass galaxies in our sample . we suggest that the more extended sources of our sample are more likely to become present - day dwarf irregulars rather than dwarf spheroidals . we have presented deimos spectra for objects detected by _ galex _ in the mis survey with imaged counterparts in sdss ; a total exposure time of 30 min per slitmask was used . galex has proven to be a sensitive instrument for wide field galaxy surveys . we have shown that the _ galex _ medim imaging survey followed up with a 30 m integration with keck / deimos yields redshifts and line measurements for star forming galaxies to z@xmath4 0.7 , and has yielded 4 qso s 3 of which are previously un cataloged . the matched sample reaches approximately 3 magnitudes fainter in _ r _ than the sdss spectroscopic survey limits . the sample is not a homogenous sample , but is indicative of the types of galaxies forming stars out to z@xmath106 . we have derived physical parameters for these galaxies from the seds and compared this sample to a sample of 50,000 sdss galaxies with galex detections . we find that roughly one - third of the galaxies are starforming late type disk galaxies , four are qsos at @xmath3 , and the remaining galaxies are faint blue low mass starbursts . approximately 3 out of 14 star formimg galaxies show emission line ratios indicative of an agn . a similar fraction was found by @xcite in their uv selected sample . the masses of the galaxies are typically lower than what is found locally in the sdss spectroscopic sample . the range of @xmath107 for the deimos sample spans from @xmath108 to @xmath109 , whereas the median of the sdss spectroscopic sample is @xmath110 . + + 4 . the sfrs of the galaxies at @xmath62 are roughly an order of magnitude greater than the sfrs for the galaxies at @xmath61 . besides three of the most massive , and reddest in nuv -r color , the remaining galaxies show evidence of a starburst in the last 100 myr , with @xmath111 . fifteen of the galaxies in the lower mass range of the sample could have formed more than 20% of their stellar mass in bursts of star formation within the last 100 myr . our sample has similar velocity dispersions , sfrs , and b luminosities to previous samples of faint blue galaxies , though the galaxies in our present study are 2 mag fainter in surface brightess . _ galex _ is a nasa small explorer , launched in april 2003 . we gratefully acknowledge nasa s support for construction , operation , and science analysis for the galex mission , developed in cooperation with the centre national detudes spatiales of france and the korean ministry of science and technology . the authors wish to recognize and acknowledge the very significant cultural role and reverence that the summit of mauna kea has always had within the indigenous hawaiian community . we are most fortunate to have the opportunity to conduct observations from this mountain . the analysis pipeline used to reduce the deimos data was developed at uc berkeley with support from nsf grant ast-0071048 . abazajian , k. et al . 2004 , , 128 , 502 baldwin , j. a. , phillips , m. m. , & terlevich , r. 1981 , pasp , 93 , 5 bertin , e. , & arnouts , s. 1996 , a&as , 117 , 393 blanton , m. et al . 2003 , , 125 , 2276 brinchmann , j. , charlot , s. , white , s. d. m. , tremonti , c. , kauffmann , g. , heckman , t. , & brinkmann , j. 2004 , mnras , 351 , 1151 bruzual , g. & charlot , s. 2003 , mnras , 344 , 1000 charlot , s. & longhetti 2000 , mnras , 323 887 charlot , s. & fall , s. m. 2000 , apj 539 , 718 davis m. et al . in discoveries and research prospects from 6- to 10-meter - class telescopes ii . edited by guhathakurta , puragra . proceedings of the spie , volume 4834 , pp . 161 - 172 ( 2003 ) ellis , r. , s. 1007 , araa , 35 , 389 faber , s. m. , et al . 2003 , , 4841 , 1657 fukugita , m. , ichikawa , t. , gunn , j. e. , doi , m. , shimasaku , k. , & schneider , d. p. 1996 , , 111 , 1748 guzm@xmath112n , r. , koo , d. c. , faber , s. m. , illingworth , g. d. , takamiya , m. , kron , r. g. , & bershady , m. a. 1996 , apj , 460 , 5 guzm@xmath112n et al . 1997 , apj , 489 , 559 guzm@xmath112n , r. , jangren , a. , koo , d. c. , bershady , m. a. , & simard , l. 1998 apj 495 , 13 hammer , f. , gruel , n. , thuan , t. x. , flores , h. , & infante , l. 2000 , apj , 550 , 570 hogg , d. w. , finkbeiner , d. p. , schlegel , d. j. , and gunn , j. e. 2001 , , 122 , 2129 kauffmann , f. et al . 2003 , mnras , 341,33 kennicutt , r. c. , ara&a , 36 , 189 kennicutt r. c. , bresolin , f. & garnett , d. r. 2003 , apj , 591 , 801 kewley , l. j. , dopita , m. a. , sutherland , r. s. , heisler , c. a. , & trevena , j. 2001 , apj , 556,121 kobulnicky , h. a. & kewley , l. 2004 , apj , 617 , 240 koo , d. c. , guzm@xmath112n , r. , faber , s. m. , illingworth , g. d. , bershady , m. a. , kron , r. g. , & takamiya , m. 1995 , apj , 440 , 49 martin , c. d. et al . 2005 , apj , 619 , 1 martin , c. d. et al . 2005 , apj , 619 , 59 milliard , b. , donas , j. , laget , m. , armand , c. , vuillemin , a. 1992 a&a , 257 , 24 morrissey , p. et al . 2005 , apj 619 , 7 mallen - ornelas , g. , lilly , s. j. , crampton , d. , & schade , d. 1999 , apj , 518 , 83 osterbrock , d. e. 1989 , astrophysics of gaseous nebulae and active edmunds , m. g. , blackwell , d. e. , chun , m. s. , & smith , g. 1979 , mnras , 189 , 95 phillips , a. c. et al . 1997 , apj , 489 , 543 pier , j. r. , munn , j. a. , hindsley , r. b. , hennessy , g. s. , kent , s. m. , lupton , r. h. , and ivezic , z. 2003 , 125 , 1559 salim , s. et al . 2005 , apj , 619 , 39 salim , s. et al 2007 , arxiv:0704.3611 scalo , j. m. 1986 . _ fundam . _ 11:1 - 278 schiminovich , d. et al . 2005 , apj , 619,47 d.j . schlegel , d.p . finkbeiner , & m. davis , apj , 500 , 525 ( 20 june 1998 ) smith , j. a. et al . 2002 , , 123 , 2121 stoughton , c. et al . 2002 , , 123 , 485 strauss , m. a. et al . 2002 , aj , 124 , 181 sullivan , m. , treyer , m. a. , ellis , s. r. , bridges , t. j. , milliard , b. , donas , j. 2000 , mnras , 312 , 442 tremonti , c. a. et al . 2004 , apj , 613 , 898 treyer , m. a. et al . 2005 , apj , 619 , 15 wyder , t. k et al . 2005 , apj , 619 , 15 wyder , t. k. 2006 , in preparation 587725578037494738 & 0.474 & 7.23 @xmath81 5.29 & 2.11 @xmath81 0.51 & 2.54 @xmath81 0.29 & * * * @xmath81 * * * & * * * @xmath81 * * * & 2.83 @xmath81 0.72 & * * * @xmath81 * * * + 587725578037494286 & 0.134 & * * * @xmath81 * * * & 2.05 @xmath81 0.20 & 2.57 @xmath81 0.22 & 4.51 @xmath81 0.05 & 3.46 @xmath81 0.13 & 1.37 @xmath81 0.18 & 0.25 @xmath81 0.01 + 587725591459201372 & 0.615 & 4.63 @xmath81 1.24 & 3.02 @xmath81 0.50 & 3.74 @xmath81 0.33 & * * * @xmath81 * * * & * * * @xmath81 * * * & 1.46 @xmath81 0.28 & * * * @xmath81 * * * + 587725578037495014 & 0.213 & * * * @xmath81 * * * & 6.16 @xmath81 4.88 & 4.83 @xmath81 4.78 & 58.32 @xmath81 30.70 & 3.91 @xmath81 16.30 & 1.12 @xmath81 1.38 & 0.98 @xmath81 0.60 + 587725578037494283 & 0.084 & * * * @xmath81 * * * & 0.82 @xmath81 0.25 & 1.17 @xmath81 0.25 & 2.36 @xmath81 0.05 & 1.68 @xmath81 0.10 & 1.84 @xmath81 0.69 & 0.27 @xmath81 0.02 + 587725591459201726 & 0.439 & 5.21 @xmath81 0.35 & 3.17 @xmath81 0.18 & 3.34 @xmath81 0.06 & * * * @xmath81 * * * & * * * @xmath81 * * * & 4.95 @xmath81 0.30 & * * * @xmath81 * * * + 587725591459201125@xmath113 & 0.066 & * * * @xmath81 * * * & 7.74 @xmath81 0.19 & 2.74 @xmath81 0.06 & 3.57 @xmath81 0.01 & 3.13 @xmath81 0.02 & 1.40 @xmath81 0.05 & 0.29 @xmath81 0.01 + 587725591459201562 & 0.068 & * * * @xmath81 * * * & 0.81 @xmath81 0.82 & 1.51 @xmath81 0.75 & 2.66 @xmath81 0.19 & 1.17 @xmath81 0.51 & 2.93 @xmath81 3.32 & 0.14 @xmath81 0.06 + 587725591459201436 & 0.066 & * * * @xmath81 * * * & * * * @xmath81 * * * & 2.28 @xmath81 0.10 & 2.25 @xmath81 0.03 & 1.65 @xmath81 0.12 & * * * @xmath81 * * * & 0.14 @xmath81 0.010 + 587725591459201670 & 0.651 & 6.22 @xmath81 0.58 & 3.66 @xmath81 0.77 & 5.06 @xmath81 0.21 & * * * @xmath81 * * * & * * * @xmath81 * * * & 3.98 @xmath81 0.84 & * * * @xmath81 * * * + 587725591459201656 & 0.157 & * * * @xmath81 * * * & 2.04 @xmath81 0.51 & 2.10 @xmath81 0.11 & 2.53 @xmath81 0.12 & 1.60 @xmath81 1.65 & 5.81 @xmath81 1.47 & 0.02 @xmath81 0.02 + 587725591459201666 & 0.078 & * * * @xmath81 * * * & 1.03 @xmath81 1.80 & 1.08 @xmath81 0.62 & 2.10 @xmath81 0.11 & 1.50 @xmath81 0.51 & 2.36 @xmath81 4.33 & 0.18 @xmath81 0.06 + 587725591459201614 & 0.188 & * * * @xmath81 * * * & 2.62 @xmath81 0.40 & 2.69 @xmath81 0.15 & 3.38 @xmath81 0.10 & 2.02 @xmath81 0.83 & 3.01 @xmath81 0.49 & 0.08 @xmath81 0.03 + 587725591459136404 & 0.220 & * * * @xmath81 * * * & 0.74 @xmath81 0.59 & 3.43 @xmath81 0.57 & 4.95 @xmath81 0.23 & 2.77 @xmath81 0.67 & 7.21 @xmath81 5.93 & 0.18 @xmath81 0.04 + 587725591459136351 & 0.319 & 1.34 @xmath81 1.16 & 2.38 @xmath81 0.72 & 0.76 @xmath81 0.45 & * * * @xmath81 * * * & * * * @xmath81 * * * & 0.34 @xmath81 0.23 & * * * @xmath81 * * * + 587725591459136031 & 0.192 & * * * @xmath81 * * * & 0.25 @xmath81 0.22 & 0.59 @xmath81 0.34 & 2.98 @xmath81 0.12 & 2.79 @xmath81 0.18 & 2.42 @xmath81 2.55 & 0.63 @xmath81 0.05 + 587725591459136024@xmath113 & 0.245 & * * * @xmath81 * * * & 2.00 @xmath81 0.44 & 1.81 @xmath81 0.47 & 3.87 @xmath81 0.12 & 2.54 @xmath81 0.32 & 0.85 @xmath81 0.29 & 0.23 @xmath81 0.03 + 587725591459136330 & 0.565 & 8.01 @xmath81 0.95 & 3.13 @xmath81 0.32 & 3.43 @xmath81 0.11 & * * * @xmath81 * * * & * * * @xmath81 * * * & 18.69 @xmath81 1.78 & * * * @xmath81 * * * + 587725591459135913 & 0.293 & * * * @xmath81 * * * & 2.32 @xmath81 0.61 & 2.23 @xmath81 0.47 & 2.76 @xmath81 0.09 & 2.18 @xmath81 0.26 & 1.29 @xmath81 0.44 & 0.19 @xmath81 0.02 + 587725591459136277 & 0.290 & * * * @xmath81 * * * & * * * @xmath81 * * * & 1.81 @xmath81 1.07 & 0.82 @xmath81 0.44 & 2.80 @xmath81 0.41 & * * * @xmath81 * * * & 3.97 @xmath81 2.21 + 587725591459136339 & 0.188 & * * * @xmath81 * * * & 2.97 @xmath81 0.35 & 2.69 @xmath81 0.35 & 5.10 @xmath81 0.07 & 5.15 @xmath81 0.17 & 0.95 @xmath81 0.17 & 0.36 @xmath81 0.01 + 587725578037166749 & 0.320 & * * * @xmath81 * * * & 2.30 @xmath81 0.35 & 2.45 @xmath81 0.86 & * * * @xmath81 * * * & * * * @xmath81 * * * & 0.59 @xmath81 0.22 & * * * @xmath81 * * * + 587725578037166468 & 0.330 & * * * @xmath81 * * * & 1.49 @xmath81 0.20 & 0.94 @xmath81 0.17 & * * * @xmath81 * * * & * * * @xmath81 * * * & 0.62 @xmath81 0.14 & * * * @xmath81 * * * + 587725578037166557 & 0.168 & * * * @xmath81 * * * & 2.47 @xmath81 0.11 & 3.06 @xmath81 0.07 & 4.11 @xmath81 0.03 & 3.06 @xmath81 0.09 & 2.39 @xmath81 0.12 & 0.20 @xmath81 0.01 + 587725578037166705 & 0.192 & * * * @xmath81 * * * & 1.54 @xmath81 0.35 & 2.71 @xmath81 0.29 & 2.50 @xmath81 0.11 & 1.24 @xmath81 0.50 & 2.94 @xmath81 0.73 & 0.06 @xmath81 0.03 + 587725578037166672 & 0.189 & * * * @xmath81 * * * & 3.85 @xmath81 0.15 & 2.92 @xmath81 0.20 & 5.48 @xmath81 0.03 & 5.96 @xmath81 0.07 & 0.59 @xmath81 0.05 & 0.34 @xmath81 0.01 + 587725578037166295 & 0.560 & 7.13 @xmath81 0.52 & 5.94 @xmath81 0.38 & 5.37 @xmath81 0.29 & * * * @xmath81 * * * & * * * @xmath81 * * * & 1.29 @xmath81 0.11 & * * * @xmath81 * * * + 587725578037166628 & 0.652 & 0.95 @xmath81 0.65 & 3.54 @xmath81 0.43 & 0.86 @xmath81 0.45 & * * * @xmath81 * * * & * * * @xmath81 * * * & 0.17 @xmath81 0.09 & * * * @xmath81 * * * + 587725578037166629 & 0.321 & 2.15 @xmath81 4.43 & 2.93 @xmath81 0.17 & 3.21 @xmath81 0.07 & * * * @xmath81 * * * & * * * @xmath81 * * * & 3.02 @xmath81 0.19 & * * * @xmath81 * * * + 587725578037100839 & 0.561 & 5.25 @xmath81 0.34 & 2.84 @xmath81 0.15 & 4.40 @xmath81 0.25 & * * * @xmath81 * * * & * * * @xmath81 * * * & 1.70 @xmath81 0.13 & * * * @xmath81 * * * + 587725578037101267 & 0.293 & * * * @xmath81 * * * & 7.99 @xmath81 8.21 & 6.65 @xmath81 0.68 & 2.60 @xmath81 1.68 & 0.17 @xmath81 0.51 & * * * @xmath81 * * * & 0.10 @xmath81 0.30 + 587725578037101221 & 0.518 & 5.79 @xmath81 0.95 & 3.31 @xmath81 0.35 & * * * @xmath81 * * * & * * * @xmath81 * * * & * * * @xmath81 * * * & * * * @xmath81 * * * & * * * @xmath81 * * * + 587725578037100922 & 0.168 & * * * @xmath81 * * * & 1.24 @xmath81 0.44 & 1.06 @xmath81 0.45 & 3.54 @xmath81 0.11 & 2.82 @xmath81 0.20 & 0.85 @xmath81 0.47 & 0.48 @xmath81 0.04 + 587725578037101176 & 0.281 & * * * @xmath81 * * * & 2.28 @xmath81 0.29 & 3.52 @xmath81 0.27 & 2.49 @xmath81 0.08 & * * * @xmath81 * * * & 2.07 @xmath81 0.31 & * * * @xmath81 * * * + 587725591458873892 & 0.281 & * * * @xmath81 * * * & 2.49 @xmath81 0.21 & 3.14 @xmath81 0.26 & * * * @xmath81 * * * & * * * @xmath81 * * * & 1.21 @xmath81 0.14 & * * * @xmath81 * * * + 587725578037100775@xmath114 & 0.176 & * * * @xmath81 * * * & * * * @xmath81 * * * & * * * @xmath81 * * * & 0.86 @xmath81 0.11 & 1.07 @xmath81 0.08 & * * * @xmath81 * * * & 1.30 @xmath81 0.20 + 587725591458808530 & 0.427 & 2.53 @xmath81 0.96 & 3.58 @xmath81 0.27 & 3.92 @xmath81 0.38 & * * * @xmath81 * * * & * * * @xmath81 * * * & 0.94 @xmath81 0.11 & * * * @xmath81 * * * + 587725591458808528 & 0.598 & 5.42 @xmath81 0.31 & 3.57 @xmath81 0.20 & * * * @xmath81 * * * & * * * @xmath81 * * * & * * * @xmath81 * * * & * * * @xmath81 * * * & * * * @xmath81 * * * + 587725578037101077 & 0.387 & 6.51 @xmath81 1.41 & 3.96 @xmath81 0.33 & 3.12 @xmath81 0.13 & * * * @xmath81 * * * & * * * @xmath81 * * * & 2.40 @xmath81 0.22 & * * * @xmath81 * * * + 587725578037101473 & 0.496 & 4.81 @xmath81 1.31 & 2.44 @xmath81 0.52 & 2.95 @xmath81 0.13 & * * * @xmath81 * * * & * * * @xmath81 * * * & 4.70 @xmath81 1.01 & * * * @xmath81 * * * + 587725578037166675 & 0.498 & 7.79 @xmath81 0.92 & 3.15 @xmath81 0.23 & 3.32 @xmath81 0.14 & * * * @xmath81 * * * & * * * @xmath81 * * * & 1.39 @xmath81 0.11 & * * * @xmath81 * * * lccllcclll 587725578037166705 & 0.192 & 17@xmath27 37@xmath28 30.681@xmath29 & 57@xmath31 20@xmath32 8.70@xmath33 & 22.53 & 23.11 & 1.51 & 8.69 & -0.42 & -0.09 + 587725578037166675 & 0.498 & 17@xmath27 37@xmath28 34.409@xmath29 & 57@xmath31 21@xmath32 13.71@xmath33 & * * * & 22.62 & 0.97 & 9.50 & 0.50 & -0.07 + 587725578037166672 & 0.189 & 17@xmath27 37@xmath28 39.053@xmath29 & 57@xmath31 21@xmath32 30.10@xmath33 & * * * & 23.04 & 3.36 & 10.05 & 0.15 & -0.55 + 587725578037166749 & 0.320 & 17@xmath27 37@xmath28 41.177@xmath29 & 57@xmath31 18@xmath32 46.29@xmath33 & 22.87 & 22.12 & 1.88 & 9.91 & 0.65 & -0.14 + 587725578037100839 & 0.561 & 17@xmath27 37@xmath28 43.542@xmath29 & 57@xmath31 24@xmath32 5.51@xmath33 & * * * & 22.19 & 1.19 & 10.40 & 0.85 & -0.32 + 587725578037166468 & 0.330 & 17@xmath27 37@xmath28 45.923@xmath29 & 57@xmath31 19@xmath32 17.19@xmath33 & 22.38 & 21.50 & 2.15 & 10.85 & 0.48 & -0.84 + 587725578037166557 & 0.168 & 17@xmath27 37@xmath28 58.250@xmath29 & 57@xmath31 20@xmath32 16.80@xmath33 & 20.61 & 20.12 & 1.18 & 9.47 & 0.53 & -0.08 + 587725578037166295 & 0.560 & 17@xmath27 37@xmath28 59.495@xmath29 & 57@xmath31 23@xmath32 21.80@xmath33 & * * * & 22.16 & 0.89 & 9.70 & 0.79 & -0.07 + 587725578037166628 & 0.652 & 17@xmath27 38@xmath28 2.622@xmath29 & 57@xmath31 23@xmath32 26.99@xmath33 & * * * & 22.73 & 1.51 & 10.51 & 1.11 & -0.24 + 587725578037166629 & 0.321 & 17@xmath27 38@xmath28 7.756@xmath29 & 57@xmath31 23@xmath32 34.90@xmath33 & * * * & 22.32 & 1.04 & 9.09 & 0.28 & -0.05 + 587725578037101267 & 0.293 & 17@xmath27 38@xmath28 22.192@xmath29 & 57@xmath31 25@xmath32 2.82@xmath33 & * * * & 23.36 & 1.66 & 9.16 & -0.06 & -0.18 + 587725578037100922 & 0.168 & 17@xmath27 38@xmath28 23.628@xmath29 & 57@xmath31 27@xmath32 56.42@xmath33 & 21.48 & 21.97 & 2.67 & 9.97 & -0.24 & -0.76 + 587725578037101473 & 0.496 & 17@xmath27 38@xmath28 27.876@xmath29 & 57@xmath31 26@xmath32 33.90@xmath33 & 22.91 & 22.28 & 1.22 & 9.76 & 0.68 & -0.10 + 587725578037101077 & 0.387 & 17@xmath27 38@xmath28 36.526@xmath29 & 57@xmath31 30@xmath32 38.59@xmath33 & * * * & 22.32 & 0.52 & 9.11 & 0.51 & -0.01 + 587725578037101221 & 0.518 & 17@xmath27 38@xmath28 37.317@xmath29 & 57@xmath31 26@xmath32 53.92@xmath33 & * * * & 22.50 & 1.09 & 10.07 & 0.62 & -0.26 + 587725578037100775@xmath114 & 0.176 & 17@xmath27 38@xmath28 37.646@xmath29 & 57@xmath31 29@xmath32 13.99@xmath33 & 21.03 & 20.34 & 3.08 & 11.08 & 0.69 & -0.88 + 587725578037101176 & 0.281 & 17@xmath27 38@xmath28 41.873@xmath29 & 57@xmath31 28@xmath32 9.30@xmath33 & * * * & 22.33 & 0.99 & 9.24 & -0.09 & -0.19 + 587725591458873892 & 0.281 & 17@xmath27 38@xmath28 48.259@xmath29 & 57@xmath31 29@xmath32 10.40@xmath33 & 22.69 & 22.23 & 1.80 & 9.76 & 0.43 & -0.13 + 587725591458808530 & 0.427 & 17@xmath27 38@xmath28 50.596@xmath29 & 57@xmath31 29@xmath32 54.31@xmath33 & * * * & 22.88 & 2.89 & 10.50 & 2.12 & 0.00 + 587725591458808528 & 0.598 & 17@xmath27 38@xmath28 54.873@xmath29 & 57@xmath31 30@xmath32 5.80@xmath33 & * * * & 22.58 & 1.83 & 10.37 & 1.33 & -0.09 + 587725578037494286 & 0.134 & 17@xmath27 39@xmath28 45.205@xmath29 & 56@xmath31 40@xmath32 2.20@xmath33 & 20.87 & 20.47 & 1.17 & 9.66 & -0.15 & -0.28 + 587725578037494409 & 0.355 & 17@xmath27 39@xmath28 51.035@xmath29 & 56@xmath31 39@xmath32 13.11@xmath33 & * * * & 23.23 & 1.78 & 9.14 & 0.36 & -0.04 + 587725578037495014 & 0.213 & 17@xmath27 39@xmath28 51.343@xmath29 & 56@xmath31 39@xmath32 58.89@xmath33 & * * * & 22.73 & 0.95 & 8.18 & -0.66 & -0.02 + 587725578037494283 & 0.084 & 17@xmath27 39@xmath28 53.130@xmath29 & 56@xmath31 40@xmath32 18.91@xmath33 & 22.24 & 22.00 & 2.34 & 9.17 & -0.57 & -0.35 + 587725578037494050 & 0.077 & 17@xmath27 39@xmath28 56.243@xmath29 & 56@xmath31 38@xmath32 17.20@xmath33 & * * * & 22.11 & 5.46 & 10.76 & -0.80 & -1.91 + 587725578037494494 & 0.356 & 17@xmath27 39@xmath28 56.287@xmath29 & 56@xmath31 37@xmath32 21.00@xmath33 & * * * & 22.20 & 1.96 & 9.99 & 1.14 & -0.03 + 587725578037494738 & 0.474 & 17@xmath27 39@xmath28 56.704@xmath29 & 56@xmath31 37@xmath32 53.80@xmath33 & * * * & 22.72 & 0.86 & 9.67 & 0.38 & -0.16 + 587725591459201436 & 0.066 & 17@xmath27 40@xmath28 1.663@xmath29 & 56@xmath31 44@xmath32 3.12@xmath33 & 20.89 & 20.74 & 2.20 & 9.21 & -0.41 & -0.35 + 587725591459201125@xmath113 & 0.066 & 17@xmath27 40@xmath28 3.853@xmath29 & 56@xmath31 41@xmath32 59.82@xmath33 & 20.37 & 19.99 & 1.96 & * * * & * * * & * * * + 587725591459201656 & 0.157 & 17@xmath27 40@xmath28 6.724@xmath29 & 56@xmath31 43@xmath32 49.51@xmath33 & 22.64 & 22.12 & -0.52 & 8.68 & -0.97 & -0.24 + 587725591459201627 & 0.079 & 17@xmath27 40@xmath28 8.196@xmath29 & 56@xmath31 44@xmath32 57.80@xmath33 & 22.18 & 21.68 & 1.51 & 8.48 & -0.61 & -0.09 + 587725591459201372 & 0.615 & 17@xmath27 40@xmath28 8.328@xmath29 & 56@xmath31 39@xmath32 39.82@xmath33 & * * * & 22.68 & 1.80 & 10.53 & 0.81 & -0.51 + 587725591459201929 & 0.436 & 17@xmath27 40@xmath28 15.117@xmath29 & 56@xmath31 46@xmath32 4.70@xmath33 & 22.36 & 22.16 & 0.24 & 9.27 & 0.22 & -0.07 + 587725591459201726 & 0.439 & 17@xmath27 40@xmath28 17.307@xmath29 & 56@xmath31 41@xmath32 31.09@xmath33 & * * * & 22.42 & 0.94 & 9.98 & 0.37 & -0.22 + 587725591459136339 & 0.188 & 17@xmath27 40@xmath28 17.468@xmath29 & 56@xmath31 48@xmath32 5.50@xmath33 & 21.23 & 20.57 & 0.76 & 9.98 & -0.06 & -0.47 + 587725591459201670 & 0.651 & 17@xmath27 40@xmath28 19.175@xmath29 & 56@xmath31 43@xmath32 39.18@xmath33 & * * * & 22.46 & 0.14 & 8.93 & 0.61 & 0.29 + 587725591459201562 & 0.068 & 17@xmath27 40@xmath28 20.969@xmath29 & 56@xmath31 42@xmath32 43.30@xmath33 & * * * & 22.77 & 2.90 & 8.77 & -1.06 & -0.53 + 587725591459201614 & 0.188 & 17@xmath27 40@xmath28 23.225@xmath29 & 56@xmath31 45@xmath32 42.52@xmath33 & 22.37 & 21.50 & 0.59 & 8.72 & -0.06 & -0.04 + 587725591459136404 & 0.220 & 17@xmath27 40@xmath28 23.708@xmath29 & 56@xmath31 46@xmath32 34.10@xmath33 & 22.53 & 21.76 & 0.79 & 9.59 & -0.47 & -0.55 + 587725591459135913 & 0.293 & 17@xmath27 40@xmath28 29.685@xmath29 & 56@xmath31 50@xmath32 27.71@xmath33 & 21.64 & 21.93 & 1.18 & 9.55 & 0.12 & -0.21 + 587725591459136024@xmath113 & 0.245 & 17@xmath27 40@xmath28 32.498@xmath29 & 56@xmath31 49@xmath32 12.51@xmath33 & 22.73 & 20.99 & -0.08 & * * * & * * * & * * * + 587725591459201666 & 0.078 & 17@xmath27 40@xmath28 33.193@xmath29 & 56@xmath31 44@xmath32 5.32@xmath33 & * * * & 20.82 & 1.06 & 9.00 & -0.90 & -0.42 + 587725591459136277 & 0.290 & 17@xmath27 40@xmath28 35.691@xmath29 & 56@xmath31 50@xmath32 29.29@xmath33 & 22.66 & 22.23 & 3.27 & 10.99 & 0.49 & -0.98 + 587725591459136330 & 0.565 & 17@xmath27 40@xmath28 46.362@xmath29 & 56@xmath31 49@xmath32 13.40@xmath33 & * * * & 22.18 & 1.33 & 10.03 & 1.06 & -0.08 + 587725591459136031 & 0.192 & 17@xmath27 40@xmath28 56.177@xmath29 & 56@xmath31 49@xmath32 15.38@xmath33 & 20.94 & 20.09 & 2.45 & 10.92 & 0.97 & -0.47 + 587725591459136351 & 0.319 & 17@xmath27 40@xmath28 56.594@xmath29 & 56@xmath31 48@xmath32 52.20@xmath33 & * * * & 21.35 & 2.15 & 10.90 & 0.75 & -0.60 + + + 587725578037100834 & 1.515 & 17@xmath27 38@xmath28 9.360@xmath29 & 57@xmath31 25@xmath32 21.36@xmath33 & * * * & 23.40 & 2.63 & * * * & * * * & * * * + 587725591459201473 & 1.387 & 17@xmath27 40@xmath28 9.836@xmath29 & 56@xmath31 40@xmath32 7.68@xmath33 & * * * & 22.67 & 1.14 & * * * & * * * & * * * + 587725591459136108 & 1.632 & 17@xmath27 40@xmath28 43.682@xmath29 & 56@xmath31 48@xmath32 45.36@xmath33 & * * * & 23.05 & 1.67 & * * * & * * * & * * * + 587725591459135981@xmath114 & 1.028 & 17@xmath27 40@xmath28 49.197@xmath29 & 56@xmath31 47@xmath32 23.65@xmath33 & 22.62 & 21.20 & 1.94 & * * * & * * * & * * *

we theoretically propose how to observe topological effects in a generic classical system of coupled harmonic oscillators , such as classical pendula or lumped - element electric circuits , whose oscillation frequency is modulated fast in time . making use of floquet theory in the high frequency limit , we identify a regime in which the system is accurately described by a harper - hofstadter model where the synthetic magnetic field can be externally tuned via the phase of the frequency - modulation of the different oscillators . we illustrate how the topologically - protected chiral edge states , as well as the hofstadter butterfly of bulk bands , can be observed in the driven - dissipative steady state under a monochromatic drive . in analogy with the integer quantum hall effect , we show how the topological chern numbers of the bands can be extracted from the mean transverse shift of the steady - state oscillation amplitude distribution . finally we discuss the regime where the analogy with the harper - hofstadter model breaks down .

1510.04697

since topological concepts were first introduced to understand the behaviour of electrons in the quantum hall effect @xcite , such ideas have become a powerful theme in condensed matter physics , applicable to a wide - range of different systems @xcite . in topological materials , the eigenstates making up an electronic energy band can be characterised by nontrivial topological invariants , which are robust against any perturbations that do not close the energy band gaps . these invariants can be directly related to the existence of remarkable topological edge states localised near the system boundary , which lead , for example , to the robust and precise quantisation of conductance in the quantum hall effect @xcite . while the study of real materials is usually complicated , the essential physics of various topological phenomena can be captured through simple lattice models . perhaps the most famous of these is the harper - hofstadter ( hh ) model of a charged electron hopping on a 2d tight - binding square lattice in a uniform magnetic field @xcite . not only does this model exhibit the quantum hall effect thanks to its topological energy bands and unidirectionally - propagating edge states , but it has a rich fractal energy spectrum known as the hofstadter butterfly . most importantly , this model has opened up the way for the study of quantum hall physics beyond traditional solid - state materials , beginning with its realisation in solid - state superlattice devices @xcite , ultracold atomic gases @xcite and silicon photonics @xcite . recently , topological concepts have entered the realm of classical physics with specially - designed mechanical meta - materials @xcite and arrays of coupled classical harmonic oscillators such as pendula @xcite , gyroscopes @xcite and electrical circuits @xcite . for many of these systems , topological properties are encoded in the normal - mode dynamical matrix , in analogy with the electronic hamiltonian of a topological lattice model . the demonstration of topological edge states in these classical systems has also ignited the hope that such set - ups could find practical applications , for instance , as loss - free propagating acoustic waveguides @xcite . in neutral systems , such as atoms , photons and classical oscillators , the simulation of topological phenomena has been made possible by important interdisciplinary developments in the creation of so - called synthetic gauge fields @xcite . in the case of the hh model , the effect of the magnetic field on a charged electron is to modify the tight - binding tunnelling amplitudes by complex spatially - dependent phases called the peierls phases @xcite . the route to simulating this model with a neutral system is therefore to find an alternative mechanism to generate suitable complex hopping phases . in ultracold gases , this can be , for example , through the control of internal atomic states , by laser - assisted tunneling or lattice shaking @xcite , while in photonics , this has been realised through the insertion of link resonators in ring - resonator arrays @xcite or by twisting waveguides helically along the direction of light propagation to simulate a time - dependent modulation @xcite . finally , in classical systems , analogous effects arise naturally for coupled gyroscopes @xcite , with angular - momentum biasing @xcite , or can be hard - wired into the coupling - connections in lattices of pendula @xcite or electrical elements @xcite . in this paper , we propose a different conceptual approach for how to introduce non - trivial coupling phases , and hence how to simulate topological phenomena , in classical harmonic oscillator lattices through the application of temporally - periodic modulations to system parameters . in this approach , we build on advances and methods from ultracold gases @xcite and photonics @xcite , where time - periodic modulations were previously introduced to realise topological lattice models . for classical harmonic oscillators , the first steps in this direction were presented in @xcite for the case of two coupled oscillators ; here , we develop these ideas further to propose a modulation scheme that simulates the analogue of the hh model in a large array of coupled oscillators . even though our discussion is focused on mechanical pendulum systems , the same ideas straightforwardly extend to other types of oscillators such as lumped - element electric circuits of @xcite . as in the gyroscopic lattices of @xcite , our proposal breaks time - reversal symmetry and can be associated with energy bands describing the collective excitation modes of the coupled pendula system , which are characterised by a nontrivial chern number : the topological invariant of the quantum hall effect . we go beyond these previous works to show how , not only the edge states , but the hallmarks of the chern number itself could be seen experimentally in a system of classical harmonic oscillators . while these systems focus on the haldane model @xcite , our scheme focuses on the hh model , where , thanks to the easy tunability of our synthetic flux , we show that the full hofstadter butterfly of states could also be observed . finally , we note that our proposal stands in contrast to the recent experiments on pendula @xcite and electrical circuits @xcite that did not break time - reversal symmetry , but instead realised two copies of the hh model with opposite synthetic flux : one copy for each of the two pseudo - spin " degrees of freedom defined in these systems . in the quantum analogy , this leads to the quantum spin hall effect in which the edge states are not topologically robust as they are unprotected against perturbations which flip the pseudo - spin " . the paper is organized as follows . in sec . [ sec : nem ] we introduce the system of coupled classical harmonic oscillators and highlight the limit in which newton s equations of motion are analogous to the heisenberg equations for a tight - binding model . in sec . [ sec : hhm ] , we introduce the specific frequency modulation scheme from which we recover the hh model in the high frequency expansion of floquet theory . the efficiency of our idea is quantitatively assessed in sec . [ sec : results ] where numerical simulations of the full classical equations of motion , are compared to the predictions of the hh model . we find that the oscillation amplitude spectra exhibit wide peaks corresponding to the bulk energy bands ; we show that these peaks build up to give the hofstadter butterfly as a function of the synthetic magnetic flux . we excite topologically - protected edge states showing that these are chiral and immune from back - scattering . we then discuss a classical counterpart of the integer quantum hall effect from which we can extract the chern number . in sec . [ sec : instability ] , we investigate the dynamical stability of the system and the role of the counter - rotating - wave terms , together with the consequent limitations of the mapping onto the hh model . we comment in sec . [ sec : experimental ] on possible experimental implementations of our proposal using arrays of either mechanical pendula or electric circuits , before drawing conclusions in sec . [ sec : conclusions ] . and @xmath0 . pendula on the edges are coupled to a rigid wall.,scaledwidth=40.0% ] we consider a square lattice of @xmath1 sites . each lattice site , labelled by two indexes @xmath2 , hosts a pendulum of mass @xmath3 and frequency @xmath4 , whose oscillation plane is rigidly fixed , for example , along the unit - cell diagonal , as shown in fig . [ fig : system ] . we assume that the natural frequencies of the pendula are periodically modulated in time around their mean value @xmath5 . the displacement of each pendulum from its equilibrium point along the fixed direction is denoted by @xmath6 . all the pendula are coupled to their nearest neighbours , while the pendula on the edges of the lattice are coupled to a fixed wall . the coupling is provided , for example , by springs of rest length equal to the lattice spacing and of constant @xmath7 and @xmath0 along the two @xmath8 and @xmath9 directions respectively . in the small oscillation regime the elongation of each spring is proportional to the difference of the spatial displacement of the pendula that it connects @xmath10 and , in absence of any pre - tension , the force is determined by the standard hooke s law . including loss mechanisms with a friction coefficient @xmath11 , newton s equations of motion for a pendulum driven by the external force @xmath12 read : @xmath13 the goal of this section is to show that in a suitable regime newton s equations of motion - can be rewritten in the same form as the heisenberg equation for a tight - binding hamiltonian , where the bosonic annihilation and creation operators are replaced by @xmath14-numbers . starting from the displacements @xmath6 and the corresponding momenta @xmath15 , classical complex variables corresponding to the annihilation operators of quantum harmonic oscillator are defined as : @xmath16 it is straightforward to see that the square modulus @xmath17 is proportional to the oscillation energy of the @xmath2-th pendulum . assuming for simplicity that the product @xmath18 is constant for every site @xmath19 and defining @xmath20 and @xmath21 , we can combine eq . and eq . into equations for the complex amplitudes @xmath22 of the form : @xmath23 \alpha_{i , j } \\ & -{\mathrm{i}}\left [ \frac{\tilde{\omega}_{i , j}}{2}- \frac{\omega_{i , j}^2(t)}{2\tilde{\omega}_{i , j}}+{\mathrm{i}}\gamma_{i , j } + 2\omega_x+2\omega_y\right ] \alpha^*_{i , j}\\ & + { \mathrm{i}}\sum_{\pm1}\omega_x\left(\alpha_{i\pm1,j}+\alpha^*_{i\pm1 , j}\right ) \\ & + { \mathrm{i}}\sum_{\pm1}\omega_y\left(\alpha_{i , j\pm1}+\alpha^*_{i , j\pm1}\right)+ { \mathrm{i}}f_{i , j}^\text{ex}(t ) , \end{split } \label{alphadrivendot}\ ] ] where the damping rate @xmath24 . the external driving force is taken to be monochromatic at frequency @xmath25 with amplitude proportional to @xmath26 , @xmath27 equations can be simplified under the assumption that the natural frequencies @xmath28 of the pendula are much larger than all other frequencies in the problem and that the driving frequency @xmath25 is comparable to the @xmath28 s . in this case , the fastest time dependence of @xmath22 is proportional to @xmath29 , while that of the conjugate variables @xmath30 is proportional to @xmath31 . it can be seen that in the limit @xmath32 the contribution of the @xmath33 s to the motion equation for @xmath34 is negligible as the fast oscillations quickly average to zero , in which the @xmath34 terms only have a ( relatively ) slow time - evolution while the @xmath33 terms oscillate at @xmath35 and therefore quickly average to zero . ] . as a result , it is legitimate to neglect in eq . the terms involving the complex conjugate variables @xmath33 , an approximation that in quantum optics is well - known under the name of the rotating wave approximation ( rwa ) . applying the rwa , we interpret eq . without driving and dissipation , as the classical analogue of the heisenberg equation of motion obtained from the following hamiltonian : @xmath36 this is a bosonic tight - binding hamiltonian , where the terms on the first line account for the on - site energy and those on the second line describe the hopping . in the hamiltonian , the rwa consists of neglecting all non - energy - conserving terms proportional to @xmath37 and @xmath38 and only keep the energy - conserving ones @xmath39 . in the previous section , we have shown how newton s equations of motion can be rewritten in the rwa in the same form as the heisenberg equation of a tight - binding model . in this section we proceed to identify a suitable modulation scheme for the @xmath4 that reduces to an effective hh model in the floquet picture . inspired by the realization of the hh model with ultra - cold atoms in optical lattices @xcite and by related proposals in the photonic context @xcite , we take the natural frequency of the pendula , _ i.e. _ a part of the single - particle on - site energy in the tight - binding analogy , to be temporally and spatially modulated according to : @xmath40 with a spatial dependence @xmath41 direction , sketching the spatial modulation of the natural frequencies of the pendula with period @xmath42.,scaledwidth=45.0% ] the temporal modulation @xmath43 is taken as periodic with frequency @xmath44 , and the static spatial modulation : @xmath45 is a saw - tooth function along the @xmath8-direction , of period @xmath46 , as schematically shown in fig . [ fig : system_x ] . while hopping along @xmath9 is unaffected by both the spatial and temporal modulations , bare hopping along @xmath8 would be strongly suppressed by the spatial modulation eq . . hopping in this direction is however restored by the temporal modulation : when the frequency of the modulation is chosen to be on resonance with the detuning of neighbouring sites , a uniform coupling is restored over the whole lattice @xcite . given the specific sawtooth form of eq . chosen for @xmath47 , a bi - harmonic temporal modulation is needed : @xmath48 \end{split } \label{timemodulation}\ ] ] where the amplitude @xmath49 is position independent . the first component at frequency @xmath44 restores a uniform hopping between the pairs of pendula with natural frequency difference @xmath44 , while the second component at frequency @xmath50 addresses the pairs with a frequency difference @xmath50 . as we shall see in the next subsection , the position - dependence of the phase @xmath51 offers us a way to control the phase of the hopping along @xmath8 @xcite . in order to obtain the desired hh model , a possible choice is the following : @xmath52 with the modulations introduced in the previous paragraph , we now show how the equations of motion in eq . take the form of the heisenberg equations of motion for the hh model by applying the floquet formalism of time - periodic systems at high - frequency @xcite . substituting the chosen forms of the temporal and static modulation eq . and eq . into eq . and performing the rwa , we straightforwardly obtain : @xmath53 \alpha_{i , j}\\ & + { \mathrm{i}}\sum_{\pm 1 } \left[\omega_x \alpha_{i\pm1 , j}+ \omega_y \alpha_{i , j\pm1}\right ] + { \mathrm{i}}f_{i , j}^\text{ex}(t ) . \end{split } \label{alpharwadot}\ ] ] in the following we focus on the case where a single @xmath54 site is driven . we also assume that the fastest frequency of the temporal modulation @xmath55 is fast compared to all other frequencies except the bare natural oscillation frequency : @xmath56 the first inequality is required by the rwa , while the latter justifies our calculating the effective dynamics after a time - average over the period @xmath57 of the temporal modulation . in this regime , a simple expression for the time - evolution of the whole system at stroboscopic times @xmath58 can be obtained using floquet theory @xcite . to this purpose we introduce new variables : @xmath59 we then look for the steady - state solution , where @xmath60 oscillates with the same frequency as the external driving force , which is tuned in the vicinity of the natural frequency of the driven pendulum @xmath61 . an effective equation of motion can be obtained for the @xmath60 variables by time - averaging eq . over one period of the fast temporal modulation , that is by applying the magnus expansion to the lowest order @xcite . after some calculations , time - averaged equations for @xmath60 are found . the full forms of the coefficients of the effective equations are expressed as sums of bessel functions and can be found in the appendix [ app : fulleff ] . a much simpler expression for the effective equations of motion is obtained by considering only the terms that give the largest contribution in these sums : @xmath62 where , with the definition of the modulation phase in eq . , we have introduced the following quantities : @xmath63 as detailed in appendix [ app : fulleff ] , the condition of validity for eq . is that the next order of the bessel functions can be neglected , @xmath64 from this expression , it is apparent that a very large period of the spatial modulation @xmath46 would be optimal , with a linear ramp in the bare frequencies of the harmonic oscillators in eq . . while this configuration is possible for ultracold atomic systems @xcite , it is not possible for these classical harmonic oscillators as the condition @xmath65 for the rwa does not allow us to use an arbitrarily large value of @xmath46 . more details of these issues can be found in sec . [ sec : experimental ] . in the following , we shall see that a good compromise can be found already for @xmath46 as small as @xmath66 . in eq . , the bare coupling frequencies @xmath67 and @xmath68 are renormalised by the bessel functions , and the effective couplings along the two directions are different . in order to recover the usual hh model with equal hopping in the two directions , it is useful to start from suitably chosen @xmath69 that will give the same effective @xmath70 in the two directions : @xmath71 one must of course be careful in choosing the amplitude of the modulation @xmath49 , such that the frequencies in eq . satisfy the inequality of eq . . as with the hopping terms , so is the amplitude of the driving term in eq . renormalised by the bessel functions . we choose the bare driving force on the pumped site as : @xmath72 in order to keep the effective driving intensity @xmath73 constant as we vary @xmath49 or @xmath44 . without the last two terms describing driving and dissipation , eq . is the classical version of the heisenberg equations derived from the hh hamiltonian : @xmath74 \end{split } \label{hhh}\ ] ] where @xmath75 is the energy at which the hh model is coherently probed . the non - trivial hopping phase in eq . is determined by the modulation phase in eq . and controlled by the parameter @xmath76 . as usual , the magnetic flux enclosed within a plaquette is calculated by summing the phases accumulated on each link when hopping around a plaquette of the lattice . with the definition in eq . , the sum of the phases gained by the complex hopping elements in eq . , is uniform for the whole lattice and equal to : @xmath77 as a result , our model of coupled pendula is a classical simulator of the hh model and allows us to generate the hofstadter butterfly by simply tuning the phase of the temporal modulations according to eq . . the tunability of this artificial magnetic field is a main advantage of our scheme compared to other systems , where the magnetic flux is fixed @xcite . we note that any other choice of the temporal modulation phase @xmath51 leading to the same flux per plaquette would give an equivalent hh model that only differs by a gauge transformation . before proceeding , it is important to note that the analysis we have done so far is strictly valid only in the deep rwa limit @xmath78 where the counter - rotating - wave terms @xmath79 in eq . can be safely neglected and effective equations such as eq . can be derived . for lower values of @xmath80 , the first effect beyond the rwa is a global shift in the detuning frequency : @xmath81 , with : @xmath82 this shift is predicted by including the counter - rotating - wave part of the @xmath79 in eq . , that in the frame rotating at @xmath80 corresponds to a non - rotating term and so includes the leading order of the counter - rotating - wave effects . as discussed in @xcite , this has the form of a classical analogue of the bloch - siegert shift . for even smaller @xmath80 , the counter - rotating - wave terms can also lead to dynamical instabilities and will be discussed in sec . [ sec : instability ] . in order to assess the validity of the effective hh model , obtained via the floquet approach , we have performed numerical simulations of the full equations of motion including the counter - rotating - wave terms as well as the full spatial and temporal modulations of the oscillation frequencies . we have quantitatively compared these predictions to those from the heisenberg equation of motion for the hh model in eq . . we have numerically integrated the set of @xmath83 time - dependent equations for a square lattice of @xmath1 sites with a standard fourth - order runge - kutta until a steady state is achieved at long evolution times @xmath84 . to match with the floquet picture , we have performed a stroboscopic sampling at times @xmath85 . at these times @xmath86 , under a monochromatic drive at @xmath25 the steady - state amplitudes have the form @xmath87 , where @xmath88 are the oscillation amplitudes . the total intensity is obtained as the sum of the oscillation amplitudes @xmath89 of all the pendula @xmath90 . the response spectra describe the total intensity as a function of the detuning @xmath91 . depending on the value of @xmath92 and the position @xmath54 of the driven site , different behaviours are expected . for @xmath92 within a band of the bulk hh model and @xmath54 located far from the edges , the bulk of the system is excited and so the response is dominated by delocalized band states . when @xmath92 belongs to a gap between two energy bands , the total oscillation is suppressed unless @xmath54 is located close to an edge of the system , so that edge states can be excited . a detailed study of these regimes will be the subject of the next subsection . to facilitate the reader in following the discussion , it is useful to first review some fundamental facts about the hh model . in the energy - flux plane , the states organize themselves in a recursive and self - similar structure , known as the hofstadter butterfly . the structure of the bands and the gaps is the simplest when the magnetic flux is a rational number @xmath93 with @xmath94 being co - prime positive integers @xcite . in this special case , one has exactly @xmath95 sub - bands separated by @xmath96 energy gaps . when @xmath95 is even , the two middle bands touch at dirac points . the eigenstates making up the hofstadter bands are associated with a non - trivial topological invariant , called the chern number @xmath97 . this invariant is related to the transverse conductivity of the integer quantum hall effect @xcite . in a finite system with non - zero @xmath97 , unidirectionally propagating states exist at the edges of the system at energies located in the gaps between bands . since there are no states available for backscattering , these states are topologically protected from disorder @xcite . in most optical @xcite and mechanical @xcite experiments so far , the presence of such chiral edge states represents the smoking gun of the non - trivial topology . we now show that such edge states can be identified also in our scheme . lattice with a flux per plaquette @xmath98 and losses @xmath99 where the central lattice site is pumped . the blue circles and orange triangles correspond to the spectra calculated from the steady state of eq . as a function of the detuning from the external driving frequency @xmath100 , including the shift in eq . . parameters are : @xmath101 , @xmath102 , @xmath103 for the blue circles , and @xmath104 , @xmath105 , @xmath103 for the orange triangles . black curve shows the spectra as obtained from the driven - dissipative hh model @xcite . green areas show the position of the hh bands . frequencies indicated by arrows are used in fig . [ fig : topologicalstate].,scaledwidth=48.0% ] the response spectra presented in fig . [ fig : spectra ] are calculated by driving the central site of a lattice of @xmath106 pendula with a flux per plaquette of @xmath98 in units of the flux quantum . losses are chosen to be @xmath99 , which is large enough to ensure that the excitation does not reach the edges of the system and that the response is only determined by the bulk properties . the position of the hh bands is highlighted by the green shaded area . the accuracy of the floquet effective dynamics is assessed by comparing the numerical result with the prediction of the driven - dissipative hh model studied in @xcite . the blue circles show the response spectrum of a system of bare frequency @xmath101 , temporal modulation frequency @xmath102 and amplitude such that @xmath103 from eq . . this corresponds to @xmath107 , @xmath108 and @xmath109 . the orange triangles are instead calculated for @xmath104 , @xmath105 , and again @xmath103 , which corresponds to @xmath110 , and the same @xmath67 and @xmath68 as before . for the latter set of parameters , the inequality in eq . is better satisfied and indeed the peak amplitudes of the pendula - spectra agree very well with the driven - dissipative hh model depicted with a solid black line . with the former set of parameters , the two central peaks are difficult to resolve but the external regions of high intensity still coincide well with the position of the hh bands . for even smaller values of @xmath80 the deviation from the hh model would be much more dramatic , as the rwa is no longer valid and the contribution of the counter - rotating - wave terms is not negligible any more . pendula with @xmath98 , @xmath111 , @xmath103 , @xmath102 and @xmath101 , obtained by pumping a single site on the lower edge for different values of the detuning of the pumping frequency @xmath112 . panel * a * and * c * are for @xmath113 located in a gap and show the excitation of an edge state with clockwise chirality . in panel * c * , the blue square indicates the position of a defect composed of @xmath114 missing pendula on the left edge . panel * b * is for @xmath115 located again in a gap and shows the excitation of an edge state with counter - clockwise chirality . panel * d * is for @xmath116 located in an energy band and shows the excitation of bulk states . the arrow indicates the position of the pumped site on the lower edge.,title="fig:",scaledwidth=24.0% ] pendula with @xmath98 , @xmath111 , @xmath103 , @xmath102 and @xmath101 , obtained by pumping a single site on the lower edge for different values of the detuning of the pumping frequency @xmath112 . panel * a * and * c * are for @xmath113 located in a gap and show the excitation of an edge state with clockwise chirality . in panel * c * , the blue square indicates the position of a defect composed of @xmath114 missing pendula on the left edge . panel * b * is for @xmath115 located again in a gap and shows the excitation of an edge state with counter - clockwise chirality . panel * d * is for @xmath116 located in an energy band and shows the excitation of bulk states . the arrow indicates the position of the pumped site on the lower edge.,title="fig:",scaledwidth=24.0% ] pendula with @xmath98 , @xmath111 , @xmath103 , @xmath102 and @xmath101 , obtained by pumping a single site on the lower edge for different values of the detuning of the pumping frequency @xmath112 . panel * a * and * c * are for @xmath113 located in a gap and show the excitation of an edge state with clockwise chirality . in panel * c * , the blue square indicates the position of a defect composed of @xmath114 missing pendula on the left edge . panel * b * is for @xmath115 located again in a gap and shows the excitation of an edge state with counter - clockwise chirality . panel * d * is for @xmath116 located in an energy band and shows the excitation of bulk states . the arrow indicates the position of the pumped site on the lower edge.,title="fig:",scaledwidth=24.0% ] pendula with @xmath98 , @xmath111 , @xmath103 , @xmath102 and @xmath101 , obtained by pumping a single site on the lower edge for different values of the detuning of the pumping frequency @xmath112 . panel * a * and * c * are for @xmath113 located in a gap and show the excitation of an edge state with clockwise chirality . in panel * c * , the blue square indicates the position of a defect composed of @xmath114 missing pendula on the left edge . panel * b * is for @xmath115 located again in a gap and shows the excitation of an edge state with counter - clockwise chirality . panel * d * is for @xmath116 located in an energy band and shows the excitation of bulk states . the arrow indicates the position of the pumped site on the lower edge.,title="fig:",scaledwidth=24.0% ] in fig . [ fig : topologicalstate ] we show the steady state intensity distribution obtained by pumping one site in the middle of the bottom edge with different values of the detuning @xmath112 corresponding to the arrows in fig . [ fig : spectra ] . system parameters are @xmath101 , @xmath102 and @xmath103 as for the blue spectra of fig . [ fig : spectra ] , while losses @xmath111 are smaller in order to allow the steady state amplitude oscillations to propagate over an adequate distance . the intensity distributions shown in panels * a * , * b * and * c * are obtained by driving the system in the band gap where the chiral edge - states are expected . the pendula with the biggest oscillation amplitudes are indeed localized on the edge of the system . as one can see by comparing panels * a * and * b * , the direction of unidirectional propagation changes when the sign of @xmath92 is changed , as expected from the hh model . the distance covered along the edge before decaying is set by the ratio of the group velocity of the edge state over the loss rate , as usual in such driven - dissipative systems @xcite . the topological robustness of such edge states is highlighted in panel * c * , where a defect modelled as @xmath117 missing pendula was introduced on the left edge of the lattice . the edge state is found to propagate around the defect , without being scattered into the bulk . finally , panel * d * shows the steady - state intensity distribution when the detuning corresponds to a hh band : in this case , the excitation penetrates into the bulk and shows no preferred chirality . pendula with @xmath99 , @xmath103 , @xmath105 and @xmath104 , obtained by pumping the central site . panels * a * and * c * are obtained from full numerical integration of newton s equation of motion , while panels * b * and * d * are obtained from a driven - dissipative hh model as in @xcite , and are shown for comparison . the first row is for @xmath98 and @xmath118 corresponding to the center of the highest band . the second row is for @xmath119 and @xmath120 corresponding to the middle of the central band.,title="fig:",scaledwidth=24.0% ] pendula with @xmath99 , @xmath103 , @xmath105 and @xmath104 , obtained by pumping the central site . panels * a * and * c * are obtained from full numerical integration of newton s equation of motion , while panels * b * and * d * are obtained from a driven - dissipative hh model as in @xcite , and are shown for comparison . the first row is for @xmath98 and @xmath118 corresponding to the center of the highest band . the second row is for @xmath119 and @xmath120 corresponding to the middle of the central band.,title="fig:",scaledwidth=24.0% ] + pendula with @xmath99 , @xmath103 , @xmath105 and @xmath104 , obtained by pumping the central site . panels * a * and * c * are obtained from full numerical integration of newton s equation of motion , while panels * b * and * d * are obtained from a driven - dissipative hh model as in @xcite , and are shown for comparison . the first row is for @xmath98 and @xmath118 corresponding to the center of the highest band . the second row is for @xmath119 and @xmath120 corresponding to the middle of the central band.,title="fig:",scaledwidth=24.0% ] pendula with @xmath99 , @xmath103 , @xmath105 and @xmath104 , obtained by pumping the central site . panels * a * and * c * are obtained from full numerical integration of newton s equation of motion , while panels * b * and * d * are obtained from a driven - dissipative hh model as in @xcite , and are shown for comparison . the first row is for @xmath98 and @xmath118 corresponding to the center of the highest band . the second row is for @xmath119 and @xmath120 corresponding to the middle of the central band.,title="fig:",scaledwidth=24.0% ] while the existence of the chiral edge states is one hallmark of a quantum hall system , we now demonstrate that characteristic features of the driven - dissipative hh model can also be observed in the response of the bulk of the system . in fig . [ fig : steadystate ] , we present the intensity distribution of the system when the central site of the lattice is driven with a detuning frequency resonant with an energy band . for the panels in the upper row , the magnetic flux is @xmath98 , while @xmath119 for the panels in the lower row ; losses are @xmath99 in all panels . panels * a * and * c * show the distributions of the full numerical integration with @xmath104 , @xmath105 , @xmath103 and can be compared with panels * b * and * d * that are obtained from the driven - dissipative hh model as in @xcite . we notice that the two pattern distributions are in excellent qualitative agreement , in particular where the periodicity of the pattern is concerned . as expected , when the natural frequency @xmath80 is reduced away from the rwa limit , the spatially symmetric structure of the oscillation amplitude pattern is broken and the agreement with the driven - dissipative hh model is less good ( not shown ) . and frequency detuning @xmath121 the total intensity of the oscillation is depicted with the color scale , where lower intensities are in darker colors . black regions indicating negligible intensities correspond to the energy gaps between energy bands . the peculiar structure of the hofstadter butterfly pattern is clearly visible . system parameters are : @xmath104 , @xmath105 , @xmath99 , @xmath103 , for a lattice of @xmath106 sites where only the central one is pumped.,scaledwidth=48.0% ] the scheme that we have proposed in the previous sections allows us to change the magnetic flux @xmath76 per plaquette by simply adjusting the phase of the modulation of each pendula in eq . . in this way , our scheme is sufficiently flexible to explore a wide portion of the flux - frequency plane . collecting a large number of numerically calculated spectra of the total intensity for different values of @xmath76 in a single color - plot , we obtain the pattern shown in fig . [ fig : butterfly ] . although the presence of dissipation does not allow us to resolve details of the fractal structure that are smaller than the loss rate , the emerging picture shows a close resemblance to the hofstadter butterfly @xcite . such a result would constitute the first experimental visualization of the hofstadter butterfly in classical physics . in order to speed up the numerical calculations needed to plot fig . [ fig : butterfly ] , the equations of motion in eq . have been solved by expanding the solution in the fourier basis : @xmath122 and then truncating the series to a value @xmath123 that is sufficiently large to ensure the convergence of the solution . solving the linear system for the fourier components coupled by the time - dependent modulation is equivalent to , but much faster than , solving the full differential equations until reaching the steady state . more details on this decomposition method are found in appendix [ app : fourier ] . it is interesting to note that the agreement with the well - known hofstadter butterfly gets worse in the top and bottom regions of the spectra and eventually breaks down . those areas correspond to very low values of the magnetic flux @xmath76 for which the number of bands is large and the smoothing has most dramatic consequences . to understand the break - down from the butterfly structure , we checked the solutions with the full numerical runge - kutta integration method and we observed that in these regions the system is dynamically unstable . this means that a stationary steady state can not exist and therefore the hofstadter butterfly pattern completely breaks down . these deviations stem from the strength of the counter - rotating - wave - terms as controlled by the amplitude of the temporal modulation @xmath49 . in our simulation , we take @xmath124 to be fixed which , from eq . , results in an increase of @xmath49 as the flux is decreased . larger @xmath49 implies a larger contribution from counter - rotating - wave terms , that results in the emergence of a parametric instability @xcite . we note that the spectra obtained from the fourier decomposition method appear to be still stable despite the exponential growth of the runge - kutta solution because such a method begins from the ansatz of a steady - state solution and so can not predict the instability . more insights into the instability will be given later in the paper . in the previous sections , we have demonstrated that the system of frequency - modulated coupled harmonic oscillators can be a classical simulator of the hh model . as a final point , we now show that an analogue of the integer quantum hall effect ( iqhe ) can be observed even in a classical mechanical context . a main feature of the iqhe is that a current appears in the direction orthogonal to the applied electric field where the associated conductivity is an integer multiple of the conductance quantum @xmath125 . this integer is related to the chern number of the populated bands of the hh model . for the bosonic model considered here , these bands can be populated by a suitable excitation scheme . our proposal is inspired by a recent work on photonic systems @xcite . pendula with @xmath98 , @xmath99 , @xmath103 , @xmath105 and @xmath104 , obtained by pumping the central site with a detuning corresponding to the middle of the lowest band @xmath126 . for panel * a * there is no external force @xmath127 , and the centre of the intensity distribution indeed remains at the centre of the system . for panel * b * , the force @xmath128 is directed along the @xmath9-direction as indicated by the arrow and induces a sizeable rightward shift of the centre of the distribution towards the positive @xmath8 direction as expected from the iqhe.,title="fig:",scaledwidth=24.0% ] pendula with @xmath98 , @xmath99 , @xmath103 , @xmath105 and @xmath104 , obtained by pumping the central site with a detuning corresponding to the middle of the lowest band @xmath126 . for panel * a * there is no external force @xmath127 , and the centre of the intensity distribution indeed remains at the centre of the system . for panel * b * , the force @xmath128 is directed along the @xmath9-direction as indicated by the arrow and induces a sizeable rightward shift of the centre of the distribution towards the positive @xmath8 direction as expected from the iqhe.,title="fig:",scaledwidth=24.0% ] in order to introduce an analogue of the electric field in our model , we need to generate a linear potential acting on the excitation field @xmath22 . applying such a potential along the positive @xmath9-direction would correspond to a term of the form @xmath129 in eq . . equivalently , and perhaps more straightforwardly , the analogue electric field could be introduced through an additional linear gradient of the bare oscillation frequency along the @xmath9-direction in eq . , with @xmath130 . in fig . [ fig : halleffect]-*a * we show the intensity distribution of a system of @xmath131 pendula with @xmath98 and @xmath99 in the absence of the external force and for a detuning @xmath126 such that only the lowest band is excited with a resulting population that is approximately uniform . as discussed in @xcite , these conditions can be fulfilled provided the energy width of the band @xmath132 is much smaller than the band gap @xmath133 and the loss rate falls in between these two energy scales . the intensity pattern is symmetric both in @xmath8 and in @xmath9 and is centred around the middle point . when a non - zero force @xmath128 along the vertical direction is applied as in panel * b * , the oscillation pattern clearly shifts to the right in the direction orthogonal to the applied force . this is a manifestation of the lorentz force and a clear signature of a hall - like effect . in order to assess the quantized nature of this hall effect , we can quantify the lateral shift as @xmath134 and compare it to the prediction for the integer quantum hall effect in driven - dissipative systems @xcite , @xmath135 where @xmath95 is the denominator of the rational flux @xmath93 and @xmath97 is the chern number of the excited band . the real number @xmath136 quantifies the ( spurious ) contribution of the bandwidth , that is responsible for a non - uniform population of the band , as well of the other neighbouring bands . the quantized nature of the iqhe is most apparent in the small @xmath137 limit in the coefficient of the leading term . & @xmath138 & @xmath139 + & & @xmath140 & & @xmath141 + & & @xmath142 & @xmath143 & @xmath142 + & & @xmath144 & & + & & @xmath142 & @xmath145 & @xmath145 + & & @xmath146 & & + & & @xmath142 & @xmath142 & @xmath147 + & & & & @xmath148 + & & @xmath143 & @xmath149 & @xmath143 + in table [ table : chern ] we summarize the estimated chern numbers @xmath150 of different bands for four different @xmath76 as calculated from eq . for a system of @xmath106 lattice sites , @xmath104 , @xmath105 , @xmath151 and @xmath152 . the constant @xmath136 , that does not depend on the losses @xmath137 , is eliminated by calculating the linear coefficient in the shift given in eq . for two values of @xmath153 choosing any value in this frequency range only affects the calculation of the chern number to the least - significant digit . when the method is not applicable because @xmath154 , the corresponding chern number of these bands was not calculated . in all other cases , the agreement with the chern number @xmath97 of the hh model is good and confirms the expected integer quantum hall effect in a classical mechanics context . in this section we discuss in more detail the instability observed in the small flux region of the hofstadter butterfly of fig.[fig : butterfly ] and the corrections to the hh model due to the counter - rotating - wave terms in the equations of motion eq . . as the amplitude of these terms is proportional to @xmath155 , their importance for a fixed @xmath156 increases as the magnetic flux @xmath76 is lowered according to eq . , explaining why the instability was only observed close to the upper and lower edges of the butterfly pattern . in order to more easily access this physics , we have introduced a cleaner strategy to reinforce the counter - rotating - wave terms by decreasing the bare frequency of the pendula @xmath80 while keeping other parameters fixed . this approach comes with the great advantage that we can compare energy spectra with the same value of the magnetic flux as the strength of the counter - rotating - wave terms increases . in the following we allow the bare frequency to be @xmath157 , by relaxing the constraint in eq . . recently , there have been several works discussing the role of the counter - rotating - wave terms in bosonic hamiltonians on topology @xcite . however , exception made for a work on edge state instabilities @xcite , only a little attention has been so far devoted to the interplay between topology and parametric instabilities . we now investigate how counter - rotating - wave terms lead to the appearance of complex energy modes in the spectrum and to a dramatic distortion of the hh bands which is accompanied by a disappearance of topological edge states . to this end , it is convenient to recast the equations in eq . , together with the corresponding equations for @xmath33 , without driving and dissipation , in the matrix form : @xmath158 where @xmath159 is a column vector . the eigenvalues of the @xmath160 time periodic matrix @xmath161 represent the oscillation frequencies of the system . due to the specific form of the coupling between the variables @xmath162 and @xmath163 , we notice that this matrix is not hermitian with respect to the standard positive - definite inner product . its block - structure : @xmath164 where @xmath165 and @xmath166 are @xmath167 real symmetric matrices , periodic with the modulation period @xmath57 , guarantees that it is hermitian with respect to the indefinite - inner product defined by the matrix : @xmath168 and of common use in the bogoliubov theory of dilute bose gases @xcite . the non - hermiticity of @xmath161 with respect to the standard inner product is crucial for the appearance of the complex energy - modes and the emergence of the instability . to treat the time - dependent problem , we can apply the floquet theory of time - periodic systems . the evolution within one period @xmath169 of the modulation is then expressed as that with a time - independent stroboscopic floquet matrix @xmath170 , defined as the time - ordered exponential @xcite : @xmath171 if @xmath161 were a hermitian matrix , then @xmath172 would be unitary and the eigenvalues of the stroboscopic floquet matrix @xmath170 would always be real @xcite . in our system , instead , the matrix @xmath173 describes , in terms of the complex variables in eq . , the one - period evolution of the system which is symplectic in the original @xmath6 and @xmath15 variables . as a result , this matrix is not hermitian with respect to the standard inner product , which implies that , in general , the eigenvalues of the floquet matrix can also be complex , so the system can be dynamically unstable @xcite . the matrix @xmath172 is the _ mapping at a period _ of the system with periodically varying parameters @xcite , and it is the classical analogue of the stroboscopic time - evolution operator : @xmath174 . it is known from classical mechanics textbooks that if all the eigenvalues @xmath175 of the mapping at a period @xmath172 are distinct and lie on the unit circle in the complex plane , the system is strongly stable @xcite . since the floquet matrix is related to @xmath172 by eq . , the stability of the system can be checked just by looking at the eigenvalues of @xmath170 . notice that the condition on the eigenvalues of @xmath172 to lie on the unit circle implies that the floquet spectra has to be real . in fact , if @xmath176 is an eigenvalue of the floquet matrix @xmath170 , then it is related to the eigenvalue of @xmath172 as : @xmath177 from eq . . it is straightforward to check that if @xmath178 , then @xmath179 is on the unit circle . in analogy with the above - mentioned bogoliubov theory @xcite , it directly follows from the specific form of the matrix @xmath161 that if @xmath180 is an eigenvector of @xmath170 with eigenvalue @xmath176 , then @xmath181 is also an eigenvector but with opposite eigenvalue @xmath182 . the @xmath183 eigenvalues of @xmath170 can be then distinguished according to their _ krein signature _ @xcite : @xmath184 that is the sign of the norm corresponding to the @xmath136 scalar product of the above - mentioned bogoliubov theory @xcite . by definition , the krein signature is either @xmath185 . all stable eigenvalues of @xmath170 have either positive or negative krein signature , while all unstable eigenvalues have zero krein signature ; this also holds for the eigenvalues of @xmath172 . in the complex plane . red points corresponds to eigenvalues with krein signature @xmath186 , points in blue are the eigenvalues with @xmath187 and green points are unstable eigenvalues with @xmath188 . panel * a * shows a stable configuration for @xmath189 , where all eigenvalues lies on the unit circles . panel * b * shows the unstable configuration @xmath190 , where some eigenvalues are clearly not lying on the unit circle . parameters are : @xmath191 , @xmath42 , @xmath105.,title="fig:",scaledwidth=24.0% ] in the complex plane . red points corresponds to eigenvalues with krein signature @xmath186 , points in blue are the eigenvalues with @xmath187 and green points are unstable eigenvalues with @xmath188 . panel * a * shows a stable configuration for @xmath189 , where all eigenvalues lies on the unit circles . panel * b * shows the unstable configuration @xmath190 , where some eigenvalues are clearly not lying on the unit circle . parameters are : @xmath191 , @xmath42 , @xmath105.,title="fig:",scaledwidth=24.0% ] in fig . [ fig : krein ] we show the representation on the unit circle of the eigenvalues of the matrix @xmath172 , coloured according to their krein signature . eigenvalues with positive @xmath192 are shown in red , eigenvalues with negative @xmath192 are displayed in blue and unstable eigenvalues with zero krein signature are displayed in green . as the ratio @xmath193 is changed , the eigenvalues @xmath175 of @xmath172 rotate on the unit circle . whenever eigenvalues with opposite krein signature meet , they can leave the unit circle , meaning that the system becomes unstable . this instability happens every half - round - trip : since the eigenvalues @xmath175 perform a full rotation on the unit circle when @xmath193 is increased by @xmath194 , the instability occurs each time @xmath195 , with @xmath196 . , modulo @xmath44 , associated with either a positive krein - signature @xmath192 or with @xmath188 and @xmath197 . the system has a magnetic flux @xmath191 , with periodic boundary conditions ; driving and dissipation are not included . the energy spectra are presented as a function of @xmath193 , for a fixed value of the amplitude modulation @xmath198 . when all the eigenvalues at a particular frequency are stable , we denote them in black , otherwise we denote them either in red or blue . red points are purely real modes , while blue points are the unstable modes , plotted according to the real part of their complex energy . panel * a * shows the eigenvalues in units of @xmath44 , while in panel * b * the same eigenvalues are shifted by @xmath199 as defined in eq . . parameters are : @xmath191 , @xmath42 , @xmath105.,title="fig:",scaledwidth=50.0% ] , modulo @xmath44 , associated with either a positive krein - signature @xmath192 or with @xmath188 and @xmath197 . the system has a magnetic flux @xmath191 , with periodic boundary conditions ; driving and dissipation are not included . the energy spectra are presented as a function of @xmath193 , for a fixed value of the amplitude modulation @xmath198 . when all the eigenvalues at a particular frequency are stable , we denote them in black , otherwise we denote them either in red or blue . red points are purely real modes , while blue points are the unstable modes , plotted according to the real part of their complex energy . panel * a * shows the eigenvalues in units of @xmath44 , while in panel * b * the same eigenvalues are shifted by @xmath199 as defined in eq . . parameters are : @xmath191 , @xmath42 , @xmath105.,title="fig:",scaledwidth=50.0% ] in fig . [ fig : stability ] the floquet energy spectra @xmath176 of the system are shown as the bare frequency @xmath80 is changed for a fixed value of @xmath44 . the magnetic flux is @xmath191 and periodic boundary conditions are used , so that the edge states do not appear in the spectra . with no loss of information , in the figure we only present stable eigenvalues associated with a positive krein - signature and unstable eigenvalues with @xmath197 . in panel * a * , the eigenvalues are displayed modulo @xmath44 , while in panel * b * the same eigenvalues @xmath176 are shifted by @xmath199 as defined in eq . . when all the eigenvalues at a particular frequency are stable , we denote them in black , otherwise we denote them either in red or blue . the points in blue correspond to the modes which have become unstable , plotted here according to the real part of the energy . the points in red are eigenmodes with purely real energies . we refer to frequencies for which there are blue points as instability regions . as anticipated above , we observe that the instability regions occur every integer and half - integer times of @xmath193 . in particular , purely imaginary modes start to develop around @xmath200 and @xmath201 . as @xmath193 is increased , the size of the instability regions along @xmath80 gets thinner and the deviation from @xmath179 correspondingly weaker . still we find that the instabilities persist even for a very high ratio @xmath193 in the conservative case considered here . however , once losses are included in the model , stability improves @xcite . this explains why in the large @xmath80 calculations shown in the previous sections instabilities were only observed at very small values of the magnetic flux . finally , in panel * b * , we show how the detailed structure of the floquet quasi - energy spectra changes as the ratio @xmath193 is varied . to allow a direct comparison of the spectra , we have shifted each of these by the quantity @xmath199 as defined in eq . . for large values of @xmath193 the spectra consists of @xmath66 bands in agreement with the hh model . as @xmath193 is decreased , the band structure of the effective floquet hamiltonian remains initially qualitatively similar to the one of the hh hamiltonian with minor deviations due to the counter - rotating - wave terms . as @xmath193 is reduced further , the deviations become more significant and the band - gaps eventually close . we have numerically checked that the topological edge states are no longer present once the bands - gaps have closed around @xmath202 . we conclude the work by commenting on possible experimental realizations of our proposal . as already mentioned , even though our attention was focussed on a classical mechanical implementation based on coupled pendula , all our results hold for harmonic oscillators of any type . in all cases , the main challenge is to have a sufficiently high quality factor @xmath203 . in fact , a high q is crucial to have enough freedom to choose the frequencies of the system to satisfy well the following inequality chain : @xmath204 these inequalities are required to be in the rotating wave limit , to find simple forms for the effective coupling amplitudes between neighbouring pendula , and to have clear access to the different spectroscopic features . as discussed around fig . [ fig : system_x ] , the ratio @xmath193 also controls the number of sites @xmath46 that can be accommodated in a period of the spatial modulation . in particular , the sawtooth profile is needed for the scalability of the system of classical oscillators with a fixed quality factor @xmath205 . this can be seen by instead considering a simple linear ramp of the bare frequency of the pendula for a system of length @xmath206 . in this case , the frequency difference between pendula would range up to @xmath207 which , for a sufficiently large system , could exceed @xmath80 and thus violate eq . . the sawtooth modulation with finite @xmath46 is instead scalable to any arbitrarily large system while keeping the bare frequency @xmath80 to satisfy eq . . we note that the atomic systems of @xcite can use a linear ramp of an external potential because they do not suffer from these difficulties , since counter - rotating - wave terms are not present . for optical cavities , instead , the bare cavity frequency and the @xmath205 factor are typically large enough for this constraint to be practically irrelevant for realistic system sizes @xcite . while the spatial modulation of the oscillation frequency is rather straightforward to implement in pendula , a practical implementation of the temporal modulation needed for the floquet scheme may be obtained , as sketched in @xcite , by mounting a magnet on each pendulum and letting it interact with an externally - controlled time - dependent magnetic field . quantitatively we need a minimum @xmath205-factor of around @xmath208 to observe chiral edge states in the scheme that we propose , before encountering the band - closing of fig . [ fig : stability]-*b*. while reaching such a value of the q factor in coupled pendula systems requires a significant , though still not dramatic , improvement of recent experimental set - ups @xcite , it should be much more achievable using conventional rlc circuits . to go even further , the spectra of the hh model in fig . [ fig : spectra ] and the hofstadter butterfly in fig . [ fig : butterfly ] were obtained using higher q factors of @xmath209 and @xmath210 , respectively . such high quality factors can be achieved in mechanical systems using micro - resonators as done in optomechanical oscillators @xcite . in fact , recently there has been a related proposal of realizing topological models using such optomechanical systems in the limit of the rotating - wave - approximation @xcite . in this work , we have theoretically proposed a scheme to realize non - trivial topological models in classical systems of coupled pendula or coupled lumped - element rlc circuits through a periodic time - modulation of system parameters . within a floquet framework , we have shown that suitably - designed temporal modulations of the natural oscillation frequencies effectively produce the complex coupling amplitudes of the hh model , where the synthetic magnetic flux is easily tunable . we have assessed the accuracy of the floquet picture by numerically solving the full equations of motion and we have shown that hallmarks of the hh model , such as the topologically - robust chiral edge states and the hofstadter butterfly pattern , can be observed in the excitations of the system . furthermore , we have illustrated a protocol to extract the topological chern number of the bands in a classical counterpart of the integer quantum hall effect . finally , we extended our investigation to the regime where the analogy with the quantum lattice model breaks down , investigating the crucial role played by the counter - rotating - wave terms in the stability of the system and the persistence of topological effects . the exploration of topological effect in classical - mechanical systems is opening up many interesting new avenues of research . the use of floquet modulation schemes to engineer topological models brings with it particular advantages . firstly , time - modulation can naturally be used to break time - reversal symmetry and so to create edge states that are fully topologically protected . secondly , the scheme discussed here is only one of many possible floquet modulation schemes that could be applied to arrays of classical oscillators ; other schemes may open up the way to build classical simulators of topological models with other interesting properties and with even more exotic topological invariants @xcite . looking to the future , our proposal can be naturally extended to include a confining external potential , leading to novel oscillation features associated with analogue momentum - space magnetism @xcite . beyond the small oscillation regime considered so far , systems of pendula also show nonlinear effects such as amplitude - dependent oscillation frequencies . the interplay of such nonlinearities with topological features may lead to a variety of new and unexpected effects . in this appendix , we give more details on the derivation of the hofstadter - like effective equation of motion in eq . . from the magnus expansion applied to time - periodic hamiltonians at first order @xcite , one gets the effective hamiltonian as : @xmath211 for high modulation frequency . the effective heisenberg equations that are derived from this effective hamiltonian are @xmath212=-{\mathrm{i}}\frac{1}{t}\int_0^t [ \beta_{i , j},{\hat{\mathcal{h}}}(t)]$ ] . therefore , from the equations of motion eq . , substituting the transformed variables of eq . and integrating over a period of the temporal modulation , the following effective equations of motion are found : @xmath213 the couplings along the @xmath8 and @xmath9 direction get renormalised , as does the external driving force . the expression for the effective coupling along @xmath9 is : @xmath214 where we have used the anger - jacobi expansion to obtain the bessel functions @xmath215 of @xmath216-th order with argument @xmath8 . @xmath156 and @xmath217 are defined as in eq . and eq . . the effective coupling along @xmath8 depends on the position along the static spatial modulation @xmath47 we first focus on the couplings between pendula that have a natural - frequency difference of @xmath218 , where the @xmath219 indicates that the hopping is calculated going towards the left ( @xmath145 ) or towards the right ( @xmath142 ) . we have that : @xmath220 this equation describes the hopping along the small steps of the static spatial modulation . the coupling between pendula that have a natural - frequency difference of @xmath221 is : @xmath222 where this time @xmath223 indicates the hopping calculated going towards the left ( @xmath142 ) or towards the right ( @xmath145 ) . this is because the frequency - difference along the `` big step '' of the static modulation has an opposite sign to the ones along the `` small steps '' . finally , the effective driving force : @xmath224 we now assume that @xmath225 , with @xmath226 and take only the largest term in the sums . in eq . , , and we have that @xmath227 , while in eq . we must take @xmath228 . remarkably , in this approximation , the hopping along @xmath8 is uniform and does not depend on the position along the static - spatial modulation , thus allowing for clear definitions of @xmath229 , @xmath230 and @xmath231 : @xmath232 by combining eq . with the effective equations of motion in eq . to lowest order in @xmath156 and @xmath49 , we obtain exactly the equations in eq . . from eq . , and we notice that the second largest term in the sums is proportional to @xmath233 , that is very small if @xmath234 or when considering a large period of the static modulation @xmath235 . we found that an optimal combination of these two requirements that fulfils the inequality eq . with the renormalisation of eq . is to have @xmath103 with @xmath42 . we now comment on the fourier decomposition that was used as an alternative method to the full - time integration with runge - kutta for solving the set of differential equations in eq . to produce the butterfly spectra in fig . [ fig : butterfly ] . from the properties of periodically driven systems @xcite , we know that , during a floquet evolution between two stroboscopic times @xmath236 , the system exhibits micro - motion , with a period set by the frequency of the temporal modulation @xmath44 . we are interested in the steady state , therefore , we search for a solution of eq . that oscillates at the frequency of the external driving @xmath25 . to this end , we expand the @xmath237 in a fourier series , using both @xmath25 and @xmath44 as harmonics : @xmath238 where @xmath239 are the time - independent fourier amplitudes , assuming that all the time - dependencies are in the exponential term . the sum over @xmath240 is truncated to a finite number @xmath123 , that is large enough to ensure a convergent solution . the sum over @xmath241 takes only two values @xmath242 , since we have assumed that the external driving force is a cosine with a defined frequency @xmath25 . by substituting eq . in eq . and isolating the component proportional to @xmath243 , we have a set of linear algebraic equations that can be inverted to find the coefficients @xmath239 . in order to simulate the real experimental situation , we performed all the calculations using the full numerical integral of eq . , except for hofstadter butterfly where the full numerical integration would have been computationally demanding . we have verified the good agreement between the two methods finding a mean error that is less than @xmath244 . we note that the two methods are in fact equivalent when the system is dynamically stable . however , when there is an instability , the system does not reach the steady state . this is clearly seen in the full numerical integration method , where the solution shows the typical exponential growth of an unstable system . in the fourier method the steady state is imposed by the decomposition eq . itself and so the emergence of an instability can not be predicted , as discussed in the main text . g. l. yu , r. v. gorbachev , j. s. tu , a. v. kretinin , y. cao , r. jalil , f. withers , l. a. ponomarenko , b. a. piot , m. potemski , d. c. elias , x. chen , k. watanabe , t. taniguchi , i. v. grigorieva , k. s. novoselov , v. i. falko , a. k. geim and a. mishchenko , _ hierarchy of hofstadter states and replica quantum hall ferromagnetism in graphene superlattices _ , nat . phys . * 10 * , 525 ( 2014 ) . c. r. dean , l. wang , p. maher , c. forsythe , f. ghahari , y. gao , j. katoch , m. ishigami , p. moon , m. koshino , t. taniguchi , k. watanabe , k. l. shepard , j. hone and p. kim , _ hofstadter s butterfly and the fractal quantum hall effect in moir superlattices _ , nature * 497 * , 598 ( 2013 ) . m. aidelsburger , m. atala , m. lohse , j. t. barreiro , b. paredes and i. bloch , _ realization of the hofstadter hamiltonian with ultracold atoms in optical lattices _ , phys . lett . * 111 * , 185301 ( 2013 ) . m. mancini , g. pagano , g. cappellini , l. livi , m. rider , j. catani , c. sias , p. zoller , m. inguscio , m. dalmonte and l. fallani , _ observation of chiral edge states with neutral fermions in synthetic hall ribbons _ , science * 349 * , 1510 ( 2015 ) . j. struck , c. lschlger , m. weinberg , p. hauke , j. simonet , a. eckardt , m. lewenstein , k. sengstock and p. windpassinger , _ tunable gauge potential for neutral and spinless particles in driven optical lattices _ , phys . * 108 * , 225304 ( 2012 ) . m. aidelsburger , m. atala , s. nascimbne , s. trotzky , y .- a . chen and i. bloch , _ experimental realization of strong effective magnetic fields in an optical lattice _ , phys . * 107 * , 255301 ( 2011 ) . p. jain , a. s. bradley and c. w. gardiner , _ quantum de laval nozzle : stability and quantum dynamics of sonic horizons in a toroidally trapped bose gas containing a superflow _ , phys . a * 76 * , 023617 ( 2007 ) .

in this paper we investigate the role of inelastic collisions in the kinetic evolution of a highly overpopulated gluon system starting from glasma - type initial condition . using the gunion - bertsch formula we derive the inelastic collision kernel under the collinear and small angle approximations . with both numerics and analytic analysis , we show that the inelastic process has two effects : globally changing ( mostly reducing ) the total particle number , while locally at small momentum regime always filling up the infrared modes extremely quickly . this latter effect is found to significantly speed up the emergence of a local thermal distribution in the infrared regime with vanishing local `` chemical potential '' and thus catalyze the onset of dynamical bose - einstein condensation to occur faster ( as compared with the purely elastic case ) in the overpopulated glasma . 1 . physics department and center for exploration of energy and matter , indiana university , 2401 n milo b. sampson lane , bloomington , in 47408 , usa . 2 . physics department and center for particle physics and field theory , fudan university , shanghai 200433 , china . 3 . riken bnl research center , bldg . 510a , brookhaven national laboratory , upton , ny 11973 , usa .

1303.7214

thermalization of the quark - gluon plasma is one of the most challenging problems in current heavy ion physics . starting with two colliding nuclei in a form of color glass condensate with high gluon occupation @xmath0 below saturation scale @xmath1 @xcite and following the initial impact , a subsequent strong field evolution stage ( likely with instabilities @xcite ) till about the time @xmath2 is then succeeded by a far - from - equilibrium gluon - dominant matter , the glasma @xcite . the evolution of this glasma stage toward a quark - gluon plasma ( qgp ) that is close to local equilibratium and exhibits viscous - hydrodynamic behavior , is indicated by phenomenology to be reached on the order of a fermi over c time ( see e.g. @xcite ) . precisely how this occurs remains to be fully understood . describing the pre - equilibrium evolution with kinetic equations is a very useful approach , based on which the so - called `` bottom - up '' thermalization scenario was developed @xcite . there is however the complication of instability driven by anisotropy that may change this picture ( see e.g. @xcite ) . there are also other kinetic - based approaches , see e.g. @xcite . more recently an alternative thermalization scenario , based on crucial role of high initial overpopulation in the glasma and kinetic evolution dominated by elastic collisions , has been proposed in @xcite . in this scenario , while the initial scale @xmath3 is large compared with @xmath4 and thus the coupling @xmath5 is small , the high occupation @xmath6 elevates the elastic scattering rate to be of the order @xmath7 rather than the usual @xmath8 , and the glasma is essentially an emergent strongly interacting matter with weak coupling albeit large aggregate of constituents . two important scales are introduced to characterize the distribution , the hard cut - off scale @xmath9 beyond which @xmath10 and the soft high - occupation scale @xmath11 below which @xmath6 . while the initial glasma has the two scales overlapping @xmath12 , during the course of thermalization the two scales are separated eventually toward @xmath13 upon thermalization . one particularly nontrivial observation in the elastic - dominant picture is that the high initial overpopulation @xmath14 and the conservation of both energy and particle number will necessarily require the formation of a bose - einstein condensate that absorbs the excess gluons . this has been explicitly shown to occur by numerically solving the elastic kinetic equation derived under small angle approximation @xcite . there have been intensive discussions related to this picture from a variety of approaches , see e.g. @xcite . strong evidences for the formation of such a bose - einstein condensate have been reported for similar thermalization problem in the classical - statistical lattice simulation of scalar field theory @xcite . the case for non - abelian gauge theory is more complicated and still under investigation @xcite . one important question that has not been addressed in the above scenario is the role of inelastic processes . this issue could indeed be critical for at least two reasons ( see discussions in e.g. @xcite ) . first of all the inelastic processes will spoil the particle number conservation , and one might naively argue that the excessive gluons in the overpopulated glasma could simply be eliminated by very fast inelastic collisions . secondly , to make it even worse , the inelastic processes are parametrically at the same order as the elastic processes ( as opposed to naive power counting ) , so there appears no apparent dominance of the elastic over the inelastic and one may indeed worry that the inelastic could efficiently reduce total particle number . in such a situation , an explicit evaluation including both elastic and inelastic collisions becomes mandatory to clarify what will happen after including both types of collisions . to be precise , once the inelastic processes are included , one does not expect any condensation in the ultimate thermal equilibrium because with long enough time the inelastic processes will always remove any excessive particles . the interesting question , instead , is what changes the inelastic collisions bring to the dynamical evolution of the system . in particular , it is found @xcite that with purely elastic scatterings the overpopulated system is driven toward a dynamical onset of condensation in a finite time via critical scaling behavior in the infrared regime . it is extremely interesting to know , upon including the inelastic processes , how such dynamical evolution may be modified and whether the transient off - equilibrium condensation would still occur or not . in this paper , we aim to address this important question by studying the kinetic evolution of a highly overpopulated system starting from glasma - type initial condition with both @xmath15 and @xmath16 scatterings . in section 2 we will derive the inelastic collision kernel under the collinear and small angle approximations using the gunion - bertsch formula for the @xmath17 matrix element . in section 3 we will use numerical solutions as well as analytic analysis to understand the role of the inelastic process for both the global particle number change and the local behavior at small momentum region . finally we will conclude in section 4 . as a first step toward understanding the inelastic contributions and for simplicity and unambiguity , we will focus on the static box case with isotropic distribution in this work and leave the study of expanding case for future work . in this section we will derive the kinetic evolution equation with both elastic and inelastic collisions . the kinetic equation deals with the gluon distribution function defined as @xmath18 where @xmath19 denotes the spin and color degeneracy factor . the boltzmann equation for @xmath20 reads @xmath21+{\cal c}_{2\leftrightarrow 3}[f_p],\end{aligned}\ ] ] where we denote @xmath20 by @xmath22 and @xmath23 with @xmath24 and @xmath25 . for later convenience , we also introduce the following notations : @xmath26 in what follows we will separately discuss the elastic term @xmath27 and the inelastic term @xmath28 . the collision kernel from the @xmath29 process with full nonlinearity has been studied in @xcite . here we only briefly summarize the main results . we have the @xmath15 collision kernel given by @xmath30&=&\frac{1}{n_g}\frac{1}{2}\int_{123}\frac{1}{2e_p}|m_{12\leftrightarrow3p}|^2 ( 2{\pi})^4{\delta}^4(p_1+p_2-p_3-p){\nonumber\\}&&\times[(1+f_p)(1+f_3)f_1f_2-f_pf_3(1+f_1)(1+f_2 ) ] , \label{c22_full}\end{aligned}\ ] ] where @xmath31 and @xmath32 collision matrix element with @xmath33 being the usual mandelstam variables . the pre - factor @xmath34 in eq . ( [ c22_full ] ) is a symmetry factor counterweighing the permutation of @xmath35 and @xmath36 while the pre - factor @xmath37 cancels the summation over the spin and color of gluon @xmath38 " in the matrix element ( [ m22 ] ) . the dominant contribution of @xmath15 scattering in ( [ m22 ] ) comes from very small exchange momentum in @xmath39 or @xmath40 kinematic regimes , for which the incoming momenta only gets `` deflected '' by very small angle . if one uses this small angle approximation , then a rather neat kernel can be derived @xcite : @xmath41}\rc , \label{c22_sa}\end{aligned}\ ] ] with @xmath42 coming from the leading - log contribution . the hard scale @xmath43 and soft scale @xmath44 in the above are defined via global integrals : @xmath45 for later convenience we also introduce the debye scale defined as @xcite @xmath46 it is interesting to notice that in a weakly coupled thermal qgp one has the well - defined separation of scales , @xmath47 , @xmath48 , @xmath49 . the matter becomes strongly interacting when the scales `` collapse '' together . one way for that to happen is to have the system become really strongly coupled @xmath50 which likely will be accompanied by change of underlying degrees of freedom @xcite . the other possibility , as in the case of glasma , is when the system is highly off - equilibrium and overpopulated @xmath51 in this case all the scales also become of the same order @xmath52 and make the system emerge as a strongly interacting matter . clearly , both the full form @xmath27 in ( [ c22_full ] ) and the small angle approximation form in ( [ c22_sa ] ) conserve the energy as well as particle number , as they should . in addition the bose - einstein distribution @xmath53 $ ] with any @xmath54 and @xmath55 ( in correspondence to the two conserved quantities ) is the fixed point solution that makes both ( [ c22_full ] ) and ( [ c22_sa ] ) vanish . we now turn to the collision kernel from the @xmath16 process as depicted in fig.[f2 ] . we denote the particle we are watching with momentum @xmath38 , the softest external momentum with @xmath56 , the exchange internal four - momentum with @xmath57 , and then the rest external momenta with @xmath58 . the @xmath16 collision kernel can then be split into two pieces in which the particle @xmath38 is on the two - particle side or three - particle side respectively ( see fig.[f2 ] ) : @xmath59&=&{\cal c}^a_{2 \leftrightarrow 3}+{\cal c}_{2 \leftrightarrow 3}^b,\\ { \cal c}_{2\leftrightarrow 3}^a&=&\frac{1}{n_g}\frac{1}{6}\int_{123k}\frac{1}{2e_p}|m_{1p \leftrightarrow 23k}|^2 ( 2{\pi})^4{\delta}^4(p+p_1-p_2-p_3-k){\nonumber\\}&&\times[(1+f_p)(1+f_1)f_2f_3f_k - f_pf_1(1+f_2)(1+f_3)(1+f_k)],{\nonumber\\}{\cal c}_{2 \leftrightarrow 3}^b&=&\frac{1}{n_g}\frac{1}{4}\int_{123k}\frac{1}{2e_p}|m_{23 \leftrightarrow 1kp}|^2 ( 2{\pi})^4{\delta}^4(p+p_1+k - p_2-p_3){\nonumber\\}&&\times[(1+f_p)(1+f_1)(1+f_k)f_3f_2-f_pf_1f_k(1+f_3)(1+f_2)],\end{aligned}\ ] ] where the gluon labeled by @xmath56 will be treated as the soft emitted or absorbed gluon . the factor @xmath60 counteracts the @xmath61 equivalent permutations in @xmath62 in process @xmath63 ( see fig . [ f2 ] left panel ) and the factor @xmath64 counteracts the @xmath65 equivalent permutations in @xmath66 and @xmath67 in process @xmath68 ( see fig . [ f2 ] right panel ) . we note that the graphs in fig . [ f2 ] are used to make the kinematics clear and it does not mean that only these two diagrams contribute : there are actually 25 different diagrams for @xmath69 and 25 diagrams for @xmath70 . so the full matrix element @xmath71 is obtained by calculating 25 feynman diagrams and it contains 6 equivalent kinematic setups in accordance with 6 permutations in @xmath62 ( see appendix [ amplitude ] ) . we can then choose the kinematic setup corresponding to the fig . [ f2 ] left panel , and multiply 6 to account other 5 kinematic setups . similarly , we can fix the kinematics for @xmath72 as in the fig . [ f2 ] right panel , and multiply 4 to get @xmath73 . thus we obtain @xmath74,{\nonumber\\}{\cal c}_{2 \leftrightarrow 3}^b&=&\frac{1}{n_g}\int_{123k}\frac{1}{2e_p}|m^b_{23\leftrightarrow1kp}|^2 ( 2{\pi})^4{\delta}^4(p+p_1+k - p_2-p_3){\nonumber\\}&&\times[(1+f_p)(1+f_1)(1+f_k)f_2f_3-f_pf_1f_k(1+f_2)(1+f_3)],\end{aligned}\ ] ] where @xmath75 are the matrix element with the kinematics fixed according to fig . while the exact @xmath16 matrix element is known @xcite , it is hard to be directly used in a kinetic approach . following many previous studies involving this process @xcite , we will use the so - called gunion - bertsch formula which is the collinear approximation and small angle approximation form of the exact matrix element and has been shown to give the dominant contribution in many cases . leaving the technical details to the appendix [ amplitude ] , we here quote the gunion - bertsch matrix element which is at the leading order in soft @xmath76 and @xmath56 expansion : @xmath77 : @xmath78 . ( right ) a typical feynman diagram contributing to @xmath79 : @xmath80.,title="fig:",width=207 ] : @xmath78 . ( right ) a typical feynman diagram contributing to @xmath79 : @xmath80.,title="fig:",width=207 ] like in the @xmath81 case , the matrix elements are dominated by the regime of very soft @xmath76 and @xmath56 . we can thus further simplify the collision kernel using similar small angle approximation as in the elastic case . there is though additional subtlety as now there are two soft scales . in fact , as shown by the detailed analysis in appendix [ colli ] , the whole collision kernel can be separated into two pieces corresponding to contributions from different kinetic domains . ( note that both graphs in fig . [ f2 ] contribute to each of these domains . ) in the domain with @xmath56 being the softest scale , i.e. the ultra - soft emission and absorption , the @xmath16 essentially reduces to an effective @xmath29 scattering with a collinear splitting / merging , and the resulting contribution to the collision kernel becomes @xmath82 where the momentum labels @xmath83 and @xmath76 are as shown in fig . [ f2 ] , and we also introduced the term @xmath84 arising from the splitting function integration . its explicit form is given in appendix [ colli ] . for the above form , one can further simplify the effective @xmath81 part using small angle approximation as done in the elastic case . all the details are presented in appendix [ colli ] . on the other hand in the domain with @xmath76 being the softest scale , the @xmath85 process effectively becomes a nearly collinear @xmath86 emission / combining process preceded by a small angle @xmath81 scattering that brings one incoming particle slightly off - shell . this part contributes the following to the collision kernel : @xmath87}\nonumber \\ & & \qquad + \frac{1}{(1-z)^4 } { \left[}g_pg_{zp/(1-z)}f_{p/(1-z)}-f_pf_{zp/(1-z)}g_{p/(1-z)}){\right]}\bigg\}.\end{aligned}\ ] ] note that in the @xmath88-integration we introduce an upper cut @xmath89 : physically this is because that @xmath56 is the softest external momentum so the @xmath90 condition would require @xmath91 . in both kernels above , a number of infrared divergences appear . we will use the debye scale @xmath92 as the infrared cut , e.g. @xmath93 . we will also treat all leading logs as order one constant . again more detailed discussions are included in the appendix [ colli ] . lastly , one can show that the inelastic kernel conserves energy while not particle number , and the fixed point solution ( i.e. the equilibrium distribution ) is the bose - einstein distribution _ without _ chemical potential , @xmath94 . finally we combine the @xmath95 and @xmath96 kernels , and the final kinetic equation under small angle approximation and collinear approximation reads @xmath97+{\cal c}^{\rm eff}_{1 \leftrightarrow 2}[f_p],\end{aligned}\ ] ] where @xmath98 the expression for @xmath99 is @xmath100}\rc,\end{aligned}\ ] ] with @xmath101 parameterizing the contribution from @xmath102 to effective @xmath15 kernel . the expression for @xmath103 is @xmath104}{\nonumber\\}&&+\int_{0}^{z_c } \frac{dz}{(1-z)^4z } { \left[}g_pg_{zp/(1-z)}f_{p/(1-z)}-f_pf_{zp/(1-z)}g_{p/(1-z)}){\right]}\bigg\},\end{aligned}\ ] ] where the constant @xmath105 parameterizes the relative ratio of the order one constants between the elastic and inelastic kernels and the cut @xmath89 in @xmath88-integration should be small to be consistent with the kinematics @xmath106 . the expressions for @xmath107 and @xmath108 can be found in appendix [ colli ] . a few comments are in order here : + 1 ) the kernel ( [ kernel_22 ] ) conserves energy and particle number with fixed point solution @xmath109 $ ] , while the kernel ( [ kernel_12 ] ) only conserves energy with fixed point solution @xmath110 $ ] , so the total kernel ( [ kernel_all ] ) conserves only energy and the equilibrium solution should be @xmath110 $ ] without any chemical potential which is different from the pure elastic case ; + 2 ) in the nearly equilibrium case with @xmath111 , @xmath112 and @xmath113 , the elastic collision rate scales as @xmath114 and the inelastic rate scales also as @xmath114 so they are at parametrically the same order ; + 3 ) in the glasma - like overpopulated case with @xmath115 , @xmath116 and @xmath117 , the elastic collision rate scales as @xmath118 and the inelastic rate scales also as @xmath119 so again they are parametrically at the same order . + we therefore see that the effect of the inelastic collision is parametrically as important as the elastic one both near and far from equilibrium , and including the inelastic collision qualitatively changes the ultimate equilibrium solution . it is clear that even starting from highly overpopulated initial condition , eventually there will be no chemical potential nor any condensate in the final thermal distribution with the presence of inelastic collision . however , the very important question that has not been understood , is how the inelastic collision will affect the transient dynamical off - equilibrium condensation driven by the pure elastic evolution starting from initial high overpopulation . will the system still reach the onset of such condensation ? will the inelastic collision speed up , delay , or completely eliminate such an onset ? we will address these questions by numerically solving the above kinetic equations . in this section , we numerically solve the kinetic equation ( [ kernel_all ] ) , starting with a glasma - type initial condition as follows : @xmath120 with @xmath3 the saturation scale . we use @xmath3 as unit for all momenta / enegy etc and use @xmath121 as unit for time . as studied in @xcite , with a given initial occupation @xmath122 , the overpopulation parameter is @xmath123 and when @xmath124 , the system is overpopulated as compared with the stefan - boltzmann limit and the system will reach onset of condensation when there is only elastic collision . for simplicity we fix the initial occupation @xmath125 which is in the overpopulated regime . note that the constant @xmath126 can be absorbed by a redefinition of time variable , @xmath127 which we will use from now on . we therefore are left with three parameters , @xmath108 , @xmath107 , and @xmath89 . the inelastic contribution will increase with increasing @xmath108 and @xmath89 while the elastic will increase with increasing @xmath107 . we will study the effect of inelastic collision by comparison with the purely elastic case ( @xmath128 versus @xmath129 ) and by varying the strength of the inelastic kernel . let us first study the kinetic equation with only the inelastic kernel ( [ kernel_12 ] ) . although this is not a realistic modeling of the glasma system , it is a very interesting problem on its own and it also serves as a very useful check of whether the derived inelastic kernel produces the physically expected dynamics . furthermore it is a useful benchmark for a contrast with the evolution driven by both elastic and inelastic processes . we will choose @xmath130 without loss of generality because @xmath108 can be absorbed by redefining the time variable , @xmath131 . the kinetic equation is then solved numerically with only the inelastic kernel and with overpopulated initial condition @xmath125 . in fig . [ fp_inf ] ( left panel ) , we show the distribution function @xmath132 at various time moments . it can be seen that the @xmath132 , starting from the initial glasma - type shape , smoothly evolves into a bose - einstein distribution after about @xmath133 , with a temperature being the supposed value required by energy conservation and a vanishing chemical potential . very different from the evolution driven by elastic kernel with the same initial condition ( see ref . @xcite ) , in the present purely inelastic case , the system is simply thermalized and during the thermalization there appears no onset of singularity ( bose - einstein condensate ) in the distribution . this can be explicitly checked by looking at the occupation at the smallest grid point @xmath134 as a function of time ( fig . [ fp_inf ] right panel ) : its value has a transient behavior of rapid rise and fall and then settles to the supposed thermal value . clearly , even with overpopulated initial condition , the inelastic process alone does not generate a dynamical onset of bose - einstein condensate , as one may reasonably expect . we have also studied the evolution of global quantities , with the results shown in fig . [ den_inf ] for the number density ( left panel ) as well as the entropy density ( right panel ) . the initial high overpopulation in gluon number is efficiently reduced by the inelastic processes , and the number density drops toward the supposed thermal value determined by equilibrium temperature . the entropy density on the other hand grows rapidly and approaches the thermal value as well . again all these features provide clear indication that with the purely inelastic kernel the system is simply thermalized as it should be . from this study , we conclude that the @xmath135 inelastic processes as described by our derived kernel ( [ kernel_12 ] ) thermalize the system efficiently and eliminate excessive gluons from initial conditions effectively , and by these processes _ alone _ no dynamic onset of bec is to occur . with such benchmark case understood , it is thus tempting to see what will happen when the elastic @xmath136 processes are also included in addition to the inelastic . as we will show in the next subsections , the evolution of the system gets dramatically changed . at different time moments during the evolution for purely inelastic collisions ; ( right ) the occupation at the smallest grid point @xmath134 as a function of time for purely inelastic collisions.,title="fig:",width=209 ] at different time moments during the evolution for purely inelastic collisions ; ( right ) the occupation at the smallest grid point @xmath134 as a function of time for purely inelastic collisions.,title="fig:",width=209 ] ( left ) and entropy density @xmath137 ( right ) , both normalized by the corresponding equilibrium values , as a function of time for purely inelastic collisions.,title="fig:",width=209 ] ( left ) and entropy density @xmath137 ( right ) , both normalized by the corresponding equilibrium values , as a function of time for purely inelastic collisions.,title="fig:",width=209 ] let us now briefly summarize the kinetic evolution from overpopulation toward onset of condensation in the purely elastic case , as reported in @xcite . the elastic kernel can be re - written in terms of flux as @xmath138\end{aligned}\ ] ] the strong overpopulation leads to a particle flux cascade toward the infrared regime . analysis of the small @xmath38 regime shows that it will quickly develop a local thermal form @xmath139 $ ] with @xmath140 , and the incoming flux will drive the ( negative ) @xmath141 to eventually vanish and reach the onset of a dynamical condensation . this picture is numerically verified in great details in @xcite . our discussion of the onset of condensation will stay in this picture ( as the elastic term is still present and its flux drives the small @xmath38 behavior ) and we will study how the inelastic process modifies it . starting with the overpopulated initial condition ( [ initial ] ) we have numerically evolved the kinetic equations ( [ kernel_all ] ) , ( [ kernel_22 ] ) , and ( [ kernel_12 ] ) with given set of parameters . shown in fig . [ fz1_fp ] is the solution with @xmath130 . in fig . [ fz1_fp ] left we show the distribution function @xmath22 at different time moments , and one can see that even with the presence of inelastic term , the small @xmath38 part of the distribution is quickly filled up and becomes a local thermal form @xmath139 $ ] despite that the distribution in the wide range of ( bigger ) momentum region is still far from equilibrium shape , and the small @xmath38 part becomes steeper and steeper with time ( meaning decreasing @xmath142 ) . in fig . [ fz1_fp ] right we show the corresponding flux @xmath143 from the elastic kernel . again the flux behaves very similarly to the purely elastic case : one see a linear behavior at small p , @xmath144 and eventually upon onset of condensation the flux diverges ( see the blue curve near @xmath145 ) . to get an intuitive idea of the contribution of the inelastic kernel , we plot the @xmath146 and @xmath147 in fig . [ fz1_c12 ] . one can see that the kernel is large and positive at small @xmath38 , small and positive at large @xmath38 , while negative at intermediate @xmath38 . this could be qualitatively understood : significant number of particles with intermediate momenta will merge toward high momenta and split toward low momenta that will fill up uv and ir region while decrease the occupation at intermediate momenta . we also notice that upon onset ( the blue dashed curve in the right panel ) the inelastic kernel near @xmath145 shows a divergent behavior in consistency with the elastic flux behavior . one can directly examine the locally determined @xmath148 and @xmath141 ( see @xcite for details ) at each time moment during the evolution : these results are shown in fig . [ fz1_tmu ] . here we also compare the results for different strength of the inelastic collision @xmath149 ( noting that the @xmath128 case corresponds to purely elastic collision ) . in all cases we can see that the local `` chemical potential '' @xmath141 decreases rather rapidly toward zero . we also show the distribution @xmath132 at the smallest grid point in our calculation @xmath150 as a function of time in fig . [ fz1_thermal ] , which shows very rapid increase of the occupation in consistence with the vanishing of @xmath141 . what is most striking is that with increase values of @xmath108 this evolution toward the onset of condensation @xmath151 becomes faster and faster . the @xmath130 case is already much faster than the purely elastic case . this is to say , contrary to expectation that the inelastic process may `` kill '' the strong overpopulation quickly , the existence of inelastic collision actually * speeds up * significantly the process of populating the infrared regime and building up a local thermal form with vanishing @xmath141 , which when combined with the structure of elastic kernel will then lead to the onset of condensation . at different time moments during evolution ; ( right ) the flux @xmath143 defined in elastic kernel at different time moments during evolution.,title="fig:",width=209 ] at different time moments during evolution ; ( right ) the flux @xmath143 defined in elastic kernel at different time moments during evolution.,title="fig:",width=209 ] ( left ) and @xmath152 ( right ) at different time moments during evolution.,title="fig:",width=209 ] ( left ) and @xmath152 ( right ) at different time moments during evolution.,title="fig:",width=228 ] ( left ) and @xmath141 ( right ) as functions of time for different values of parameter @xmath108.,title="fig:",width=209 ] ( left ) and @xmath141 ( right ) as functions of time for different values of parameter @xmath108.,title="fig:",width=209 ] to understand better the influence of inelastic collision on the small @xmath38 region , let us examine the kernel ( [ kernel_12 ] ) for @xmath153 limit before the onset of condensation . provided that @xmath154 and that its derivatives with respect to @xmath38 at @xmath145 is also finite , we can have the expansion @xmath155 . if we look at a small enough @xmath38 regime , then all the involved momenta ( @xmath156 ) in the kernel ( [ kernel_12 ] ) can be considered small and we can use the expansion for them . this leads to @xmath157},\end{aligned}\ ] ] where we have introduced the constants @xmath158 all these @xmath159 s are positive for @xmath160 . clearly for sufficiently small @xmath38 the leading term in the inelastic kernel @xmath161 is _ always positive _ and becomes bigger and bigger with increasing @xmath122 ( which is a kind of `` self - amplification '' ) . this will tend to increase the particle number near @xmath145 very rapidly and the effect becomes stronger with increasing values of @xmath108 , which explains the behavior seen in fig . [ fz1_thermal ] . as a function of time for different values of parameter @xmath108.,width=209 ] physically this behavior may be understood in two ways . first note that the inelastic kernel has its fixed point to be @xmath162 which at small @xmath38 is @xmath163 so as long as @xmath164 is finite yet the inelastic kernel will try to fill it up toward @xmath165 . second , this is also related to the boson nature : if all involved particles are from small @xmath38 , then the merging rate is like @xmath166 while the splitting rate is like @xmath167 so the splitting `` wins '' due to bose enhancement for the final state particles and it increases particle number at small @xmath38 . to conclude , the inelastic kernel contribution is always positive at very small @xmath38 and it will catalyze and speed up the onset of a bose condensation ( which itself is driven by the elastic term at @xmath151 ) . . ( right ) the entropy density as a function of time for @xmath130 and @xmath128.,title="fig:",width=209 ] . ( right ) the entropy density as a function of time for @xmath130 and @xmath128.,title="fig:",width=209 ] while the inelastic kernel always increases the occupation at sufficiently small @xmath38 , it may still decrease the the total particle number . indeed as shown in fig . [ fz1_density ] ( left panel ) , the total particle number decreases when @xmath168 , and it decreases more rapidly for larger @xmath108 . to understand the change of particle number @xmath169 , one can integrate the two sides of the kinetic equation ( [ kernel_all ] ) and obtain @xmath170}\nonumber \\ & & = r\frac{i_a}{2{\pi}^2i_b } \int dp p^2\int_{0}^{z_c } \frac{dz}{1-z } { \left[}f_p + f_p f_{(1-z)p } + f_p f_{zp } - f_{(1-z)p}f_{zp}{\right]}.\end{aligned}\ ] ] from the above one can see the for the region @xmath171 the leading order in the @xmath88-integrand becomes @xmath172 and the contribution is positive , i.e. increasing particle number . for general @xmath88 , the @xmath88-integrand can be rewritten as @xmath173}.\end{aligned}\ ] ] we see that for not too small @xmath88 , the momenta @xmath174 , @xmath175 become well separated from @xmath38 and the second term in the above integrand becomes important and its contribution is negative which decreases the particle number . in fig . [ fz1_density ] ( right panel ) we also show the entropy density as a function of time and compare the case with @xmath130 and the purely elastic case with @xmath128 . one can see that with inelastic collision included , the entropy density increases much faster . that is , the inelastic process tends to accelerate the thermalization . ( left ) and @xmath141 ( right ) as functions of time for @xmath176 respectively.,title="fig:",width=209 ] ( left ) and @xmath141 ( right ) as functions of time for @xmath176 respectively.,title="fig:",width=209 ] finally we study the dependence on the parameter @xmath89 which is the kinematic cut to make sure the validity of the approximations used for the matrix element . generally speaking , with larger @xmath89 we include more effects from the inelastic process . to see how the results depend on @xmath89 , we fix other parameters and compare the results for different choices of @xmath89 . in fig . [ fz1_varyz1 ] , we show the local thermal form parameters @xmath148 ( left panel ) and @xmath141 ( right panel ) as functions of time for @xmath176 respectively . in fig . [ fz1_varyz2 ] , we show the occupation at the smallest grid point ( left panel ) and the total particle number ( right panel ) as functions of time for @xmath176 respectively . from the plots we can see that indeed with larger @xmath89 the @xmath122 increases faster and @xmath141 vanishes faster as expected for stronger inelastic effect . for the particle number , the case with @xmath177 actually has @xmath178 increasing with time , which can be understood fro m the analysis in the previous subsection . the particle number in both @xmath179 and @xmath180 cases drops with time and does so faster for larger @xmath89 . in passing let us mention that we have also studied the dependence on the parameter @xmath107 : basically increasing @xmath107 will enhance the effect from elastic collision and also speed up the thermalization in general , as well as reach onset of condensation at earlier times compared with @xmath181 case . ( left ) and the particle number density ( right ) as functions of time for @xmath176 respectively , title="fig:",width=209 ] ( left ) and the particle number density ( right ) as functions of time for @xmath176 respectively , title="fig:",width=209 ] in summary , we have studied the kinetic evolution of a highly overpopulated system starting from glasma - type initial condition with the presence of both elastic and inelastic collisions . using the gunion - bertsch formula for the @xmath16 matrix element , we have derived the inelastic collision kernel under the collinear and small angle approximations . putting together the inelastic kernel together with the previously obtained elastic kernel , we have then numerically solved the kinetic evolution for varied choices of parameters . our main finding is that the inelastic process has two effects : globally changing ( mostly reducing ) the total particle number , while locally at small @xmath38 always filling up the infrared regime extremely quickly . the latter effect is shown both from numerics and by analytic analysis . this effect significantly speeds up the emergence of local thermal form near @xmath145 and the vanishing of local `` chemical potential '' @xmath141 as previously found in the purely elastic collision case to lead to the onset of dynamical bose condensation . therefore in our present approach of including the inelastic scattering , we conclude that , contrary to some previously discussed expectations about the role of number non - conserving processes , the inelastic collision actually helps to build up the local `` critical form '' @xmath182 much faster and catalyzes the onset of condensation in the overpopulated glasma . our finding may sound counter - intuitive at first , as the usual conception would suggest that increasing the strength of the inelastic collisions tends to obstruct more effectively the formation of any condensate . it should however be emphasized that the evolution toward onset of bec that has been studied thus far is not the end of the story . our analysis addresses the evolution up to the onset of bec while does not treat the evolution afterwards . as is well known in the bec literature ( see e.g. @xcite ) , in order to describe the kinetic evolution of the system with the presence of condensate , a new set of kinetic equations is needed for an explicit description of the coupled evolution for a condensate plus a regular distribution . efforts are underway to derive these equations , and so far a kinetic study of the stage after bec onset for the glasma system has not been achieved to our best knowledge . however , it appears very plausible that the subsequent evolutions may develop as follows : immediately after onset , the strong ir flux will not cease right away but continue for a while and thus drive the condensate to grow in time ; at certain point , the time would be long enough to allow the inelastic processes to decrease the total number density adequately and cause the condensate to decay thus decreasing in time ; eventually the inelastic processes will be able to remove all excess gluons and lead to the thermal equilibrium state with neither condensate nor any chemical potential . while the detailed understanding of such dynamic processes can only be achieved through solving the new set of kinetic equations , one can reasonably expect that with increasing strength of the inelastic processes the whole evolution would be faster . thus the following overall picture may likely be the case : with increasing strength , the inelastic processes on one hand catalyze the onset of condensation initially , while on the other hand eliminate the fully formed condensate faster , thus limiting the time duration for the presence of condensate to be shorter . a schematic picture of such conjectured full evolution is shown in fig . [ conje ] , which is in line with the usual conception . it is worth mentioning that recent analysis in @xcite has shown that the the @xmath17 inelastic cross section from exact matrix element becomes significantly smaller than that from the gunion - bertsch formula , and amounts to @xmath183 of the @xmath136 cross section . it therefore seems very plausible that a realistic choice of @xmath108 value would be rather modest , which may imply a considerable time window for the condensate to be sizable and play an important role for the evolution . a complete investigation of the evolution including the condensate will be future project to be reported elsewhere . the authors are particularly grateful to l. mclerran for very helpful discussions . j.l . also thanks j .- blaizot , f. gelis , and r. venugopalan for collaborations and communications that motivated and benefited this work . j.l . acknowledges the riken bnl research center for partial support . the invariant @xmath184 or @xmath17 ( squared ) matrix element summed over all final states and also summed over all initial states is computed by considering 25 different feynman diagrams @xcite . we quote it here : @xmath185,{\nonumber\\}\end{aligned}\ ] ] where @xmath186 is the number of color , @xmath19 is the gluon degeneracy number , and other notations are defined as @xmath187 because @xmath188 is completely symmetry in @xmath189 , let s take @xmath190 and @xmath191 as the hard momenta in the entrance channel , @xmath192 and @xmath193 as the hard momenta in the exit channel , and @xmath194 as the emitted soft gluon . we denote the exchanging momentum as @xmath195 . a typical feynman diagram illuminating this setup is shown in fig . [ f1 ] . .,width=207 ] define the mandelstam variables as @xcite @xmath196 in addition , the following relations involving @xmath194 hold : @xmath197 in terms of the mandelstam variables , @xmath188 can be written as @xmath198,\end{aligned}\ ] ] where @xmath199.\nonumber\end{aligned}\ ] ] because @xmath188 is very singular when , for example , @xmath200 , we can expand it around these singularities order by order in some small momenta . to this end , let s assume that the exchanging momentum @xmath195 is small ( @xmath194 is also small ) . in this case @xmath201 are small , while @xmath202 are large , and @xmath203 . with other choices of picking the small exchanging momenta and soft emitted momenta , we can get other while equivalent expansions . we will come to this point latter . keeping only leading order and subleading order terms in @xmath76 and @xmath194 , we have @xmath204 in addition , we have @xmath205 then @xmath206 and @xmath188 can be simplified as @xmath207+o{\left(}\frac{t}{s}{\right)}{\nonumber\\}&=&32g^6n_c^3n_g\frac{{\cal n}}{tt'}\frac{s+(k_1+k_2)\cdot k_5}{s^2(k_1\cdot k_5)(k_2\cdot k_5)}+o{\left(}\frac{k_5 ^ 2}{s},\frac{q^2}{s},\frac{k_5\cdot q}{s}{\right)}{\nonumber\\}&=&8g^6n_c^3n_g\frac{s^2}{tt'}\frac{s-3(k_1+k_2)\cdot k_5}{(k_1\cdot k_5)(k_2\cdot k_5)}+o{\left(}\frac{k_5 ^ 2}{s},\frac{q^2}{s},\frac{k_5\cdot q}{s}{\right)}{\nonumber\\}&=&32g^6n_c^3n_g\frac{(k_1\cdot k_2)^2}{q^2(q - k_5)^2}\frac{2k_1\cdot k_2 - 3(k_1+k_2)\cdot k_5}{(k_1\cdot k_5)(k_2\cdot k_5)}+o{\left(}\frac{k_5 ^ 2}{s},\frac{q^2}{s},\frac{k_5\cdot q}{s}{\right)}.{\nonumber\\}\end{aligned}\ ] ] in the center - of - mass frame of @xmath190 and @xmath191 , it goes to @xmath208 and @xmath209 is the gunion - bertsch formula @xcite . here we have used the fact that @xmath210 and @xmath211 ^ 2+o[({{\mathbf q}}_\perp-{{\mathbf k}}_{5\perp})^2(q_0-k_{50})/|{{\mathbf k}}_1|]$ ] for soft @xmath76 and @xmath194 . thus , the gunion - bertsch formula is the leading order result in soft @xmath76 and @xmath56 expansion ; and eq . ( [ m23 ] ) is the result including both the leading ( gunion - bertsch ) and the subleading order terms . higher order terms can also be obtained , but we will not use them . note that one can naively boosts the gunion - bertsch formula from the center - of - mass frame to a general frame by using the replacements @xmath212 , @xmath213 , and @xmath214 : @xmath215 this expression coincides with eq . ( [ m23 ] ) at leading order but not at next to leading order in soft @xmath76 and @xmath194 expansion . the collision kernel @xmath216 has a very complicated structure , in this section , we simplify it by taking the collinear approximation , i.e. , @xmath217 or @xmath218 . we rewrite the collision kernels @xmath219 and @xmath220 as ( we denote @xmath221 by @xmath222 ) @xmath223 where we introduced two auxiliary integrations over @xmath83 and @xmath76 . the kinematics is shown in fig . [ f2 ] , and the expressions for @xmath224 and @xmath225 are given by @xmath77 second , if @xmath231 , then , under collinear approximation , @xmath232 is nearly parallel to either @xmath233 or @xmath234 . for @xmath217 , @xmath235\sim-|{{\mathbf k}}|{{{\mathbf q}}'}^2_\perp / e_1 + |{{\mathbf k}}||{{\mathbf q}}|{\theta}_{1k}\ll q^2 $ ] ; for @xmath218 , @xmath236 . thus in the small angle approximation plus the collinear approximation , if @xmath231 , we can approximate @xmath237 as @xmath238 , up to @xmath239 order , we can neglect @xmath240 in the denominator and @xmath56 as well as @xmath241 and @xmath242 in the numerator , @xmath243[(p_1+q)\cdot k]}{\nonumber\\}&\approx & 64g^6n_c^3n_g\frac{[(p_1+q)\cdot ( p - q)]^3}{(q^2)^2 [ ( p - q)\cdot k][(p_1+q)\cdot k]}{\nonumber\\}&\approx & 64g^6n_c^3n_g\frac{(p_1\cdot p)^3}{(q^2)^2 ( p\cdot k)(p_1\cdot k)}{\nonumber\\}&= & third , if @xmath244 , under the collinear approximation , for @xmath217 , @xmath245 because @xmath246 ; for @xmath218 , @xmath247 because @xmath248 . thus the collinear approximation simplifies the matrix element also when @xmath244 : @xmath249 fourth , the whole kinematic ( phase ) space can be separated into two parts , one for @xmath231 and another for @xmath250 . we would expect that at kinematic region with @xmath231 , the @xmath17 process may be regarded as a @xmath136 hard " process with one additional soft " gluon emitted or absorbed by one of the hard " gluons ; with @xmath250 the @xmath17 process can be regarded as an effective @xmath135 process with a spectator gluon joined to make the effective @xmath135 matrix element nonzero ( the matrix element of the @xmath135 process is zero for massless gluons ) . thus we separate the collision kernel as @xmath251 with @xmath252,{\nonumber\\}{\cal c}^<_{2\leftrightarrow3}&=&\frac{1}{n_g}\int\frac{d^4q}{(2{\pi})^4 } \int_{123l}\int_k^{q < k}\frac{2e_l}{2e_p}\int\frac{dl_0 } { 2{\pi}}(2{\pi})^8{\delta}^4(p+q - l){\delta}^4(p_1-q - p_2){\nonumber\\}&&\times [ ( 2{\pi})^4{\delta}^4(l - k - p_3)|m_{1p\leftrightarrow23k}^a|^2(g_pg_1f_2f_3f_k - f_pf_1g_2g_3g_k){\nonumber\\ } & & + ( 2{\pi})^4{\delta}^4(l+k - p_3)|m_{23\leftrightarrow1kp}^b|^2(g_pg_1g_kf_2f_3-f_pf_1f_kg_2g_3)].\nonumber\end{aligned}\ ] ] expand the integrand of @xmath253 in terms of @xmath56 and keep the leading order terms : @xmath254{\nonumber\\}&\approx&\frac{1}{n_g}\int\frac{d^4q}{(2{\pi})^4}\int_{12l}\int_k^{q > k}\frac{1}{2e_p}\int\frac{dl_0 } { 2{\pi}}(2{\pi})^8{\delta}^4(p+q - l){\delta}^4(p_1-q - p_2){\nonumber\\}&&\times|m_{1p\leftrightarrow23k}^a|^2 [ ( 2{\pi}){\delta}(l_0-|{{\bf l}}|)(g_pg_1f_2f_lf_k - f_pf_1g_2g_lg_k){\nonumber\\ } & & + ( 2{\pi}){\delta}(l_0-|{{\bf l}}|)(g_pg_1g_kf_2f_l - f_pf_1f_kg_2g_l)]{\nonumber\\}&=&\frac{1}{n_g}\int\frac{d^4q}{(2{\pi})^4}\int_{12l}\int_k^{q > k}\frac{1}{2e_p } ( 2{\pi})^4{\delta}^4(p+q - l)(2{\pi})^4{\delta}^4(p_1-q - p_2){\nonumber\\}&&\times|m_{1p\leftrightarrow2l}|^2 \frac{2g^2n_c(p\cdot p_1)}{(p\cdot k)(p_1\cdot k ) } ( 1 + 2f_k)(g_pg_1f_2f_l - f_pf_1g_2g_l),\end{aligned}\ ] ] where in the last equality @xmath255 is on - shell , and the @xmath136 matrix element is latexmath:[\ ] ] it can be shown that the collision kernel @xmath293 conserves energy , i.e , @xmath360=0 $ ] . l. d. mclerran and r. venugopalan , phys . d * 49 * , 2233 ( 1994 ) [ hep - ph/9309289 ] . l. d. mclerran and r. venugopalan , phys . d * 49 * , 3352 ( 1994 ) [ hep - ph/9311205 ] . j. p. blaizot and a. h. mueller , nucl . * b289 * , 847 ( 1987 ) . e. iancu and r. venugopalan , in * hwa , r.c . et al . : quark gluon plasma * 249 - 3363 . [ hep - ph/0303204 ] ; h. weigert , prog . phys . * 55 * , 461 - 565 ( 2005 ) . [ hep - ph/0501087 ] ; f. gelis , e. iancu , j. jalilian - marian and r. venugopalan , ann . nucl . part . sci . * 60 * , 463 - 489 ( 2010 ) . [ arxiv:1002.0333 [ hep - ph ] ] . p. romatschke and r. venugopalan , phys . lett . * 96 * , 062302 ( 2006 ) [ arxiv : hep - ph/0510121 ] ; p. romatschke and r. venugopalan , phys . d * 74 * , 045011 ( 2006 ) [ arxiv : hep - ph/0605045 ] ; p. romatschke and r. venugopalan , eur . j. * a29 * , 71 - 75 ( 2006 ) . [ hep - ph/0510292 ] . u. w. heinz , in relativistic heavy ion physics , landolt - boernstein new series , i/23 , edited by r. stock ( springer verlag , new york,2010 ) chap . 5 [ arxiv:0901.4355 [ nucl - th ] ] . u. heinz and r. snellings , ann . nucl . part . sci . * 63 * , 123 ( 2013 ) [ arxiv:1301.2826 [ nucl - th ] ] . a. h. mueller , a. i. shoshi , and s. m. h. wong , phys . lett . * b632 * , 257 - 260 ( 2006 ) . [ hep - ph/0505164 ] ; a. h. mueller , a. i. shoshi , and s. m. h. wong , eur . j. * a29 * , 49 - 52 ( 2006 ) . [ hep - ph/0512045 ] . a. h. mueller , a. i. shoshi , and s. m. h. wong , nucl . phys . * b760 * , 145 - 165 ( 2007 ) . [ hep - ph/0607136 ] . j. -p . blaizot , f. gelis , j. -f . liao , l. mclerran , and r. venugopalan , nucl . a * 873 * , 68 ( 2012 ) . [ arxiv:1107.5296 [ hep - ph ] ] . j. -p . blaizot , j. liao , and l. mclerran , nucl . a * 920 * , 58 ( 2013 ) [ arxiv:1305.2119 [ hep - ph ] ] . a. kurkela and g. d. moore , jhep * 1112 * , 044 ( 2011 ) [ arxiv:1107.5050 [ hep - ph ] ] . a. kurkela and g. d. moore , jhep * 1111 * , 120 ( 2011 ) [ arxiv:1108.4684 [ hep - ph ] ] . t. epelbaum and f. gelis , nucl . a * 872 * , 210 ( 2011 ) [ arxiv:1107.0668 [ hep - ph ] ] . f. gelis , j. phys . * 381 * , 012021 ( 2012 ) [ arxiv:1110.1544 [ hep - ph ] ] . j. berges and d. sexty , phys . lett . * 108 * , 161601 ( 2012 ) [ arxiv:1201.0687 [ hep - ph ] ] . a. kurkela and g. d. moore , phys . d * 86 * , 056008 ( 2012 ) [ arxiv:1207.1663 [ hep - ph ] ] . s. schlichting , phys . d * 86 * , 065008 ( 2012 ) [ arxiv:1207.1450 [ hep - ph ] ] . j. berges , j. -p . blaizot , and f. gelis , j. phys . g * 39 * , 085115 ( 2012 ) [ arxiv:1203.2042 [ hep - ph ] ] . j. berges , k. boguslavski , s. schlichting , and r. venugopalan , arxiv:1303.5650 [ hep - ph ] . t. epelbaum and f. gelis , phys . lett . * 111 * , 232301 ( 2013 ) [ arxiv:1307.2214 [ hep - ph ] , arxiv:1307.2214 [ hep - ph ] ] . m. attems , a. rebhan , and m. strickland , phys . d * 87 * , 025010 ( 2013 ) [ arxiv:1207.5795 [ hep - ph ] ] . m. chiu , t. k. hemmick , v. khachatryan , a. leonidov , j. liao , and l. mclerran , nucl . a * 900 * , 16 ( 2013 ) [ arxiv:1202.3679 [ nucl - th ] ] . j. -p . blaizot , f. gelis , j. liao , l. mclerran , and r. venugopalan , nucl . a904 - 905 * 2013 * , 829c ( 2013 ) [ arxiv:1210.6838 [ hep - ph ] ] . j. liao , j. phys . ser . * 420 * , 012031 ( 2013 ) [ arxiv:1209.2998 [ hep - ph ] ] . m. ruggieri , f. scardina , s. plumari and v. greco , arxiv:1312.6060 [ nucl - th ] . j. -p . blaizot and e. iancu , phys . rept . * 359 * , 355 ( 2002 ) [ hep - ph/0101103 ] . j. liao and e. shuryak , phys . c * 75 * , 054907 ( 2007 ) ; phys . * 101 * , 162302 ( 2008 ) ; phys . lett . * 102 * , 202302 ( 2009 ) ; phys . lett . * 109 * , 152001 ( 2012 ) . j. f. gunion and g. bertsch , phys . d * 25 * , 746 ( 1982 ) . b. arnold , g. d. moore , and l. g. yaffe , jhep * 0011 * , 001 ( 2000 ) [ hep - ph/0010177 ] . b. arnold , g. d. moore , and l. g. yaffe , jhep * 0305 * , 051 ( 2003 ) [ hep - ph/0302165 ] . b. arnold , c. dogan , and g. d. moore , phys . d * 74 * , 085021 ( 2006 ) [ hep - ph/0608012 ] . z. xu , c. greiner , and h. stocker , phys . * 101 * , 082302 ( 2008 ) [ arxiv:0711.0961 [ nucl - th ] ] . o. fochler , j. uphoff , z. xu , and c. greiner , phys . d * 88 * , 014018 ( 2013 ) [ arxiv:1302.5250 [ hep - ph ] ] . a. el , f. lauciello , c. wesp , z. xu , and c. greiner , arxiv:1207.5331 [ hep - th ] . j. -w . chen , h. dong , k. ohnishi , and q. wang , phys . b * 685 * , 277 ( 2010 ) [ arxiv:0907.2486 [ nucl - th ] ] . j. -w . chen , j. deng , h. dong , and q. wang , phys . c * 87 * , 024910 ( 2013 ) [ arxiv:1107.0522 [ hep - ph ] ] . j. -w . chen , y. -f . liu , y. -k . song , and q. wang , phys . d * 87 * , no . 3 , 036002 ( 2013 ) [ arxiv:1212.5308 [ hep - ph ] ] . f. a. berends , r. kleiss , p. de causmaecker , r. gastmans , and t. t. wu , phys . b * 103 * , 124 ( 1981 ) . d. v. semikoz and i. i. tkachev , phys . lett . * 74 * , 3093 ( 1995 ) . d. v. semikoz and i. i. tkachev , phys . d * 55 * , 489 ( 1997 ) . b. zhang , j. phys . * 420 * , 012035 ( 2013 ) [ arxiv:1208.1224 [ nucl - th ] ] . b. zhang , arxiv:1307.6234 [ nucl - th ] . r. abir , c. greiner , m. martinez , and m. g. mustafa , phys . d * 83 * , 011501 ( 2011 ) [ arxiv:1011.4638 [ nucl - th ] ] . t. bhattacharyya , s. mazumder , s. k. das , and j. alam , phys . d * 85 * , 034033 ( 2012 ) [ arxiv:1106.0609 [ nucl - th ] ] . g. altarelli and g. parisi , nucl . b * 126 * , 298 ( 1977 ) . v. n. gribov and l. n. lipatov , sov . j. nucl . phys . * 15 * , 438 ( 1972 ) [ yad . * 15 * , 781 ( 1972 ) ] . y. l. dokshitzer , sov . jetp * 46 * , 641 ( 1977 ) [ zh . eksp . teor . fiz . * 73 * , 1216 ( 1977 ) ] .

we study the dissipative dynamics of a harmonic oscillator which couples linearly through its position and its momentum to two independent heat baths at the same temperature . we argue that this model describes a large spin in a ferromagnet . we find that some effects of the two heat baths partially cancel each other . this leads to unexpected features such as underdamped oscillations and long relaxation times in the strong coupling regime . such a partial frustration of dissipation can be ascribed to the canonically conjugate character of position and momentum . we compare this model to the scenario where a single heat bath couples linearly to both the position and the momentum of the central oscillator . in that case less surprising behavior occurs for strong coupling . the dynamical evolution of the quantum purity for a single and a double wave packet is also investigated .

cond-mat0608484

the dissipative harmonic oscillator has for long attracted considerable interest as a prototype of open quantum system . the early work of magalinskii @xcite and ullersma @xcite focusing mainly on an ohmic environment and the weak coupling limit has been later extended in many different aspects mol70,dek77,tal81,schm82,grab83,grab84,lind84,ris85,haak85,grab88,unr89,har90 as , for instance , to strong coupling @xcite , to non markovian noise @xcite or to nonlinear coupling @xcite . much activity was boosted by the work of caldeira and leggett @xcite on the dissipative mechanics of a macroscopic quantum variable . quantum decoherence has also been addressed in refs . @xcite . a comprehensive review can be found in the textbook by weiss @xcite__. _ _ in the above context relatively little attention has been paid to effects arising from the coupling to different system variables . the system variable which couples to the heat bath is most often assumed to be the position _ q_. in the sequel we refer to this model as the _ _ q__oscillator . the choice of @xmath0 as the coupling variable was favoured in ref . there it was argued that a complex dissipative environment can be modelled by a bath of harmonic oscillators , with the couplings parameters chosen to yield a langevin equation for the @xmath0 variable in the semiclassical limit . it was shown later @xcite ( see also ref . kohl04 ) that , in a superconducting josepshon juntion , the oscillator bath model can be derived microscopically from the coupling of the phase variable to the quasiparticle bath , with the phase playing the role of position . at that time the possibility of having a second bath coupled to the momentum @xmath1 variable was ruled out . however , in a superconducting weak link , the electromagnetic field couples to the relative number variable , which in the above scheme plays the role of momentum @xcite . if the coupling to the phase is neglected , the semiclassical behavior of the particle number is governed by an abraham lorentz equation kohl04,jack75 . the coupling to the momentum has been also been considered by leggett @xcite and by cuccoli _ a systematic study of the combined role of the electromagnetic and quasiparticle fields is presented in ref . thus josephson junctions provide a scenario where the distinction between the different baths and their different coupling mechanisms becomes imperious . our study has also been motivated by the work of castro neto , novais and coworkers @xcite . to investigate the problem of a spin @xmath2 impurity in an antiferromagnetic environment , they studied a generalized spin boson problem using renormalization group techniques . the localized spin couples with two components to the environment . they found a crossover from incoherent to coherent transitions in the spin correlation functions in certain parameter ranges . there the term _ quantum frustration _ was coined , which refers to the reduction of the effective interaction with the environment due to the non commutativity of the coupled operators . in this article , we argue that a harmonic oscillator coupled through position and momentum to two independent baths is a suitable representation of a large spin impurity in a ferromagnetic system . we find a weaker cancellation of the effects of the two baths . since that situation corresponds to a large ( quasiclassical ) spin , we have referred to this weaker form of frustration as _ quasiclassical frustration _ @xcite . here we wish to investigate systematically the scenario where a particle couples through its position and momentum to two baths , which are independent but have the same temperature . as a paradigmatic system we choose a harmonic oscillator . this choice is motivated not only by mathematical convenience . a more fundamental reason is that the symmetrical roles played by position and momentum makes the harmonic oscillator the natural ground to study the difference and the interplay ( when both are present ) between position and momentum coupling . an explanatory comment is in order . as is well known , a coupling to @xmath0 can be converted into a coupling to @xmath1 through a suitable canonical transformation . however , the two representations bear an important difference . only in one of them is it possible to represent the system bath interaction as the coupling of the quantum variable to a system of _ otherwise independent _ oscillators . a characteristic example is provided by a particle of charge @xmath3 in an electromagnetic field , where a velocity coupling model " @xcite seems to apply . minimal coupling ( @xmath4 ) generates not only an interaction term @xmath5 but also a diamagnetic term @xmath6 which can be rightly interpreted as an interaction between the effective oscillators . a unitary transformation @xmath7 with @xmath8_{-}=i\delta _ { ij}$ ] , acting on @xmath9 , removes the coupling @xmath5 . this happens at the expense of generating a coupling @xmath10 between the position and the electric field @xmath11 . importantly , in this new representation no quadratic field term is left , i.e. the charge couples to a set of independent photons . thus , in the precise language which we propose here , a charged particle couples to the electromagnetic field through its position @xmath12 . in sec . [ two independent baths ] we study the model of an oscillator coupled to two different baths . the general formalism is sketched in sec . [ general results ] . in sec . [ ohm ] we investigate the case where the two independent ohmic baths couple linearly to position and momentum . a related feature is the possibility of inducing a crossover from overdamped to underdamped oscillations by _ increasing _ the damping coefficient . surprisingly , we obtain underdamped equilibrium oscillations for arbitrarily strong coupling . [ gencase ] is devoted to the properties of spectral functions under rather general combinations of double bath systems . time dependent phenomena and implications for quantum decoherence are discussed in sec . [ timeev ] and sec . [ timedep ] for the case of a double ohmic bath . in sec . [ onebath ] we consider the situation where an oscillator couples to a single bath with the most general form of linear coupling to both position and momentum . we find that the most prominent feature of the double bath model , namely , the existence of underdamped oscillations for arbitrarily strong coupling , disappears in the case of single bath dissipation . the general form of a hamiltonian describing an oscillator coupled to two independent baths is @xcite @xmath13where @xmath14_{-}=i$ ] and all operators are dimensionless and @xmath15 . the form of the hamiltonian highlights the symmetry between @xmath0 and @xmath1 . the notion of mass is avoided by @xmath16 @xmath17 @xmath18 . for the fluctuating pieces we assume that they are linear in the bath variables and independent of @xmath1 and @xmath0 , @xmath19after two unitary transformations @xmath20 and @xmath21 one arrives at the hamiltonian @xmath22where the short - hand notation @xmath23 is used . the model is a caldeira leggett type of hamiltonian @xcite . it describes a harmonic oscillator with momentum @xmath1 and position @xmath0 , each variable being coupled to a different oscillator bath . the frequency of the central oscillator is @xmath24 . the baths are described by the spectral densities @xmath25although one could have started directly with eq . ( [ mod5 ] ) , we prefer to present the hamiltonian eq . ( [ mod1 ] ) as the starting point in order to provide a natural justification for the renormalization terms @xmath26 @xmath27 and @xmath28 @xmath29 in eq . ( [ mod5 ] ) which otherwise would have to be introduced _ ad hoc_. for low - lying excitations , and thus for low temperatures , the hamiltonian ( [ mod5 ] ) becomes equivalent to that of a large spin @xmath30 in a ferromagnetic environment . this can be seen as follows : the hamiltonian for a spin in a large magnetic field along the @xmath31 direction is @xmath32where the fluctuating terms model the low - lying bosonic ( magnon ) excitations of the ferromagnet at the site of the large spin . the fluctuations in the @xmath31 direction are neglected since they are quadratic in @xmath33 and @xmath34 with @xmath33 , @xmath35 . in other words @xmath36 is approximately a constant of motion @xcite . for @xmath37 the first term in ( [ magnet1 ] ) is a harmonic oscillator @xmath38 . the action of @xmath39 on the eigenstates of this harmonic oscillator is @xmath40and @xmath41=2\hbar ^{2}s$ ] . this amounts to keeping only the leading order term in the holstein primakoff transform of @xmath42 , @xmath43 kit89 . the rescaled hamiltonian @xmath44 becomes formally identical to eq . ( [ mod5 ] ) in the limit @xmath37 , with @xmath45 @xmath46 and the fluctuating pieces scaling as @xmath47 @xmath48 and @xmath49 @xmath50 . such large spins have been observed in magnetic particles @xcite . in the following we assume a power - law behavior at @xmath51 for both spectral densities @xmath52@xmath53@xmath54 and @xmath55@xmath56@xmath57 . we introduce the phononic " frequency @xmath58 and the dimensionless coupling constants @xmath59 , with @xmath60 @xmath61 . moreover we introduce a cutoff frequency @xmath62 for both baths . unless stated otherwise , we assume @xmath63 @xmath64 . elimination of the bath variables yields the heisenberg equations of motion for @xmath0 and @xmath1 @xmath65the response kernel is defined as@xmath66and the force operator @xmath67 . , with @xmath68 defined accordingly . in fourier space , eq . ( [ equations of motion ] ) reads@xmath69 q+i\omega p & = f_{q}(\omega ) \label{sist - fourier } \\ -i\omega q+\left [ \widetilde{j}_{p}(\omega ) -\omega _ { p}\right ] p & = f_{p}(\omega ) ~,\end{aligned}\]]where @xmath70 is the symmetrized riemann transform abr72 : @xmath71the oscillation modes are given by the zeros of the function@xmath72where @xmath73 is the generalized susceptibility . for the two baths at the same temperature the symmetrized correlation functions for the position @xmath74_{+}\rangle $ ] and the momentum @xmath75_{+}\rangle $ ] are obtained from @xmath73 as follows @xmath76the corresponding expression for @xmath77 is obtained by interchanging @xmath78 in eq . ( [ symmetrized autocorrelationfkt ] ) . the antisymmetrized correlation functions for position and momentum @xmath79_{-}\rangle $ ] ( with @xmath80 ) are obtained from eq . ( [ symmetrized autocorrelationfkt ] ) in the standard way @xcite , i.e. by substituting @xmath81 for @xmath82 . we wish to emphasize that , within the dissipation model studied here , the equations of motion ( [ equations of motion ] ) are symmetric in @xmath0 and @xmath1 for arbitrary coupling strength . thus the situation is different from that of a @xmath0oscillator when described within the rotating wave approximation , which is known to yield a model symmetric in @xmath0 and @xmath1 but which holds only in the weak coupling limit @xcite . first we focus on the important special case that both spectral densities are ohmic , @xmath83 @xmath17 @xmath84 @xmath17 @xmath85 . then the real parts of @xmath86 and @xmath87 vanish to lowest order in @xmath88 . the susceptibility has poles at the roots of the quadratic polynomial @xcite @xmath89which are @xmath90where @xmath91the solutions of eq . ( [ charpol ] ) are either purely imaginary or a pair of complex conjugates depending on whether @xmath92 is greater or smaller than 1 . thus , @xmath93is the commonly accepted criterion to distinguish between underdamped and overdamped oscillations . the underdamped region lies in a stripe of width @xmath94 , with @xmath95 , limited by the graphs of the functions @xmath96 in fig . [ fig1 ] the stripes of underdamped oscillations , marked in the @xmath97 plane are plotted for three different values of @xmath98 . for large @xmath99 , corresponding to a large mass of the central oscillator , the stripe of underdamped oscillations becomes smaller . in the limit @xmath100 , the introduction of an infinesimal coupling to the momentum can induce a transition from overdamped to underdamped oscillations . however the range of values of @xmath101 allowing for underdamped oscillations also becomes increasingly small as @xmath102 . the behavior of the system as a function of @xmath101 for fixed @xmath103 is also interesting . we set @xmath104 . the inverse damping time @xmath105 is a monotonously increasing function of @xmath101 for @xmath106 and a monotonously decreasing function for @xmath107 . therefore an additional bath coupling to @xmath1 has opposite effects on the damping time @xmath108 depending on whether the original @xmath0oscillator starts in the underdamped ( @xmath109 ) or in the overdamped ( @xmath110 ) regime . in the first case , the additional bath always reduces the damping time , leading to infinitely strong damping in the limit @xmath111 . however , in the regime @xmath107 we can drive the system from the overdamped into the underdamped regime by _ increasing _ the coupling strength @xmath101 . what is more , we can take @xmath112 @xmath113 , which corresponds to a completely symmetric hamiltonian in @xmath1 and @xmath0 . at this point the system is _ always _ in the underdamped regime and for large @xmath114 the oscillator frequency is close to its maximum . we find for the inverse damping time @xmath115therefore we are led to the paradoxical situation that the higher the friction coefficient @xmath114 the larger @xmath108 becomes . in particular , for @xmath116 one gets @xmath117 . this time dilatation is in itself a remarkable effect , since it contrasts with the behaviour of the @xmath0oscillator , for which @xmath118 as @xmath119 [ see eq . ( [ solu ] ) ] . in the general , non symmetric case , we note that for a given finite @xmath120 the damping time @xmath108 can not be made to acquire arbitrary values by tuning @xmath121 . rather , it is bounded between @xmath122 and @xmath123 , which is higher depending on @xmath124 . however , from these striking observations one should not conclude that in the limit of infinite @xmath114 the central oscillator recovers the dynamics of a free oscillator . it rather leads to a dilatation of all time scales . for instance , the renormalized oscillator frequency @xmath125 vanishes even faster than the inverse damping time , since @xmath126 . an analysis of the correlator in eq . ( [ symmetrized autocorrelationfkt ] ) shows that for @xmath127 the particle behaves as a free ballistic ( though very slow ) particle , @xmath128 . instead of using criterion a , we may focus on @xmath129 . as a spectral function , @xmath130 is independent of temperature . of special interest is the slope of @xmath131 near @xmath51 . since @xmath132 , the condition for the existence for @xmath131 displaying a maximum can be written as@xmath133which may be viewed as an indicator of underdamped oscillations or , equivalently , coherent transitions . it is often employed for the spin boson model @xcite . for ohmic damping , one finds @xmath134^{2}+(\gamma _ { q}\omega _ { p}+\gamma _ { p}\omega _ { q})^{2}\omega ^{2}}~ , \label{crit2}\]]and the critical curve for @xmath101 is thus given by the relation @xmath135 @xmath120 @xmath136 @xmath137 , so that criterion b is satisfied for @xmath138 . both being based on exact expressions , we observe that criterion b differs substantially from criterion a , see fig . [ fig1 ] . surprisingly , even for the @xmath0oscillator there is a region @xmath139 where criterion a and criterion b are different . in fig . [ fig2 ] @xmath140 is plotted for different coupling strengths @xmath120 , @xmath101 . for large couplings @xmath101 , we observe that , although we are ( according to criterion b ) in the underdamped region , the maxima of the curves are not very pronounced . this is related to the fact that the renormalized oscillator frequency @xmath125 vanishes for large @xmath101 faster than the inverse damping time @xmath141 . the foregoing discussion suggests the introduction of a third , more restrictive criterion for underdamped oscillations , namely @xcite , @xmath142according to this condition , the region of underdamped oscillations is convex , i.e. by increasing the coupling strength no transition from overdamped to underdamped oscillations can be realized . .condition for coherent dynamics according to three different criteria ( left column , labeled a , b , c ) , for three particular cases ( upper row ) . the symmetric limit includes the assumption @xmath143 . general case is given in main text . [ cols="^,^,^,^,^",options="header " , ] table i summarizes the different criteria for coherent , underdamped dynamics as they are applied to three prototypical cases . for high temperatures ( @xmath144 ) the integral for the coordinate autocorrelation function eq . ( [ symmetrized autocorrelationfkt ] ) can easily be solved yielding the classical solutions of the heisenberg equations of motion . for @xmath145 they read @xmath146for @xmath147 , eq . ( [ classical ] ) describes the corresponding non - oscillating solution . analytical expressions for @xmath148 are also available for @xmath149 . in this case the integral in eq . ( [ symmetrized autocorrelationfkt ] ) can be expressed in terms of exponential integrals weis99 . we do not show the result here and limit ourselves to note that for long times only the lowest order term , i. e. the term linear in @xmath150 , contributes in the numerator of the r.h.s . ( [ symmetrized autocorrelationfkt ] ) . thus the long time behaviour is the same as for the @xmath0oscillator , namely , we have @xmath151 for the classical solution eq . ( [ hightemp ] ) criterion a [ eq . ( [ crit1 ] ) ] is the only natural criterion to differentiate between underdamped and overdamped oscillations , since it distinguishes solutions with infinitely many zeros from solutions with no zeros at all . for zero temperature the number of zeros of @xmath152 is always finite and criterion a is less informative . the zero temperature mean squares @xmath153 , @xmath154 can be calculated exactly . they are given by @xmath155 \notag \label{meanq } \\ & & + \frac{\gamma _ { p}}{\pi } \left ( 1+\gamma _ { p}\gamma _ { q}\right ) \ln \frac{\omega _ { p}\sqrt{1+\gamma _ { q}\gamma _ { p}}}{\omega _ { 0}}\ + \mathcal{o}(\omega _ { p}^{-1},\omega _ { q}^{-1 } ) , \notag \\ f(\kappa ) & = & \frac{1}{\pi \sqrt{\kappa ^{2}-1}}\ln \frac{\kappa + \sqrt{\kappa ^{2}-1}}{\kappa -\sqrt{\kappa ^{2}-1}}\ .\end{aligned}\]]the corresponding expressions for @xmath156 are obtained by interchanging @xmath78 in eqs . ( [ meanq ] ) and ( [ kappa ] ) . for small @xmath157 we have @xmath158 and the position mean square becomes @xmath159and , correspondingly,@xmath160for @xmath161 , @xmath162 we recover the results for the @xmath0oscillator , with the characteristic logarithmic dependency of @xmath154 on the cutoff @xcite . in the general case , the heisenberg uncertainty product diverges as @xmath163 . thus the reduced density matrix at equilibrium becomes approximately the identity , its off diagonal elements being essentially zero in both the position and the momentum representation . for general spectral densities @xmath164 and @xmath165 are arbitrary positive real numbers . while we always have @xmath166we find @xmath167 first , we briefly recall the behavior of @xmath153 and @xmath156 as functions of @xmath164 and @xmath120 in the case of the @xmath0oscillator . for @xmath153 it may be summarized by the formula @xcite @xmath168a measurement of the position , i.e. a diagonalization of the density matrix in the position basis takes place only for @xmath169 . the behaviour of @xmath156 is opposed to that of @xmath153 in such a way that the product @xmath170 is always @xmath171 , as required by the heisenberg relation . for @xmath154 we may write @xmath172 in the forthcoming discussion we focus on @xmath153 . we have a contribution from @xmath52 , which reduces @xmath153 , and another contribution from @xmath55 enhancing @xmath173 [ see eq . ( [ meanqper ] ) ] . from the results for the @xmath0oscillator we expect that the @xmath55 contribution should always dominate for @xmath174 and the @xmath52 contribution should be negligible for @xmath175 . the opposite regime @xmath169 and @xmath176 is more interesting ( the case @xmath177 and @xmath178 has been studied in ref . @xcite ) . there we have a sum of two terms of order @xmath85 in @xmath88 with opposite sign . in the following we focus on that case . we assume the susceptibility @xmath73 to be an algebraic function . this entails rational exponents @xmath179 with integer @xmath180 and @xmath16 . the integrand in eq . ( [ symmetrized autocorrelationfkt ] ) has @xmath181 poles on the @xmath16 riemann sheets which split into two classes . for @xmath182 there are @xmath183 poles close to the oscillator frequency @xmath184 on each sheet . in addition there are @xmath185 poles of the order of @xmath186 . if either @xmath187 or @xmath165 becomes equal or smaller one , the @xmath183 poles detach from @xmath188 moving into the complex plane . the equilibrium mean square can be written as a sum of the contributions from the two types of poles . @xmath189for the second term we find ( @xmath120 , @xmath101 small ) @xmath190for a sketch of the derivation of eq . ( [ highcont ] ) see [ appa ] . therefore , the high energy poles yield , to first order in @xmath101 , a finite contribution to the position mean square . this contribution is positive for superohmic coupling to the momentum but vanishes for ohmic and subohmic coupling . it has a singularity for @xmath191 which marks the transition to the logarithmic dependence on the cutoff frequency , see eq . ( [ meanq ] ) . we note that the right hand side of eq . ( [ highcont ] ) does not depend on the oscillator frequency @xmath192 , i. e. it does not depend on the properties of the system itself . [ fig4 ] the contribution from the poles close to @xmath192 , @xmath193 , can in principle be calculated in a similar way as @xmath194 for arbitrary exponents @xmath164 , @xmath165 ; however , the calculations become increasingly messy . a general treatment is also hampered by the wide variety of casuistic behavior , with different regimes defined by the conditions @xmath195 , @xmath196 and @xmath197 , @xmath198 . we focus on the specific case @xmath199 and @xmath176 , which is the most relevant , since most baths occuring in nature are either ohmic ( markovian approximation , electron gas @xcite ) or superohmic ( photon or acoustic phonon baths ) . in that case , the low energy poles are essentially determined by the lower spectral exponent ( here , @xmath164 ) . this reflects the general property of dissipative quantum systems that an environment becomes increasingly efficient with decreasing spectral exponent @xmath200 . to leading order in @xmath120 , we find @xmath201 + \mathcal{o}(\gamma _ { q}^{2})\ . \label{lowcont}\]]if we compare @xmath202 with eq . ( [ highcont ] ) , we observe that the two contributions have opposite sign . in particular , the contributions of @xmath203 and @xmath204 _ cancel each other _ to first order in @xmath120 and @xmath101 provided that @xmath205 \gamma _ { q}\ . \label{equalgam}\]]in fig . [ fig4 ] @xmath153 is plotted against the coupling strengths @xmath101 and @xmath120 for @xmath177 and @xmath206 . @xmath153 is monotonously increasing as a function of @xmath101 . for @xmath162 it is a monotonously decreasing function of @xmath120 . for the parameters chosen in , holds on the diagonal @xmath120 @xmath207 . there @xmath173 remains close to its unperturbed value @xmath208 if @xmath209 . finally , on the right hand side of the position mean square @xmath153 is plotted as a function of @xmath120 for different spectral exponents @xmath165 . for small @xmath120 the enhancement of @xmath153 due to the coupling to the momentum is larger for smaller values of the spectral exponent @xmath165 . this is the expected behaviour . this behavior , however , is inverted when @xmath210 increases . then the relative enhancement of @xmath153 as compared with the @xmath0oscillator is bigger for higher spectral exponent @xmath165 . a crossing occurs always for some value of the coupling constant @xmath120 , although some times this can be very large . to study the nonequilibrium properties and in particular the loss of coherence for an initially pure state , the calculation of @xmath148 is not sufficient . instead , one should perform a full nonequilibrium calculation with inclusion of specific initial conditions by means of a laplace transformation . the position - momentum symmetry of the oscillator hamiltonian suggests the use of the reduced wigner function @xmath211 instead of the reduced density matrix . it was shown in ref . @xcite for the @xmath0 oscillator that the time evolution of @xmath211 can only for certain initial conditions be described by an exact master equation . here we consider decoupled initial conditions @xmath212 , which fall into this class . in this case we are able to derive not only an exact master equation for the reduced wigner function but a solution in terms of a two fold convolution integral . generally @xmath211 can be expressed as @xmath213where the bracket denotes the average over initial conditions @xcite @xmath214and @xmath215 are the classical solutions of the equations of motion eq . ( [ equations of motion ] ) . the wigner function of the thermalized baths is given by @xmath216the system itself being in an arbitrary pure state characterized by @xmath217 . in this case one can express eq . ( [ reduced wgnerfunktion ] ) as a twofold convolution integral @xmath218where @xmath219 is the fourier transform of @xmath217 in both arguments . we have introduced the auxiliary functions @xmath220and the temperature - dependent quantities @xmath221 \ . \label{tempdep}\end{aligned}\]]the functions @xmath222 , @xmath223 and @xmath224 are defined accordingly by interchanging @xmath1 and @xmath0 . for a better understanding of the forthcoming discussion in sec . [ timedep ] we notice that the functions defined in eq . ( [ definition of auxiliary functions phi_0,phi_p ] ) are the time dependent coefficients of the solutions @xmath225 , @xmath226 of the initial value problem eq . ( [ equations of motion ] ) . for instance @xmath227this necessarily requires @xmath228 and @xmath229 . finally , we note that @xmath230 fulfills a fokker planck type of equation @xcite , @xmath231 w(q , p , t)\ , \label{fokker -- planck equation}\]]where @xmath232 , with @xmath233 the phase space drift term , @xmath234and with ( the @xmath235 matrix ) @xmath236 the state independent phase space diffusion term , @xmath237@xmath238 is obtained from @xmath239 by exchanging @xmath0 and @xmath1 . to derive the coefficients ( [ driftterm ] ) and ( [ diffusionterm ] ) we have employed the ansatz eq . ( [ fokker planck equation ] ) for the fokker planck operator . acting with yet undefined functions @xmath240@xmath241,@xmath242@xmath243 ( @xmath244 ) on the r.h.s . of eq . ( [ fokker planck equation ] ) , and comparing with its time derivative , yields conditions for the functions @xmath240@xmath245 . an exact fokker planck equation in the form of eq . ( fokker planck equation ) for the @xmath246oscillator has been first derived in ref . @xcite and later by a different method in refs . unr89,hu92 . a convenient quantity to measure the degree of global decoherence is the purity @xmath247 , defined as the average of the density matrix itself @xcite,@xmath248which is basis independent . of special interest is the equilibrium value @xmath249 which measures the efficiency of the environment in destroying quantum coherence . here we implicitly assume ergodic behavior which , as we shall see , applies in the presence of ohmic baths . for a harmonic oscillator in thermal equilibrium , the reduced wigner function is @xcite @xmath250 ^{1/2}}\exp \left ( -\frac{q^{2}}{2\langle q^{2}\rangle _ { \beta } } -\frac{p^{2}}{2\langle p^{2}\rangle _ { \beta } } \right ) , \]]which leads to @xmath251 ^{1/2}}\ . \label{purityversuncertainty}\ ] ] although the equilibrium decoherence ( as measured by the product @xmath252 ) is enhanced by the additional noise term , one may wonder whether for low temperatures the decoherence time becomes larger than for the damped @xmath0oscillator . to answer this question exhaustively one would have to calculate the time evolution of the purity for an arbitrarily pure initial condition . we consider two cases . first , we choose a coherent ( gaussian ) state as initial state . this case should present the greatest robustness against decoherence @xcite . second , we choose a superposition of two gaussian wave packets . then a new aspect of decoherence comes into play , namely , the fast vanishing of the relative coherence between the two gaussian wave packets . we distinguish the two manifestations of decoherence by introducing a new quantity , which we call _ relative purity_. we consider here the case where the system starts in a coherent state at @xmath253 . the wigner function for a gaussian wave packet is @xmath254 \ , \label{wignerfuction of coherent state}\]]where @xmath255 and @xmath256 are defined in terms of the complex eigenvalue @xmath257 ( amplitude ) of the coherent state ( recall @xmath258 ) . since @xmath259 is gaussian , the convolution integral of eq . ( [ wigsolution ] ) can be performed . the integrals become particularly simple when both baths are equal ( @xmath112 @xmath113 and @xmath104 ) , corresponding to the completely _ symmetric case_. then we have @xmath260 @xmath261 and , by the same token , @xmath262 in eq . ( [ definition of auxiliary functions phi_0,phi_p ] ) and also @xmath263 @xmath264 in eq . ( tempdep ) . the crossed terms in eq . ( [ wigsolution ] ) vanish and the wigner function becomes a product of two gaussians for all times @xmath265 . \label{wigner2}\end{aligned}\]]for the purity one obtains the simple expression @xmath266the remaining task is to calculate the quantities @xmath267 , @xmath268 , @xmath269 , @xmath270 from their definition in eqs . ( [ definition of auxiliary functions phi_0,phi_p ] ) and ( [ tempdep ] ) . using the nave form ( @xmath271 ) of the spectral function of an ohmic bath as given in sec . [ two independent baths ] , we would find @xmath272 @xmath273 @xmath274 leading to inconsistencies [ see the discussion after eq . ] . the reason for this lies in an initial slippage caused by the somewhat unphysical character of the decoupled initial condition @xcite . technically it stems from the fact that for @xmath275 @xmath276 the integrals in eqs . ( [ definition of auxiliary functions phi_0,phi_p ] ) do not converge at @xmath253 @xcite . this problem is overcome by regularizing the ohmic spectral functions , i.e. by reintroducing a finite cutoff @xmath277 . the explicit calculation with a drude regularized spectral function is sketched in appb . * * * * here we state only the main result : at @xmath149 the purity decays on two time scales , given by @xmath88 and @xmath278 in ( [ tauspez ] ) , @xmath279 , & \quad \ \text{for ~ } \omega ^{-1}\ll t\rightarrow \infty ~,\end{array}\right . \label{purityresult}\]]with @xmath280 and the oscillator frequency @xmath281 . coherence is reduced immediately after the start of the coupling . although afterwards it decreases more slowly , on a time scale @xmath108 , for larger couplings the curves of @xmath247 for different values of @xmath114 never cross , i.e. , @xmath247 is a monotonously decreasing function of @xmath114 for all @xmath282 . an initial slip similar to that discussed in this section also occurs for the @xmath0oscillator @xcite . however in that case its effect on purity is much less severe . specifically , out of the set of functions @xmath283 , @xmath284 and @xmath285 defined in eq . ( [ definition of auxiliary functions phi_0,phi_p ] ) , only @xmath283 is affected by the initial slip . after a time @xmath88 it becomes @xmath286 @xmath287 @xcite . however , we note that @xmath288 appears in the temperature - dependent terms @xmath223 , @xmath289 , @xmath224 , @xmath290 of eq . ( [ tempdep ] ) only in combination with the spectral density of the momentum coupling @xmath55 , which vanishes for the @xmath0oscillator by definition . we thus reach the important conclusion that the purity evolution of the @xmath0oscillator is insensitive to the initial slip stemming from the use of decoupled initial conditions : at @xmath291 the purity is still approximately unity , decreasing afterwards at a rate @xmath292 . isar and coworkers @xcite studied in detail the decay of purity for the @xmath0oscillator . in particular they found constraints that must be satisfied by the bath if the purity is to remain constant and close to unity during the whole time evolution of the oscillator . such a high purity is not possible in the present case due to the initial slip [ see eq . ( [ purityresult ] ] . another possible preparation of the initial state is in a constrained equilibrium . the expectation values @xmath293 @xmath17 @xmath294 and @xmath295 @xmath17 @xmath296 are held fixed with the bath equilibrated around those values . then the wigner function is @xcite @xmath297 ^{1/2}}\exp \left ( -\frac{(q-\langle q(t)\rangle)^{2}}{2\langle q^{2}\rangle _ { \beta } } -\frac{(p-\langle p(t)\rangle)^{2}}{2\langle p^{2}\rangle _ { \beta } } \right)\ ] ] while the purity is given by its equilibrium value eq . ( [ purityversuncertainty ] ) . a more detailed discussion of the purity decay of the @xmath0oscillator , together with the general formulae , is given in [ appb ] . formula ( [ wigsolution ] ) also allows us to investigate more complicated initial conditions such as as , for example , the superposition of two gaussian wave packets . this case has been studied for a single bath by caldeira and leggett @xcite . it is an interesting case study because it displays two different aspects of decoherence . on the one hand , there is the decoherence which either wave packet would experience alone . this part is essentially described by @xmath247 , which we will call _ gaussian purity _ [ see eqs . ( [ purity2 ] ) and ( [ purityresult ] ) ] . on the other hand there is the decoherence due to the spatial separation of the two packets . this second contribution is expected to become increasingly important when the distance @xmath298 between the two packets becomes large . as initial wave function we choose @xmath299{2\pi \sigma ^{2}}}\left ( e^{-\frac{(x+a/2)^{2}}{4\sigma ^{2}}}+e^{-\frac{(x - a/2)^{2}}{4\sigma ^{2}}}\right ) \ , \]]which translates into an initial wigner function @xmath300here @xmath301 ^{1/2}$ ] is a normalization constant . we assume the most symmetric case and set @xmath302 . plugging ( [ wig2 ] ) into ( [ wigsolution ] ) we find that the purity can be expressed in terms of the gaussian purity as @xmath303 } { \cosh ^{2}\left ( a^{2}/8\right ) } \right\ } . \label{two - purity}\]]the function @xmath304 is given by @xmath305where @xmath267 and @xmath268 are defined in ( [ definition of auxiliary functions phi_0,phi_p ] ) and computed in ( [ phifunc ] ) for the symmetric case . @xmath304 evolves from @xmath306 to @xmath307 . we define the _ relative purity _ as the ratio @xmath308as expected , @xmath309 as @xmath310 , and @xmath311 as @xmath312 . interestingly , the structure of ( [ two - purity ] ) is such that , as time passes and @xmath313 evolves from @xmath85 to @xmath314 , the ratio @xmath315 starts at unity , as corresponds to a pure state , then decreases and finally , at long times , goes back to unity . when @xmath298 is large @xmath315 decays rapidly to @xmath208 , on a timescale @xmath316 . there it stays for a time which increases with distance as @xmath317 . afterwards it returns to one . the ratio 1/2 can be rightly interpreted as resulting from the incoherent mixture of the two wave packets . thus it comes as a relative surprise that @xmath315 becomes unity again at long times , as if coherence among the two wave packets were eventually recovered . the physical explanation lies in the ergodic character of the long time evolution , with both wave packets evolving towards the equilibrium configuration [ see eq . ( [ meanq ] ) and the subsequent discussion ] . it is instructive to compare eq . ( [ mod5 ] ) with the hamiltonian of a harmonic oscillator interacting with a single heat bath in the most general form of linear coupling , @xmath318with complex parameters @xmath319 . this model is different from that discussed in the previous sections in that here time reversal invariance is broken . by this we mean the following . in the models described by eq . ( [ mod5 ] ) , the bath modes might describe for example a magnetic field coupled to the momentum of a charged particle , which clearly would break time reversal invariance . however in that case , such a symmetry breaking is somewhat fictitious since , due to the linear nature of the coupling and to the modelling of the bath as a set of harmonic oscillators , one can always find a unitary transformation which restores time reversal invariance , i.e. which renders all parameters in the hamiltonian real quantities . it is easy to see that , for general ( complex ) @xmath320 and @xmath321 , such a unitary transformation can not be found for eq . ( [ onebathonlyhamiltonian ] ) . by an analysis similar to that of one finds general expressions for the symmetrized equilibrium correlation functions : @xmath322 ^{2}\right . \notag \label{symmetrized correlation function for one bath } \\ & & \lo\quad + \mathrm{im}\widetilde{j}_{p}(\omega ) \left [ \left ( \omega + \mathrm{re\,}\widehat{j}_{-}(\omega ) \right ) ^{2}+\left ( \mathrm{re\,}\widetilde{j}_{+}(\omega ) \right ) ^{2}\right ] \notag \\ & & \lo\quad -\left . \left [ \omega _ { p}+\mathrm{re\,}\widetilde{j}_{p}(\omega ) \right ] \mathrm{im}\left [ 2\omega \widehat{j}_{-}(\omega ) -\widehat{j}_{-}^{2}(\omega ) -\widetilde{j}_{+}^{2}(\omega ) \right ] \right\ } \ .\end{aligned}\]]the same expression applies for @xmath77 with the indexes @xmath323 interchanged everywhere . as before , we define a generalized susceptibility `` for the system which now reads @xmath324 \left [ \omega _ { p}-\widetilde{j}_{p}(\omega ) \right ] -\omega ^{2}-\widetilde{j}_{+}^{2}(\omega ) -\widehat{j}_{-}^{2}(\omega ) + 2\omega \widehat{j}_{-}(\omega ) ~. \label{generalized susceptibility for one bath}\]]eqs . ( [ symmetrized correlation function for one bath ] ) and ( generalized susceptibility for one bath ) involves four different spectral functions . @xmath86 and @xmath87 are defined as in eqs . ( [ mod6 ] ) and ( [ tildetransformed ] ) . now we introduce the spectral functions @xmath325which reflect the mixing of the bath modes . the hat symbol denotes the transformation @xmath326 in eq . ( [ hattransformation ] ) the real part is an antisymmetric function and the imaginary part is symmetric . this is exactly reverse to the tilde '' ( riemann ) transform , defined in eq . ( tildetransformed ) . before we calculate eq . ( [ symmetrized correlation function for one bath ] ) for a specific case , we analyze some generic features . first we notice that , by setting @xmath327 and @xmath328 to zero , we recover the autocorrelation function for two independent baths [ see eq . ( [ symmetrized autocorrelationfkt ] ) ] . on the other hand , the two spectral densities @xmath329 and @xmath330 are not independent from @xmath55 and @xmath52 ; rather , they satisfy@xmath331actually this relation has already been used to simplify the integrand of eq . ( [ symmetrized correlation function for one bath ] ) . we note that it holds for the spectral densities themselves but in general _ not _ for their riemann transforms @xmath328 and @xmath332 . apart from the above constraint one can , at least in principle , freely choose three of the four spectral functions . if , in addition , we make the physically reasonable assumption that the mixing angle `` in eq . ( [ definitions of j_sym and j_asy ] ) is frequency independent ( @xmath333 ) , condition ( [ relation between j_q , j_p and j_sym , j_asy ] ) fixes @xmath329 and @xmath330 completely . @xmath334 ^{1/2}\cos \theta \notag \label{j_{sym } and j_{asy } for fixed eta } \\ j_{-}(\omega ) & = & \left [ j_{q}(\omega ) j_{p}(\omega ) \right ] ^{1/2}\sin \theta \ .\end{aligned}\]]therefore , in the important case where both @xmath52 and @xmath55 have the same spectral exponent ( @xmath335 ) , @xmath336 and @xmath337 also obey the same power law . using eq . ( [ relation between j_q , j_p and j_sym , j_asy ] ) we observe that the term @xmath338 implicit in eq . ( [ generalized susceptibility for one bath ] ) drops out . we recall that this interference term has been responsible for the non trivial phase space '' diagram fig . . what is more , only @xmath339 appears in eqs . ( [ symmetrized correlation function for one bath ] ) and ( [ generalized susceptibility for one bath ] ) . that means , according to eq . ( [ rejn ] ) , that for @xmath340 the symmetric spectral function @xmath336 does not contribute at all . the relation between the hat `` and the tilde '' transformation is @xmath341from which all properties of @xmath342 can easily be deduced . for example from eq . ( [ hattilde ] ) and eqs . ( [ rejn ] ) and ( [ gen1 ] ) it follows that @xmath343 only for @xmath344 . it is illustrative to look at the generalized susceptibility eq . ( [ generalized susceptibility for one bath ] ) for the case @xmath344 , where all real parts of the spectral functions vanish . we obtain @xmath345 \ . \label{generalized susceptibility for one bath1}\end{aligned}\]]this is exactly the suceptibility of a @xmath0oscillator coupled to a single bath with an additive spectral function @xmath346we conclude that , when only one bath is involved , the particular structure of the coupling of the bath to the position or momentum variable can always be modelled by a @xmath0oscillator with an appropriately chosen spectral function . in this context the combination @xmath347 can be considered as stemming from an effective additional noise source that couples to position . note however , that this is not the same as saying that a single bath coupling to @xmath0 and @xmath1 can always be unitarily transformed to a coupling only to @xmath0 ( @xmath0oscillator ) , since this is not true in general . for one thing , the noise properties , which are not uniquely given by @xmath348 , would be different . we also wish to emphasize that _ a double - bath dissipative oscillator can not be described in terms of an oscillator coupled to an effective single bath_. in particular , it can never be modelled by a susceptibility like ( [ effsingle ] ) . this is the reason why the physics explored in is so different from that which could be found in any possible single - bath scenario . we do not repeat here the analysis of for arbitrary spectral functions but focus instead on the case of ohmic coupling ( @xmath349 ) in order to compare with the results obtained for two independent baths . as @xmath350 vanish [ see ] , we obtain @xmath351 ^{2}-2\omega _ { p}\widehat{j}_{-}(\omega ) \left [ \omega -\mathrm{re\,}\widehat{j}_{-}(\omega ) \right ] \right\ } .\end{aligned}\]]the form of the susceptibility varies depending on whether the mixing angle is a multiple of @xmath352 or not . we discuss the two cases separately . for @xmath354 or @xmath355 one can bring , by an appropriate redefinition of @xmath356 , @xmath357 , the interaction part of the hamiltonian eq . ( [ onebathonlyhamiltonian ] ) into the form @xmath358so that the bath couples to the main oscillator either through the position of their oscillators or through their momentum , but not through both . from eq . ( [ symandasym ] ) it can be seen that the antisymmetric spectral function vanishes identically . apart from the susceptibility , the integrand of eq . ( [ ohmic symmetrized correlation ] ) becomes equal to that of eq . ( [ symmetrized autocorrelationfkt ] ) . thus all results of sec . [ ohm ] , in particular eqs . ( [ classical ] ) and ( [ meanq ] ) , carry formally over . the only but crucial difference lies in the poles of the generalized susceptibility @xmath73 . the set of zeros can be written as @xmath359 with @xmath92 defined as @xmath360according to criterion a , eq . ( [ crit1 ] ) , the regions of overdamped and underdamped oscillations in the @xmath97 plane are separated by a straight line with negative slope given as before by the condition @xmath361 [ see fig . [ fig3 ] ] . therefore , a crossover from overdamped to underdamped oscillations by increasing one of the coupling constants is now impossible . we also note that , in the symmetric case ( @xmath362 @xmath363 @xmath364 ) , the relaxation time is the same as for the @xmath0oscillator [ @xmath365 @xmath366 . increasing @xmath120 or @xmath101 leads inevitably to an enhancement of dissipation . this is in blatant _ contrast _ to the results obtained previously for the double - bath model and one of the most remarkable results of this work . it best illustrates the importance of the specific structure of the bath and justifies a posteriori the detailed study of a model with two independent baths . we briefly consider the case of non vanishing mixing angle . the effects should be largest for @xmath368 , corresponding to an interaction hamiltonian @xmath369this form of coupling is called amplitude coupling in @xcite . now @xmath330 is not longer zero but we have @xmath370 \label{ohmicjtilde}\]]which is exactly the spectral function of a superohmic bath with exponent @xmath371 . the logarithm in @xmath372 inhibits further analytic treatment of eq . ( [ ohmic symmetrized correlation ] ) even in the high temperature limit ; thus we limit ourselves to a qualitative analysis . for simplicity , we omit the terms containing logarithms in the generalized suceptibility eq . ( [ generalized susceptibility for one bath ] ) . the contributions to @xmath152 stemming from those terms should decay on a time scale @xmath373 . the remaining terms define a suceptibility which is identical to the suceptibility of a @xmath0oscillator coupled linearly to one bath with spectral density@xmath374it is well known @xcite that , in the case of a polynomial spectral function , the term with the lowest exponent dominates [ see also the discussion in sec . [ gencase ] ] . therefore we may conclude that the crossover diagram remains unchanged by a mixing angle @xmath375 . we have discussed the behavior of a quantum brownian particle in a harmonic potential subject to two independent noise sources , one of which couples to its position and the other one to its momentum . in the symmetric case where both baths are ohmic and their coupling strength is the same , we find underdamped oscillations of the central oscillator for all coupling strengths . this indicates that the two baths partially cancel each other . the effect is due to the mutually conjugate character of position and momentum . it was first noted in ref . @xcite for the ( cylindrically ) symmetric spin boson model with spin @xmath376 , i.e. in the deep quantum regime . quantum frustration `` of the spin can pictorially be described as the result of having two observers attempting to measure simultaneously , with equal efficiency , two non - commuting components of the spin . because of the uncertainty principle , both of them fail to measure anything . here we have investigated the analogous effect for a quantum oscillator coupled to two independent baths . we have found a moderate form of cancellation which we have labelled quasiclassical frustration '' because our dissipative oscillator may describe a large spin impurity coupled to the magnon bath in a ferromagnetic medium . the most notable features of quasiclassical frustration become manifest in the strong coupling limit , where underdamped dynamics survives and all time scales diverge . it remains to be investigated whether the occurrence of frustration in the classical regime is a general property or an artifact of the harmonic oscillator . we have compared the double - bath model with the case where a single bath couples linearly to the position and momentum of the oscillator . in the latter case the phase space diagram is simpler in that transitions from overdamped to underdamped oscillations can never occur . comparison of the two models indicates that bath correlations can qualitatively change the behaviour of a dissipative system . a point of caution is needed in the interpretation of our results . we have definitely ruled out the at first sight enticing but at closer inspection unphysical conjecture that the effects of the two observers cancel each other completely and the particle is not affected by the environment . indeed we have seen in the specific example of a decoupled initial state that , in destroying quantum coherence , two baths are always more efficient than one . this is true for both the overdamped and the underdamped regime . this demonstrates that , at least for the harmonic oscillator , underdamped dynamics is by no means a reliable signature of high global coherence , which here we identify with purity [ see sec . [ timedep ] ] . our work provides further evidence that decoherence and dissipation are not necessarily correlated . by dissipation we mean here the net transfer of energy from the central oscillator to the thermal baths . it is characterized by the classical equation of motion or , equivalently , by the properties of the spectral function . we have seen that , depending on the situation , an increase of dissipation can be accompanied by either a reduction or an increase of decoherence . vice versa , a source of decoherence may or may not lead to dissipation . this is the case in the so called pure dephasing ( not considered here ) , where the interaction part of the hamiltonian commutes with the system hamiltonian and no dissipation occurs at all , see i. e. gor04 . the fact that we have focused on a quantum oscillator coupled linearly to oscillator baths has allowed us to investigate analytically the equilibrium and dynamical properties . the question of frustration is also raised in other dissipative quantum systems , such as a small spin coupled to a boson @xcite or a fermion @xcite bath , which are not amenable to an exact analytical treatment . the extension of the present study to less tractable physical scenarios provides a theoretical challenge . we are indebted to f. guinea for useful discussions . this work has been supported by ministerio de ciencia y tecnologa ( spain ) under grants no . bfm2001 - 0172 and fis2004 - 05120 , and by the ramn areces foundation . one of us ( h.k . ) acknowledges financial support from the rtn network of the european union under grant no . hprn ct2000 - 00144 . we calculate the high frequency contribution @xmath194 to the position mean square in eq . ( [ expression for general spectralfunctions ] ) . this contribution vanishes for @xmath377 but yields a finite contribution otherwise . we assume again @xmath378 and @xmath379 , ( @xmath380,@xmath381,@xmath16 @xmath382 ) to be rational numbers . then this contribution can be written in a determinant form @xmath383here we use the vandermonde determinant @xmath384 . we recall that @xmath385 is the total number of poles , while @xmath386 is the number of high energy poles . the function @xmath387 is given by @xmath388where the function @xmath389 is a rather complicated function involving a special function called lerch transcendent @xcite . it can , for @xmath30 odd , be expressed in terms of a logarithm plus a polynomial @xmath390 \ . \label{phifunction}\]]in order to use this formula we assume @xmath16 and @xmath180 odd . consequently we formally exclude @xmath200 even . however , that case can be included by considering the limit @xmath391 with @xmath180 @xmath392 . the argument of @xmath387 is the square of one of the @xmath385 roots of the characteristic polynomial @xmath393 \left [ \frac{2\theta ( \alpha _ { p}-2)\lambda ^{m}}{\pi ( \alpha _ { p}-2)}-i\lambda ^{\beta _ { p}-m}\right ] -\frac{1}{\gamma _ { q}\gamma _ { p}}=0\ . \label{highfrequcharacteristicpoly}\]]the asymptotic behavior of @xmath194 for @xmath394 can be derived by expanding @xmath395 in eq . ( [ phifunction ] ) . for @xmath396 we find to lowest order @xmath397keeping only the first term for small @xmath120 and small @xmath101 yields eq . ( [ highcont ] ) . we regularize the ohmic spectral functions by a drude cutoff @xmath398assume @xmath399 @xmath400 @xmath401 , @xmath402 @xmath403 . for @xmath267 and @xmath268 we obtain @xmath404 e^{-t/\tau } \notag \\ & & \qquad \quad + \frac{\gamma ^{2}}{1+\gamma ^{2}}\left [ \cos ( \gamma ^{2}\omega t)+\frac{1}{\gamma } \sin ( \gamma ^{2}\omega t)\right ] e^{-\omega t } \notag \\ \phi _ { 1}(t ) & = & \frac{1}{1+\gamma ^{2}}\left [ \gamma \cos \left ( \lambda t\right ) + \sin \left ( \lambda t\right ) \right ] e^{-t/\tau } \notag \\ & & \qquad \quad + \frac{\gamma ^{2}}{1+\gamma ^{2}}\left [ -\cos ( \gamma ^{2}\omega t)+\gamma \sin ( \gamma ^{2}\omega t)\right ] e^{-\omega t}\ , \label{phifunc}\end{aligned}\]]where no ambiguity is left [ @xmath405 and , we recall , @xmath141 @xmath406 . in a time of order @xmath407 after the connection ( with @xmath271 ) , both functions have dropped on average to a value @xmath408 . this is the value one would have obtained by using directly @xmath275 @xmath409 . after this short initial period , @xmath267 and @xmath268 decay much more slowly , on a time scale @xmath108 . in the expression ( [ purity2 ] ) for the purity decay after the initial slip , we may safely neglect the first two terms in the denominator . the functions @xmath267 , @xmath268 however still govern through @xmath269 , @xmath270 the time evolution of the purity @xmath247 [ see eq . ( [ tempdep ] ) ] . we notice that both @xmath269 and @xmath270 are zero at @xmath253 . however in the initial time interval of order @xmath410 they both increase rapidly . after the initial slip , they settle to a value @xmath411 @xmath412 and increase much more slowly afterwards . this can be seen if we write for instance @xmath413 as follows @xmath414where @xmath415 is the principal value integral @xmath416eq . ( [ g0aux ] ) can be evaluated straightforwardly using the residue theorem . however for our purposes the most important point is that @xmath417 @xmath418 and that @xmath415 behaves in time essentially in the same way as @xmath285 in eq . ( [ phifunc ] ) . thus we may write @xmath419 e^{-t/\tau } \notag \\ \fl & & \qquad \quad + \frac{\gamma ^{2}}{1+\gamma ^{2}}\left [ a_{\omega } ( \omega ) \cos ( \gamma ^{2}\omega t)+b_{\omega } ( \omega ) \sin ( \gamma ^{2}\omega t)\right ] e^{-\omega t}\ . \label{aux1}\end{aligned}\]]here @xmath420 , @xmath421 are some algebraic functions of @xmath150 whose exact form is not important for the present discussion , since after the initial slip of time @xmath88 the second line of eq . ( [ aux1 ] ) can be neglected . proceeding in the same way with @xmath422 we arrive at @xmath423where @xmath148 is given in eq . ( [ symmetrized autocorrelationfkt ] ) . the temperature dependence is encoded in @xmath424 and @xmath148 . of course the r.h.s . of eq . ( [ endresult ] ) is not zero any more at @xmath253 due to the neglected initial transient . at zero temperature the autocorrelation function decays algebraically @xmath425 . then we reproduce eq . ( [ purityresult ] ) . here we state the general formulae for the purity decay of a coherent ( gaussian ) initial state . to calculate the purity decay for two ohmic baths , both with decoupled initial conditions but with different coupling strengths , we proceed as in sec . [ coherent state ] . performing a twofold fourier transform of eq . ( [ wignerfuction of coherent state ] ) and plugging the result into eq . ( [ wigsolution ] ) yields for the reduced wigner function @xmath426here , @xmath427for the purity we have with the above definitions @xmath428as expected , this expression reduces to eq . ( [ purity2 ] ) for @xmath429 and @xmath104 . for @xmath430 we have the case of the @xmath0oscillator . from the definitions ( [ tempdep ] ) we get @xmath431 @xmath432 and obtain a closed expression for the purity decay of the @xmath0oscillator , @xmath433 ^{-1/2}\end{aligned}\]]the explicit form of the functions @xmath285 , @xmath434 with drude regularized ohmic spectral function can be found in ref . @xcite . here we give only their form after taking the limit @xmath271 , i. e. for @xmath435 . @xmath436 e^{-t/\tau } \notag \label{appb5 } \\ \phi _ { p}(t ) & = & \frac{\omega _ { p}}{\zeta } \sin ( \zeta t)e^{-t/\tau } \notag \\ \phi _ { q}(t ) & = & -\gamma _ { q}-\gamma _ { q}\left [ \cos ( \zeta t)-\frac{1}{\zeta \tau } \sin ( \zeta t)\right ] e^{-t/\tau } \\end{aligned}\]]as in the main text @xmath125 is the renormalized oscillator frequency . with these functions we can calculate @xmath437 , @xmath438 and @xmath439 . we obtain , @xmath440 -\phi _ { 0}(t)c_{qq}^{(+)}(t)+\frac{\dot{\phi}_{0}^{2}(t)}{2\omega _ { p}^{2}}\langle q^{2}\rangle _ { \beta } + \frac{\dot{\phi}_{0}^{2}}{\omega _ { p}^{2}}\dot{c}_{qq}^{(+)}(t ) \notag \label{appb6 } \\ x_{qp}(t ) & = & \frac{1}{2}\langle q^{2}\rangle _ { \beta } \left [ 1+\phi _ { 0}^{2}(t)\right ] -\phi _ { 0}(t)c_{qq}(t)+\frac{1}{2}\phi _ { p}^{2}(t)\langle p^{2}\rangle _ { \beta } -\frac{\phi _ { p}(t)}{\omega _ { p}}\dot{c}_{qq}^{(+)}(t ) \notag \\ y_{qp}(t ) & = & \frac{\dot{\phi}_{0}(t)}{2\omega _ { p}}\left [ \phi _ { 0}(t)\langle q^{2}\rangle _ { \beta } -c_{qq}^{(+)}(t)\right ] + \frac{\phi _ { p}}{2}\left [ \langle p^{2}\rangle _ { \beta } \phi _ { 0}(t)-c_{pp}^{(+)}(t)\right]\end{aligned}\]]we notice that @xmath441therefore we find from eq . ( [ appb4 ] ) , @xmath442 , i.e. , the purity is _ unaffected _ by the initial slip . 99 v. b. magalinskii , soviet . jetp 9 , ( 1959 ) 1381 .

we present the results of a deep , wide - area , optical and near - ir survey of massive high - redshift galaxies . the prime focus camera ( suprime - cam ) on the subaru telescope was used to obtain imaging over arcmin fields , while imaging was provided by the sofi camera at the new technology telescope ( ntt ) for a subset of the area , partly from the eso imaging survey ( eis ) . in this paper , we report on the properties of-band selected galaxies , identified from a total area of arcmin to , of which 320 arcmin are complete to . the selection technique was used to assemble complete samples of about 500 candidate massive star - forming galaxies ( ) and about 160 candidate massive passively evolving galaxies ( ) at ; and the color criterion was used to assemble a sample of about 850 extremely red objects ( eros ) . we accurately measure surface densities of arcmin and arcmin for the and the , respectively . both and are strongly clustered , at a level at least comparable to that of eros , with appearing more clustered than . we estimate the reddening , star formation rates ( sfrs ) and stellar masses ( ) for the ensemble of , confirming that to typical ( median ) values are , , and . a correlation is detected such that the most massive galaxies at are also the most actively star - forming , an effect that can be seen as a manifestation of _ downsizing _ at early epochs . the space density of massive at that we derive is 20%% that of similarly massive early - type galaxies at . adding this space density to that of our massive star forming class , , in the same redshift range produces a closer comparison with the local early - type galaxy population , naturally implying that we are detecting star formation in a sizable fraction of massive galaxies at , which has been quenched by the present day . follow - up optical and near infrared spectroscopy is in progress at the eso very large telescope ( vlt ) and at the subaru telescope , in order to elucidate more thoroughly the formation and evolution of massive galaxies .

astro-ph0510299

despite the recent extraordinary progress in observational cosmology and the successful convergence on a single cosmological model , galaxy formation and evolution largely remain an open issue . one critical aspect is how and when the present - day most massive galaxies ( e.g. elliptical galaxies and bulges with @xmath24 ) were built up and what type of evolution characterized their growth over cosmic time ( e.g. , cimatti 2004 ; glazebrook 2004 , and references therein ) . indeed , various current renditions of the @xmath25cdm hierarchical merging paradigm differ enormously in this respect , with some models predicting the virtually complete disappearance of such galaxies by @xmath26 ( e.g. , cole 2000 ; menci 2002 ; somerville 2004a ) and other models predicting a quite mild evolution , more in line with observations ( e.g. , nagamine 2001 ; 2005 ; granato 2004 ; somerville 2004b ; a direct comparison of such models can be found in fig . 9 of fontana 2004 ) . moreover , models that provide an acceptable fit to the galaxy stellar mass function at @xmath27 may differ considerably in the actual properties of the galaxies with @xmath24 at @xmath28 , with some models predicting very few , if any , passively evolving galaxies at these redshifts , at variance with recent findings ( cimatti 2004 ; mccarthy 2004 ; daddi 2005a ; saracco 2005 ) . while various @xmath25cdm models may agree with each other at @xmath22 ( where they all are tuned ) their dramatic divergence with increasing redshift gives us powerful leverage to restrict the choice among them , thus aiding understanding of the physics of galaxy formation and evolution . hence , a direct observational mapping of galaxy evolution through cosmic time is particularly important and rewarding , especially if a significant number of massive galaxies at @xmath29 can be identified and studied . in this regard , the critical questions concern the evolution with redshift of the number density of massive galaxies and their star formation histories , as reflected by their colors and spectral energy distributions ( seds ) . these questions have just started to be addressed by various spectroscopy projects , such as the k20 survey ( cimatti 2002a , 52 arcmin@xmath2 ) , the hubble deep fields ( hdfs ; ferguson 2000 , 5.3 arcmin@xmath2 in the hdf - north and 4.4 arcmin@xmath2 in the hdf - south ) , the great observatories origins deep survey ( goods ; giavalisco 2004 , 320 arcmin@xmath2 in the north and south fields combined ) the hst / acs ultra deep field ( s. beckwith 2006 , in preparation ; 12 arcmin@xmath2 ) , the gemini deep deep survey ( abraham 2004 , 121 arcmin@xmath2 ) , and the extension down to @xmath30 of the lyman break galaxy ( lbg ) project ( steidel 2004 , @xmath31 arcmin@xmath2 ) . however , massive galaxies are quite rare and likely highly clustered at all redshifts , and hence small areas such as those explored so far are subject to large cosmic variance ( daddi 2000 ; bell 2004 ; somerville 2004c ) . therefore , although these observation have demonstrated that old , passive and massive galaxies do exist in the field out to @xmath32 , it remains to be firmly established how their number and evolutionary properties evolve with redshift up to @xmath30 and beyond . to make a major step forward we are undertaking fairly deep , wide - field imaging with the suprime - cam on subaru of two fields of 940 arcmin@xmath2 each for part of which near - ir data are available from eso new technology telescope ( ntt ) observations . the extensive imaging has supported the spectroscopic follow - up with the vlt and the subaru telescopes , for which part of the data have already been secured . the prime aim of this survey is to understand how and when the present - day massive galaxies formed , and to this end , the imaging observations have been optimized for the use of optical / near - ir multi - color selection criteria to identify both star - forming and passive galaxies at @xmath33 . color criteria are quite efficient in singling out high redshift galaxies . the best - known example is the dropout technique for selecting lbgs ( steidel 1996 ) . besides targeting lbgs , color criteria have also been used to search for passively evolving galaxies at high redshifts , such as extremely red objects ( eros ) at redshifts @xmath34 ( thompson 1999 ; mccarthy 2004 ) and distant red galaxies ( drgs ) at redshifts @xmath35 ( franx 2003 ) . recently , using the highly complete spectroscopic redshift database of the k20 survey , daddi ( 2004a ) introduced a new criterion for obtaining virtually complete samples of galaxies in the redshift range @xmath36 , based on @xmath37 , @xmath38 and @xmath39 band for short ] imaging : star - forming galaxies are identified requiring @xmath40 ( for convenience , we use the term for galaxies selected in this way ) ; and passively evolving galaxies at @xmath41 requiring @xmath42 and @xmath43 ( hereafter ) . this criterion is reddening independent for star - forming galaxies in the selected redshift range , thus allowing us also to select the reddest most dust - extinguished galaxies , together with those that are old and passively evolving . this should allow for a relatively unbiased selection of @xmath19 galaxies within the magnitude limit of the samples studied . in this paper observations , data reduction and galaxy photometry are described , together with the first results on k - band selected samples of distant , high redshift massive galaxies . compared to optical , the near - ir selection ( in particular in the @xmath4 band ) offers several advantages , including the relative insensitivity of the k - corrections to galaxy type , even at high redshift , the less severe dust extinction effects , the weaker dependence on the instantaneous star formation activity , and a tighter correlation with the stellar mass of the galaxies . therefore , the study of faint galaxy samples selected in the near - infrared have long been recognized as ideal tools to study the process of mass assembly at high redshift ( broadhurst 1992 ; kauffmann & charlot 1998 ; cimatti 2002a ) . the paper is organized as follows : section 2 describes the observations and the data reduction . section 3 discusses the photometric calibration of the images . section 4 presents the selection and number counts for eros , , and . section 5 presents the analysis of the clustering of field galaxies , eros , , and . the properties of are presented in section 6 . finally , a brief summary is presented in section 7 . throughout the paper , we use the salpeter imf extending between 0.1 and 100 @xmath44 and a cosmology with @xmath45 , and @xmath46(km s@xmath47 mpc@xmath47)@xmath48 . for the sake of comparison with previous works , magnitudes and colors in both ab and vega systems have to be used . , @xmath49 , @xmath50 , and @xmath51 . ] two widely separated fields were imaged as a part of our survey : one centered at @xmath52(j2000)@xmath53 , @xmath54(j2000)@xmath55 ( hereafter deep3a - f ) , and the the second , the so - called `` daddi field '' ( hereafter daddi - f ; daddi 2000 ) centered at @xmath52(j2000)@xmath56 , @xmath54(j2000)@xmath57 . details of the optical and near - ir observations are shown in table 1 . figure [ fig : area ] shows the layout of the two areas observed . 0.9@llrcccr filter & telescope & obs . date & exps . & seeing&@xmath58 & area + & & & ( sec ) & ( @xmath59)&(mag ) & ( arcmin@xmath2 ) + + + & subaru & mar.5 , 03 & 3900 & 0.77 & 27.4 & 940 + _ r@xmath60 _ & subaru & mar.4 - 5 , 03 & 7320 & 0.85 & 26.9 & 940 + _ i _ & subaru & mar.4 - 5 , 03 & 5700 & 0.77 & 26.5 & 940 + _ z@xmath61 _ & subaru & mar.4 - 5 , 03 & 9900 & 0.80 & 26.0 & 940 + _ j _ & ntt & jan.00-feb.01 & 3600 & 0.76 & 23.4 & 320 + _ @xmath39_&ntt & jan.00-feb.01 & 4800 & 0.76 & 22.7 & 320 + + + & subaru & mar . 5 , 03 & 1500 & 0.75 & 27.0 & 940 + _ r_&wht & may 19 - 21 , 98 & 3600 & 0.70 & 25.6 & 715 + _ i _ & subaru & mar . 5 , 03 & 1800 & 0.90 & 26.0 & 940 + _ z@xmath61 _ & subaru & mar . 4 - 5 , 03 & 2610 & 0.80 & 25.5 & 940 + _ @xmath39_&ntt & mar . 27 - 30 , 99 & 720 & 0.90 & 21.5 & 600 + infrared observations in the near - infrared passband @xmath62 and @xmath39 were obtained using the sofi camera ( moorwood , cuby & lidman 1998 ) mounted on the new technology telescope ( ntt ) at la silla . sofi is equipped with a rockwell 1024@xmath2 detector , which , when used together with its large field objective , provides images with a pixel scale of @xmath63 and a field of view of @xmath64 arcmin@xmath2 . deep3a - f is part of the eso deep public survey ( dps ) carried out by the eso imaging survey ( eis ) program ( renzini & da costa 1999 ) ( see http://www.eso.org/science/eis/ ) . the deep3a sofi observations cover a total area of about 920 arcmin@xmath2 in the k band , most at the relatively shallow limits of @xmath6519.5 . about 320 arcmin@xmath2 , the region used in the present paper , have much deeper integrations with a minimum 3600s per sky pixel ( and up to 2 hr ) reaching to @xmath66 and @xmath67 . the ntt @xmath62- and @xmath4-band images of deep3a - f were retrieved from the eso science archive and reduced using the eis / mvm pipeline for automated reduction of optical / infrared images ( vandame 2002 ) . the software produces fully reduced images and weight - maps carrying out bias subtraction , flat - fielding , de - fringing , background subtraction , first - order pixel - based image stacking ( allowing for translation , rotation and stretching of the image ) and astrometric calibration . mosaicking of individual sofi fields was based on the astrometric solution . photometric calibration was performed using standard stars from persson ( 1998 ) , and the calibration performed as linear fits in airmass and color index whenever the airmass and color coverage allowed for it . the reduced ntt @xmath4-band data and wht @xmath68-band data for daddi - f were taken from daddi ( 2000 ) . the average seeing and the size of the final coadded images are reported in table 1 . deep optical imaging was obtained with the prime focus camera on the subaru telescope , suprime - cam , which with its 10 @xmath69k @xmath70k mit / ll ccds covers a contiguous area of @xmath71 with a pixel scale of @xmath72 pixel@xmath47 ( miyazaki 2002 ) . deep3a - f was observed with the four standard broad - band filters , @xmath37 , @xmath73 ( hereafter @xmath68 band ) , @xmath74 , and @xmath75 ( hereafter @xmath38 band ) on the two nights of 2003 march 45 with @xmath76 seeing . during the same nights also daddi - f was imaged in @xmath37 , @xmath74 , and @xmath38 to a somewhat shallower magnitude limit to match the shallower @xmath68 and @xmath4 data from daddi ( 2000 ) . a relatively long unit exposure time of several hundred seconds was used in order to reach background - noise - dominated levels . for this reason bright stars , which are saturated in the optical images , have been excluded by subsequent analysis . during the same nights the photometric standard - star field sa95 ( landolt 1992 ) was observed for @xmath77 , @xmath78 , and @xmath79band flux calibration , and the sdss standard - star fields sa95 - 190 and sa95 - 193 were observed for @xmath38-band flux calibration ( smith 2002 ) . the subaru imaging was reduced using the pipeline package sdfred ( yagi 2002 ; ouchi 2004 ) . the package includes overscan correction , bias subtraction , flat - fielding , correction for image distortion , psf matching ( by gaussian smoothing ) , sky subtraction , and mosaicking . bias subtraction and flat fielding were processed in the same manner as for the conventional single chip ccd . in mosaicking , the relative positions ( shifts and rotations ) and relative throughput between frames taken with different ccds and exposures are calculated using stars common to adjacent frames and running ( bertin & arnout 1996 ) with an s / n = 10 threshold . we obtained k - selected catalogs of objects in our survey by detecting sources in the k - band mosaics . for the daddi - f we used the sample of @xmath4-selected galaxies defined in daddi ( 2000 ) . ( bertin & arnouts 1996 ) was used to perform the image analysis and source detection in deep3a - f . the total magnitudes were then defined as the brightest between the kron automatic aperture magnitudes and the corrected aperture magnitude . multicolor photometry in all the available bands was obtained by running in double image mode , after aligning all imaging to the k - band mosaic . colors were measured using 2@xmath59 diameter aperture magnitudes , corrected for the flux loss of stars . the aperture corrections were estimated from the difference between the kron automatic aperture magnitudes ( mag_auto ) and the 2@xmath80 aperture magnitudes , resulting in a range of 0.10 0.30 mag , depending on the seeing . all magnitudes were corrected for galactic extinction ( @xmath81 and @xmath82 for deep3a - f and daddi - f , respectively ) taken from schlegel ( 1998 ) , using the empirical selective extinction function of cardelli ( 1989 ) , with @xmath83 . in deep3a - f we selected objects to @xmath84 , over a total sky area of 320 arcmin@xmath2 . simulations of point sources show that in all the area the completeness is well above 90% at these k - band levels . we recall that objects in daddi - f were selected to completeness limits of @xmath85 over an area of 700 arcmin@xmath2 ( but we limit our discussion in this paper to the 600 arcmin@xmath2 covered by the subaru observations ) and to @xmath86 over a sub - area of 440 arcmin@xmath2 ( see daddi 2000 for more details ) . the total area surveyed , as discussed in this paper , therefore ranges from a combined area of 920 arcmin@xmath2 ( daddi - f and deep3a - f ) at @xmath85 to 320 arcmin@xmath2 ( deep3a - f ) at @xmath84 . objects in deep3a - f and daddi - f were cross - correlated with those available from the 2mass survey ( cutri 2003 ) in the j and k bands , resulting in good photometric agreement at better than the 3% level . in order to further verify the photometric zero points we checked the colors of stars ( fig . [ fig : ccd ] ) , selected from the @xmath8 diagram following daddi ( 2004a ; see sect.[sec : s / g ] ) . from these color - color planes , we find that the colors of stellar objects in our data are consistent with those of pickles ( 1998 ) , with offsets , if any , of @xmath87 mag at most . similar agreement is found with the lejeune ( 1997 ) models . figure [ fig : knumc ] shows a comparison of k - band number counts in our survey with a compilation of literature counts . no attempt was made to correct for different filters ( @xmath39 or @xmath4 ) . no corrections for incompleteness were applied to our data , and we excluded the stars , using the method in sect . [ sec : s / g ] . the filled circles and filled squares correspond , respectively , to the counts of deep3a - f and daddi - f . as shown in the figure , our counts are in good agreement with those of previous surveys . stellar objects are isolated with the color criterion ( daddi 2004a ) @xmath88 . in fig . [ fig : sandg ] we compare the efficiency of such a color - based star - galaxy classification with the one based on the parameter class_star , which is based on the shape of the object s profile in the imaging data . it is clear that the color classification is superior , allowing us to reliably classify stars up to the faintest limits in the survey . however , appear to find resolved in the imaging data a small fraction of objects that are color - classified as stars . most likely , these are blue galaxies scattered into the stellar color boundaries by photometric uncertainties . in this section , we select , and eros in the deep3a - f and daddi - f , using the multicolor catalog based on the nir k - band image ( see sect . [ sec : photo ] ) . forthcoming papers will discuss the selection of drgs ( @xmath92 objects ) and lbgs , using the @xmath93 photometry from our database . in order to apply the @xmath8 selection criteria consistently with daddi ( 2004a ) , we first accounted for the different shapes of the filters used , and applied a correction term to the b - band . the b - band filter used at the subaru telescope is significantly redder than that used at the vlt by daddi ( 2004a ) . we then carefully compared the stellar sequence in our survey to that of daddi ( 2004a ) , using the pickles ( 1998 ) stellar spectra and the lejeune ( 1997 ) corrected models as a guide , and applied small color terms to @xmath94 and @xmath95 ( smaller than @xmath96 mags in all cases ) , in order to obtain a fully consistent match . in the following we refer to @xmath8 photometry for the system defined in this way , consistent with the original @xmath8 definition by daddi ( 2004a ) . figure [ fig : bzk ] shows the @xmath8 color diagram of k - selected objects in deep3a - f and daddi - f . using the color criterion from daddi ( 2004a ) , @xmath97 , 387 galaxies with @xmath98 were selected in deep3a - f as , which occupy a narrow range on the left of the solid line in fig . [ fig : bzk]a . using @xmath99 and @xmath43 , 121 objects were selected as candidate , which lie in the top - right part of fig . [ fig : bzk]a . to @xmath84 a surface density of @xmath11 arcmin@xmath12 and of @xmath13 arcmin@xmath12 is derived separately for and , respectively ( poisson s errors only ) . the surface density of is larger but still consistent within 2@xmath100 with the @xmath101 arcmin@xmath12 found in the 52 arcmin@xmath2 of the k20 field ( daddi 2004a ) , and with the @xmath102 arcmin@xmath12 found in the goods north field ( daddi 2005b ) . instead , the surface density of recovered here is significantly larger than what found in both fields . this may well be the result of cosmic variance , given the strong clustering of ( see section [ sec : clustering ] ) , and their lower overall surface density . using the same criteria , we select and in the daddi - f field . in daddi - f 108 and 48 are selected , and they are plotted in fig . [ fig : bzk]c . the density of and in daddi - f is consistent with that in deep3a , if limited at @xmath103 . eros were first identified in @xmath4-band surveys by elston , rieke , & rieke ( 1988 ) , and are defined here as objects having red optical - to - infrared colors such that @xmath1046 , corresponding to @xmath1054.35 . eros are known to be a mixture of mainly two different populations at @xmath106 ; passively evolving old elliptical galaxies and dusty starburst ( or edge - on spiral ) galaxies whose uv luminosities are strongly absorbed by internal dust ( cimatti 2002b ; yan 2004 , moustakas 2004 ) . in daddi - f , eros were selected and studied by daddi ( 2000 ) using various @xmath107 thresholds . in order to apply a consistent ero selection in deep3a - f , we considered the filter shapes and transmission curves . while the same k - band filters ( and the same telescope and instrument ) were used for k - band imaging in the two fields , the r - band filters used in the two fields differ substantially . in the daddi - f the wht r - band filter was used , which is very similar , e.g. , to the r - band filter of fors at the vlt used by the k20 survey ( cimatti 2002a ) . the subaru+suprime - cam @xmath68-band filter is much narrower than the above , although it does have a very close effective wavelength . as a result , distant @xmath108 early - type galaxy spectra as well as m - type stars appear to have much redder @xmath107 color , by about @xmath1090.3 mag , depending on exact redshift and spectral shape . therefore , we selected eros in deep3a - f with the criterion @xmath110 ( ab magnitudes ) , corresponding closely to @xmath111 ( ab magnitudes ) or @xmath112 ( vega magnitudes ) . in deep3a - f , 513 eros were selected to @xmath84 , and they are plotted in fig . [ fig : bzk]b with solid red triangles , for a surface density of 1.6 arcmin@xmath12 . to the same @xmath84 limit , this agrees well with the density found , e.g. , in the k20 survey ( @xmath113 arcmin@xmath12 ) , or in the 180 arcmin@xmath2 survey by georgakakis ( 2005 ) . in the daddi - f , 337 eros were selected with the criterion @xmath111 , consistent with what done in deep3a , and are plotted in fig . [ fig : bzk]d as red solid triangles . the surface density of eros in both fields at @xmath114 can be compared , with overall good consistency , to the one derived from the large 1 deg@xmath2 survey by brown ( 2005 ) . the peak of the eros redshift distribution is at @xmath108 ( e.g. , cimatti 2002a ) . by looking at the @xmath8 properties of eros we can estimate how many of them lie in the high-@xmath38 tail @xmath115 , thus testing the shape of their redshift distribution . in the deep3a - f to @xmath84 some 90 of the eros are also , thus likely belong to the category of dusty starburst eros at @xmath115 , while 121 eros are classified as . totally , @xmath116% of eros are selected with the @xmath8 criteria , thus expected to lie in the high - z tail ( @xmath115 ) for @xmath84 sample . this result is consistent with the value of 35% found in the 52 arcmin@xmath2 of the k20 field ( daddi 2004a ) , and with the similar estimates of moustakas ( 2004 ) for the full goods - south area . in the daddi - f , to @xmath86 , 49 of the eros are also ; and 48 of them are also . about 29% of eros at @xmath86 are in the high - z tail at @xmath115 . 1.0@lccccrcccc & + k bin center&galaxies&eros & & & k bin center&galaxies&eros & & + 16.75 & 2.981 & 1.353 & & & 16.75 & 2.888 & 1.254 & & + 17.00 & 3.109 & 1.654 & & & 17.00 & 3.023 & 1.555 & & + 17.25 & 3.213 & 1.830 & & & 17.25 & 3.165 & 1.891 & & + 17.50 & 3.337 & 2.198 & & & 17.50 & 3.323 & 2.120 & & + 17.75 & 3.470 & 2.483 & 1.353 & & 17.75 & 3.426 & 2.321 & 1.254 & + 18.00 & 3.565 & 2.675 & 1.654 & 1.052 & 18.00 & 3.498 & 2.590 & 1.622 & 1.078 + 18.25 & 3.678 & 2.822 & 2.006 & 1.830 & 18.25 & 3.569 & 2.786 & 1.923 & 1.379 + 18.50 & 3.764 & 3.025 & 2.228 & 2.353 & 18.50 & 3.669 & 2.888 & 2.209 & 1.891 + 18.75 & 3.802 & 3.138 & 2.529 & 2.467 & 18.65 & 3.708 & 2.990 & 2.342 & 2.175 + 19.00 & 3.859 & 3.145 & 2.724 & 2.596 & 19.00 & 3.783 & 3.063 & 2.834 & 2.461 + 19.25 & 3.911 & 3.162 & 2.971 & 2.608 & & & & & + 19.50 & 3.983 & 3.201 & 3.228 & 2.675 & & & & & + 19.75 & 4.072 & 3.297 & 3.470 & 2.759 & & & & & + simple surface densities provide limited insight into the nature of different kind of galaxies . however , number - magnitude relations , commonly called number counts , provide a statistical probe of both the space distribution of galaxies and its evolution . for this reason , we derived @xmath4-band differential number counts for eros , and in our fields , and plotted them in figure [ fig : numcz ] . the differential number counts in 0.5 mag bins are shown in table [ tab : numc_highz ] . also shown are the number counts for all @xmath4-selected field galaxies in deep3a - f ( circles with solid line ) and daddi - f ( squares with dot - dashed line ) , with the same as in fig . [ fig : knumc ] for comparison . a distinguishable characteristic in fig . [ fig : numcz ] is that all of high redshift galaxies ( eros , , and ) have faint nir apparent magnitudes ( @xmath11717 mag ) , and the slopes of the counts for eros , and are steeper than that of the full @xmath4-selected sample . the open squares with a solid line in fig . [ fig : numcz ] shows the number counts for in deep3a - f . the fraction of in deep3a - f increases very steeply towards fainter magnitudes . the triangles with a solid line and crosses with a solid line show , respectively , the number counts of eros and in deep3a - f . the open squares with a dot - dashed line in fig . [ fig : numcz ] show the number counts for in daddi - f . the counts of in daddi - f are almost identical to those in deep3a - f , to their limit of @xmath118 . the triangles with a dot - dashed line and crosses with a dot - dashed line in fig . [ fig : numcz ] show respectively the number counts for eros and in daddi - f . for eros , the slope of the number counts is variable , being steeper at bright magnitudes and flattening out toward faint magnitudes . a break in the counts is present at @xmath119 , very similar to the break in the ero number counts observed by mccarthy ( 2001 ) and smith ( 2002 ) . the number counts have a similar shape , but the break in the counts slope is apparently shifted @xmath1201.5 mag fainter . there are indications that eros and have fairly narrow redshift distributions : peaked at @xmath108 for eros ( cimatti 2002b ; yan 2004 ; doherty 2005 ) and at @xmath121 for ( daddi 2004b ; 2005a ) . the number counts might therefore be direct probes of their respective luminosity function , and the shift in the counts is indeed consistent with the different typical redshifts of the two population of galaxies . the counts of have roughly the same slope at all k - band magnitudes . this is consistent with the much wider redshift distribution of this class of galaxies . however , we expect that at bright magnitudes agn contamination might be more relevant than at @xmath122 . correcting for this , the counts of non - agn galaxies at bright magnitudes might be intrinsically steeper . measuring the clustering of galaxies provides an additional tool for studying the evolution of galaxies and the formation of structures . in this section we estimate over the two fields the angular correlation of the general galaxy population as well as of the eros , and . in order to measure the angular correlation function of the various galaxy samples , we apply the landy & szalay technique ( landy & szalay 1993 ; kerscher 2000 ) , following the approach already described in daddi ( 2000 ) , to which we refer for formulae and definitions . in our analysis a fixed slope @xmath123 0.8 was assumed for the two - point correlation function [ @xmath124 . this is consistent with the typical slopes measured in both faint and bright surveys and furthermore makes it possible to directly compare our results with the published ones that were obtained adopting the same slope ( see daddi 2000 for more details ) . in figures [ fig : dad_clu ] and [ fig : d3a_clu ] the bias - corrected two - point correlation functions @xmath125 of daddi - f and deep3a - f are shown as squares ; the bins have a constant logarithmic width ( @xmath126 ) , with the bin centers ranging from 5.6@xmath80 to 11.8@xmath127 for both fields . these values of @xmath128 are large enough to avoid problems of under - counting caused by the crowding of galaxy isophotes and yet are much smaller than the extent of the individual fields . the dashed line shows the power - law correlation function given by a least square fit to the measured correlations . we clearly detect a positive correlation signal for both fields with an angular dependence broadly consistent with the adopted slope @xmath123 0.8 . the derived clustering amplitudes ( where @xmath129 is the amplitude of the true angular correlation at 1@xmath130 ) are presented in the third column of table [ tab : dad_clu ] and of table [ tab : d3a_clu ] for daddi - f and deep3a - f , respectively , and shown in fig . [ fig : angular ] . we compare our result on daddi - f with those previously reported by daddi ( 2000 ) , since we use the same @xmath4-band images for this field . the present @xmath4-selected galaxy sample is however different from that of daddi ( 2000 ) in thta : ( 1 ) the area ( 600 @xmath131 ) is smaller than that ( 701 @xmath131 ) in daddi ( 2000 ) , because we limit ours to the area with all the @xmath132-band data ; and ( 2 ) the star - galaxy separation was done by different methods . in daddi ( 2000 ) the class_star _ morphological _ parameter was used , while the _ photometric _ criterion of daddi ( 2004a ) is used here . column 10 of table [ tab : dad_clu ] lists the clustering amplitudes of all the @xmath4-selected galaxies in daddi ( 2000 ) . for @xmath133 and fainter bins the @xmath129-values in the two samples are in fair agreement , but for the brightest bin ( @xmath134 ) the @xmath129-value in daddi ( 2000 ) is @xmath135 , 45% smaller than that found here ( @xmath136 ) , probably mainly due to the more efficient star - galaxy separation employed here ( sect . [ sec : s / g ] ) . apart from this small discrepancy , we find good agreement between our results and those of daddi ( 2000 ) and oliver ( 2004 ) . the clustering amplitudes in deep3a - f tend to be slightly but systematically smaller than in daddi - f , which is likely caused by the intrinsic variance among different fields , depending on the survey geometry , surface density and clustering properties ( see sect . [ sec : cos ] ) . 1.0@lrrrrrrrrrr & & & & & + k limit & num . & a[10@xmath137 & num . & a[10@xmath137 & num . & a[10@xmath137 & num . & a[10@xmath137 & gal . & eros + 18.0 & 978 & [email protected] & 51&[email protected] & 7 & & 1 & &1.3&24 + 18.5 & 1589 & [email protected] & 132&[email protected] & 21&[email protected] & 5 & &1.6&22 + 18.8 & 2089 & [email protected] & 228&[email protected] & 43&[email protected] & 20&[email protected]&1.5&14 + 19.2 & 2081 & [email protected] & 264&[email protected] & 92&[email protected] & 40&[email protected]&1.6&13 + 1.0@lrrrrrrrrrrr & & & & & & & + k limit & num . & a[10@xmath137 & & num . & a[10@xmath137 & & num . & a[10@xmath137 & & num . & a[10@xmath137 + 18.5 & 997 & [email protected] & & 95&[email protected] & & 13 & & & 8 & + 18.8 & 1332 & [email protected] & & 166 & [email protected] & & 27&[email protected] & & 26&[email protected] + 19.5 & 2284 & [email protected] & & 340 & [email protected] & & 129 & [email protected] & & 71&[email protected] + 20.0 & 3333 & [email protected] & & 513 & [email protected] & & 387 & [email protected] & & 121&[email protected] + we find a smooth decline in amplitude with k - band magnitude that is consistent with the results from roche ( 2003 ) , fang ( 2004 ) , and oliver ( 2004 ) . the decline is not as steep as in the range @xmath139 ( see e.g. roche 1999 ) . the flattening , which extends up to @xmath14024 ( carlberg 1997 ; daddi 2003 ) has been interpreted as due to the existence of strongly clustered k - selected galaxy populations extending to redshifts @xmath1083 ( daddi 2003 ) . beside the well known @xmath141 eros strongly clustered galaxy populations , other populations with high angular clustering indeed exist with redshift extending to @xmath142 at least , as discussed in the next sections . we estimate the clustering properties of the eros , using the large sample of eros derived from our two fields . figures [ fig : distrib_obj]a ) and [ fig : distrib_obj]b ) clearly show , from both fields , that the sky distribution of eros is very inhomogeneous , as first noted by daddi ( 2000 ) . the two - point correlation functions are shown in fig . [ fig : dad_clu ] and [ fig : d3a_clu ] for daddi - f and deep3a - f , and the dot - dashed lines show the power - law correlation function given by a least squares fit to the measured values . the correlations are well fitted by a @xmath143 power law . a strong clustering of the eros is indeed present at all scales that could be studied , and its amplitude is about one order of magnitude higher than that of the field population at the same @xmath144 limits , in agreement with previous findings ( daddi 2000 ; firth 2002 ; brown 2005 ; georgakakis 2005 ) . the derived clustering amplitudes are reported in column ( 5 ) of table [ tab : dad_clu ] and table [ tab : d3a_clu ] for daddi - f and deep3a - f , respectively . the amplitudes shown in fig . [ fig : angular ] suggest a trend of decreasing strength of the clustering for fainter eros in both fields . for the @xmath145 and @xmath146 mag subsamples of daddi - f , the correlation function signal is significant at the 7 @xmath100 level with clustering amplitudes @xmath147 and @xmath148 , respectively . for the subsample with @xmath149 mag , the detected signal is significant at the 10 @xmath100 confidence level ( @xmath150 ) . in column 11 of table [ tab : dad_clu ] we also list the clustering amplitudes of eros in daddi ( 2000 ) , which are in good agreement with the present findings . using a @xmath151 arcmin@xmath2 ks - band survey of a region within the phoenix deep survey , georgakakis ( 2005 ) have analyzed a sample of 100 eros brighter than @xmath152 mag , and estimated an amplitude @xmath153 , consistent with our results . as for the @xmath4-selected galaxies , the clustering amplitude of eros in deep3a - f is slightly smaller than that in daddi - f . the clustering amplitudes are @xmath154 for @xmath85 eros , and @xmath155 for @xmath156 eros , which is weaker than the @xmath157 for @xmath84 eros in georgakakis ( 2005 ) . field - to - field variation is one of possible reasons for this discrepancy , which is discussed in sect . [ sec : cos ] . we notice that our results are solid against possible contamination by stars . contamination by unclustered populations ( i.e. stars ) would reduce the amplitude of the angular correlation function by @xmath158 , where @xmath159 is the fractional contamination of the sample . the prime candidates for contamination among eros are red foreground galactic stars ( note that stars are not contaminating bzk samples ) , which can have red @xmath107 colors . however , we have rejected stars among eros using the photometric criterion for star - galaxy separations ( see sect . [ sec : s / g ] ) . therefore , our clustering measurements refer to extragalactic eros only . figures [ fig : distrib_obj]c ) and [ fig : distrib_obj]d ) display the sky distribution of the in daddi - f and deep3a - f , and shows that also these galaxies have a quite inhomogeneous distribution . this is not an artifact of variations of the detection limits over the fields , because the monte carlo simulations show that differences in detection completeness within small ( @xmath160 ) areas in the image are very small , and that the detection completeness does not correlate with the distribution of the . the resulting angular correlation functions for the are shown in fig . [ fig : dad_clu ] and fig . [ fig : d3a_clu ] for the two fields . again , a slope @xmath161 provides a good fit to the data . the best fit values of @xmath129 are reported in column 7 of table [ tab : dad_clu ] and table [ tab : d3a_clu ] . the @xmath162 amplitudes of in daddi - f are @xmath163 , @xmath164 and @xmath165 at @xmath166 , 18.8 and 19.2 , respectively . for in deep3a - f , the @xmath125 amplitudes become @xmath167 , @xmath168 and @xmath169 at @xmath170 , 19.5 and 20.0 , respectively . the appear to be strongly clustered in both fields , and the clustering strength increases with the @xmath4-band flux . actually , they are as strongly clustered as the eros . strong clustering of the was also inferred by daddi ( 2004b ) , by detecting significant redshift spikes in a sample of just nine @xmath84 with spectroscopic redshifts in the range @xmath171 . albeit without spectroscopic redshifts , the present result is instead based on a sample of 500 . adelberger ( 2005 ) also found that uv - selected , star forming galaxies with @xmath172 in the redshift range @xmath173 are strongly clustered . our results are in good agreement with these previous findings . figures [ fig : distrib_obj]e ) and [ fig : distrib_obj]f ) display the sky distribution of the in daddi - f and deep3a - f , and show that also these galaxies have a very inhomogeneous distribution . we then derive the angular two - point correlation function of , using the same method as in the previous subsections . the resulting angular correlation functions for the are shown in fig . [ fig : dad_clu ] and fig . [ fig : d3a_clu ] for the two fields . again , a slope @xmath161 provides a good fit to the data . the best fit values of @xmath129 are reported in column ( 9 ) of table [ tab : dad_clu ] and table [ tab : d3a_clu ] . the appear to be the most strongly clustered galaxy population in both fields ( with @xmath174 for @xmath86 in daddi - f and @xmath175 for @xmath176 in deep3a - f ) , and the clustering strength increases with increasing @xmath4-band flux . [ fig : angular ] summarizes the clustering measurements for the populations examined ( field galaxies , eros , and bzks ) , as a function of the @xmath4-band limiting magnitudes of the samples . clear trends with @xmath4 are present for all samples , showing that fainter galaxies have likely lower intrinsic ( real space ) clustering , consistent with the fact that objects with fainter @xmath4 are less massive , or have wider redshift distributions , or both . for the eros ( see sect . [ sec : erosel ] ) we have found evidence that faint @xmath84 samples have indeed a higher proportion of galaxies in the @xmath115 tail , with respect to brighter @xmath177 objects , and thus a wider redshift distribution . as already noted , all color - selected high - redshift populations are substantially more clustered than field galaxies , at all the magnitudes probed here . the reason for the stronger angular clustering of , compared , e.g , to or eros , is likely ( at least in part ) their narrower redshift distributions @xmath178 ( daddi 2004a ; 2005a ) . in future papers , we will use the limber equation , with knowledge of the redshift distributions , to compare the real space correlation length of the different populations . for strongly clustered populations , with angular clustering amplitudes @xmath179 , a large cosmic variance of clustering is expected , which is relevant also on fields of the size of the ones studied here ( see daddi 2001 ; 2003 ) . there are in fact variations in our clustering measurements for high - redshift objects between the two fields , sometimes larger than expected on the basis of the errors on each clustering measurements . we remind the reader that realistic _ external _ errors on the angular clustering of eros , as well as and are likely larger than the poissonian ones that we quote . following the recipes by daddi ( 2001 ) we estimate that the typical total accuracy for our measurements of @xmath180 , when including _ external _ variance , is on the order of 30% . the presence of strong clustering will also produce substantial field to field variations in the number counts . given the available measurements of angular clustering , presented in the previous sections , we are able to quantify the expected variance in the galaxies counts , following e.g. eq . ( 8) of daddi ( 2000 ) . we estimate that , for the deep3a - f limit of @xmath84 , the integrated numbers of objects are measured to @xmath181% precision for eros and , and to 30% precision for . the accurate analysis of the physical properties of high redshift galaxies ( such as sfr , stellar mass , etc . ) requires the knowledge of their spectroscopic redshift . vlt vimos spectra for objects culled by the present sample of and eros have been recently secured and are now being analyzed , and will be used in future publications . in the meantime , estimates of these quantities can be derived on the basis of the present photometric data , following the recipes calibrated in daddi ( 2004a ) , to which we refer for definitions and a more detailed discussion of the recipes . while errors by a factor of 2 or more may affect individual estimates done in this way , when applied to whole population of @xmath8-selected galaxies these estimates should be relatively robust , on average , because the daddi et al.s ( 2004a ) relations were derived from a sample of galaxies with spectroscopic redshifts . the estimates presented here thus represent a significant improvement on the similar ones provided by daddi ( 2004a ) because of the 6 to 20 times larger area probed ( depending on magnitude ) with respect to the k20 survey , which should help to significantly reduce the impact of cosmic variance . following daddi ( 2004a ) , estimates of the reddening @xmath182 and sfr for can be obtained from the @xmath8 colors and fluxes alone ( see daddi 2004a for more details ) . the reddening can be estimated by the @xmath94 color , providing a measure of the uv slope . the daddi s recipe is consistent with the recipes by meurer ( 1999 ) and kong ( 2004 ) for a calzetti ( 2000 ) extinction law , based on the uv continuum slope . daddi et al . ( 2004a ) showed that for @xmath8 galaxies this method provides @xmath182 with an rms dispersion of the residuals of about 0.06 , if compared to values derived with knowledge of the redshifts , and with the use of the full multicolor sed . with knowledge of reddening , the reddening corrected b - band flux is used to estimate the 1500 rest - frame luminosity , assuming an average redshift of 1.9 , which can be translated into sfr on the basis , e.g. , of the bruzual & charlot ( 2003 ) models . daddi et al . ( 2005b ) showed that sfrs derived in this way are consistent with radio and far - ir based estimates , for the average . the @xmath182 and sfr histograms of the in daddi - f and deep3a - f are shown in fig . [ fig : bzk_nat ] . about 95% of the in daddi - f ( @xmath149 ) have sfr@xmath183yr@xmath47 , and the median sfr is about @xmath184yr@xmath47 . about 90% of the in deep3a - f ( @xmath185 ) have sfr@xmath18670@xmath187yr@xmath47 , and the median sfr is @xmath188 @xmath44yr@xmath47 . the median reddening for the @xmath84 is estimated to be @xmath189 , consistently with daddi ( 2004a ; 2005b ) . of 55% have @xmath190 , the limit at which we estimate the uv - based criteria of steidel ( 2004 ) would fail at selecting @xmath19 galaxies . therefore , we estimate that @xmath191% of the @xmath19 galaxies would be missed by the uv criteria . this is similar to , but higher , than the 40% estimated by reddy ( 2005 ) . the probable reason for the small discrepancy is that reddy ( 2005 ) excluded from their sample the reddest for which the optical magnitudes could not be accurately measured in their data . using bc03 models , spectroscopic redshifts of individual k20 galaxies , and their whole @xmath192@xmath193@xmath194 sed , fontana ( 2004 ) have estimated the stellar - mass content for all the k20 galaxies ( using a salpeter imf from 0.1 to 100 m@xmath195 , as adopted in this paper ) . the individual mass estimates for 31 non - agn and objects with @xmath115 have been used by daddi ( 2004a ) to calibrate an empirical relation giving the stellar mass for both and as a function of their observed @xmath4-band total magnitude and @xmath95 color . this relation allows one to estimate the stellar mass with uncertainties on single objects of about 60% compared to the estimates based on knowledge of redshifts and using the full multicolor seds . the relatively small variance is introduced by intrinsic differences in the luminosity distance , in the @xmath196 ratio for given magnitudes and/or colors . the histograms for the stellar mass of the derived in this way are shown in fig . [ fig : bzk_nat]e and fig . [ fig : bzk_nat]f . about 95% of the in daddi - f have @xmath197 , and the median stellar mass is @xmath198 ; in deep3a - f @xmath199 of the have @xmath197 , the median stellar mass is @xmath200 , a difference due to the different limiting @xmath4 magnitude in the two fields . using the same method , we also estimate the stellar mass of the in both fields , and plot them in fig . [ fig : bzk_nat]e and fig . [ fig : bzk_nat]f as the dotted line . the median stellar mass of in daddi - f is @xmath201 , and in deep3a - f is @xmath202 , respectively . again , the higher masses for in daddi - f compared to deep3a - f result from the shallower @xmath4-band limit . it is worth noting that in the @xmath203 sample there are barely any less massive than @xmath204 ( see fig . [ fig : bzk_nat]f ) , while over 50% of are less massive than this limit . this is primarily a result of being by construction redder than @xmath205 , and hence eq . ( 6 ) and eq . ( 7 ) in daddi ( 2004a ) with @xmath84 implies @xmath206 . note that above @xmath207 ( above which our sample should be reasonably complete ) the numbers of and are similar . we return to this point in the last section . figures [ fig : bzk_cor]a ) and [ fig : bzk_cor]b ) show the correlation between color excess @xmath182 and sfr for the in daddi - f and deep3a - f , respectively . the spearman rank correlation coefficients are @xmath208 for daddi - f and @xmath209 for deep3a - f . this implies that the sfr is significantly correlated to @xmath182 , at a @xmath210 level , and the reddest galaxies have the highest sfr . part of this correlation can arise from simple error propagation , as an overestimate ( underestimate ) of the reddening automatically leads to an overestimate ( underestimate ) of the sfr of a galaxy . in fig . [ fig : bzk_cor]a ) and fig . [ fig : bzk_cor]b ) the arrow shows the resulting slope of the correlated reddening errors , which indeed is parallel to the apparent correlation . however , the scatter in the original sfr-@xmath182 correlation [ @xmath211 ; daddi 2004a ] is much smaller than what needed to produce the full correlation seen in fig . [ fig : bzk_cor]a ) and [ fig : bzk_cor]b ) . we conclude that there is evidence for an _ intrinsic _ correlation between sfr and reddening for @xmath30 star - forming galaxies , with galaxies with higher star formation rates having more dust obscuration . a positive correlation between sfr and reddening also exists in the local universe ( see fig . 1 of calzetti 2004 ) , and was also found by adelberger & steidel ( 2000 ) for @xmath212 lbgs , on a smaller range of reddening . in figures [ fig : bzk_cor]c ) and [ fig : bzk_cor]d ) , we plot the relation between color excess @xmath182 and stellar mass of the in daddi - f and deep3a - f . the spearman rank correlation coefficient is @xmath213 for daddi - f and @xmath214 for deep3a - f , indicating that the correlation between @xmath182 and stellar mass is significant at the @xmath215 level in both fields . in this case the estimate of the stellar mass depends only mildly on the assumed reddening , and therefore the correlation is likely to be intrinsic , with more massive galaxies being also more absorbed . given the previous two correlation , not surprisingly we also find a correlation between sfr and stellar mass ( figure [ fig : bzk_cor]e and [ fig : bzk_cor]f ) . the spearman rank correlation coefficient is @xmath216 for daddi - f , and @xmath217 for deep3a - f , indicating that the correlation between sfr and stellar mass is significant at the @xmath218 level in both fields . the origin of the sharp edge in fig . [ fig : bzk_cor]e ) and fig . [ fig : bzk_cor]f ) is caused by the color limit @xmath219 for selecting . to show this clearly , we plot galaxies with @xmath220 as open squares in fig . [ fig : bzk_cor]f ) . however , no or very few @xmath19 galaxies exist below the @xmath219 line , and therefore the upper edge shown in the figure appears to be intrinsic , showing a limit on the maximum sfr that is likely to be present in a galaxy of a given mass . at @xmath221 the vast majority of massive galaxies ( @xmath24 ) are passively evolving , `` red '' galaxies ( e.g. , baldry et al . 2004 ) , while instead at @xmath30 , actively star - forming ( ) and passive ( ) galaxies exist in similar numbers , and fig . [ fig : bzk_cor ] shows that the most massive tend also to be the most actively star forming . this can be seen as yet another manifestation of the _ downsizing _ effect ( e.g. , cowie et al . 1996 ; kodama et al . 2004 ; treu et al . 2005 ) , with massive galaxies completing their star formation at an earlier epoch compared to less massive galaxies , which instead have more prolonged star formation histories . because of the correlations discussed above , uv - selected samples of @xmath19 galaxies ( steidel et al . 2004 ) will tend to preferentially miss the most star - forming and most massive galaxies . still , because of the large scattering in the correlations , some of the latter galaxies will also be selected in the uv , as can be seen in fig . [ fig : bzk_cor ] and as emphasized in shapley et al . ( 2004 ; 2005 ) . in this subsection we derive the contribution of the to the integrated star formation rate density ( sfrd ) at @xmath19 and of and to the stellar mass density at @xmath19 . some fraction of the galaxies are known to be agn - dominated galaxies ( daddi et al . 2004b ; 2005b ; reddy et al . 2005 ) . to estimate the agn contamination , we have used the 80 ks xmm-@xmath222 x - ray data that are available for daddi - f ( brusa 2005 ) . a circular region of 11@xmath61 radius from the point of maximum exposure time include 70 18 of which are identified with x - ray sources using a 5@xmath223 radius error circle ( see fig . [ fig : bzk_cor ] ) . this fraction is comparable to the one estimated in the cdfs field ( daddi et al . 2004b ) and in the goods - n field ( daddi et al . 2005b ; reddy et al . 2005 ) , for which @xmath224 ms chandra data are available . based also on the latter result , we assume the agn contamination is about 25% , and we adopt this fraction to statistically correct properties derived from our samples . the left panel of fig . [ fig : sfrdmd ] shows the differential contribution to the sfr density at @xmath225 from as a function of their @xmath4-band magnitude . using the volume in the redshift range @xmath9 ( daddi et al . 2004a ; see also reddy et al . 2005 ) , an sfrd of @xmath226 @xmath227 yr@xmath47 mpc@xmath228 is derived from the ( @xmath229 ) in deep3a - f , and an sfrd of @xmath230 @xmath227 yr@xmath47 mpc@xmath228 is derived from the ( @xmath149 ) in daddi - f . these estimates are reduced , respectively , to @xmath231 @xmath227 yr@xmath47 mpc@xmath228 ( @xmath229 ) and 0.018 @xmath227 yr@xmath47 mpc@xmath228 ( @xmath149 ) when subtracting an estimated @xmath232 agn contamination . using the same method and the 24 in the k20/goods - s sample , daddi ( 2004a ) derived an sfrd @xmath233 @xmath227 yr@xmath47 mpc@xmath228 for the volume in the redshift range @xmath9 ; i.e. , @xmath234 lower than that derived here , possibly due to cosmic variance . however , note that there appears to be just a hint for the increasing trend in sfrd with @xmath4 magnitude to flatten out at @xmath15 , indicating that a substantial contribution to the total sfrd is likely to come from @xmath235 , and therefore the values derived here should be regarded as lower limits . recently , reddy et al . ( 2005 ) provided an estimate for the sfrd of @xmath84 of @xmath236 @xmath227 yr@xmath47 mpc@xmath228 , about one - third of the estimate that we have obtained here . part of the difference is due to the absence of the reddest from the reddy et al . ( 2005 ) sample , as already noticed . however , most of the difference is likely due to the fact that the reddy et al . ( 2005 ) sfr estimate is based primarily on the x - ray emission interpreted with the ranalli et al . ( 2003 ) relation . as shown by daddi et al . ( 2005b ) , the x - ray emission interpreted in this way typically underestimates the sfr of by factors of 23 , with respect to the radio- , mid - ir- and far - ir - based sfr estimates , all of which are also in reasonable agreement with the uv - corrected sfr estimate . the right panel of fig . [ fig : sfrdmd ] shows the differential contribution to the stellar mass density @xmath237 at @xmath225 from and as a function of their @xmath4-band magnitude . the open squares and triangles represent values that were calculated from all , the solid symbols represent values corrected for the agn contamination . the stars and crosses represent the mass density contributed by . the stellar mass density in deep3a - f , integrated to our @xmath84 catalog limit , is log@xmath238 @xmath227 mpc@xmath228 , in excellent agreement with the value reported in table 4 of fontana ( 2004 ) , i.e. , log@xmath239 @xmath227 mpc@xmath228 for @xmath240 galaxies , and log@xmath241 @xmath227 mpc@xmath228 for @xmath242 galaxies , but now from a much bigger sample . these estimates also agree with the log@xmath243 @xmath227 mpc@xmath228 estimate at @xmath30 by dickinson ( 2003 ) , although their selection is much deeper ( @xmath244 ) , although it extends over a much smaller field ( hdf ) . so , while our sample is likely to miss the contribution of low - mass galaxies , the dickinson sample is likely to underestimate the contribution of high - mass galaxies due to the small field and cosmic variance . there is little evidence for flattening of log@xmath245 by @xmath7 . as already noted , the total stellar mass density at @xmath19 has to be significantly larger than that estimated here , i.e. , only from the contribution of @xmath84 @xmath8-selected galaxies . there are 121 in deep3a - f , and for @xmath31 of them we derive @xmath246 @xmath227 . correspondingly , the number density of with @xmath246 @xmath227 over the range @xmath247 is @xmath248 mpc@xmath228 ( poisson error only ) . this compares to @xmath249 mpc@xmath228 over the same redshift range as estimated by daddi ( 2005a ) using six objects in the hubble ultra deep field ( hudf ) with spectroscopic redshift . while the daddi et al . ( 2005a ) hudf sample is important to establish that most are indeed passively evolving galaxies at @xmath250 , their density measurements is fairly uncertain due to cosmic variance . being derived from an area which is @xmath251 times larger than hudf , the results presented here for the number density of massive , passively evolving galaxies in deep3a - f in the quoted redshift range should be much less prone to cosmic variance . hence , we estimate that , compared to the local value at @xmath22 ( @xmath252 mpc@xmath228 , baldry 2004 ) , at @xmath178 there appears to be about 20%@xmath1387% of massive ( @xmath253 ) , passively evolving galaxies , with the error above accounting also for cosmic variance . this paper presents the results of a survey based on @xmath93 photometry obtained by combining subaru optical and eso near - ir data over two separate fields ( deep3a - f and daddi - f ) . complete k - selected samples of galaxies were selected to @xmath84 in the deep3a - f over 320 arcmin@xmath2 , and to @xmath118 in the daddi - f over a field roughly twice the area . deep multicolor photometry in the @xmath254 bands were obtained for the objects in both fields . object catalogs constructed from these deep data contain more than @xmath255 objects in the nir bandpasses . galaxy @xmath4-band number counts were derived and found to be in excellent agreement with previous survey results . we have used color criteria to select candidate massive galaxies at high redshift , such as @xmath8-selected star - forming ( ) and passively evolving ( ) galaxies at @xmath9 , and eros , and derived their number counts . the main results can be summarized as follows . \1 . down to the @xmath4-band limit of the survey the log of the number counts of increases linearly with the @xmath4 magnitude , while that of flattens out by @xmath256 . over the deep3a - f we select 387 and 121 down to @xmath7 , roughly a factor of 10 more than over the 52 arcmin@xmath2 fields of the k20 survey . this corresponds to a @xmath257 higher surface density , quite possibly the result of cosmic variance . over daddi - f we select 108 and 48 down to @xmath258 . the clustering properties ( angular two - point correlation function ) of eros and @xmath8-selected galaxies ( both and ) are very similar , and their clustering amplitudes are about a factor of 10 higher than those of generic galaxies in the same magnitude range . the most strongly clustered populations at each redshift are likely to be connected to each other in evolutionary terms , and therefore the strong clustering of eros and bzks makes quite plausible an evolutionary link between bzks at @xmath30 and eros at @xmath34 , with star formation in subsiding by @xmath34 thus producing passively evolving eros . while some may well experience secondary , stochastic starbursts at lower redshift , the global evolutionary trend of the galaxy population is dominated by star formation being progressively quenched in massive galaxies , with the quenching epoch of galaxies depending on environmental density , being earlier in high - density regions . \3 . using approximate relations from daddi et al . ( 2004a ) and multicolor photometry , we estimated the color excess , sfr and stellar mass of . these @xmath203 galaxies have median reddening @xmath18 , average sfr@xmath259 , and typical stellar masses @xmath260 . correlations between physical quantities are detected : the most massive galaxies are those with the largest sfrs and optical reddening @xmath182 . the high sfrs and masses of these galaxies add further support to the notion that these @xmath261 star - forming galaxies are among the precursors of @xmath262 passive eros and @xmath263 early - type galaxies . the contribution to the total star formation rate density at @xmath19 was estimated for the @xmath84 in our fields . these vigorous starbursts produce an sfrd @xmath264 @xmath227 yr@xmath47 mpc@xmath228 , which is already comparable to the global sfrd at @xmath30 as estimated from other surveys and simulations ( e.g. springel & hernqwist 2003 ; heavens et al . 2004 ) . however , a sizable additional contribution is expected from @xmath265 . \5 . in a similar fashion , the stellar mass of was obtained , with the result that the number density of @xmath203 more massive than @xmath207 is about 20%@xmath1387% of that of similarly massive , early - type galaxies at @xmath221 , indicating that additional activity and subsequent quenching of star - formation in @xmath266 star - forming galaxies must account for increasing the number of massive passive galaxies by a factor of about 5 from @xmath267 . the number density of @xmath266 @xmath268s is similar to that of @xmath269s . given their strong star - formation activity , it seems that by @xmath1081.4 the full population of local @xmath266 passive galaxies could be eventually assembled as a result . this result , advocated also in daddi et al . ( 2005b ) , may appear in contradiction with the recent finding by bell et al . ( 2004 ) of a factor of 2 decrease in the number density of early - type galaxies at @xmath108 , with respect to the local value ( see also faber et al . 2005 ) . however , our analysis of the bell et al ( 2004 ) results shows that most of this evolution is to be ascribed to the progressive disappearence with increasing redshift of the fainter galaxies , while the population of the brightest , most massive galaxies being substantially stable . this would be , in fact , another manifestation of the _ downsizing _ effect . a future publication will address this point in its full details and implications . mapping the metamorphosis of active star - forming galaxies into passively evolving , early - type galaxies from high to low redshifts , and as a function of galaxy mass and environment is one of the primary goals of the main ongoing galaxy surveys . using subaru and vlt telescopes , optical and near - infrared spectra are being obtained , with targets from the present database having been selected according to the same criteria adopted in this paper . future papers in this series will present further scientific results from this _ pilot _ survey , along with a variety of data products . we thank the anonymous referee for useful and constructive comments that resulted in a significant improvement of this paper . the work is partly supported by a grant - in - aid for scientific research ( 16540223 ) by the japanese ministry of education , culture , sports , science and technology and the chinese national science foundation ( 10573014 ) . x.k . gratefully acknowledges financial support from the jsps . acknowledge support from nasa through the spitzer fellowship program , under award 1268429 . thanks the poincar fellowship program at observatoire de la cte dazur and the danish natural science research council for financial support .

we study the distribution of galaxies and galaxy clusters in a field in the aquarius region . in addition to 63 clusters in the literature , we have found 39 new candidate clusters using a matched - filter technique and a counts - in - cells analysis . from redshift measurements of galaxies in the direction of these cluster candidates , we present new mean redshifts for 31 previously unobserved clusters , while improved mean redshifts are presented for 35 other systems . about 45% of the projected density enhancements are due to the superposition of clusters and/or groups of galaxies along the line of sight , but we could confirm for 72% of the cases that the candidates are real physical associations similar to the ones classified as rich galaxy clusters . on the other hand , the contamination due to galaxies not belonging to any concentration or located only in small groups along the line of sight is 10% . using a percolation radius of 10 mpc ( spatial density contrast of about 10 ) , we detect two superclusters of galaxies in aquarius , at 0.086 and at 0.112 , respectively with 5 and 14 clusters . the latter supercluster may represent a space overdensity of about 160 times the average cluster density as measured from the abell et al . ( 1989 ) cluster catalog , and is possibly connected to a 40 mpc filament from 0.11 to 0.14 .

astro-ph0111541

superclusters of galaxies are the largest known systems of galaxies , and are representative of the largest expected fluctuations in primordial spectrum . furthermore , since perturbations on supercluster scales are likely still in a linear growth regime , they have not reached a state of equilibrium and are probably in the phase of condensing out of the hubble flow , so they may have imprints of the processes that occurred during their formation @xcite . thus , their study may ultimately give some clues about the nature of density fluctuations and also of the formation and evolution of galaxies and clusters . the orientation of galaxies relative to other galaxies as well as the system within which they are embedded also produce estimates of the amount of dark matter inside larger volumes . by characterizing the observed supercluster properties it is possible to place constraints that must be satisfied by any successful theory that explains the formation and evolution of galaxies and the large scale structures they are embedded in . so far there have been two approaches used in the identification of superclusters . one searches for significant density enhancements in the spatial distribution of galaxies ( e.g. ; * ? ? ? * ; * ? ? ? * ) , which requires a relatively high redshift sampling rate , while the other uses clusters of galaxies ( or even of quasars ) to delineate these structures . in either case , because of the large scale sizes of superclusters , as well as the difficulty of defining supercluster membership , the characteristic properties of these systems are very uncertain , and thus somewhat inconclusive . for instance , it is not clear whether superclusters have already reached their maximum expansion phase and are now collapsing . superclusters masses are also highly uncertain , estimates ranging from @xmath4 to @xmath5 ( @xmath6 km s@xmath7 mpc@xmath7 ) ( e.g. ; * ? ? ? * ; * ? ? ? * ; * ? ? ? the derived estimates for mass - light ratios imply @xmath8 of about 0.2 to 0.4 of the critical value @xcite . in this paper we analyze the distribution of galaxy clusters in a region of @xmath9 located in the direction of aquarius constellation . this region contains two large concentrations of such systems identified by @xcite , named sc-16 and sc-17 , which are among the richest superclusters in that catalog . this concentration of galaxies has also been identified by more recent works using objective criteria applied to catalogs of clusters ( e.g. ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? in general , these works have used cluster catalogs for which the redshifts are either based on the 10@xmath10 brightest galaxy ( @xmath11 relation ) or on a very small number of spectroscopic measures per cluster . the first targeted study of clusters in aquarius region was carried out by @xcite , who , by using spectroscopic redshifts of the brightest cluster galaxies , claimed that sc-16 was a superposition of 22 clusters with @xmath12 . a more extensive study of the distribution of galaxies in this region was carried out by ( * ? ? ? * hereafter b99 ) , who obtained redshifts for a sample of clusters in a @xmath13 strip of sky , including aquarius and eridanus supercluster candidate regions . in the region of aquarius that we are considering in this work b99 measured about 200 redshifts around 11 clusters . using the cluster redshifts available at that time , b99 find a filamentary supercluster made up of 14 @xmath14 abell clusters ( where @xmath15 is the richness class , as defined by * ? ? ? * ) , with an estimated extension of about 110@xmath2 mpc , oriented almost along the line of sight . they also find a `` knot '' of 5 clusters at @xmath3 0.11 that represents an overdensity about 150 times greater than the mean spatial density for @xmath14 clusters . until now , effectively , only the richest abell clusters were used to study this region . our goal in this work is to study the distribution of galaxies in this region , by considering systems not only in rich clusters but also in lower density - contrast candidate systems , identified using objective criteria . in this paper we analyze the galaxy distribution in the region enclosed by the limits in right ascension of @xmath16 , and declination of @xmath17 , located in the aquarius constellation , containing sc-16 and sc-17 . the catalog is derived from apm scans of @xmath18-band films taken with the eso schmidt telescope , supplemented by @xmath19 data from the cosmos / ukst southern sky object catalog ( ssc , * ? ? ? * ; * ? ? ? * ) , which was obtained on - line from the anglo - australian observatory and the naval research laboratory / royal observatory of edinburgh . catalogs characteristics are summarized in table 1 . the @xmath18 magnitudes are defined from the combination of the 4415 tech - pan emulsion plus rg630 filter . prior to the photometric calibration , the plates were placed on a uniform instrumental system following @xcite , and then corrected to a common zero point . the uncertainty in the zero point determination is @xmath1 0.13 mag . the instrumental magnitudes were then calibrated with ccd data for a sequence of 18 galaxies , 5 measured at the 1.60 m telescope of observatrio do pico dos dias ( opd , braspolis , brazil ) and 13 from @xcite . the five galaxies measured at opd are listed on table 2 . the ( 2og570 + 3kg3 ) filters reproduce the cousins @xmath20 magnitude system @xcite . the @xcite galaxies are also in this system . the rms in the calibration relation is of about 0.1 mag , for the magnitude range @xmath21 , as can be seen in figure 1 . as an independent check on the photometric calibration , in figure 2 we plot the number counts in the aquarius region and those of @xcite , @xcite , @xcite and @xcite . the completeness of this catalog ( i.e. , unambiguous identification of objects as galaxies ) is estimated to be of 90% around @xmath22 and 80% around 19.0 . most of the misclassified galaxies at these limits are objects classified by the apm algorithm as `` merged '' . the detection limit is around @xmath23 , while the estimated optimal range of magnitudes for this data is @xmath24 . the figure suggests that there might be a lack of galaxies at magnitudes brighter than @xmath25 . this is partly due to the misclassification of brighter galaxies as merged objects , as well as to a low space density of nearby galaxies . on the other hand , for magnitudes fainter than this limit the counts are higher than the expected , what may indicate the presence of large scale structures . the @xmath19 magnitudes result from the combination of iiia - j emulsion with the gg395 filter used in the ukst survey . the calibration of galaxy magnitudes was done using existing ccd photometry in b and v @xcite . by comparing such magnitudes with eso imaging survey ( eis , * ? ? ? * ) data , @xcite find that they have a rms error of 0.2 mag in the range @xmath26 . the completeness level varies from about 90% at @xmath27 19.5 - 20.0 , to @xmath1 80% at 20.0 - 20.5 . so , we consider in this work only ssc galaxies brighter than @xmath28 , limiting the loss of galaxies at about 15% . the distribution of the galaxy @xmath19 magnitudes is shown in figure 3 . also plotted are the expected galaxy number counts as estimated for the edinburgh - durham southern galaxy catalogue ( edsgc , * ? ? ? * ) , covering the southern galactic cap ; for the eis , covering the southern galactic pole region ; for the northern ecliptic pole region ( nep , * ? ? ? * ) ; and for the sloan digital sky survey ( sdss ) commissioning data @xcite . the same conclusion taken from fig . 2 can be drawn from fig . 3 , since the excess in the distribution for galaxies fainter than @xmath29 is also present . we matched the @xmath19 galaxies with all objects in @xmath18 catalog using a search radius of @xmath30 , which is an optimal tolerance given the density of the catalogs , which simultaneously increased the match success and almost eliminated the possibility of double matches . astrometric uncertainties in both catalogs are smaller than @xmath31 , while systematic deviations between them are not larger than @xmath32 in @xmath33 and @xmath34 in @xmath35 . about 90% of the ssc galaxies had counterparts in the @xmath18 catalog . from these , about 70% were also classified as galaxies in @xmath18 catalog and 25% as merged . the distribution of color indices , @xmath36 , for aquarius galaxies is shown in figure 4 . given the large number of potential candidates for spectroscopic observations , we narrowed down the sample of galaxies to objects that are likely members of groups and clusters . this was done by using catalogs of clusters presented in the literature , as well as by applying 2 different algorithms to the projected distribution of galaxies , in order to detect slightly lower density enhancements from which potential poor clusters and groups of galaxies might be identified . for each identified aggregate of galaxies ( cluster / group ) we have searched for surface density peaks inside the estimated abell radius . from a @xmath37 field centered on each peak , we selected about the 15 brightest galaxies as spectroscopic targets . the aquarius region contains 48 clusters originally identified by @xcite , comprising systems from all richness ( @xmath15 ) and distance classes . from ( * aco ) 7 additional rich southern clusters and 3 supplementary ones are found . two other catalogs , both machine - based , have clusters in the aquarius region : the edinburgh - durham cluster catalog ( edcc , * ? ? ? * ) and the automatic plate measuring machine cluster catalog ( apmcc , * ? ? ? * ) . the former covers only about half of the aquarius region ( 35 @xmath38 ) , with 18 catalogued clusters , most of them ( 15 ) corresponding to abell / aco clusters . the apmcc , on the other hand , fully covers the aquarius region , where 17 clusters are found , 15 of which are also in the abell / aco catalogs . although x - rays are one of the most efficient means of detecting rich clusters of galaxies , there are still few cluster candidates in the aquarius region with confirmed x - ray emission . x - rays have been detected in 13 abell / aco clusters by the _ rosat _ all - sky survey @xcite , the heao-1 @xcite and the einstein observatory ( heao-2 ) @xcite . in general , these x - ray detections may be considered as corroborative to the reality of such galaxy clusters . in order to identify new significant galaxy density enhancements , we applied the matched - filter technique described by @xcite , to the galaxy catalogs discussed in section 2 . the matched - filter is a maximum likelihood - based method that objectively identifies two - dimensional density enhancements by considering projected positions and apparent magnitudes . basically it uses a filter which suppresses galaxy fluctuations that are not due to galaxy clusters . as discussed by @xcite , this method is optimized to detect weak signals in a noise - dominated background and has a good dynamic range , besides being able to suppress false detections . the price that is payed for this is that one must assume a parametric form both for the cluster luminosity function and its radial profile . in the case of the galaxy distribution in aquarius , the filter assumes a spherically symmetric @xcite model with core ( @xmath39 ) and tidal ( @xmath40 ) radii such that @xmath41 @xcite . the schechter function parameters were fixed as follows : @xmath42 , @xmath43 , and @xmath44 , these values being typical of poor clusters of galaxies @xcite . k_-corrections , the fitting formulae for e / s0 galaxies defined by @xcite were used . other parameters that are considered by the matched - filter are the cluster redshift , @xmath45 , and its richness , @xmath46 . the @xmath46 is defined as the number of member galaxies with magnitudes brighter than ( @xmath47 ) and within central 1.5@xmath2 mpc . in the present analyses , all parameters except @xmath45 , @xmath46 , and @xmath39 are basically fixed . when computing the likelihood for the aquarius region , the model of the spatial and luminosity distributions of the filter were compared to the actual galaxy distribution considering only those galaxies within a circular region with @xmath48 radius and in the magnitude range of @xmath49 and @xmath50 for the @xmath19 and @xmath18 data , respectively . in the first step of the procedure , we fix ( @xmath51 ) at ( 0.2 , 50@xmath2 kpc ) and tune only @xmath46 in order to maximize the likelihood at each given point and to simplify the calculation . the likelihood and corresponding @xmath46 were computed at all lattice points separated by 0.02@xmath52 to make a `` likelihood map '' and a `` richness map '' . because of simpler appearance of clusters in the richness map , we use the latter to detect clusters ( see fig . 2 of * ? ? ? next , we smooth the raw richness map with a gaussian filter with @xmath53 ( figures 5_a _ and 5_b _ ) . cluster candidates are then detected as local peaks with @xmath54 ( @xmath55 ) in the smoothed richness map . then @xmath45 and @xmath39 were surveyed in the range of @xmath56 and @xmath57 for @xmath19 data and @xmath58 and @xmath59 for @xmath18 data , respectively , to estimate the redshift and richness for each candidate . an abell - like richness , @xmath60 , was calculated from @xmath46 , using the relationship between them obtained from monte - carlo simulations . uncertainties in these estimated quantities , also obtained from monte - carlo simulations , are of 0.03 in redshifts and 20% for both @xmath46 and @xmath60 . a total of 57 cluster candidates were identified in the @xmath19 band galaxy catalog . two of these were discarded , having resulted from obvious contamination caused by bright stars . of the remaining 55 cluster candidates , 18 are new identifications , though 1 is outside the region considered in this work . the richness map of this sample is shown in figure 5_a_. selected peaks , those with @xmath54 , are marked in the figure with pluses . since for the @xmath19 band galaxy catalog the optimal magnitude range covers @xmath61 , redshifts of clusters at @xmath62 should have the best estimation . the @xmath18 band galaxy catalog shows 44 detections , of which 26 are in common with the @xmath19 band matched - filter catalog . of these , only one has not been previously detected by the abell , aco , edcc and apmcc catalogs . the number of new detections in the @xmath18 catalog is of 9 clusters . figure 5_b _ shows the richness map of this catalog , where again peaks above @xmath54 are marked with pluses . although the matched - filter does provide an estimate of the cluster redshift , given the somewhat low-@xmath63 of the objects in aquarius , the measured uncertainty ( @xmath64 ) is too large to provide a meaningful estimate of the cluster distance . as the catalogs and methods presented above were designed to search mainly for rich clusters of galaxies , we tried an additional technique to identify smaller potential aggregates of galaxies , such as poor clusters and groups . the idea is that these structures may trace lower density contrast structures such as filaments and walls , as observed in the great wall @xcite . the main limitation of this procedure is that only positional information is used , so that the rate of false detections due to foreground / background contamination is larger than with the matched - filter technique . in order to minimize the contamination from interlopers and spurious detections , we optimized the search so that the cells would cover @xmath65 with a step of @xmath66 , equivalent to a resolution of 0.2@xmath2 mpc at the distance in which we expect to find the most representative supercluster ( @xmath3 0.11 ) . in addition to the cell size , we considered three samples : @xmath19 and @xmath18 data and a sample that contain only galaxies redder than ( @xmath67 ) , typically redder than _ sab _ @xcite , taking advantage of the fact that the morphology - density relation is also statistically valid for groups of galaxies @xcite and that early - type galaxies are generally red . such an approach would preferentially detect objects located in the centers of clusters and rich groups ( e.g. ; * ? ? ? the efficiency of the counts - in - cells approach was estimated by running the algorithm on the updated zwicky catalog ( uzc , * ? ? ? * ) , where the cells were optimized for a distance similar to that of the great wall . by comparing the results from the counts - in - cells analysis against the catalog of groups identified in the uzc by @xcite , we find that the overall detection rate is 80% for all uzc groups ( @xmath68 ) and greater than 90% for groups located at the distance of the great wall ( @xmath69 ) . the result from this analysis for aquarius is shown in figure 6 where surface density isocontour maps for @xmath70 , @xmath71 and @xmath72 , are presented in panels _ ( a ) _ , _ ( b ) _ and _ ( c ) _ , respectively . from the three maps we selected 38 peaks presenting projected densities higher than 3 times the standard deviation of background density ( @xmath73 ) in the three plots ( hereafter sample sd-1 ) . besides these conspicuous overdensities , other 27 additional clumps , presenting a density contrast greater than 3@xmath73 in two of the maps ( hereafter sample sd-2 ) , were also identified . by combining all the detections described above , we are now able to construct a catalog of galaxy aggregates , potential clusters and groups in the region we are studying . this is presented in table 3 where we show in column ( 1 ) the identifier in our catalog ( aquarius cluster catalog - aqrcc ) and , in columns ( 2 ) and ( 3 ) , the j2000.0 coordinates from the first catalog that the cluster was identified . in column ( 4 ) we list the abell identification and in column ( 5 ) the corrected number of galaxies , @xmath74 , from the aco catalog . the edcc numbers are listed in column ( 6 ) , while column ( 7 ) contains the corrected number of galaxies according to @xcite , @xmath75 . the apmcc identification is presented in column ( 8) , followed in column ( 9 ) by the cluster richness , @xmath76 , and the estimated redshift , @xmath77 , in column ( 10 ) . the matched - filter richness and estimated redshift are presented in columns ( 11 ) and ( 12 ) , ( 13 ) and ( 14 ) , for the @xmath19 and @xmath18 data , respectively . column ( 15 ) notes clusters that have been identified as x - ray sources ; and column ( 16 ) notes objects identified by the counts - in - cells analysis . there are 102 cluster or group candidates identified in table 3 , of which 39 are new detections . a map showing the projected distribution of objects in table 3 is presented in figure 7 , where it can be seen that most objects are contained in at least two catalogs of objects . a comparison of the percentual overlap between different catalogs is shown in table 4 . the aco and sd-2 catalogs detect , on average , about 75% and 85% of objects in the other catalogs , respectively . the sd-1 , mf - b and mf - r catalogs , on the other hand , detect about 60 - 65% , while the edcc and apmcc detect on average , respectively , 50% and 25% of systems in the other catalogues . redshift observations now exist for 72 of the 102 aggregates in table 3 . cluster redshifts , both from the literature and from our new measurements , are presented in columns ( 17 ) to ( 22 ) . column ( 17 ) lists the mean cluster redshift from ned , followed by the number of redshifts used to estimate that value in column ( 18 ) , and reference ( 19 ) ; new cluster redshifts obtained in this work are shown in column ( 20 ) , while column ( 21 ) presents the number of cluster galaxies with redshifts ; and finally , column ( 22 ) indicates notes to individual clusters . the detailed parameters obtained for each observed cluster , like velocity dispersions , virial masses , richnesses , core radii , luminosities , etc . , will be subject of a forthcoming paper . from the 72 observed cluster / group candidates , 69 resulted in a positive identification in redshift space . of these , 31 are new cluster redshift measurements , while of the 38 with previous redshifts from the literature , 35 had additional galaxies observed in the present survey . for 12 of these clusters , the inclusion of new observations gives a mean redshift that differs from the previous published values by more than 2000 km s@xmath7 . in the case of 9 of these clusters , the published redshift was based on a single galaxy , which is likely to be an interloper . in two of these cases , the observed overdensity is a superposition of two systems and , since we chose the richer of them as representing the cluster , the previous redshift is of a galaxy belonging to the poorer system . about half of the observed aggregates show a single significant peak in the distribution of observed galaxy redshifts up to @xmath3 0.2 . by a significant peak we mean that there are no gaps larger than 1500 km s@xmath7 amongst the member galaxy velocities . for 32 aggregates we find more than one significant peak in redshift space . this high fraction of superpositions is not unexpected in a direction that possibly intercepts more than one supercluster . there are 3 cases in which we failed to detect a redshift peak . therefore , we identified 109 significant peaks in the redshift distributions of 69 candidates . these concentrations in redshift space , that we call generically `` galaxy systems '' , have richness classes that vary from rich clusters to small groups . we estimate the actual fraction of overdensities in the projected distribution of galaxies that result from chance alignments , by adding the fraction of detection failures in redshift space and the fraction of superposition of only small groups of galaxies . for that , we separated the 72 observed candidates according to the fraction of observed galaxies that turned out to be system members and visual inspection of system images . for the 51% of the cases that a single system was found , the fraction of system members to observed galaxies ( @xmath78 ) is @xmath790.55 ( mean 0.8 ) , i.e. , most of the observed galaxies were converted to system members . the mean value for the velocity dispersion of these systems is 789@xmath80319 , typical value for rich galaxy clusters @xcite . so , we considered such systems as rich galaxy clusters . the other 45% that turned out to be more than one galaxy system in redshift space , were separated in three categories . the first is composed by clusters with superimposed ( or background ) groups , for which the main clump has @xmath81 0.4 and the other clumps have @xmath82 0.3 . the richness of the main system is slightly contaminated , but we also consider it as a rich cluster , since the mean velocity dispersion is also close to what is expected for such systems , 767@xmath80411 . these correspond to 21% of the aggregates with redshift , pushing the fraction of probable clusters to 72% . the other two categories are cases of significant superposition ( 24% ) . although two or more galaxy systems may exist in these directions , the available information does not allow a great deal of confidence in classifying them as poor clusters or groups . instead , we separated them in cases of two significant systems , when both have @xmath81 0.3 ( 11% of cases ) , or superimposed smaller groups , when all have @xmath83 0.3 ( 13% ) . thus , we estimate the contamination in the aqrcc could range between 4% , from the cases in which we failed in detecting a system , to 17% , adding the cases of only superimposed smaller groups . this is a conservative range , since some of the systems classified as groups may turn to be undersampled rich clusters at @xmath3 0.2 , as their images suggest . this is the case of aqrcc_002 ( a2509 ) , aqrcc_018 ( a2536 ) , aqrcc_033 ( a2550 ) and aqrcc_080 ( a2604 ) . for one of these , a2536 , we have deeper photometry that confirms its higher cluster richness , while a2550 has confirmed x - ray emission . thus , removing these deeper clusters from the maximum contamination , the fraction of chance alignments becomes 10% . the same analysis can be done for each of the original catalogs , which is shown in table 5 . previous estimates of contamination for abell / aco , edcc and apmcc are , respectively , 10 - 15% , 8 - 13% and 3 - 5% ( e.g. ; * ? ? ? we find for abell 2 - 15% and for aco 3 - 15% , in accordance to the above estimates . for edcc and apmcc the small number of objects in aquarius does not allow making an evaluation of the contamination . since we have new mean redshifts for most of the clusters in the abell , edcc and apmcc catalogs , we can also make a rough estimate of the depth of each of these cluster catalogs by using the mean redshift of the most distant quartile of clusters . for aco we have spectroscopic redshifts for 98% ( 57 ) of the clusters and we confirm the expected depth of this catalog to be about @xmath3 0.2 ; while for edcc and apmcc we find @xmath3 0.19 and @xmath3 0.13 , respectively , for the average of the 25% more distant clusters , both 100% complete on redshift in aquarius . the observed overdensity of abell clusters in the region may be taken as preliminary evidence that there might be superclusters of galaxies in this part of the sky . the projected density of abell and aco clusters in aquarius ( excluding the supplementary catalog objects ) is 0.83 clusters/@xmath38 , which is an overdensity of about 4 when compared to a slice of 15@xmath52 at the same galactic latitude . similarly , for the edcc the mean surface density of the overall catalog is 0.45 clusters/@xmath38 , while in aquarius this increases to 0.54 clusters/@xmath38 . for the apm clusters the projected densities are 0.22 clusters/@xmath38 and is 0.29 clusters/@xmath38 , respectively . a redshift cone diagram showing the 109 systems with available redshifts , is presented in figure 8 , with different symbols for the four categories described on the previous section . concentrations of clusters at @xmath3 0.08 and @xmath3 0.11 are easily seen . it is also noticeable that most of the identified groups follow the structures defined by the clusters ( in fact , 55% of them seem to be part of concentrations at 0.08 and 0.11 ) . details of figure 8 are presented in figures 9 , 10 and 11 , which also show the cluster names . to test whether the distribution of clusters seen in figure 8 could be forming larger systems , we applied a percolation analysis to the data . the results of the percolation are presented in table 6 , in two blocks : the first ( top ) where only the 56 rich clusters are considered , and the second ( bottom ) where all systems are considered . the first column of table 6 shows the search radius in mpc , while column 2 shows the corresponding space overdensity . the mean spatial density is that estimated from the aco catalog ( e.g. ; * ? ? ? * ; * ? ? ? * ; * ? ? ? @xmath84 mpc@xmath85 corrected for the number of aco clusters in the sample over their total number in the region ( 58 ) . even though few rich clusters are lost in the identification process , it is likely that the number of less dense systems is underestimated . as can be seen in table 6 , with a small percolation radius ( @xmath86 ) , corresponding to an overdensity of about @xmath87 , some supercluster core seeds emerge at about @xmath3 0.08 , 0.11 and 0.15 . as the @xmath86 is increased , the supercluster cores grow fast , particularly when the small clusters and groups are considered . the concentration at @xmath3 0.086 has two cores , that coalesce at @xmath86= 10@xmath2 mpc with the groups ( only at 15@xmath2 mpc without them ) , forming a supercluster of at least 5 clusters and 12 groups . the largest structure , on the other hand , has a main concentration at @xmath3 0.11 and may extend up to @xmath3 0.14 . at @xmath3 0.112 there are 3 cores that coalesce at @xmath88 10@xmath2 mpc , with 14 clusters and at least 10 groups . from @xmath3 0.11 to 0.14 there is a filament of about 40@xmath2 mpc , with at least 7 clusters and 8 groups , connecting with the concentration at @xmath3 0.11 with @xmath88 15@xmath2 mpc . besides these two large superclusters , an additional 4 potential concentrations appear beyond @xmath3 0.14 , respectively at 0.147 , 0.171 , 0.201 and 0.212 . these concentrations might be severely undersampled in our survey , as most of their brighter galaxies are are close to the photometric limits of our survey . consequently we have , for example , a small number of redshifts per cluster ( only 7 on average for @xmath89 0.14 clusters against 14 for @xmath90 0.14 ) and a very small number of groups detected ( only 8 of 35 have @xmath89 0.14 ) . because the present work probes more densely the galaxy distribution in aquarius , some of our results differ from those of b99 . the main discrepancy is for 4 clusters that have a mean redshift which differ significantly from the values of b99 . all of them are cases for which b99 observed galaxies in a larger area around the aco cluster position and almost all the galaxies they used to calculate the mean redshifts are outside the estimated abell radius , being generally galaxies belonging to the superclusters dispersed component superimposed to the observed cluster . in contrast , all the galaxies we observed are inside the abell radius and have a greater chance of being a representative sample to measure the mean cluster redshift . in one specific case , aqrcc_058 ( a3996 ) , we do not find a cluster around aco position , but only the superposition of the dispersed component of both the 0.08 and 0.11 superclusters . in the case of four clusters for which b99 used data from the literature ( all of them based on a single galaxy redshift ) we also have new mean redshift measurements . however , two are cases of superposition of two systems ( poor clusters or groups ) where we choose the richer to quote the redshift but the first observation was for a galaxy of the other system . in consequence of these redshift re - evaluations two clusters were removed from the b99 `` knot '' , aqrcc_058 and aqrcc_036 ( a2553 ) , though we now find 14 clusters for this 0.11 concentration ( their 4 plus 10 new ones , including 6 aco clusters , 2 apmcc ones , 1 edcc and 1 new detection ) . another thing that can be noted , both from figure 8 and table 6 , is that the aquarius supercluster of b99 was split in two distinct structures . for b99 the clusters from @xmath3 0.08 to 0.12 formed a supercluster at a @xmath91 8 . nevertheless , one of the clusters that made the link for this filament , a2541 ( aqrcc_025 ) , is one of the cases of re - evaluation which had only one observed galaxy previously , and has since shown to be more distant it is , in fact , part of the core of 0.11 supercluster . thus , even when considering a much smaller density contrast ( @xmath92 3 ) or using groups , we could not connect both superclusters . moreover , the 30 candidates of aqrcc that are yet unobserved have small estimated richnesses and also estimated redshifts that could not turn them to be potential links between the superclusters . considering the spatial distribution of the 5 clusters in the @xmath3 0.086 supercluster , we are able to estimate a space overdensity of about 130 times the considered mean density , for an equivalent volume of 6@xmath9319@xmath9310 @xmath94 mpc@xmath95 . similarly , for the 14 clusters in the @xmath3 0.112 supercluster , we find a spatial overdensity of 100@xmath96 or , if we remove the cluster aqrcc_063 ( a2583 ) from the supercluster , which is slightly far from the main concentration , we finally find a space overdensity of 160@xmath96 for a volume of 12@xmath9329@xmath9318 @xmath94 mpc@xmath95 . in terms of mass , if we consider the mean velocity dispersion we obtained for rich clusters , a mean mass for groups ( e.g. ; * ? ? * ) , and a fraction of about one third for the dispersed component ( e.g. ; * ? ? ? * ) , we can estimate for the @xmath3 0.086 and 0.112 superclusters the masses of at least @xmath97 and @xmath98 , respectively . in this work we combine publicly available as well as new data to study the distribution of galaxies in a @xmath9 area in the aquarius constellation . \(i ) a compilation of galaxy clusters from the literature reveals the presence of 63 clusters in the region , corresponding to a projected density of 0.95 clusters@xmath99 . if only rich aco clusters are considered , this represents an overdensity of more than 4 in the surface density of such systems . \(iii ) we present new redshifts for 31 clusters in the aquarius region , and provide more robust estimates for 23 systems confirming previous redshifts from the literature . we also report 12 new measurements , typically from 7 galaxies observed per cluster , that differ by more than 2000 km s@xmath7 from previous measurements . the completeness of aquarius cluster catalog in redshifts is 70% . \(iv ) from the 72 observed cluster candidates , 51% revealed a single significant peak in redshift space , while 45% showed more than one and 4% did not show any . moreover , 72% of them were found to be probable real clusters , while 24% may possibly be poor clusters or groups . \(vi ) we find 2 rich superclusters in aquarius region , at @xmath3 0.086 and @xmath3 0.112 , respectively with 5 and 14 clusters at a spatial number density contrast of about 10 ( @xmath88 10@xmath2 mpc ) . for both of them , a number of smaller galaxy systems ( at least 10 for each ) , possibly poor clusters or groups , were also found to be part of the superclusters at the same percolation radius . the @xmath3 0.11 supercluster may even be connected to a 40@xmath2 mpc filament of at least 7 clusters from 0.11 and 0.14 . we also find possible cluster concentrations at 0.15 , 0.17 , 0.20 and 0.21 , that need deeper photometry and spectroscopy to be confirmed . \(vii ) with the cluster redshifts available in this region , we estimate that the characteristic depth of the most distant quartile of the abell / aco , edcc and apmcc catalogs are respectively 0.20 , 0.19 and 0.13 . the redshift completenesses in aquarius for these catalogs are 98% for abell / aco and 100% for the others . \(viii ) our analysis presents some differences relative to b99 . the re - evaluation of some cluster redshifts , used by b99 to detect the supercluster , revealed 2 significant superclusters in this part of the sky , rather than only one structure that extends along the line of sight , as they suggested . their conclusion was probably based on an incorrect redshift for the cluster a2541 , which led them to connect both structures . \(ix ) our data support the interpretation of b99 that the @xmath100 0.11 supercluster in aquarius is very likely a significant concentration of matter , representing an overdensity of @xmath101 160 , and a mass of at least @xmath98 . we are grateful to the staff and night assistants of opd / lna 1.6 m , eso 1.52 m and casleo 2.15 m ; the aao and nrl / roe for providing the ssc ; g. pizzaro for taking our eso plates , and m. irwin for providing the apm digitization . this research has also made use of nasa / ipac extragalactic database ( ned ) . the authors acknowledge use of the ccd and data acquisition system supported under u.s . national science foundation grant ast 90 - 15827 to r.m . rich . c.a.c . acknowledges financial support from capes scholarship , m.a.g.m . to cnpq grant 301366/86 - 1 , w.k . to financial support from japan society for the promotion of science ( jsps ) research fellowship , and c.n.a.w . to cnpq grant 301364/86 - 9 , nsf ast 95 - 29028 and nsf ast 00 - 71198 . prandoni , i. , wichmann , r. , da costa , l. , benoist , c. , mndez , r. , nonino , m. , olsen , l. , wicenec , a. , zaggia , s. , bertin , e. , deul , e. , erben , t. , guarnieri , m. d. , hook . i. , hook , r. , scodeggio , m. , slijkhuis , r. 1999 . 345 , 448 yentis , d. j. , cruddace , r. g. , gursky , h. , stuart , b. v. , wallin , j. f. , macgillivray , h. t. , collins , c. a. 1992 . in _ digitised optical sky surveys , _ eds . h. t. macgillivray & e. b. thomson ( kluwer : dordrecht ) , p. 67 cccccccccr 1 & & 23 07 41.46 & -22 42 39.0 & & 16.27 & 18.20 & 1.96 & 0.1998 & 1,2 + 2 & & 23 07 37.93 & -22 43 05.2 & & 16.81 & 18.12 & 1.18 & 0.1916 & 1,3 + 3 & & 23 07 42.49 & -22 43 49.9 & & 16.97 & 18.86 & 1.81 & 0.1698 & 3 + 4 & & 23 07 37.78 & -22 42 15.3 & & 17.92 & 19.64 & 1.69 & 0.1999 & 4 + 5 & & 23 07 41.29 & -22 44 05.9 & & 18.06 & 19.89 & 1.94 & 0.2009 & 4 + lcccccrccrccrrcrrcrrcccccrrrcrrc 001 & & 22 57 52.6 & -24 55 19 & & & & & & & & & & & & 31 & 0.080 & & & & & & & o & & & & & & & & + 002 & & 22 57 53.1 & -21 43 56 & & a2509 & 70(1 ) & & o & & & & & & & & & & & & & & & o & & 0.2306 & 1 & 3 & & 0.2305 & 3 & + 003 & & 22 58 00.7 & -20 50 56 & & & & & o & & & & & & & 32 & 0.090 & & & & & & & o & & & & & & & & + 004 & & 22 58 46.4 & -19 58 55 & & a3949 & 43(0 ) & & o & & & & & & & 61 & 0.114 & & & & & & & o & & & & & & 0.1580 & 6 & + 005 & & 22 59 27.6 & -23 49 01 & & & & & & & & & & & & 107 & 0.177 & & & & & & & b & & & & & & 0.1177 & 11 & + 006 & & 23 00 03.6 & -25 35 56 & & & & & & & & & & & & 55 & 0.100 & & & & & & & o & & & & & & & & + 007 & & 23 00 17.3 & -23 11 53 & & a2514 & 64(1 ) & & & & & & & & & & & & 64 & 0.216 & & & & r & & & & & & 0.0000 & & 1 + 008 & & 23 00 31.9 & -22 26 49 & & & & & & & & & & & & & & & 105 & 0.252 & & & & r & & & & & & & & + 009 & & 23 00 47.6 & -24 09 52 & & a2518 & 78(1 ) & & 231 & 21 & & & & & & 57 & 0.083 & & 80 & 0.207 & & & & b , r , c & & 0.1351 & 1 & 3 & & 0.1342 & 7 & 2 + 010 & & 23 02 12.7 & -22 01 12 & & a2521 & 103(2 ) & & o & & & 845 & 77 & 0.117 & & 90 & 0.121 & & 50 & 0.109 & & 2,5 & & b , r , c & & 0.1340 & 2 & 5 & & 0.1364 & 17 & + 011 & & 23 04 04.3 & -24 41 20 & & & & & & & & & & & & 58 & 0.133 & & & & & & & & & & & & & & & + 012 & & 23 04 05.2 & -24 01 49 & & a2526 & 53(1 ) & & 242 & 73 & & & & & & 55 & 0.157 & & 55 & 0.201 & & & & r , c & & & & & & 0.2043 & 6 & + 013 & & 23 04 21.8 & -19 54 48 & & a3964 & 42(0 ) & & o & & & & & & & 44 & 0.100 & & & & & & & o & & & & & & 0.1325 & 4 & 3 + 014 & & 23 05 11.5 & -25 18 48 & & a2527 & 68(1 ) & & 244 & 34 & & & & & & & & & & & & & & & & & & & & 0.1630 & 7 & + 015 & & 23 05 40.1 & -21 23 47 & & a2528 & 39(0 ) & & o & & & 852 & 66 & 0.099 & & 66 & 0.077 & & 49 & 0.134 & & & & b , r , c & & 0.0955 & 1 & 3 & & 0.0949 & 12 & + 016 & & 23 06 58.0 & -21 40 46 & & a2531 & 73(1 ) & & o & & & & & & & & & & 53 & 0.116 & & & & b , r , c & & 0.1741 & 1 & 3 & & 0.1731 & 7 & + 017 & & 23 07 34.3 & -22 39 45 & & a2534 & 110(2 ) & & 253 & 73 & & & & & & 90 & 0.149 & & 74 & 0.169 & & 5 & & b , r , c & & 0.1976 & 3 & 1,3 & & 0.2014 & 16 & + 018 & & 23 07 46.2 & -22 25 45 & & a2536 & 102(2 ) & & & & & & & & & & & & & & & 5 & & b , r , c & & 0.1971 & 1 & 3 & & 0.1973 & 4 & + 019 & & 23 08 27.6 & -25 46 19 & & & & & & & & & & & & 69 & 0.121 & & & & & & & o & & & & & & & & + 020 & & 23 08 35.1 & -19 52 29 & & a2538 & 72(1 ) & & o & & & 859 & 88 & 0.086 & & 93 & 0.079 & & & & & & & o & & 0.0831 & 42 & 4 & & 0.0829 & 44 & + 021 & & 23 08 45.8 & -21 28 44 & & a2539 & 66(1 ) & & o & & & & & & & & & & & & & & & b , r , c & & 0.1735 & 1 & 3 & & 0.1863 & 4 & 4 + 022 & & 23 08 53.0 & -20 47 00 & & & & & & & & & & & & & & & & & & & & b , c & & & & & & 0.0825 & 6 & 5 + 023 & & 23 09 27.9 & -22 09 43 & & a2540 & 70(1 ) & & & & & & & & & & & & 61 & 0.195 & & 1 & & r , c & & 0.1297 & 1 & 3 & & 0.1290 & 8 & + 024 & & 23 09 54.0 & -21 30 00 & & & & & & & & & & & & & & & & & & & & b , c & & & & & & 0.1109 & 7 & 6 + 025 & & 23 10 04.1 & -22 57 43 & & a2541 & 83(2 ) & & 256 & 51 & & & & & & 51 & 0.117 & & 58 & 0.159 & & & & b , r , c & & 0.1100 & 2 & 9 & & 0.1135 & 16 & 7 + 026 & & 23 10 04.5 & -24 29 43 & & a2542 & 57(1 ) & & & & & & & & & 71 & 0.167 & & 59 & 0.166 & & & & b , r , c & & 0.1603 & 1 & 3 & & 0.1684 & 4 & + 027 & & 23 10 27.6 & -24 47 27 & & & & & & & & & & & & & & & 51 & 0.157 & & & & b , c & & & & & & 0.1998 & 5 & 8 + 028 & & 23 10 45.9 & -22 39 42 & & a2546 & 90(2 ) & & 258 & 41 & & 862 & 63 & 0.108 & & 86 & 0.121 & & & & & & & b , r , c & & 0.1119 & 1 & 3 & & 0.1130 & 22 & 7 + 029 & & 23 10 51.4 & -21 07 42 & & a2547 & 84(2 ) & & o & & & & & & & 82 & 0.129 & & & & & & & b , r , c & & 0.1501 & 2 & 3,12 & & 0.1517 & 15 & + 030 & & 23 11 15.0 & -20 47 03 & & & & & o & & & & & & & 81 & 0.155 & & & & & & & b , r , c & & & & & & 0.1076 & 9 & 9 + 031 & & 23 11 21.2 & -20 25 41 & & a2548 & 65(1 ) & & o & & & & & & & & & & 23 & 0.156 & & & & b , r , c & & 0.1101 & 1 & 3 & & 0.1107 & 9 & + 032 & & 23 11 24.0 & -21 30 00 & & & & & & & & & & & & & & & & & & & & b , r , c & & & & & & & & + 033 & & 23 11 33.5 & -21 44 41 & & a2550 & 122(2 ) & & o & & & & & & & & & & & & & 1 & & b , r , c & & 0.1543 & 1 & 2 & & 0.1226 & 6 & + 034 & & 23 12 10.0 & -22 50 00 & & & & & & & & & & & & & & & & & & & & b , c & & & & & & & & + 035 & & 23 12 15.1 & -21 33 56 & & a2554 & 159(3 ) & & o & & & 866 & 112 & 0.102 & & 77 & 0.083 & & 64 & 0.100 & & 1 & & b , r , c & & 0.1111 & 28 & 4 & & 0.1108 & 35 & + 036 & & 23 12 22.4 & -24 56 40 & & a2553 & 76(1 ) & & 264 & 56 & & & & & & 100 & 0.141 & & 70 & 0.167 & & & & b , r , c & & 0.1496 & 2 & 10 & & 0.1481 & 4 & + 037 & & 23 12 45.5 & -22 12 40 & & a2555 & 72(1 ) & & & & & & & & & 53 & 0.091 & & 42 & 0.101 & & 1 & & b , r , c & & 0.1385 & 1 & 3 & & 0.1106 & 11 & + 038 & & 23 13 03.3 & -21 37 40 & & a2556 & 67(1 ) & & o & & & & & & & & & & & & & 1,4,5,6 & & b , c & & 0.0865 & 2 & 2,3 & & 0.0871 & 9 & + 039 & & 23 13 15.7 & -23 08 40 & & s1099 & 17(0 ) & & & & & & & & & & & & & & & & & b & & & & & & 0.1104 & 12 & + 040 & & 23 14 37.0 & -23 26 02 & & & & & & & & & & & & & & & 49 & 0.210 & & & & b , r & & & & & & 0.0910 & 6 & 10 + 041 & & 23 15 50.9 & -21 06 37 & & a2565 & 4 & & o & & & & & & & 46 & 0.089 & & 65 & 0.199 & & & & b , r , c & & 0.1271 & 1 & 3 & & 0.0825 & 12 & 11 + 042 & & 23 15 57.5 & -23 19 37 & & a3985 & 36(0 ) & & 270 & 21 & & 874 & 53 & 0.080 & & 52 & 0.068 & & 60 & 0.157 & & & & b , r , c & & & & & & 0.1094 & 17 & + 043 & & 23 16 00.0 & -20 51 00 & & & & & & & & & & & & & & & & & & & & b , c & & & & & & & & + 044 & & 23 16 03.2 & -20 27 41 & & a2566 & 51(1 ) & & o & & & 872 & 42 & 0.086 & & 39 & 0.066 & & & & & 6 & & b , c & & 0.0821 & 1 & 3 & & 0.0822 & 11 & + 045 & & 23 16 33.6 & -24 58 37 & & & & & & & & & & & & 52 & 0.110 & & & & & & & b & & & & & & & & + 046 & & 23 16 57.0 & -22 11 36 & & a2568 & 35(0 ) & & & & & 876 & 45 & 0.112 & & & & & & & & & & b , c & & 0.1398 & 1 & 3 & & 0.1397 & 6 & + 047 & & 23 17 32.3 & -25 20 45 & & & & & 275 & 8 & & & & & & 37 & 0.080 & & & & & & & b & & & & & & 0.1453 & 6 & + 048 & & 23 17 35.2 & -22 37 05 & & & & & & & & & & & & & & & & & & & & b , r & & & & & & 0.0827 & 5 & 12 + 049 & & 23 19 44.6 & -19 30 04 & & & & & o & & & & & & & 83 & 0.131 & & & & & & & o & & & & & & & & + 050 & & 23 19 50.7 & -22 04 34 & & a2575 & 80(2 ) & & & & & & & & & & & & & & & & & r & & & & & & & & + 051 & & 23 19 56.8 & -22 30 34 & & a2576 & 92(2 ) & & 284 & 31 & & & & & & 68 & 0.102 & & 63 & 0.142 & & 2 & & b , r , c & & 0.1875 & 10 & 8 & & 0.1876 & 10 & + 052 & & 23 20 03.1 & -24 07 34 & & s1113 & 8(0 ) & & & & & & & & & & & & & & & & & b & & & & & & 0.1468 & 9 & 13 + 053 & & 23 20 21.1 & -25 11 56 & & & & & & & & & & & & & & & 148 & 0.309 & & & & b , r & & & & & & & & + 054 & & 23 20 21.1 & -24 42 00 & & & & & & & & & & & & 18 & 0.041 & & & & & & & & & & & & & & & + 055 & & 23 20 44.8 & -22 57 33 & & a2577 & 73(1 ) & & & & & & & & & & & & & & & 2 & & b , r , c & & 0.1251 & 1 & 10 & & 0.1248 & 7 & + 056 & & 23 20 24.0 & -21 49 00 & & & & & & & & & & & & & & & & & & & & c & & & & & & & & + 057 & & 23 21 08.4 & -21 33 33 & & a2579 & 66(1 ) & & o & & & 887 & 62 & 0.116 & & 34 & 0.087 & & 45 & 0.127 & & & & b , r , c & & 0.1117 & 1 & 10 & & 0.1114 & 9 & + 058 & & 23 21 20.5 & -22 06 33 & & a3996 & 59(1 ) & & & & & 889 & 41 & 0.109 & & & & & 49 & 0.162 & & & & r , c & & 0.1155 & 8 & 14 & & 0.0854 & 6 & 14 + 059 & & 23 21 21.0 & -24 10 33 & & a3997 & 39(0 ) & & & & & & & & & & & & 65 & 0.183 & & & & b , r & & & & & & 0.1478 & 8 & 13 + 060 & & 23 21 54.0 & -23 48 00 & & & & & & & & & & & & & & & & & & & & b , r & & & & & & & & + 061 & & 23 21 23.2 & -23 11 16 & & a2580 & 62(1 ) & & 287 & 20 & & 888 & 76 & 0.116 & & 89 & 0.104 & & 74 & 0.153 & & 2,3,5,6 & & b , r , c & & 0.1870 & 1 & 6 & & 0.0890 & 17 & 15 + 062 & & 23 21 24.8 & -22 33 10 & & & & & & & & & & & & & & & & & & & & b , r & & & & & & 0.0000 & & 16 + 063 & & 23 22 19.2 & -20 24 52 & & a2583 & 52(1 ) & & o & & & 890 & 57 & 0.100 & & & & & 62 & 0.176 & & & & b , r , c & & 0.1160 & 1 & 11 & & 0.1145 & 8 & + 064 & & 23 23 14.4 & -22 49 31 & & s1117 & 19(0 ) & & & & & & & & & & & & & & & & & b , r , c & & & & & & 0.0000 & & 17 + 065 & & 23 23 25.9 & -20 25 31 & & a2586 & 46(0 ) & & o & & & & & & & 62 & 0.116 & & 44 & 0.108 & & & & b , r , c & & & & & & 0.1448 & 11 & + 066 & & 23 23 32.3 & -22 24 31 & & a2587 & 97(2 ) & & & & & & & & & & & & & & & & & c & & & & & & 0.2157 & 6 & + 067 & & 23 24 08.3 & -22 33 47 & & & & & 291 & 20 & & & & & & 167 & 0.175 & & 54 & 0.190 & & & & b , r , c & & & & & & 0.1224 & 15 & + 068 & & 23 24 44.3 & -23 06 30 & & a4003 & 49(0 ) & & & & & & & & & & & & & & & & & & & 0.0866 & 1 & 13 & & 0.2159 & 2 & + 069 & & 23 24 55.7 & -20 31 30 & & a2595 & 4 & & o & & & & & & & 106 & 0.168 & & 81 & 0.209 & & & & r , c & & & & & & 0.1803 & 5 & + 070 & & 23 24 59.0 & -23 25 03 & & a2596 & 44(0 ) & & 293 & 27 & & 893 & 46 & 0.092 & & 52 & 0.088 & & & & & & & b , r , c & & & & & & 0.0892 & 24 & 15 + 071 & & 23 25 22.8 & -22 26 50 & & & & & & & & & & & & & & & & & & & & b , r , c & & & & & & 0.1226 & 5 & 18 + 072 & & 23 25 47.5 & -24 06 38 & & & & & & & & 894 & 50 & 0.116 & & & & & & & & & & b , c & & & & & & 0.1116 & 6 & 19 + 073 & & 23 26 19.5 & -24 08 53 & & & & & & & & 895 & 50 & 0.111 & & 59 & 0.082 & & 50 & 0.142 & & & & b , r , c & & 0.0880 & 1 & 11 & & 0.1116 & 21 & + 074 & & 23 26 43.9 & -23 50 53 & & a2599 & 51(1 ) & & 297 & 60 & & 898 & 59 & 0.098 & & & & & & & & & & b , c & & 0.0889 & 4 & 7,9 & & 0.0906 & 14 & 20 + 075 & & 23 26 43.9 & -22 24 29 & & a2600 & 50(1 ) & & 298 & 25 & & 896 & 46 & 0.105 & & 58 & 0.094 & & & & & & & b , r , c & & 0.1205 & 12 & 14 & & 0.1187 & 13 & + 076 & & 23 26 44.3 & -24 25 29 & & a2601 & 62(1 ) & & & & & & & & & & & & & & & & & r & & 0.1113 & 7 & 14 & & 0.2126 & 5 & + 077 & & 23 27 56.4 & -25 20 28 & & a2603 & 64(0 ) & & & & & & & & & & & & 81 & 0.231 & & & & b , r , c & & & & & & 0.2109 & 6 & + 078 & & 23 28 11.0 & -24 54 26 & & & & & 300 & 17 & & & & & & 47 & 0.088 & & & & & & & b , c & & & & & & 0.1125 & 12 & + 079 & & 23 28 12.0 & -23 48 00 & & & & & & & & & & & & & & & & & & & & c & & & & & & & & + 080 & & 23 28 31.7 & -22 31 28 & & a2604 & 31(0 ) & & & & & & & & & & & & 19 & 0.128 & & & & r & & & & & & 0.2121 & 3 & + 081 & & 23 29 01.8 & -23 21 02 & & a2605 & 54(1 ) & & 303 & 12 & & & & & & 33 & 0.071 & & & & & & & b , c & & & & & & 0.1121 & 13 & 21 + 082 & & 23 29 37.3 & -21 12 27 & & a2606 & 78(1 ) & & o & & & & & & & 126 & 0.188 & & 118 & 0.250 & & 6 & & r , c & & 0.2800 & 1 & 12 & & 0.1431 & 4 & + 083 & & 23 30 00.0 & -20 39 40 & & & & & o & & & & & & & & & & 43 & 0.117 & & & & c & & & & & & & & + 084 & & 23 30 21.6 & -24 21 32 & & & & & & & & & & & & & & & 78 & 0.191 & & & & r & & & & & & & & + 085 & & 23 30 31.3 & -21 39 27 & & a2608 & 59(1 ) & & o & & & & & & & 62 & 0.114 & & 43 & 0.130 & & & & b , r , c & & 0.0498 & 5 & 14 & & 0.1557 & 4 & + 086 & & 23 30 41.5 & -23 03 29 & & & & & & & & & & & & 48 & 0.125 & & & & & & & b & & & & & & & & + 087 & & 23 31 30.0 & -21 55 00 & & & & & & & & & & & & & & & & & & & & r , c & & & & & & & & + 088 & & 23 31 42.2 & -20 35 04 & & & & & o & & & & & & & 58 & 0.119 & & 32 & 0.088 & & & & b , r , c & & & & & & 0.1490 & 4 & + 089 & & 23 31 42.7 & -25 45 07 & & & & & & & & & & & & 291 & 0.223 & & & & & & & o & & & & & & & & + 090 & & 23 32 17.3 & -22 24 40 & & & & & & & & & & & & 57 & 0.132 & & & & & & & c & & & & & & & & + 091 & & 23 32 25.7 & -25 29 26 & & a4014 & 35(0 ) & & & & & & & & & 55 & 0.119 & & 47 & 0.181 & & & & b , r , c & & 0.1128 & 11 & 15 & & 0.1130 & 11 & + 092 & & 23 32 41.3 & -23 01 44 & & & & & & & & & & & & 72 & 0.146 & & & & & & & & & & & & & & & + 093 & & 23 32 55.0 & -21 34 25 & & a2614 & 54(1 ) & & o & & & & & & & & & & 46 & 0.167 & & & & r & & & & & & 0.1635 & 4 & 22 + 094 & & 23 33 01.4 & -23 33 25 & & a2615 & 114(2 ) & & & & & & & & & 30 & 0.081 & & 125 & 0.237 & & & & b , r , c & & & & & & 0.2061 & 5 & + 095 & & 23 33 19.0 & -20 28 01 & & & & & o & & & & & & & & & & 64 & 0.206 & & & & & & & & & & & & + 096 & & 23 34 31.4 & -23 56 17 & & & & & & & & & & & & 67 & 0.139 & & & & & & & b , c & & & & & & & & + 097 & & 23 35 21.8 & -22 05 42 & & & & & o & & & & & & & & & & 50 & 0.180 & & & & & & & & & & & & + 098 & & 23 36 54.0 & -23 24 00 & & & & & & & & & & & & & & & & & & & & b , c & & & & & & & & + 099 & & 23 37 00.9 & -24 09 23 & & a2628 & 83(2 ) & & 324 & 33 & & & & & & 58 & 0.112 & & 65 & 0.196 & & & & b , r , c & & 0.1858 & 10 & 8 & & 0.1858 & 10 & + 100 & & 23 37 01.7 & -20 51 54 & & & & & o & & & & & & & 87 & 0.172 & & & & & & & & & & & & & & & + 101 & & 23 37 42.7 & -22 55 23 & & a2629 & 100(2 ) & & & & & & & & & 80 & 0.155 & & 111 & 0.232 & & & & r , c & & & & & & 0.2069 & 7 & + 102 & & 23 37 55.9 & -22 35 53 & & & & & & & & & & & & 78 & 0.154 & & & & & & & & & & & & & & & + lrrrrrrr aco & & 83 & 88 & 59 & 75 & 84 & 68 + edcc & 49 & & 67 & 56 & 36 & 51 & 39 + apmcc & 26 & 33 & & 22 & 21 & 29 & 25 + mf - b & 55 & 89 & 71 & & 59 & 71 & 54 + mf - r & 68 & 56 & 63 & 57 & & 68 & 55 + sd-1 & 66 & 67 & 77 & 59 & 59 & & + sd-2 & 90 & 89 & 100 & 77 & 82 & & + lrrrrrrrrr single cluster in @xmath63 & 0.49 & 0.49 & 0.55 & 0.65 & 0.56 & 0.47 & 0.51 & 0.51 & 0.51 + w/ superimposed group & 0.28 & 0.25 & 0.28 & 0.23 & 0.23 & 0.29 & 0.27 & 0.23 & 0.21 + two concentrations & 0.08 & 0.11 & 0.17 & 0.06 & 0.13 & 0.08 & 0.05 & 0.11 & 0.11 + only small groups & 0.13 & 0.12 & 0.00 & 0.06 & 0.08 & 0.14 & 0.14 & 0.12 & 0.13 + no concentration & 0.02 & 0.03 & 0.00 & 0.00 & 0.00 & 0.02 & 0.03 & 0.03 & 0.04 ccccccccccccccccccccccccc 3.5 & 250 & & & & 2 & & & 2 + 2 & 2 & 2 & 2 & & & & 2 & & & & & & & & & + + 5 & 90 & & & & 2 + 2 & & & 5 & 4 & 2 & 2 & & & & 2 & & & & & & & & & + + 10 & 10 & & & & 2 + 3 & & & & 14 & & 5 & & 3 & & 4 & & & & & & & 2 & & 2 + 2 + 3 + + 15 & 3 & & & & 5 & & & & 21 & & & & 8 & & 4 & & & 2 & & & & 3 & & 7 + + 3.5 & 200 & & & & 4 + 2 + 4 & 2 & & 2 + 2 & 4 & 2 & 2 & & 2 & & 2 & & & 2 & & 2 + 2 & & & & 2 + + 5 & 70 & & & & 7 + 2 + 6 & 2 & & 7 & 6 & 2 & 2 & 2 + 2 & 2 & & 2 & & & 2 & & 2 + 2 & & & & 2 + + 10 & 9 & & 3 + 2 & & 17 & 3 & & & 24 & & 6 & 7 & 5 & & 4 & 2 & & 2 + 3 & & 2 + 3 & & 2 + 2 & & 2 + 2 + 4 + + 15 & 3 & & 6 & & 21 & & & & & 43 & & & & & 4 & 5 & & 6 & & 5 & & 4 & & 8 +

we develop parallels between the holographic renormalization group in the bulk and the wilsonian renormalization group in the dual field theory . our philosophy differs from most previous work on the holographic rg ; the most notable feature is the key role of multi - trace operators . we work out the forms of various single- and double - trace flows . the key question , ` what cutoff on the field theory corresponds to a radial cutoff in the bulk ? ' is left unanswered , but by sharpening the analogy between the two sides we identify possible directions . holographic and wilsonian renormalization groups * idse heemskerk * _ department of physics _ _ university of california _ _ santa barbara , ca 93106 _ * joseph polchinski * _ kavli institute for theoretical physics _ _ university of california _ _ santa barbara , ca 93106 - 4030 _ = 16pt

1010.1264

in a sense , ads / cft duality reduces the construction of quantum gravity to a solved problem . gravity is written in terms of a nongravitational quantum field theory @xcite , and the technology of wilson @xcite can then be applied . in its current form , this duality reports only the measurements made by an external observer studying gravity confined to an anti - de sitter box . still , this allows one to address many of the conceptual questions of quantum gravity , most notably the purity of black hole hawking radiation . this implies in turn an extreme nonlocality of quantum gravity @xcite . formulating a theory in which locality has been so radically abandoned is a great challenge . fortunately , we have the example of ads / cft duality , and so we should understand this example as fully as possible . in retrospect , what ads / cft describes are the simplest and most sharply defined observables possible in a theory of quantum gravity . one needs go beyond these , to describe local observers within ads space , and to understand gravity outside the ads box . in this paper we study a framework suggested by the discussion above , in which we pull the wilson renormalization group back through the duality , where it takes a holographic form . of course there is a wide literature that falls under the broad heading of _ holographic renormalization group ( hrg)_. however , the parallels between the wilson and holographic perspectives have not been fully developed , and there are some simple observations and exercises that we would like to add , although we will also leave many questions unanswered for now . many of the points that we make have already been made in various forms in the literature , but it is useful to bring together here a coherent point of view . although we have emphasized the use of the duality to understand quantum gravity , clarifying the relation between the rg s should also be useful in the applications to strongly coupled field theories . indeed , there is interesting overlap between some recent work in this area and ours . wilson separates the path integral into high- and low - energy modes , and integrates out the former , @xmath0 where we use @xmath1 for generic boundary fields . we are being impressionistic here : the precise form of the cutoff at scale @xmath2 is not specified . the progressive integration of momentum shells is then described by a differential equation , and by integrating out to @xmath3 one obtains the full path integral . the intermediate stages are quite complicated : @xmath4 is necessarily a quasilocal function of length scale @xmath5 , and so depends on an infinite number of parameters , but the flow is strongly convergent in almost all directions . in essence , this flow supplies the deltas and epsilons of path integration , reducing the construction of qft to dimensional analysis . in the spirit of full disclosure we should note that it is difficult to make the mode separation while preserving explicitly important structures such as gauge invariance , supersymmetry , and duality . ultimately this may point to the need for a new way to think about qft , but for now this is the only tool of broad applicability , enabling one to reduce arbitrary calculations to algorithms . in ads / cft duality , the radial coordinate @xmath6 of the ads bulk emerges from the energy scale of the boundary field theory @xcite . this suggests a correspondence between radial evolution in the bulk and wilson renormalization group flow in the field theory . in sec . 2 we develop this idea . our approach is to postulate a mapping between the radial cutoff on the bulk , and a wilsonian cutoff on the field theory . to make the form of this mapping precise is the deep and difficult problem that we would like to solve in the future , but for the present we are interested in the consequences of this postulate for the structures on the two sides . two somewhat unexpected results are the central role played by multi - trace operators , and the need to carry out the gravitational path integral in gauge - fixed rather than wheeler - dewitt form . in sec . 3 we work out some examples , in particular the flows induced by relevant single- and double - trace operators , and also flows involving bulk gauge fields and the metric . in sec . 4 we discuss various truncations of the full wilsonian flow and relations with the earlier literature . in sec . 5 we discuss the picture from the boundary field theory point of view , and are led to a possible , but rather unusual , regulator . in sec . 6 we discuss future directions and intended applications . in the bulk , our approach requires that we write the theory as a path integral over local fields . already this leads to some tension , as the wilsonian framework requires that we keep all operators in the theory , including those dual to stringy states , and so it is not clear whether such a local representation is possible . we will gloss over this point for now , and return to it in sec . an asymptotically @xmath7 metric can be written @xmath8 we use a lower - case @xmath9 for the lapse to distinguish it from the number of colors . some flows of interest involve large deformations of the spacetime , and so we must be must deal with the dynamics of the metric . in order to get oriented , however , we will first consider flows for which the backreaction is small , and so take the path integral in the fixed metric ( [ adsmet ] ) , focussing for simplicity on scalar fields . separate the path integration over the fields @xmath10 into three parts , @xmath11 , @xmath12 , and @xmath13 : @xmath14 here @xmath15 , and @xmath16 denotes the bulk action , while @xmath17 and also @xmath18 will be used for qft actions . the dependence of the full amplitude @xmath19 and of the uv piece @xmath20 on the choice of ads boundary conditions , including any sources , is left implicit . we keep track of @xmath21 ( which for @xmath22 is @xmath23 ) in anticipation of the classical limit , and we assume that @xmath24 is of supergravity type , by which we mean that with canonical normalization it is independent of @xmath25 . this radial separation resembles a wilsonian treatment of the dual field theory . in particular , it is natural to interpret the ir factor @xmath26 as the path integral in the dual qft with a uv cutoff @xcite , on a length scale @xmath2 to be determined shortly , @xmath27 here @xmath28 are a complete set of local single trace operators built from the matrix fields @xmath1 and their derivatives . they are normalized @xmath29 where the ellipsis represents a product of the matrix fields and their derivatives . thus @xmath30 and @xmath31 are held fixed in the large-@xmath32 limit , and @xmath33 . it would be more standard to have a factor of @xmath34 in the exponent rather than @xmath35 , which is of the same order in @xmath32 , but this makes subsequent equations simpler . the action @xmath36 is included for generality , though one might absorb it into the definition of the measure . the linearity of the action in @xmath37 can be arranged by a bulk field redefinition . _ we take eq . ( [ ir ] ) as a postulate , _ and explore its consequences . of course we would like to be explicit about the precise form of the cutoff @xmath2 and action @xmath36 . this is the hard question , whose solution would allow us to discuss local physics in the bulk . for now , though , we will simply assume this and see what it implies for the rest of the rg structure . similarly , it is tempting to identify the uv factor @xmath38 with a wilsonian action , integrating out the fields above the cutoff scale . this action is not local , because propagating fields have been integrated out , but will fall exponentially beyond some length @xmath2 because the fields are confined to a finite range @xmath39 , and so there is a gap for propagation in the @xmath40 directions . this locality property has been discussed previously in ref . @xcite . the scale @xmath2 is estimated in ref . @xcite in terms of a geodesic connecting two boundary points and hanging down to radius @xmath6 . the same order of magnitude is given by the extent in @xmath40 of the lightcone of a boundary point when it touches @xmath13 @xcite , as illustrated in fig . 1 . on which the wilsonian action is localized . ] [ deltafig ] for a poincar invariant metric @xmath41 this is given by @xmath42 here @xmath43 is the proper distance , and we assume that , as in ads , @xmath44 grows rapidly toward the ir and so the integral can be estimated by the behavior near the upper limit . for pure ads space one simply has @xmath45 more generally , @xmath46 meaning that the proper length corresponding to the cutoff is of the order of the effective ads radius at the given scale . this is also consistent with one degree of freedom per planck area @xcite , @xmath47 in terms of the @xmath48 dimensional planck length . the expansion in local terms thus involves an infinite number of parameters , as in the wilsonian action . finally , in comparing the holographic and wilsonian structures ( [ holosplit ] ) and ( [ wilsplit ] ) , what is the role of the functional integral over @xmath49 ? it appears that rather than a definite rg flow we have instead some weighted average over effective couplings . to get some insight , consider a single scalar , suppose that the uv factor is a local gaussian@xmath50 using the postulate ( [ ir ] ) and carrying out the integral over @xmath37 gives @xmath51 thus we do obtain a definite effective action , but one that has double - trace as well as single - trace terms . more generally , we see that the wilsonian action is not @xmath52 directly , but rather an integral transform of it , @xmath53 the action @xmath54 has general multi - trace terms , and is localized on the scale @xmath2 . that renormalization group flow generically leads to multi - trace operators , and moreover that these contribute to connected correlators even in the planar limit , has been discussed in a number of situations . for example , in nonsupersymmetric orbifolds of the @xmath55 theory this was observed in refs . @xcite , while in the wilson rg context it was noted in refs . @xcite ; we will review the latter work in sec . 5 . nevertheless , it comes as a bit of a surprise , because most of the hrg literature focuses on flows within the space of single - trace couplings , and because of the apparently misplaced prejudice that single - trace actions are the norm in taking large-@xmath32 limits , with multi - trace actions arising in exotic applications such as refs . @xcite . the total amplitude is independent of the radius at which the division is made , @xmath56 where the angle brackets denote the matrix path integral ( [ ir ] ) , and the derivative in the last form acts on both the cutoff and the action . the ` amplitudes ' @xmath57 evolve via radial schrdinger equations @xcite @xmath58 where @xmath59 . for example , for @xmath60 @xmath61 for the wilsonian action ( [ wilson ] ) , the evolution equation becomes @xmath62 this is our main result , an equation that should be compared in form to that found in the qft . in the classical limit @xmath63 , corresponding to the planar limit of the qft , this reduces to the hamilton - jacobi ( hj ) equation @xmath64 let @xmath65 be a complete set of local operators , including multi - traces . expand @xmath66 where we allow for position - dependent couplings so as to have a generating functional for correlators . then the rg equation ( [ wilrg ] ) for the running is of the general form @xmath67 and the statement of invariance with respect to cutoff is @xmath68 where the partial derivative acts only on the cutoff . further , using functional derivatives to introduce the correlator @xmath69 we have @xmath70 where @xmath71 . in this discussion , we have taken a euclidean metric , as is usual in studies of the renormalization group , but there appears to be no obstacle to having a lorentzian metric on the poincar slices . now we consider the general case , with dynamical metric . because the metric is fixed on the ads boundary one can in a neighborhood of the boundary fix the coordinates fully such that @xmath72 these fefferman - graham coordinates have been employed extensively in the holographic renormalization program of ref . @xcite . once the coordinates are fixed , one can treat the @xmath73-dimensional metric @xmath74 on the same footing as the scalar fields . for pure einstein gravity we get the schrdinger equation @xmath75 with the upper sign for the uv and the lower sign for the ir . tildes again denote the field restricted to @xmath76 . here @xmath77 , @xmath78 is the intrinsic curvature of a fixed @xmath6 slice , and @xmath79 this approach may raise concerns on two points , stability and coordinate invariance . the fefferman - graham coordinates will in general break down at finite distance due to formation of caustics . this is actually exacerbated by the ads geometry : if we consider the radial geodesic @xmath80 , nearby geodesics @xmath81 diverge quadratically . we view this as an inessential complication . one can think of this coordinate system as defining the surface of given @xmath82 as the set of points of distance @xmath83 from the surface @xmath84 , where @xmath85 is in the asymptotic region ( [ adsmet ] ) . the coordinates @xmath40 are then assigned on the @xmath82 surface according to the closest point on the @xmath85 surface . when there are caustics , the @xmath82 surface and the @xmath40 mapping are not smooth . it should be possible to smooth these , for example by defining the distance from the @xmath85 surface in terms of some average distance to a disk of radius @xmath6 . we leave the details for future work , hopefully by others with better geometric tools . more substantively , by fixing completely the lapse and shift we have lost the corresponding equations of motion , so that the state @xmath20 does not satisfy the constraints of canonical gravity . in our approach , we are foliating the bulk according to the lapse from the @xmath85 surface , and the integral over uv metrics is constrained to those of given total lapse and zero shift . in the wheeler - dewitt ( wdw ) approach @xcite , one would integrate over all metrics with given inner and outer boundary , and identify a radial variable according to some intrinsic property of the geometry . in this case the wavefunction would satisfy the constraints . in most contexts it is the wdw wavefunction that is physical , but for the purpose of defining the wilsonian action it is the coordinate - fixed wavefunction that is appropriate . one sign of this is that the wdw wavefunction does not have the quasilocality expected of the wilsonian action , because the action for the metric component @xmath86 does not involve @xmath6-derivatives ; we will work out explicitly the corresponding effect for the bulk gauge field in sec . 3.2 . another reason that this can not work is that an ultralocal term in the metric would have to be of the form @xmath87 if the constraints are satisfied . on the other hand , the wilsonian action should contain arbitrarily ultralocal functions of @xmath88 , whose integral transform would be a correspondingly general function of @xmath89 . conceptually , the boundary qft action is not coordinate invariant , and so neither should the cutoff action obtained by integrating out high energy fields . note that in most applications of gaussian normal coordinates there is no natural surface on which to begin , and so one must impose the constraints . in ads space the boundary provides such a surface . for the ir path integral , there is no corresponding boundary surface and no condition to fix the coordinates globally . one is integrating over all metrics with given outer boundary , and so @xmath90 does satisfy the hamiltonian and momentum constraints . the equation of motion corresponding to the constraint , which is given by the matrix element of the constraint between @xmath90 and @xmath20 , is therefore satisfied . as a result of the constraint , @xmath90 has no direct dependence on @xmath91 . rather , for given @xmath74 the ir path integral has an effective cutoff given by ( [ cutoff ] ) . insertions of the energy - momentum tensor are generated by derivatives with respect to @xmath74 , but we can not linearize as in the scalar case ( [ ir ] ) . the rg equation is @xmath92 when the metric is approximately classical , the uv amplitude will be peaked at a metric @xmath93 . this defines the cutoff ( [ cutoff ] ) , and the integration over fluctuatations around @xmath93 generates multitrace interactions involving the energy - momentum tensor . incidentally , if one has a metric of finite total lapse in the ir direction , as with massive deformations of the qft , most foliations will eventually break down . here there is a physical origin : when we pass through a massive threshold we must recast the rg in terms of new degrees of freedom appropriate to that scale . more generally , the monotonicity of the energy with radius will break down , for example in multi - throated solutions corresponding to breaking of @xmath94 to a product of smaller @xmath95 s , and in semi - holographic situations where there are massless degrees of freedom in a region of finite warp factor @xcite . again , these represent thresholds in the rg . it is interesting to work out in detail the case of a single free scalar , in the limit that backreaction is negligible . we work in units @xmath96 , and take a poincar invariant background , so @xmath97 , @xmath98 , and @xmath99 \bigr\ } \,.\ ] ] we further assume that the amplitudes are gaussian , @xmath100 in the ir this follows dynamically , while in the uv it is an assumption about the initial conditions on the flow , which could be relaxed . we have allowed for a possible translation - breaking source for the linear term , but not for the quadratic one : the subscript @xmath101 is for fourier components of fields and sources , while the argument @xmath101 is for kernels . then @xmath102 with the upper sign for the uv and the lower sign for the ir . for the normalization constant , @xmath103 \,.\ ] ] for this quadratic system the full schrdinger equation differs from the hj equation only by the @xmath104 term in @xmath105 . the center of the gaussian , @xmath106 , is not a solution to the bulk field equation , but it and @xmath107 can be written @xmath108 where @xmath109 does solve the massive klein - gordon equation @xmath110 we now specialize to the case of pure ads , @xmath111 , for which @xmath112 is a bessel function of order @xmath113 . it is convenient to write it as @xmath114 in terms of@xmath115 we will need a different prescription , below . ] @xmath116 the set of flows is thus parameterized by two functions @xmath117 , @xmath118 . first let us cast the evolution equations in the form of an rg flow . according to eq . ( [ gauss ] ) , @xmath20 corresponds to the action @xmath119 define single- and double - trace couplings @xmath120 the evolution equations ( [ fphi ] ) become @xmath121 we can also interpret the evolution of @xmath105 as a running of the unit operator , @xmath122 at momenta long compared to the cutoff , where the @xmath123 term can be neglected , the double - trace coupling has two fixed points , @xmath124 at these fixed points , @xmath125 correspondingly , the single trace operator @xmath126 has dimension @xmath127 : the lower sign is the standard quantization , and the upper sign is the alternate quantization @xcite . also when @xmath128 , we can take the leading behaviors of the bessel functions , @xmath129 so that @xmath130 for generic @xmath131 this describes the flow @xcite from the alternate quantization in the uv to the standard quantization in the ir . the standard conformal quantization is @xmath132 and the alternate conformal quantization is @xmath133 . recall that for @xmath134 only the standard quantization gives a unitary theory all the way into the uv , so we focus on @xmath135 ; we could also consider the flow at general @xmath136 in an effective theory with a maximum uv scale . the uv boundary conditions on the fields are generally parameterized as @xmath137 our restriction to gaussian @xmath20 implies a linear relation @xmath138 in order to relate this parameterization to the general flow above , we can directly evaluate the path integral , which is extremized by @xmath139 , which has the asymptotic behavior ( [ asymp ] ) at @xmath140 and is equal to @xmath37 at @xmath76 . the solution to the classical equation of motion @xmath141 with given boundary conditions is @xmath142 using the classical equation of motion and integrating by parts gives @xmath143 \ , . \label{wuv}\ ] ] we have included a surface term , @xmath144 , as needed to give a good variational principle with the boundary condition ( [ linab ] ) . inserting the classical solution , we can reduce this to the earlier form ( [ fsoln ] ) , with @xmath145 the @xmath146 are by construction entire functions of @xmath101 . , i.e. @xmath147 , we need in this case to replace @xmath148 with @xmath149 . ] the uv amplitude depends only on @xmath146 and its derivative , and on the initial data @xmath150 . therefore if the latter are are localized ( analytic around @xmath151 ) then so is the wilsonian action , as expected . we can then expand @xmath152 then @xmath153 is the coefficient of @xmath154 in the lagrangian . the rg equation becomes @xmath155 note that the equation for @xmath153 involves on the right only @xmath156 with @xmath157 , allowing an iterative solution . this iterative structure in the momentum expansion is a property of the planar limit in the interacting case as well . the infrared amplitude solves the evolution equations ( [ fphi ] ) with the lower sign . in the path integral , the condition that the classical solution not blow up as @xmath158 fixes @xmath159 so @xmath160 and @xmath161 . the amplitude is @xmath162 to see how the rg acts on this , we have from ( [ ir ] ) @xmath163 in terms of the action @xmath164 with single- and double - trace couplings . at @xmath165 we simply have @xmath166 and @xmath167 . the rg ( [ rg ] ) then takes the form @xmath168 the integrand of the second term on the right can be rewritten @xmath169 and so we recover the schrdinger equation . in the hj approximation , the rhs becomes @xmath170 the ir amplitude is generated by integrating out long wavelengths , and so is not localized . using @xmath171 , we have @xmath172 in the asymptotic expansion at small @xmath101 , one obtains terms of the form @xmath173 . the local part , defined by the terms of the form @xmath174 , is @xmath175 this has no particular relation to the uv amplitude , which is also local , since the latter depends on the uv boundary conditions . in identifying the local part , we have ignored additional integer powers that appear when @xmath136 is rational ; for more general flows , picking out an part analytic in @xmath101 may not be well defined . however , the ultralocal @xmath176 term is unambiguous if @xmath177 , and in the present case is just @xmath178 the leading nonlocal behavior at small @xmath101 is @xmath179 our next example is a bulk @xmath180 gauge field @xmath181 in a fixed poincar invariant background geometry , which must couple to a conserved current @xmath182 in the boundary theory . integrating by parts along the @xmath183 directions the action can be written @xmath184 where @xmath185 is a projection operator . we will first deal with the gauge symmetry by fixing radial gauge , @xmath186 . normally this leaves a residual invariance under @xmath6-independent gauge transformations , but in ads space the gauge parameter is required to vanish at the boundary ( in the cft the @xmath180 is a global symmetry ) , so the gauge symmetry is fully fixed . we can then write a hamiltonian directly @xmath187 giving the ( momentum - space ) schrodinger equation @xmath188 where the upper / lower sign is for the . the schrdinger equation separates in the longitudinal and transverse variables @xmath189 , @xmath190 , @xmath191 we first obtain @xmath20 by direct evaluation of the path integral with @xmath192 and @xmath193 fixed . the latter is the same as the classical gauge field coupled to the the equations of motion in radial gauge reduce to @xmath194 the transverse part obeys the scalar klein - gordon equation ( [ bulkkg ] ) with the replacements @xmath195 and @xmath196 , while the longitudinal part is given by @xmath197 the classical action is then @xmath198 the longitudinal part is @xmath199 the transverse part is @xmath200 where @xmath201 here @xmath202 is the solution to the massive klein - gordon equation at @xmath203 and @xmath204 such that @xmath205 . it is clear from the path integral construction that the total classical action is localized in the boundary values of the fields . this is not manifest in the final form , because the projection @xmath206 makes the transverse and longitudinal pieces separately nonlocal : each piece separately is @xmath207 times an analytic function of @xmath208 , and a generic solution to the schrdinger equation would be nonlocal . one can verify that the pole cancels in the sum by noting that at @xmath209 , @xmath210 and @xmath211 is identically 1 , so that the nonderivative part of the action is simply @xmath212 the integral transform , which is the wilsonian action , therefore contains a a @xmath213 term and an @xmath214 term . if we had fixed @xmath215 to some generic functional form , then by a gauge transformation we can see that the action would be the same as above with the replacement @xmath216 if we integrated freely over @xmath215 , the result would be a gauge volume times an integral over @xmath217 , with a result independent of @xmath218 . that is , the uv amplitude would now satisfy the constraint @xmath219 in canonical quantization , one would normally be interested only in such physical states , but for our purposes this can not give rise to a wilsonian action : eliminating the dependence on @xmath218 necessarily leaves the amplitude nonlocal . effectively , the wilson action must remember that the @xmath180 symmetry was a global symmetry in the uv , not a gauge symmetry . all of these observations are parallel to our discussion of the coordinate choice in sec . 2.2 : the radial gauge has the same properties as the gaussian normal gauge , and gives rise to a wavefunction that does not satisfy the constraints , but gives a wilson action with the expected properties . the @xmath215-integrated amplitude is analogous to the wheeler - dewitt wavefunction and does not lead to a local action . the longitudinal part of the ir amplitude is @xmath220 when there is a horizon ( @xmath221 ) then @xmath222 and the longitudinal part vanishes ( except for @xmath223 , but this is a rather special case ) . when there is a mass gap ( so @xmath82 is finite ) , @xmath224 is finite but we must integrate over the gauge field on the lower boundary , and the longitudinal amplitude is again trivial . thus the ir amplitude always satisfies the constraint . now we consider an interacting scalar theory , taking for simplicity a single field . in this subsection we assume that the backreaction on the metric is small , in the next we discuss its inclusion . we focus on the momentum - independent terms in the action . these satisfy closed equations , as we have already seen in the example ( [ momex ] ) . thus , writing @xmath225 we have the hj equation @xmath226 suppose that @xmath227 has a stationary point @xmath228 , @xmath229 then there are scale invariant solutions @xmath230 where @xmath231 note that the equation for @xmath232 is independent of the other @xmath233 , and quadratic . this just produces the standard and alternate quantizations , subject to the usual restrictions . the remaining equations then determine the higher @xmath233 iteratively . flows that approach one of the extrema of @xmath227 in the uv correspond to perturbations by relevant operators . much of the literature on the holographic renormalization group focuses on flows that interpolate between different extrema of @xmath227 . in our framework , we look for a solution of the form @xmath234 this form is redundant , because @xmath235 can be set to an arbitrary function by redefinition of the @xmath236 s , but it is useful to present it in this way . taylor expanding the hj equation around @xmath228 gives @xmath237 the first three equations are @xmath238 we can obtain the same iterative structure ( [ iter ] ) as in the conformal case if we fix the redundancy noted above by requiring @xmath239 the @xmath240 equation is then independent of @xmath232 , and requires that @xmath241 satisfies the classical field equation , @xmath242 in the equations for the higher @xmath233 , the next term @xmath243 enters multiplied by @xmath244 , so each successive equation closes . for the case of a domain wall solution @xcite , where @xmath228 approaches distinct extrema of @xmath227 at @xmath245 , this asymptotes to the conformal solution at each end . the quadratic term @xmath232 , adjusts in response to the flow of @xmath240 ( through @xmath228 ) , and this is necessary to obtain the evolution of the connected two point - functions , as calculated directly in ref . @xcite . in particular , eq . is now solved by @xmath246 , where @xmath247 . for a quadratic potential , discussed in the previous section , @xmath248 . if we then shift @xmath249 to eliminate the linear term we find that the quadratic term is centered at zero , i.e. @xmath250 . in the flow studied in sec . 3.1 we chose instead @xmath251 . the @xmath240 equation then becomes @xmath252 here @xmath228 is not a solution to the field equation but obeys the same equation ( [ fphi ] ) as @xmath253 . however , the equation for @xmath232 depends on @xmath254 , which can not be zero for a nonquadratic potential , and so does not close as in eq . ( [ freescalareqns ] ) . in this form , the @xmath255 do not decouple from the @xmath256 . for the wilsonian action , again consider the momentum - independent terms @xmath257 expanding @xmath258 one finds that @xmath259 where the ellipsis represents terms that contain only @xmath260 such that @xmath261 . unlike the equations for @xmath262 , these are not in general iterative in @xmath9 . now let us consider the backreaction on the metric . to emphasize the main points we consider the case without scalars , with negative cosmological constant @xmath263 . for @xmath264 with @xmath265 , the hj equation is @xmath266 the solution , given by the classical action with @xmath267 as @xmath268 , is @xmath269 note the wrong - sign quadratic term , a manifestation of the sign problem of euclidean gravity @xcite . roughly speaking , the @xmath270 contour must be rotated into the imaginary direction at the saddle point . again , the uv action does not satisfy the constraint , which would allow only a term @xmath271 . for a general domain wall solution with back - reaction , ( @xmath272 ) , one can simply treat the scale factor on the same footing as the scalars , and so extend the previous discussion . if we are applying this formalism to the full bulk theory including excited string states , we must correspondingly include in the field theory all operators , and their flow is described by the wilsonian . in comparing different forms of the rg , it is important to distinguish this full wilsonian rg from various forms where the flow has been projected onto a subspace . the most familiar projection is the rg in renormalizable theories . in the wilsonian framework , nonrenormalizable couplings correspond to directions where the flow of the action is strongly convergent . ) require that we evolve the action toward the ir , and the cut - off amplitude @xmath90 toward the uv , for typical hamiltonians that are unbounded above . ] projecting onto the remaining directions , one obtains the flow of the nearly - marginal operators , those for which @xmath273 is parametrically small : this is the renormalizable flow , described by the callan - symanzik equation . superrenormalizable couplings correspond to directions in which the flow is diverging , and these must be suppressed by tuning or by symmetries to obtain a flow over a wide range of scales . in the bulk , this projection corresponds to an effective theory of nearly massless fields , @xmath274 . it is important to note that this greatly changes the form of the rg . for example , in the framework of ref . @xcite , the full rg contains only a tree level term which is quadratic in the couplings , and a one - loop term which is linear in the couplings , and these are explicitly known . the renormalizable flow , on the other hand , contains terms of all orders in the coupling , which must be calculated order by order . to see how this comes about , consider the model equations @xcite @xmath275 here @xmath276 is a nearly marginal coupling , and @xmath277 is irrelevant . as we flow toward the ir the latter approaches a value that depends only on @xmath276 and not on the initial conditions . in this simple example we can immediately write the resulting equation for @xmath276 by taking @xmath278 of the first of eqs . ( [ toyrg ] ) and then using the second , @xmath279 this is to be solved in a ` slow - roll ' approximation , where the scale derivatives are of order @xmath276 and so the second term on the right is subleading . inserting the leading approximation @xmath280 into this term gives @xmath281 continuing the iteration gives the infinite series typical of renormalizable theories . in theories having a gravity dual , there is another natural truncation . in this case there is a set of operators whose dimensions are of order one , dual to states whose masses are of order the ads scale , and there are operators whose dimensions are parametrically large , dual to string scale ( or heavier ) states . these ultra - high dimension fields represent directions in which the flow is converging extremely rapidly , and one can project onto the relatively slower flow of operators of dimension @xmath282 . in the bulk , this corresponds to truncating to the effective supergravity theory including the dependence of the fields on the compact directions ( when these are at the ads scale ) . indeed , our whole framework of separating the bulk path integral radially into two parts is much more natural in this effective theory than in the full string theory . thus it will be of interest to study this ` supergravity ' truncation of the wilsonian rg @xcite . the correspondence between radius and energy of course goes back to the first work on ads / cft @xcite , and many papers have explored the analogy between a radial cutoff and the wilson . some of our results have appeared in one form or another , but our main result ( [ wilrg ] ) does not seem to have been noted , and much of the holographic rg literature seems rather different in form , in particular dealing exclusively on the single - trace operators . the widely - employed formalism of ref . @xcite , focuses on the low energy amplitude @xmath90 and does not consider the wilsonian action as we have identified it . this amplitude satisfies the evolution equation @xmath283 it is a function only of the @xmath49 , which are in one - to - one correspondence with the single - trace operators , and so these are naturally interpreted as couplings in a purely single - trace action . however , eq . ( [ irevol ] ) does not have the form of an rg : the latter involves only first derivatives with respect to the coupling , i.e. @xmath284 , while the hamiltonian ( [ irevol ] ) will be second order and higher in @xmath285 . in our interpretation , these higher derivative terms correspond to the flow of the action out of the single - trace space . in the planar limit the hj evolution can be written in terms of first derivatives , @xmath286 but it is nonlinear in first derivatives and so not of rg form . finally , one can write an rg - like equation by expanding the hj equation around a given solution @xmath287 , @xmath288 then at linear order @xmath289 which does have the form of an in ref . @xcite , the zeroth order solution @xmath287 is taken to be the local part of the action , to some order in derivatives . we can illustrate this with the free scalar example of sec . the leading local term ( [ freescalarultralocalir ] ) gives @xmath290 and so @xmath291 this is satisfied by the leading nonlocal term ( [ freescalarnlir ] ) , ) involved nonzero @xmath292-functions only for the unit and double - trace operators , whereas here there is only a single - trace @xmath292-function . ] @xmath293 for the full nonlocal amplitude ( [ fir ] ) one must keep also the @xmath294 terms , which we have identified with the flow of multi - trace operators . this rg is in a non - wilsonian scheme : the @xmath292-function depends on the particular solution @xmath287 , which depends on boundary conditions in the this is analogous to rg schemes where couplings are defined in terms of physical correlators at given momentum points , which depend to some extent on physics at all scales and on the particular vacuum the theory is in , whereas the wilsonian couplings and evolution depend only on scales above the cutoff and are independent of the state . this state - dependence is also emphasized in the lorentzian discussion in ref . @xcite . much of the hrg literature applies the formalism of ref . we emphasize that we are taking for our purposes a very specific definition of ` wilsonian , ' and these other approaches may be quite useful in organizing the physics according to scale . other work that organizes the physics via a wilsonian splitting but uses an ir renormalization scheme includes ref . @xcite ( the appearance of the bessel function @xmath295 in the construction of the @xmath292 function indicates an ir scheme , because it is distinguished by finiteness at @xmath296 ) . various papers take an approach more similar to ours , though we believe that much of our framework is new . @xcite analyzes properties of the radial schrodinger flow . @xcite , in the context of randall - sundrum compactification , introduces a flow of the uv boundary action . ref . @xcite uses evolution of a radial slicing to study bulk divergences . ( 46 ) of that paper identifies a dilatation operator , but it involves only single - trace operators ; we are not sure of the relation to our work . some properties of @xmath297 were considered in refs . the recent work @xcite has some overlap with our general approach , in particular on the importance of multitrace operators , and with the calculations in sec . 3.1 . some relevant papers on the field theory side will be noted in the next section . recent work on critical behavior and transport in strongly coupled cft s employ ideas closely related to ours . v of ref . @xcite gives an analysis of the flow between the horizon and the boundary that is similar in form to our discussion of domain wall flow . @xcite also takes a wilsonian approach to transport calculations , and refs . @xcite discuss phase transitions driven by multi - trace flows . in discussions of the @xmath298 expansion single - trace actions are usually regarded as the norm , with multi - trace terms arising in more exotic applications . however , from the renormalization group point of view the reverse is true @xcite . the schematic form of the wilson rg @xcite , in the form used in ref . @xcite , @xmath299 here @xmath300 represents the derivative of the propagator with respect to scale . the first term arises when the integrated propagator connects two distinct vertices as in fig . 2a , and the second arises when it connects a vertex to itself as in fig . the graphs provide useful intuition , but the derivation is nonperturbative . for a matrix - valued theory , one can keep track of the effect on the multi - traces @xcite . it is convenient t o rewrite the rg as @xmath301 there are then three kinds of term , as in fig . when the derivatives act on distinct traces they combine them into one : the action is of the form @xmath302 . when they act on a the same trace , they leave the number of traces fixed if they act on fields in adjacent cyclic order , a @xmath303 term , and they add a trace if the act on fields that are separated in the cyclic order , a @xmath304 term . in all , the evolution is of the schrodinger form ( [ evolve ] ) with @xmath305 we have accounted a factors of @xmath32 from the propagator , the index loop , and the definition of @xmath126 . all terms contribute in leading planar order , where @xmath17 is of the form @xmath306 . this cubic hamiltonian does not at all resemble that of supergravity , but looks very much like string field theory , with @xmath307 destroying a string and @xmath126 creating one . this is in keeping with the remarks in sec . it suggests that the wilson rg should be related to to string theory quantized in a gauge where the ads radius is identified with world - sheet time @xcite . ) such a truncation with respect to derivatives acting on traced operators , but do not believe that one can also truncate with respect to derivatives acting on the separate fields within the trace.]@xmath308 ) , and the term @xmath309 would look like a lower order interaction . also , the rg of the cft acts not on the total action but on the interaction part , and the rg ( [ wilrg ] ) similarly acts on an action shifted by @xmath36 ; these shifts must be taken into account in comparing forms . ] the nonpolynomial supergravity action is obtained from string field theory by integrating out the stringy fields , in parallel to the procedure described in sec . the formal structure that we have developed in this paper will be useful only if we can give meaning to the relation ( [ ir ] ) between the cutoffs in the bulk and the gauge theory . of course , already on the gauge theory side it is intricate to construct a continuum cutoff that respects the gauge symmetry @xcite ; extended supersymmetry and dualities are also mangled in the wilson framework . as a final calculation in this paper , we would like to note that the wilsonian holographic rg suggests a new approach to this problem . if we begin with the cutoff amplitude @xmath26 in the bulk , whose precise cft dual is to be determined , we could try to use the schrdinger evolution ( [ evolve ] ) @xmath310 to express this in terms of @xmath311 ; for simplicity we are looking at the case that the backreaction is small . taking the limit @xmath312 , we reach the boundary and so the standard dictionary @xcite allows it to be translated into gauge theory variables . thus we identify a cft calculation that produces the bulk amplitude with radial cutoff . the problem is that we are running the evolution the wrong way : essentially , we are forming @xmath313 , which is not defined for a hamiltonian unbounded above . curiously , it seems that we can circumvent this by going to the lorentzian theory , where an @xmath314 will appear in the evolution equation , giving an ordinary schrdinger kernel . ( the construction is similar to that of bulk fields in ads @xcite , which also seems to be essentially lorentzian . ) for the quadratic scalar theory , one finds @xmath315 where @xmath316 . to interpret this , note that we can interpret the rhs of eq . ( [ runir ] ) in terms of a path integral for @xmath90 with a free boundary condition and a boundary action from @xmath317 . the surface term in the saddle point equation for this path integral is @xmath318 this fixes the normalizable @xmath292 coefficient to the value @xmath319 . in other words , the effective path integral describes the alternate conformal quantization with a single trace perturbation ( higher order terms dropped in the second line of eq . ( [ rgback ] ) give rise also to a double - trace perturbation ) . this applies to all fields , so leads to a strange result : the cutoff path integral corresponds to taking the alternate quantization for all operators ; this is unphysical for @xmath320 , but that is to be expected in a regulated theory . however , it is not clear in what sense this is a regulator . a higher covariant derivative regulator would correspond to adding higher derivative single - trace terms , but of course this does not work in yang - mills theory . the alternate quantization corresponds to adding double - trace terms , which begin at quartic order in the matrix fields and so do not affect the propagator directly . in a sense we are treating the theory with alternate quantization as a starting point , and using its deformations to define the cutoff theory . theory ( with standard quantization ) as a starting point for defining wilsonian regulators of theories of less symmetry . it may be that one should revise the wilsonian strategy along these lines . ] we should note that the required step of running the rg backwards raises suspicions , though the theory with alternate quantization is actually an attractor for this flow in almost all directions , since most single - trace operators get negative dimensions . we leave further exploration of this idea for later work . we have identified structural parallels between the wilson and holographic rg s . the hope is that this will give a framework for making precise the connection ( [ ir ] ) between the cutoff theories , pursuing the various parallels we have noted . the general direction of this work , if it succeeds , is to produce a more local version of holography , where the boundary is pulled into the bulk . of course , the radial slicing is not locally special , and so it should then be possible to discuss more general subregions of the bulk , with some sort of equivalence principle relating different slicings . possibly a more modest goal is to extend the results of ref . @xcite , which found sufficient conditions for local physics to emerge in the bulk . that approach of that paper used detailed properties of conformal blocks , and appears to rather technical to extend . we believe that there should be a more flexible and physical way to organize the argument , perhaps using the tools we have developed here . we thank abhay ashtekay , tom faulkner , steve giddings , david gross , hong liu , mukund rangamani , eva silverstein , dam son , and especially don marolf for discussions and for comments on the manuscript . this work was supported in part by nsf grants phy05 - 51164 and phy07 - 57035 . j. m. maldacena , `` the large n limit of superconformal field theories and supergravity , '' adv . * 2 * , 231 ( 1998 ) [ int . j. theor . phys . * 38 * , 1113 ( 1999 ) ] [ arxiv : hep - th/9711200 ] . s. s. gubser , i. r. klebanov and a. m. polyakov , `` gauge theory correlators from non - critical string theory , '' phys . b * 428 * , 105 ( 1998 ) [ arxiv : hep - th/9802109 ] . e. witten , `` anti - de sitter space and holography , '' adv . theor . math . * 2 * , 253 ( 1998 ) [ arxiv : hep - th/9802150 ] . k. g. wilson , `` the renormalization group and critical phenomena , '' rev . phys . * 55 * , 583 ( 1983 ) . l. susskind and e. witten , `` the holographic bound in anti - de sitter space , '' arxiv : hep - th/9805114 . mansfield and d. nolland , `` one - loop conformal anomalies from ads / cft in the schroedinger representation , '' jhep * 9907 * , 028 ( 1999 ) [ arxiv : hep - th/9906054 ] . r. bousso , `` complementarity in the multiverse , '' phys . d * 79 * , 123524 ( 2009 ) [ arxiv:0901.4806 [ hep - th ] ] . a. a. tseytlin and k. zarembo , `` effective potential in non - supersymmetric su(n ) x su(n ) gauge theory and interactions of type 0 d3-branes , '' phys . b * 457 * , 77 ( 1999 ) [ arxiv : hep - th/9902095 ] ; c. csaki , w. skiba and j. terning , `` beta functions of orbifold theories and the hierarchy problem , '' phys . d * 61 * , 025019 ( 2000 ) [ arxiv : hep - th/9906057 ] ; a. adams and e. silverstein , `` closed string tachyons , ads / cft , and large n qcd , '' phys . rev . d * 64 * , 086001 ( 2001 ) [ arxiv : hep - th/0103220 ] ; a. dymarsky , i. r. klebanov , r. roiban , `` perturbative search for fixed lines in large n gauge theories , '' jhep * 0508 * , 011 ( 2005 ) . [ hep - th/0505099 ] ; m. li , `` a note on relation between holographic rg equation and polchinski s rg equation , '' nucl . b * 579 * , 525 ( 2000 ) [ arxiv : hep - th/0001193 ] . a. c. petkou , `` boundary multi - trace deformations and opes in ads / cft correspondence , '' jhep * 0206 * , 009 ( 2002 ) [ arxiv : hep - th/0201258 ] ; e. t. akhmedov , `` notes on multi - trace operators and holographic renormalization group , '' arxiv : hep - th/0202055 ; c. becchi , s. giusto and c. imbimbo , `` the wilson - polchinski renormalization group equation in the planar limit , '' nucl . b * 633 * , 250 ( 2002 ) [ arxiv : hep - th/0202155 ] . o. aharony , m. berkooz , e. silverstein , `` multiple trace operators and nonlocal string theories , '' jhep * 0108 * , 006 ( 2001 ) . [ hep - th/0105309 ] . e. witten , `` multi - trace operators , boundary conditions , and ads / cft correspondence , '' arxiv : hep - th/0112258 . t. hertog , g. t. horowitz , `` towards a big crunch dual , '' jhep * 0407 * , 073 ( 2004 ) . [ hep - th/0406134 ] . m. henningson and k. skenderis , `` the holographic weyl anomaly , '' jhep * 9807 * , 023 ( 1998 ) [ arxiv : hep - th/9806087 ] ; k. skenderis , `` lecture notes on holographic renormalization , '' class . * 19 * , 5849 ( 2002 ) [ arxiv : hep - th/0209067 ] . s. s. lee , `` holographic description of quantum field theory , '' nucl . b * 832 * , 567 ( 2010 ) [ arxiv:0912.5223 [ hep - th ] ] . wheeler , superspace and the nature of quantum geometrodynamics , battelle rencontres ( benjamin , 1968 ) ; i. r. klebanov and e. witten , `` ads / cft correspondence and symmetry breaking , '' nucl . phys . b * 556 * , 89 ( 1999 ) [ arxiv : hep - th/9905104 ] . m. berkooz , a. sever and a. shomer , `` double - trace deformations , boundary conditions and spacetime singularities , '' jhep * 0205 * , 034 ( 2002 ) [ arxiv : hep - th/0112264 ] ; w. mueck , `` an improved correspondence formula for ads / cft with multi - trace operators , '' phys . b * 531 * , 301 ( 2002 ) [ arxiv : hep - th/0201100 ] ; p. minces , `` multi - trace operators and the generalized ads / cft prescription , '' phys . d * 68 * , 024027 ( 2003 ) [ arxiv : hep - th/0201172 ] . a. sever and a. shomer , `` a note on multi - trace deformations and ads / cft , '' jhep * 0207 * , 027 ( 2002 ) [ arxiv : hep - th/0203168 ] . d. nickel and d. t. son , `` deconstructing holographic liquids , '' arxiv:1009.3094 [ hep - th ] . i. r. klebanov and e. witten , `` superconformal field theory on threebranes at a calabi - yau singularity , '' nucl . phys . b * 536 * , 199 ( 1998 ) [ arxiv : hep - th/9807080 ] . e. t. akhmedov , `` a remark on the ads / cft correspondence and the renormalization group flow , '' phys . lett . b * 442 * , 152 ( 1998 ) [ arxiv : hep - th/9806217 ] . e. alvarez and c. gomez , `` geometric holography , the renormalization group and the c - theorem , '' nucl . b * 541 * , 441 ( 1999 ) [ arxiv : hep - th/9807226 ] . l. girardello , m. petrini , m. porrati and a. zaffaroni , `` novel local cft and exact results on perturbations of n = 4 super yang - mills from ads dynamics , '' jhep * 9812 * , 022 ( 1998 ) [ arxiv : hep - th/9810126 ] ; j. distler and f. zamora , `` non - supersymmetric conformal field theories from stable anti - de sitter spaces , '' adv . theor . math . * 2 * , 1405 ( 1999 ) [ arxiv : hep - th/9810206 ] ; d. z. freedman , s. s. gubser , k. pilch and n. p. warner , `` renormalization group flows from holography supersymmetry and a c - theorem , '' adv . theor . math . phys . * 3 * , 363 ( 1999 ) [ arxiv : hep - th/9904017 ] . m. porrati and a. starinets , `` rg fixed points in supergravity duals of 4-d field theory and asymptotically ads spaces , '' phys . b * 454 * , 77 ( 1999 ) [ arxiv : hep - th/9903085 ] ; w. muck , `` spectral functions in holographic renormalization group flows , '' jhep * 0901 * , 060 ( 2009 ) [ arxiv:0811.2072 [ hep - th ] ] . g. w. gibbons , s. w. hawking and m. j. perry , `` path integrals and the indefiniteness of the gravitational action , '' nucl . b * 138 * , 141 ( 1978 ) . j. polchinski , `` renormalization and effective lagrangians , '' nucl . b * 231 * , 269 ( 1984 ) . i. heemskerk , j. penedones , j. polchinski and j. sully , `` holography from conformal field theory , '' jhep * 0910 * , 079 ( 2009 ) [ arxiv:0907.0151 [ hep - th ] ] ; i. heemskerk and j. sully , `` more holography from conformal field theory , '' [ arxiv:1006.0976 [ hep - th ] ] ; a. l. fitzpatrick , e. katz , d. poland and d. simmons - duffin , `` effective conformal theory and the flat - space limit of ads , '' arxiv:1007.2412 [ hep - th ] . j. de boer , e. p. verlinde and h. l. verlinde , `` on the holographic renormalization group , '' jhep * 0008 * , 003 ( 2000 ) [ arxiv : hep - th/9912012 ] ; j. de boer , `` the holographic renormalization group , '' fortsch . phys . * 49 * , 339 ( 2001 ) [ arxiv : hep - th/0101026 ] ; e. p. verlinde and h. l. verlinde , `` rg - flow , gravity and the cosmological constant , '' jhep * 0005 * , 034 ( 2000 ) [ arxiv : hep - th/9912018 ] . a. lawrence and a. sever , `` holography and renormalization in lorentzian signature , '' jhep * 0610 * , 013 ( 2006 ) [ arxiv : hep - th/0606022 ] . v. balasubramanian and p. kraus , `` spacetime and the holographic renormalization group , '' phys . lett . * 83 * , 3605 ( 1999 ) [ arxiv : hep - th/9903190 ] . a. lewandowski , m. j. may and r. sundrum , `` running with the radius in rs1 , '' phys . d * 67 * , 024036 ( 2003 ) [ arxiv : hep - th/0209050 ] . i. papadimitriou and k. skenderis , `` ads / cft correspondence and geometry , '' arxiv : hep - th/0404176 . verlinde and h. l. verlinde , `` rg - flow , gravity and the cosmological constant , '' jhep * 0005 * , 034 ( 2000 ) [ arxiv : hep - th/9912018 ] . u. ellwanger , `` brane universes , ads / cft , hamiltonian formalism and the renormalization group , '' arxiv : hep - th/0009006 . l. vecchi , `` multitrace deformations , gamow states , and stability of ads / cft , '' arxiv:1005.4921 [ hep - th ] ; n. iqbal , h. liu , `` universality of the hydrodynamic limit in ads / cft and the membrane paradigm , '' phys . rev . * d79 * , 025023 ( 2009 ) . [ arxiv:0809.3808 [ hep - th ] ] . i. bredberg , c. keeler , v. lysov and a. strominger , `` wilsonian approach to fluid / gravity duality , '' arxiv:1006.1902 [ hep - th ] . t. faulkner , h. liu , m. rangamani , `` integrating out geometry : holographic wilsonian rg and the membrane paradigm , '' [ arxiv:1010.4036 [ hep - th ] ] . a. jevicki and j. p. rodrigues , `` loop space hamiltonians and field theory of noncritical strings , '' nucl . b * 421 * , 278 ( 1994 ) [ arxiv : hep - th/9312118 ] . m. fukuma , n. ishibashi , h. kawai and m. ninomiya , `` two - dimensional quantum gravity in temporal gauge , '' nucl . b * 427 * , 139 ( 1994 ) [ arxiv : hep - th/9312175 ] . g. lifschytz and v. periwal , `` schwinger - dyson = wheeler - dewitt : gauge theory observables as bulk operators , '' jhep * 0004 * , 026 ( 2000 ) [ arxiv : hep - th/0003179 ] . r. brustein and s. p. de alwis , `` renormalization group equation and nonperturbative effects in string field theory , '' nucl . b * 352 * , 451 ( 1991 ) . e. t. akhmedov and e. t. musaev , `` an exact result for wilsonian and holographic renormalization group , '' phys . d * 81 * , 085010 ( 2010 ) [ arxiv:1001.4067 [ hep - th ] ] . s. arnone , t. r. morris and o. j. rosten , `` manifestly gauge invariant exact renormalization group , '' econf * c990809 * , 174 ( 2000 ) [ arxiv : hep - th/0606181 ] . t. banks , m. r. douglas , g. t. horowitz and e. j. martinec , `` ads dynamics from conformal field theory , '' arxiv : hep - th/9808016 ; + v. balasubramanian , p. kraus , a. e. lawrence and s. p. trivedi , `` holographic probes of anti - de sitter space - times , '' phys . d * 59 * , 104021 ( 1999 ) [ arxiv : hep - th/9808017 ] ; + i. bena , `` on the construction of local fields in the bulk of ads(5 ) and other spaces , '' phys . d * 62 * , 066007 ( 2000 ) [ arxiv : hep - th/9905186 ] ; + a. hamilton , d. n. kabat , g. lifschytz and d. a. lowe , `` holographic representation of local bulk operators , '' phys . rev . d * 74 * , 066009 ( 2006 ) [ arxiv : hep - th/0606141 ] . n. arkani - hamed , h. murayama , `` holomorphy , rescaling anomalies and exact beta functions in supersymmetric gauge theories , '' jhep * 0006 * , 030 ( 2000 ) . [ hep - th/9707133 ] .

the magnetic and thermodynamic properties of spin- heisenberg diamond chains are investigated in three different cases : ( a ) , , ( frustrated ) ; ( b ) , , ( frustrated ) ; and ( c ) , , ( non - frustrated ) , where the bond coupling stands for an antiferromagnetic ( af ) interaction , and for a ferromagnetic ( f ) interaction . the density matrix renormalization group ( dmrg ) technique is invoked to study the properties of the system in the ground state , while the transfer matrix renormalization group ( tmrg ) technique is applied to explore the thermodynamic properties . the local magnetic moments , spin correlation functions , and static structure factors are discussed in the ground state for the three cases . it is shown that the static structure factor shows peaks at wavevectors for different couplings in a zero magnetic field , which , however in the magnetic fields where the magnetization plateau with pertains , exhibits the peaks only at , and , which are found to be couplings - independent . the dmrg results of the zero - field static structure factor can be nicely fitted by a linear superposition of six modes , where two fitting equations are proposed . it is observed that the six modes are closely related to the low - lying excitations of the system . at finite temperatures , the magnetization , susceptibility and specific heat show various behaviors for different couplings . the double - peak structures of the susceptibility and specific heat against temperature are obtained , where the peak positions and heights are found to depend on the competition of the couplings . it is also uncovered that the xxz anisotropy of f and af couplings leads the system of case ( c ) to display quite different behaviors . in addition , the experimental data of the susceptibility , specific heat and magnetization for the compound cu(co)(oh) are fairly compared with our tmrg results .

0704.0150

low - dimensional quantum spin systems with competing interactions have become an intriguing subject in the last decades . among many achievements in this area , the phenomenon of the topological quantization of magnetization has attracted much attention both theoretically and experimentally . a general necessary condition for the appearance of the magnetization plateaus has been proposed by oshikawa , yamanaka and affleck ( oya ) @xcite , stating that for the heisenberg antiferromagnetic ( af ) spin chain with a single - ion anisotropy , the magnetization curve may have plateaus at which the magnetization per site @xmath18 is topologically quantized by @xmath19 , where @xmath20 is the spin , and @xmath21 is the period of the ground state determined by the explicit spatial structure of the hamiltonian . as one of fascinating models which potentially possesses the magnetization plateaus , the heisenberg diamond chain , consisting of diamond - shaped topological unit along the chain , as shown in fig . [ chain ] , has also gained much attention both experimentally and theoretically ( e.g. refs . drillon , sakurai , ishii , fujisawa , cucooh1,cucooh2,tks , aaa1,aaa2,aaa3,aaa4,aaa5,faf1,faf2 ) . it has been observed that the compounds , a@xmath16cu@xmath16(po@xmath22)@xmath22 with a = ca , sr@xcite , and bi@xmath22cu@xmath16v@xmath17o@xmath23sakurai can be nicely modeled by the heisenberg diamond chain . another spin-@xmath0 compound cu@xmath16cl@xmath24(h@xmath17o)@xmath17@xmath252h@xmath26c@xmath22so@xmath17 was initially regarded as a model substance for the spin-@xmath0 diamond chain @xcite , but a later experimental research reveals that this compound should be described by a double chain model with very weak bond alternations , and the lattice of the compound is found to be cu@xmath17cl@xmath27@xmath25h@xmath26c@xmath22so@xmath17 @xcite . recently , kikuchi _ _ et al.__@xcite have reported the experimental results on a spin-@xmath0 compound cu@xmath16(co@xmath16)@xmath17(oh)@xmath17 , where local cu@xmath28 ions with spin @xmath29 are arranged along the chain direction , and the diamond - shaped units consist of a one - dimensional ( 1d ) lattice . the @xmath30 magnetization plateau and the double peaks in the magnetic susceptibility as well as the specific heat as functions of temperature have been observed experimentally@xcite , which has been discussed in terms of the spin-@xmath0 heisenberg diamond chain with af couplings @xmath1 , @xmath2 and @xmath31 . on the theoretical aspect , the frustrated diamond spin chain with af interactions @xmath1 , @xmath2 and @xmath31 was studied by a few groups . the first diamond spin chain was explored under a symmetrical condition @xmath32@xcite . owing to the competition of af interactions , the phase diagram in the ground state of the spin-@xmath0 frustrated diamond chain was found to contain different phases , in which the magnetization plateaus at @xmath12 as well as @xmath30 are predicted@xcite . another frustrated diamond chain with ferromagnetic ( f ) interactions @xmath1 , @xmath4 and af interaction @xmath5 was also investigated theoretically , which can be experimentally realized if all angles of the exchange coupling bonds are arranged to be around @xmath33 , a region where it is usually hard to determine safely the coupling constants and even about their signs@xcite . despite of these works , the investigations on the heisenberg diamond spin chain with various competing interactions are still sparse . motivated by the recent experimental observation on the azurite compound cu@xmath34(co@xmath16)@xmath17(oh)@xmath17@xcite , we shall explore systematically the magnetic and thermodynamic properties of the spin-@xmath0 heisenberg diamond chain with various competing interactions in a magnetic field , and attempt to fit into the experimental observation on the azurite in a consistent manner . the density matrix renormalization group ( dmrg ) as well as the transfer matrix renormalization group ( tmrg ) techniques will be invoked to study the ground - state properties and thermodynamics of the model under interest , respectively . the local magnetic moments , spin correlation functions , and static structure factors will be discussed for three cases at zero temperature . it is found that the static structure factor @xmath9 shows peaks in zero magnetic field at wavevectors @xmath10 @xmath35 for different couplings , while in the magnetic fields where the magnetization plateau with @xmath12 remains , the peaks appear only at wavevectors @xmath13 , @xmath14 and @xmath15 , which are found to be couplings - independent . these information could be useful for further neutron studies . the double - peak structures of the susceptibility and specific heat against temperature are obtained , where the peak positions and heights are found to depend on the competition of the couplings . it is uncovered that the xxz anisotropy of f and af couplings leads the system without frustration ( see below ) to display quite different behaviors . in addition , the experimental data of the susceptibility , specific heat and magnetization for the compound cu@xmath16(co@xmath16)@xmath17(oh)@xmath17 are fairly compared with our tmrg results . the rest of this paper is outlined as follows . in sec . ii , we shall introduce the model hamiltonian for the spin-@xmath0 heisenberg diamond chain with three couplings @xmath1 , @xmath2 and @xmath36 , where three particular cases are identified . in sec . iii , the magnetic and thermodynamic properties of a frustrated diamond chain with af interactions @xmath1 , @xmath2 and @xmath3 will be discussed . in sec . iv , the physical properties of another frustrated diamond chain with f interactions @xmath1 , @xmath4 and af interaction @xmath5 will be considered . in sec . v , the magnetism and thermodynamics of a non - frustrated diamond chain with af interactions @xmath1 , @xmath5 and f interaction @xmath4 will be explored , and a comparison to the experimental data on the azurite compound will be made . finally , a brief summary and discussion will be presented in sec . the hamiltonian of the spin-@xmath0 heisenberg diamond chain reads @xmath37where @xmath38 is the spin operator at the @xmath39 site , @xmath40 is the total number of spins in the diamond chain , @xmath41 ( @xmath42 ) stands for exchange interactions , @xmath43 is the external magnetic field , @xmath44 and @xmath45 . @xmath46 represents the af coupling while @xmath47 the f interaction . there are three different cases particularly interesting , as displayed in fig . [ chain ] , which will be considered in the present paper : ( a ) a frustrated diamond chain with @xmath1 , @xmath2 , @xmath31 ; ( b ) a frustrated diamond chain with competing interactions @xmath1 , @xmath4 , @xmath48 ; and ( c ) a diamond chain without frustration with @xmath1 , @xmath5 , @xmath4 . it should be remarked that in the case ( c ) of this model , since the two end points of the @xmath2 bond represent the two different lattice sites , it is possible that @xmath1 and @xmath36 can be different , even their signs . the magnetic properties and thermodynamics for the aforementioned three spin-@xmath0 heisenberg diamond chains in the ground states and at finite temperatures will be investigated by means of the dmrg and tmrg methods , respectively . as the dmrg and tmrg techniques were detailed in two nice reviews@xcite , we shall not repeat the technical details for concise . in the ground - state calculations , the total number of spins in the diamond chain is taken at least as @xmath49 . at finite temperatures , the thermodynamic properties presented below are calculated down to temperature @xmath50 ( in units of @xmath51 ) in the thermodynamic limit . in our calculations , the number of kept optimal states is taken as @xmath52 ; the width of the imaginary time slice is taken as @xmath53 ; the trotter - suzuki error is less than @xmath54 ; and the truncation error is smaller than @xmath55 . figure [ aaasj](a ) manifests the magnetization process of a frustrated spin-@xmath0 heisenberg diamond chain with the couplings satisfying @xmath57 at zero temperature . the plateau of magnetization per site @xmath12 is observed . according to the oya necessary condition oya , the @xmath18 = @xmath58 plateau of a spin-@xmath0 heisenberg chain corresponds to the period of the ground state @xmath59 . beyond the magnetization plateau region , the magnetic curve goes up quickly with increasing the magnetic field @xmath60 . above the upper critical field , the magnetic curve shows a s - like shape . to further look at how this magnetization plateau appears , the spatial dependence of the averaged local magnetic moment @xmath61 in the ground states under different external fields is presented , as shown in fig . [ aaasj](b ) . it is seen that in the absence of external field , the expectation value @xmath62 changes its sign at every three sites within a very small range of @xmath63 because of quantum fluctuations , resulting in the magnetization per site @xmath64 . @xmath65 increases with increasing the magnetic field , and oscillates with increasing @xmath66 , whose unit of three spins is gradually divided into a pair and a single , as displayed in fig . [ aaasj](c ) . at the field @xmath67 , as demonstrated in fig . [ aaasj](d ) , the behavior of @xmath62 falls into a perfect sequence such as @xmath68 with @xmath69 and @xmath70 , giving rise to the magnetization per site @xmath12 . in addition , such a sequence remains with increasing the magnetic field till @xmath71 , implying that the @xmath12 plateau appears in the range of @xmath72 , as manifested in fig . [ aaasj](a ) . when the field is promoted further , the sequence changes into a waved succession with smaller swing of @xmath73 , as shown in fig . [ aaasj](e ) , which corresponds to the fact that the plateau state of @xmath12 is destroyed , and gives rise to a s - like shape of @xmath74 . it is noting that when the plateau state of @xmath12 is destroyed , the increase of @xmath18 at first is mainly attributed to a rapid lift of @xmath75 , and later , the double @xmath76 start to flimsily increase till @xmath77 at the saturated field . the physical picture for the above results could be understood as follows . for the @xmath12 plateau state at @xmath78 , we note that if an approximate wave function defined by@xcite @xmath79where @xmath80 denotes spin up ( down ) on site @xmath66 , is applied , one may obtain @xmath81 , @xmath82 , @xmath83 , giving rise to a sequence \ { ... , ( @xmath84 , @xmath85 , @xmath86 ) , ... } , and @xmath87 , which is in agreement with our dmrg results \ { ... , ( @xmath88 , @xmath88 , @xmath89 ) , ... } . this observation shows that the ground state of this plateau state might be described by trimerized states . let @xmath90 and @xmath91 be the lower and upper critical magnetic field at which the magnetization plateau appears and is destructed , respectively . for @xmath92 , the magnetization @xmath93 , namely , the system falls into the magnetization plateau state . for @xmath94 and @xmath78 , the magnetization curve shows the following behavior@xmath95 , \label{eqmh-1}\]]where @xmath96 , the parameters @xmath97 and @xmath98 . obviously , when @xmath99 , @xmath100 ; @xmath101 , @xmath102 . a fair comparison of eq . ( [ eqmh-1 ] ) to the dmrg results is presented in fig . [ aaaeqmh](a ) . for @xmath103 and @xmath78 , where @xmath104 is the saturated magnetic field , the magnetization curve has the form of @xmath105\ } , \label{eqmh-2}\end{aligned}\ ] ] where @xmath106 with @xmath107 the saturation magnetization , and @xmath108 , @xmath109 the parameters . one may see that when @xmath110 , @xmath102 ; @xmath111 , @xmath112 . a nice fitting to the dmrg result gives the parameters @xmath113 , @xmath114 , @xmath115 , @xmath116 , @xmath117 and @xmath118 , as shown in fig . [ aaaeqmh](b ) . it should be remarked that from the phenomenological eqs.([eqmh-1 ] ) and ( [ eqmh-2 ] ) , we find that , away from the plateau region , the magnetic field dependence of the magnetization of this model differs from those of haldane - type spin chains where @xmath119 . to explore further the magnetic properties of the frustrated spin-@xmath0 heisenberg diamond chain in the ground states with the couplings @xmath57 at different external fields , let us look at the static structure factor @xmath9 which is defined as @xmath120where @xmath121 is the wave vector , and @xmath122 is the spin correlation function in the ground state . as demonstrated in fig . [ aaasq](a ) , in the absence of the external field , @xmath9 shows three peaks : two at @xmath123 , @xmath124 , and one at @xmath125 , which is quite different from that of the spin @xmath29 heisenberg af chain , where @xmath9 only diverges at @xmath125 . as indicated by eq . ( [ sq ] ) , the peaks of @xmath126 reflect the periods of the spin correlation function @xmath127 , i.e. , the peaks at @xmath123 ( @xmath124 ) and @xmath128 reflect the periods of @xmath129 and @xmath130 for @xmath127 , respectively . as shown in fig . [ aaasq](b ) , in the absence of the external field , @xmath122 changes sign every three sites , which corresponds really to the periods of @xmath131 and @xmath130 . with increasing the magnetic field , the small peak of @xmath9 at @xmath125 becomes a round valley while the peak at @xmath123 ( @xmath124 ) continuously shifts towards @xmath132 ( @xmath15 ) with the height enhanced , indicating the corruption of the periods of @xmath129 and @xmath130 but the emergence of the new period @xmath133 @xmath134 for @xmath127 , as shown in fig . [ aaasq](c ) . at the field @xmath135 , two peaks shift to @xmath132 and @xmath15 respectively , and merge into the peaks already existing there , showing the existence of the period 3 for @xmath122 , as clearly displayed in fig . [ aaasq](d ) . the valley and peaks of @xmath9 keep intact in the plateau state at @xmath12 . when the plateau state is destroyed at the field @xmath136 , the peaks at @xmath132 and @xmath15 are depressed dramatically while the peaks at @xmath123 , @xmath124 and @xmath137 appear again with very small heights , revealing the absence of the period @xmath138 and the slight presence of the period @xmath130 and @xmath129 , as shown in fig . [ aaasq](e ) . at the field @xmath139 , all peaks disappear and become zero , except for the peak at @xmath13 , which is the saturated state . therefore , the static structure factor @xmath9 shows different characteristics in different magnetic fields@xcite . on the other hand , it is known that @xmath9 also reflects the low - lying excitations of the system . it is thus reasonable to expect that the low - lying excitations of the frustrated diamond chain will behave differently in different magnetic fields . to investigate the zero - field static structure factor @xmath9 in the ground state for the frustrated spin-@xmath0 diamond chains with various af couplings , the four cases with @xmath140 , @xmath31 , and @xmath141 , @xmath142 , @xmath143 , and @xmath144 are shown in figs . [ aaah0](a)-[aaah0](d ) , respectively . for @xmath141 , as shown in fig . [ aaah0](a ) , @xmath9 displays a sharp peak at @xmath125 when @xmath145 , and three peaks at @xmath13 , @xmath14 and @xmath146 when @xmath147 . it is shown from the ground state phase diagram @xcite that the system is in the spin fluid ( sf ) phase when @xmath140 , @xmath148 , @xmath145 , and enters into the ferrimagnetic ( fri ) phase when @xmath149 , @xmath141 , @xmath147 . for @xmath150 , as indicated in fig . aaah0(b ) , the incommensurate peaks exist , such as the case of @xmath151 , where the system is in the dimerized ( d ) phase @xcite . for @xmath152 , as manifested in fig . [ aaah0](c ) , @xmath9 has a sharp peak at @xmath125 and two ignorable peaks at @xmath153 and @xmath124 when @xmath36 @xmath154 ; three sharp peaks at @xmath13 , @xmath14 and @xmath15 at @xmath155 ; a round valley ( @xmath156 ) or a small peak ( @xmath157 ) at @xmath125 and two mediate peaks at @xmath153 and @xmath124 when @xmath158 . the system with @xmath140 and @xmath159 is in the d phase when @xmath160 , and in the sf phase when @xmath161 or @xmath162 @xcite . for @xmath163 revealed in fig . aaah0(d ) , the situations are similar to that of fig . [ aaah0](c ) , but here only the sf phase exists for the system with @xmath140 , @xmath163 and @xmath3 @xcite . it turns out that even in the same phase , such as the sf phase , the zero - field static structure factor @xmath9 could display different characteristics for different af couplings . in fact , we note that the exotic peak of @xmath9 has been experimentally observed in the diamond - typed compound sr@xmath16cu@xmath16(po@xmath22)@xmath22 @xcite . if the spin correlation function for the spin - s chain can be expressed as @xmath164 , where @xmath165 and @xmath166 are two parameters , its static structure factor will take the form of @xmath167which can recover exactly the @xmath9 of the spin - s aklt chain @xmath168 @xcite with @xmath169 and @xmath170 . eq.([seq ] ) has a peak at wavevector @xmath125 . by noting that the zero - field static structure factor @xmath171 for the frustrated diamond chains displays peaks at wavevectors @xmath172 @xmath35 for different af couplings , the spin correlation function @xmath122 could be reasonably divided into six modes @xmath173 or @xmath174 with @xmath175 and @xmath176 , whose contributions to the static structure factor should be considered separately@xcite . thus , the static structure factor for the present systems could be mimicked by a superposition of six modes , which leads to @xmath177}{\cosh ( 6\beta ) -\cos ( 6q ) } \notag \\ & & + c_{l}\frac{\cos ( lq)-\cos [ ( 6-l)q]}{1-\cos ( 6q)}]+\frac{1}{4 } , \label{sixeq}\end{aligned}\]]or @xmath178+\frac{1}{4 } , \label{sixeqpow}\]]respectively , depending on which phase the system falls into , where @xmath179 , @xmath180 and @xmath166 are couplings - dependent parameters . as presented in fig . [ aaaeq ] , the dmrg results of the static structure factor as a function of wavevector are fitted by eqs . ( [ sixeq ] ) and ( sixeqpow ) for the spin-@xmath0 frustrated heisenberg diamond chains with various af couplings in zero magnetic field . it can be found that the characteristic peaks can be well fitted by eqs . ( [ sixeq ] ) and ( sixeqpow ) , with only a slightly quantitative deviation , showing that the main features of the static structure factor for the present systems can be reproduced by a linear superposition of six modes . the fitting results give six different modes in general , as shown in fig . [ aaasq](b ) . to further understand the above - mentioned behaviors of the zero - field static structure factor , @xmath9 , we have applied the jordan - wigner ( jw ) transformation to study the low - lying excitations of the spin-@xmath0 frustrated heisenberg diamond chain with various af couplings ( see appendix a for derivations ) . it can be seen that the zero - field low - lying fermionic excitation @xmath181 behaves differently for different af couplings , as shown in figs . [ aaaexcita](a)-(d ) . obviously , these low - lying excitations are responsible for the dmrg calculated behaviors of s(q ) , where the positions of minimums of @xmath181 for different af couplings , as indicated by arrows in figs . [ aaaexcita](a)-(c ) , are exactly consistent with the locations of the peaks of zero - field static structure factor @xmath9 shown in figs . [ aaaeq](a)-(c ) , respectively , although there is a somewhat deviation for fig . [ aaaexcita](d ) and fig . aaaeq(d ) . it also shows that the six modes suggested by eqs . ( [ sixeq ] ) and ( [ sixeqpow ] ) is closely related to the low - lying excitations of the system . the magnetization process for the spin-@xmath0 frustrated diamond chain with @xmath149 , @xmath31 , and @xmath141 and @xmath130 is shown in fig . [ aaatmrg1](a ) and [ aaatmrg1](b ) , respectively , where temperature is fixed as @xmath182 . it is found that the magnetization exhibits different behaviors for different af couplings : a plateau at @xmath12 is observed , in agreement with the ground state phase diagram @xcite ; for @xmath148 , as shown in fig . [ aaatmrg1](a ) , the larger @xmath36 is , the larger the width of the plateau at @xmath12 becomes ; for @xmath152 , as presented in fig . [ aaatmrg1](b ) , the width of the plateau at @xmath12 becomes larger with increasing @xmath183 , and then turns smaller with increasing @xmath158 ; for @xmath183 , the larger @xmath2 , the larger the width of the plateau at @xmath12 ; for @xmath184 , the larger @xmath2 , the smaller the width of the plateau at @xmath12 . the saturated field is obviously promoted with increasing af @xmath36 and @xmath2 . figures [ aaatmrg1](c ) and [ aaatmrg1](d ) give the susceptibility @xmath185 as a function of temperature @xmath186 for the spin-@xmath0 frustrated diamond chain with @xmath140 , @xmath31 and @xmath2 @xmath187 and @xmath130 , respectively , while the external field is taken as @xmath188 . for @xmath148 , as shown in fig . [ aaatmrg1](c ) , the low temperature part of @xmath189 keeps finite when @xmath145 , and becomes divergent when @xmath190 . as clearly manifested in the inset of fig . [ aaatmrg1](c ) , @xmath191 is the critical value , which is consistent with the behaviors of static structure factor @xmath9 in fig . [ aaah0](a ) . for @xmath152 , as shown in fig . [ aaatmrg1](d ) , an unobvious double - peak structure at low temperature is observed at small and large @xmath36 such as @xmath192 and @xmath138 . the temperature dependence of the specific heat @xmath193 with @xmath140 , @xmath31 and @xmath141 and @xmath130 is shown in fig . [ aaatmrg1](e ) and [ aaatmrg1](f ) , respectively , while the external field is fixed as @xmath188 . for @xmath148 , as given in fig . [ aaatmrg1](e ) , a double - peak structure of @xmath194 is observed at low temperature for small and large @xmath36 such as @xmath195 and @xmath142 . the case with @xmath152 shown in fig . [ aaatmrg1](f ) exhibits the similar characteristics . thus , the thermodynamics of the system demonstrate different behaviors for different af couplings . as manifested in fig . [ aaah0 ] , the low - lying excitations behave differently for different af couplings . the double - peak structure of the susceptibility as well as the specific heat could be attributed to the excited gaps in the low - lying excitation spectrum@xcite . figure [ fafsj](a ) shows the magnetization process of a frustrated spin-@xmath197 heisenberg diamond chain in the ground states with the couplings @xmath198 . the plateau of magnetization per site @xmath12 is clearly obtained . to understand the occurrence of the magnetization plateau , the spatial dependence of the averaged local magnetic moment @xmath65 in the ground states at different external fields is calculated . it is seen that in the absence of the magnetic field , as presented in fig . [ fafsj](b ) , the expectation values of @xmath61 change sign every three sites within a very small range of @xmath199 , resulting in the magnetization per site @xmath100 . with increasing the field , the expectation values of @xmath61 increase , whose unit of three spins is gradually divided into a pair and a single , as displayed in fig . [ fafsj](c ) . at the field @xmath200 , as illustrated in fig . [ fafsj](d ) , the behavior of @xmath65 shows a perfect sequence such as @xmath201 with @xmath202 and @xmath203 , giving rise to the magnetization per site @xmath12 . in addition , the sequence is fixed with increasing the field until @xmath204 , corresponding to the plateau of @xmath12 . as the field is enhanced further , the double @xmath75 begin to rise , and the sequence becomes a waved series with smaller swing of @xmath205 as revealed in fig . [ fafsj](e ) , which corresponds to the plateau state at @xmath12 that is destroyed . it is noting that the increase of @xmath18 is mainly attributed to the promotion of double @xmath75 , as @xmath76 is already saturated until @xmath77 at the saturated field . as discussed above , the physical picture of the @xmath12 plateau state at @xmath198 can be understood by the following approximate wave function @xmath206by use of this wave function , we have @xmath207 , @xmath208 , @xmath209 , leading to a sequence of @xmath210 , and @xmath211 this is in agreement with our dmrg results @xmath212 . the static structure factor @xmath9 of the frustrated spin-@xmath0 heisenberg diamond chain with the competing couplings @xmath213 in the ground states is considered in different external fields . at zero field , shown in fig . [ fafsq](a ) , @xmath9 has three peaks at @xmath123 , @xmath124 and @xmath137 with mediate heights , which reflects the periods of @xmath129 and @xmath130 for @xmath122 , respectively . as shown in fig . fafsq(b ) , in the absence of the external field , @xmath127 changes sign every three sites , corresponding to the periods of @xmath129 and @xmath130 . with increasing the field , the peak at @xmath125 becomes flat with height depressed forwardly , while the peak at @xmath123 ( @xmath214 ) is divided into two peaks shifting oppositely from @xmath123 ( @xmath214 ) with height decreased , indicating the corruption of the periods of @xmath129 and @xmath130 and the emergence of new periods for @xmath127 , as shown in fig . [ fafsq](c ) . at the field @xmath200 , two shifting peaks have respectively reached @xmath132 and @xmath15 , and are merged with the existing peaks , showing the occurrence of period @xmath138 for @xmath122 , as clearly displayed in fig . [ fafsq](d ) . the flat and peaks keep constant during the plateau state at @xmath12 . when the plateau state is destroyed at field @xmath215 , the peaks at @xmath132 and @xmath15 are depressed sharply , revealing the decay of the period @xmath138 , as shown in fig . [ fafsq](e ) . at the saturated field of @xmath216 , all peaks disappear and become flat with the value zero , which is the saturated state . so , the static structure factor @xmath9 shows different characteristics in different magnetic fields . similar to the discussions in fig . [ aaasq ] , the low - lying excitations of this frustrated diamond chain would also behave differently in different magnetic fields . to further investigate the zero - field static structure factor @xmath9 in the ground state for the frustrated spin-@xmath0 diamond chains with various @xmath217 , @xmath4 and @xmath5 , two cases with @xmath218 , @xmath4 and @xmath219 and @xmath144 are illustrated in fig . [ fafh0](a ) and [ fafh0](b ) , respectively . for @xmath150 , @xmath9 shows a round peak at @xmath125 , two sharp peaks at @xmath153 and @xmath124 when @xmath220 , and a very sharp peak at @xmath13 when @xmath221 . for @xmath163 , @xmath9 shows three peaks at @xmath222 , @xmath137 and @xmath124 as @xmath220 , and a very sharp peak at @xmath223 and nearly ignorable peaks at @xmath153 and @xmath124 as @xmath221 . in general , the zero - field static structure factor @xmath9 shows different characteristics with different competing couplings , whose exotic characteristics could be experimentally observed in the related diamond - typed compounds . as shown in fig . [ fafeq ] , the dmrg results of the static structure factor as a function of wavevector are fitted by eq.([sixeq ] ) for the spin-@xmath0 frustrated heisenberg diamond chains with @xmath1 , @xmath4 and @xmath48 . it can be found that the characteristic behaviors can be nicely fitted by eq.([sixeq ] ) , with only a slightly quantitative deviation , showing that the main features of the static structure factor for the present systems can be captured by a superposition of six modes . it is consistent with the fact that the spin correlation function for the present systems has six different modes , as manifested in fig . [ fafsq](b ) . _ _ _ _ similar to the case ( a ) with all af couplings in the last section , the characteristics of zero - field @xmath9 for the spin-@xmath0 frustrated heisenberg diamond chain with @xmath1 , @xmath4 and @xmath5 can be further undersood in terms of the low - lying excitations of the system ( see appendix a ) . by means of the jw transformation , the zero - field low - lying fermionic excitation @xmath181 of the present case is calculated , as shown in figs . [ fafexcita](a)-(b ) . it is found that the zero - field low - lying fermionic excitation @xmath181 differs for different couplings , but the positions of minimums of @xmath181 for different couplings , as indicated by arrows in figs . [ fafexcita](a ) and ( b ) , appear to be the same . one may see that these positions coincide exactly with the locations of peaks of the zero - field static structure factor @xmath9 manifested in figs . [ fafeq](a ) and ( b ) , respectively , showing that our fitting equation is qualitatively consistent with the low - lying excitations of the system . figures [ faftmrg1](a ) and [ faftmrg1](b ) show the magnetization process for the spin-@xmath0 frustrated diamond chain at a finite temperature @xmath224 with @xmath218 , @xmath4 , and @xmath150 and @xmath144 , respectively . it is shown that the magnetization exhibits different behaviors for different @xmath1 , @xmath4 and @xmath5 . a plateau at @xmath225 is observed at small @xmath226 ; with a fixed @xmath2 , the larger latexmath:[$% and the plateau disappears when @xmath226 exceeds the critical value ; for a fixed @xmath36 , the larger @xmath2 , the wider the width of the plateau at @xmath12 ; the saturation field is obviously depressed with the increase of @xmath226 at a fixed @xmath2 , and is enhanced with the increase of @xmath2 at a fixed @xmath228 . figures [ faftmrg1](c ) and [ faftmrg1](d ) manifest the susceptibility @xmath185 as a function of temperature @xmath186 for the spin-@xmath0 frustrated diamond chain with @xmath218 , @xmath4 and @xmath150 and @xmath144 , respectively , where the external field is taken as @xmath229 . for @xmath230 , the low temperature part of @xmath231 keeps finite when @xmath232(c ) , @xmath233 is the critical value , which is in agreement with the behaviors of static structure factor @xmath9 shown in fig . [ fafh0](a ) . for @xmath163 , a clear double - peak structure of @xmath189 is obtained at @xmath234 . the temperature dependence of the specific heat @xmath193 with @xmath218 , @xmath235 @xmath236 and @xmath150 and @xmath144 is shown in figs . [ faftmrg1](e ) and [ faftmrg1](f ) , respectively , where the external field is fixed as @xmath229 . for @xmath150 , a double - peak structure of @xmath237 is observed for the case of @xmath238 . the case with @xmath163 shown in fig . [ faftmrg1](f ) exhibits the similar characteristics . it is also found that , owing to the competitions among @xmath239 , @xmath36 and @xmath2 , the thermodynamics demonstrate rich behaviors at different couplings . as reflected in fig . [ fafh0 ] , the low - lying excitations behave differently with various f interactions @xmath1 , @xmath36 and af interaction @xmath2 , while the excitation gaps could induce the double - peak structure in the susceptibility as well as in the specific heat@xcite . figure [ aafsj](a ) shows the magnetization process of a non - frustrated spin-@xmath0 heisenberg diamond chain in the ground states with the couplings satisfying @xmath241 . the plateau of magnetization per site @xmath12 is observed . the appearance of the magnetization plateau can be understood from the spatial dependence of the averaged local magnetic moment @xmath62 in the ground states under different external fields . in absence of the external field , the expectation values of @xmath61 change sign every one site with a waved swing within a very small range of @xmath199 , giving rise to the magnetization per site @xmath100 . under a finite field , every three successive spins have gradually cooperated into a pair and a single , as shown in fig . [ aafsj](c ) . at @xmath242 , as given in fig . [ aafsj](d ) , @xmath61 shows a perfect sequence such as @xmath201 with @xmath243 and @xmath244 , resulting in the magnetization per site @xmath12 . moreover , the sequence is fixed with the field increased until @xmath245 , corresponding to the plateau state of @xmath12 . as the field is increased further , double @xmath75 begin to increase , and the sequence becomes a waved succession with a smaller swing of @xmath205 , as manifested in fig . [ aafsj](e ) , which corresponds to the fact that the plateau state at @xmath12 is destroyed . it is observed that the increase of @xmath18 at first is mainly attributed to the speedy boost of double @xmath75 , and later , @xmath76 starts to increase weakly until @xmath77 at the saturated field . for this non - frustrated case with couplings @xmath241 , the obtained perfect sequence of @xmath246 for the @xmath247 plateau state could be understood by the following approximate trimerized wave function @xmath248 according to this function , we find @xmath249 , @xmath250 , @xmath251 , _ _ _ _ giving rise to a sequence of @xmath252 . it turns out that @xmath253 . this observation implies that the ground state of the plateau state can also be described by the trimerized states . the static structure factor @xmath9 of the non - frustrated spin-@xmath0 heisenberg diamond chain in the ground states with the couplings @xmath254 is probed in different external fields . as shown in fig . [ aafsq](a ) , in absence of the external field , @xmath9 shows a sharp peak at @xmath125 , similar to the behaviors of the @xmath29 heisenberg af chain , which reflects the period of @xmath130 for @xmath127 . as displayed in fig . [ aafsq](b ) , at zero external field , @xmath122 changes sign every one lattice site , corresponding to the period of @xmath130 . when the field is increased , the peak at @xmath125 becomes a flat with the height depressed greatly , while two new peaks with small heights at @xmath132 and @xmath15 appear , indicating the corruption of the period of @xmath130 and the emergence of the new period of @xmath138 for @xmath122 , as demonstrated in fig . [ aafsq](c ) . at the field @xmath242 , the peaks at @xmath13 , @xmath14 and @xmath15 become sharper . the flat and peaks of @xmath126 keep unchanged during the plateau state at @xmath18 = @xmath58 . when the plateau state is destroyed at the field @xmath71 , the peaks at @xmath255 and @xmath15 are suppressed dramatically , revealing the decay of the period @xmath138 , as shown in fig . [ aafsq](e ) . at the field @xmath256 , all peaks disappear , and become a flat with the value zero , except for the peak at @xmath121 @xmath257 , which is the saturated state . it can be stated that the static structure factor @xmath9 shows various characteristics in different magnetic fields . similar to what discussed in figs . [ aaasq ] and [ fafsq ] , the low - lying excitations of this non - frustrated diamond chain would also behave differently in different magnetic fields . the zero - field static structure factor @xmath9 in the ground state for the present system displays a peak at @xmath125 with different couplings , but whose static correlation function @xmath122 varies with the couplings . as illustrated in fig . [ aafh0 ] , only are the values of @xmath122 larger than zero presented for convenience . in order to gain deep insight into physics , for a comparison we also include the static correlation function for the @xmath29 heisenberg antiferromagnetic ( haf ) chain in fig . [ aafh0](a ) , whose asymptotic behavior has the form of @xcite @xmath258eq . ( [ equa ] ) is depicted as solid lines in figs . [ aafh0 ] . to take the finite - size effect into account , the length of the diamond chain is taken as @xmath259 , @xmath260 and @xmath261 , respectively . as revealed in fig . aafh0(a ) , the dmrg result of the @xmath29 haf chain with an infinite length agrees well with the solid line . [ aafh0](b ) shows that the static correlation function for the spin-@xmath0 diamond chain with frustrated couplings @xmath262 decays faster than that of the haf chain . compared with figs . [ aafh0](c)-(f ) , it can be found that all the static correlation functions for the spin-@xmath0 non - frustrated diamond chain with af interactions @xmath1 , @xmath2 and f interaction @xmath36 fall more slowly than that of the @xmath29 haf chain ; for fixed af interactions @xmath1 and @xmath2 , the static correlation functions drop more leisurely with increasing the f interaction @xmath226 ; for fixed af interaction @xmath1 and f interaction @xmath36 , the static correlation functions decrease more rapidly with increasing the af interaction @xmath2 . figures [ aaftmrg1](a ) and [ aaftmrg1](b ) show the magnetization process for the spin-@xmath0 non - frustrated diamond chain at a finite temperature @xmath263 with @xmath140 , @xmath4 , and @xmath141 and @xmath130 , respectively . it is found that the magnetization behaves differently with different af interactions @xmath1 , @xmath2 and f interaction @xmath36 . a plateau at @xmath12 is obtained at small @xmath226 ; for fixed @xmath1 and @xmath264 , the larger @xmath226 , the narrower the width of the plateau at @xmath225 , and after @xmath226 exceeds a critical value , the plateau at @xmath12 is eventually smeared out ; for fixed @xmath1 and @xmath36 , the larger @xmath2 , the wider the width of the plateau at @xmath12 ; the saturated field is obviously unchanged with changing the f interaction @xmath36 . the coupling - dependence of the spin-@xmath0 non - frustrated diamond chain with af interactions @xmath1 , @xmath2 and f interaction @xmath36 is similar to that of trimerized f - f - af chains@xcite . figures [ aaftmrg1](c ) and [ aaftmrg1](d ) present the susceptibility @xmath185 as a function of temperature @xmath186 for the spin-@xmath0 frustrated diamond chain with @xmath140 , @xmath4 and @xmath141 and @xmath130 , respectively , where the external field is taken as @xmath188 . a double - peak structure of @xmath231 is observed at small f interaction latexmath:[$% dependence of the specific heat @xmath193 with @xmath1 @xmath266 , @xmath4 and @xmath141 and @xmath130 is shown in figs . [ aaftmrg1](e ) and [ aaftmrg1](f ) , respectively , where @xmath188 . it is seen that , when @xmath2 is small , @xmath267 exhibits only a single peak ; when @xmath2 is large , a double - peak structure of @xmath267 is observed . in the latter case , the double - peak structure is more obvious for small f interaction @xmath226 , and tend to disappear at large latexmath:[$% with different af interactions @xmath1 , @xmath2 and f interaction @xmath36 . some magnetic materials show different behaviors under longitudinal and transverse magnetic fields , showing that the anisotropy plays an important role in the physical properties of the system . first , let us investigate the xxz anisotropy of the af interaction @xmath2 on the properties of the spin-@xmath197 non - frustrated diamond chain with the couplings @xmath269 for various anisotropy parameter defined by @xmath270 , where the @xmath271 axis is presumed to be perpendicular to the chain direction . for @xmath272 , the magnetization @xmath273 , susceptibility @xmath231 and specific heat @xmath267 are presented in figs . [ fig - tmrg23](a ) , ( b ) and ( c ) , respectively . with increasing @xmath274 , it is found that when the magnetic field @xmath60 is along the @xmath271 direction , the width of the magnetization plateau at @xmath12 as well as the saturation field are enlarged , while those are more increased for @xmath60 along the @xmath275 direction than along the @xmath271 direction ; the peak of the susceptibility @xmath231 for @xmath60 along the @xmath271 direction at lower temperature side is promoted , and the second round peak at high temperature side is depressed with a little shift , while @xmath231 for @xmath60 along the @xmath275 direction shows the similar varying trend ; the peak of the specific heat @xmath194 for @xmath60 along the @xmath271 direction at lower temperature side leaves almost unchanged , and the second round peak at high temperature side moves towards the higher temperature side , while @xmath267 for @xmath60 along the @xmath275 direction coincides with those for @xmath60 along the @xmath271 direction . for @xmath276 , the anisotropy just shows very reverse effect on the thermodynamic properties in comparison to what we discussed above . now let us discuss the effect of the xxz anisotropy of @xmath4 on the magnetic and thermodynamic properties of the spin-@xmath0 non - frustrated diamond chain with the couplings @xmath277 . recall that as the @xmath2 bond connects two different lattice sites , as shown in fig . [ chain ] , @xmath1 and @xmath36 can be different , even their signs . define a parameter @xmath278 to characterize the anisotropy as @xmath279 , where the @xmath271 axis is perpendicular to the chain direction . for @xmath280 , the magnetization @xmath273 , susceptibility @xmath231 and specific heat @xmath267 are depicted in figs . fig - tmrg23(d ) , ( e ) and ( f ) , respectively . with increasing @xmath278 , it is seen that the width of the plateau at @xmath12 for @xmath60 along the @xmath271 direction becomes slightly wider , while it goes smaller for @xmath60 along the @xmath275 direction ; the saturation field is not changed with @xmath278 along both directions ; the peak of @xmath231 for @xmath60 along the @xmath271 direction at lower temperature side is promoted , and the second round peak at higher temperature side is slightly depressed , while the situations along the @xmath275 direction are just reverse , namely , the peak at lower temperature side is depressed , and the second peak at higher temperature side is slightly promoted ; the peak of @xmath267 for @xmath60 along the @xmath271 direction at lower temperature side leaves almost unchanged , and the second round peak at higher temperature side moves slightly to the higher temperature side , while @xmath267 along the @xmath275 direction coincides with that along the @xmath271 direction . for @xmath281 , the situation just becomes reverse in comparison to what we discussed above . recently , kikuchi _ et al . _ @xcite have performed a nice measurement on a spin-@xmath0 diamond - chain compound cu@xmath16(co@xmath16)@xmath17(oh)@xmath17 , i.e. , azurite . they have observed the @xmath30 magnetization plateau , unambiguously confirming the previous theoretical prediction . the two broad peaks both in the magnetic susceptibility and the specific heat are observed . we note that in ref . @xcite , the experimental data at finite temperatures are fitted by the zero - temperature theoretical results obtained by the exact diagonalization and dmrg methods , while the result of the high temperature series expansion fails to fit the low - temperature behavior of the susceptibility . in accordance with our preceding discussions , by using the tmrg method , we have attempted to re - analyse the experimental data presented in ref . @xcite to fit the experiments for the whole available temperature region . our fitting results for the temperature dependence of the susceptibility @xmath185 of the compound cu@xmath16(co@xmath16)@xmath17(oh)@xmath17 are presented in fig . [ exp](a ) . for a comparison , we have also included the tmrg result calculated by using the parameters given in ref . @xcite . obviously , our tmrg results with @xmath282 and @xmath283 fit very well the experimental data of @xmath185 , and the two round peaks at low temperatures are nicely reproduced , while the result with @xmath284 obtained in ref . cucooh1 can not fit the low - temperature behavior of @xmath285 @xcite . on the other hand , the fitting results for the temperature dependence of the specific heat @xmath267 of the compound cu@xmath16(co@xmath16)@xmath17(oh)@xmath17 are shown in fig . [ exp](b ) . the lattice contribution , which is included in the raw experimental data in ref.@xcite , is subtracted according to @xmath286 , where @xmath287 is a parameter . obviously , our tmrg result with the same set of parameters @xmath288 and @xmath289 fits also remarkably well the experimental data of @xmath267 , and the two round peaks at low temperatures are nicely reproduced , while the result with @xmath290 given in ref . @xcite can not fit the low - temperature behavior of @xmath267 , even qualitatively . in addition , the sharp peak of @xmath267 experimentally observed at temperature around @xmath291 can not be reproduced by both sets of the coupling parameters , which might be a three - dimensional long - range ordering due to interchain interactions . the fitting results for the magnetization @xmath273 of the compound cu@xmath16(co@xmath292)@xmath17(oh)@xmath17 are shown in fig . [ exp](c ) . we would like to point out that the quantitative fitting by our above parameters to the width of the plateau is not so good , but the qualitative behavior is quite consistent with the experiments both in the transverse and longitudinal magnetic fields , say , @xmath293 , @xmath294 , and the saturation field is fixed along both directions , suggesting that our fitting parameters capture the main characteristics . it is worth pointing out that if the anisotropy ratio is increased up to @xmath295 with the same couplings @xmath282 , the width of the @xmath30 plateau for @xmath296 will be decreased to about one - half of that for @xmath297 . therefore , our calculations show that ( i ) the best couplings obtained by fitting the experimental data of the susceptibility for the azurite could be @xmath282 with the anisotropic ratio for the ferromagnetic interaction @xmath289 , where @xmath298 ; ( ii ) the compound may not be a spin frustrated magnet ; ( iii ) the double peaks of the susceptibility and the specific heat are not caused by the spin frustration effect , but by the two kind of gapless and gapful excitations owing to the competition of the af and f interactions . one might argue that for this diamond chain compound , from the point of the lattice distance it is unlikely that @xmath1 is af without xxz anisotropy while @xmath36 is f with strong xxz anisotropy . we may offer another possibility to support our findings , namely , the case of @xmath1 and @xmath36 with opposite signs is not excluded from the lattice structure of the compound . a linear relationship exists between the exchange energy and the metal - ligand - metal bridge angle : the coupling energy , positive ( ferromagnetic ) at angles near @xmath299 , becomes increasingly smaller ( more antiferromagnetic ) as the angle increases@xcite . as the ferromagnetic coupling @xmath36 is determined by fitting the experimental low - temperature behaviors of @xmath231 and @xmath267 , this fitting coupling parameters should not be impossible if one considers the angle of @xmath1 bridge to keep the antiferromagnetic coupling while the angle of @xmath36 bridge to induce the ferromagnetic coupling . on the other hand , we note that there is another compound with cu ions , cu@xmath17(abpt)(so@xmath22)@xmath17(h@xmath17o)@xmath25h@xmath300o , whose @xmath301 factors in xy plane are different from that in @xmath271 direction @xcite . besides , someone might argue that the condition @xmath302 is necessary to explain the double peak behavior of the diamond chain . in fact , such an argument is not necessarily true , as manifested in fig . [ aaftmrg1](c ) , where the double peaks of @xmath231 at low temperatures can also be produced with the parameters @xmath303 . in other words , the double - peak behavior of the diamond chain may not depend on whether @xmath304 , @xmath36 or not , but may be strongly dependent on the competition of af and f interactions , as discussed above . in this paper , we have numerically studied the magnetic and thermodynamic properties of spin-@xmath0 heisenberg diamond chains with three different cases ( a ) @xmath1 , @xmath2 , @xmath31 ( frustrated ) , ( b ) @xmath1 , @xmath4 , @xmath48 ( frustrated ) , and ( c ) @xmath1 , @xmath5 , @xmath4 ( non - frustrated ) by means of the dmrg and tmrg methods . in the ground states , the local magnetic moment , spin correlation function , and static structure factor are explored . the static structure factor @xmath9 at zero field shows peaks at wave vector @xmath13 , @xmath153 , @xmath14 , @xmath137 , @xmath305 and @xmath124 for different couplings , in which the peaks at @xmath13 , @xmath306 and @xmath15 in the magnetization plateau state with @xmath12 are observed to be couplings independent . the dmrg results of the zero - field static structure factor can be nicely fitted by a linear superposition of six modes , where two fitting equations are proposed . it is seen that the six modes are closely related to the low - lying excitations of the system . at finite temperatures , the magnetization , susceptibility and specific heat are calculated , which show various behaviors for different couplings . the double - peak structure of the susceptibility and specific heat can be procured , whose positions and heights are found to be dependent on competing couplings . it has been shown that the xxz anisotropy of f and af couplings can have remarkable effect on the physical behaviors of the system . in addition , the experimental susceptibility , specific heat and magnetization of the diamond chain compound cu@xmath16(co@xmath16)@xmath17(oh)@xmath17cucooh1 can be nicely fitted by our tmrg results . for the spin-@xmath0 frustrated heisenberg diamond chains with af couplings @xmath239 , @xmath2 and @xmath36 , the magnetization plateau at @xmath12 in the ground state coincides with a perfect fixed sequence of the averaged local magnetic moment such as @xmath68 with @xmath307 , which might be described by trimerized states . on the other hand , the static structure factor @xmath9 shows peaks at wave vectors @xmath308 , @xmath153 ( @xmath124 ) , and @xmath14 ( @xmath15 ) for different external fields and different af couplings . we note that the similar behavior of @xmath309 has been experimentally observed in diamond - typed compound sr@xmath16cu@xmath16(po@xmath22)@xmath22 @xcite . in addition , the dmrg results of the zero - field static structure factor can be nicely fitted by a linear superposition of six modes . it is observed that the six modes are closely related to the low - lying excitations of the present case . at finite temperatures , the magnetization @xmath273 , susceptibility @xmath231 and specific heat @xmath267 demonstrate different behaviors at different af couplings , say , the magnetization plateau at @xmath12 is observed whose width is found to be dependent on the couplings ; the double peak structure is observed for the susceptibility @xmath231 and specific heat @xmath267 as a function of temperature , and the heights and positions of the peaks are found to be dependent on the af couplings . for the spin-@xmath0 frustrated heisenberg diamond chains with f couplings @xmath239 , @xmath36 and af coupling @xmath2 , the magnetization plateau at @xmath12 in the ground state corresponds to a perfect fixed sequence of the averaged local magnetic moment such as @xmath310 with @xmath311 , which could be understood by trimerized states . the static structure factor @xmath9 shows peaks also at wave vectors @xmath13 , @xmath312 ( @xmath124 ) , and @xmath14 ( @xmath15 ) for different external fields and different f couplings @xmath1 , @xmath36 and af coupling @xmath2 , which is expected to be experimentally observed in the related diamond - type compound . in addition , the dmrg results of the zero - field static structure factor can be nicely fitted by a linear superposition of six modes with the fitting equations mentioned above . the six modes are closely related to the low - lying excitations of the system . at finite temperatures , the magnetization @xmath273 , susceptibility @xmath231 and specific heat @xmath267 demonstrate various behaviors for different couplings , namely , the magnetization plateau at @xmath18 = @xmath58 is observed whose width is found to depend on the couplings ; the double - peak structure is also observed for the susceptibility @xmath231 and specific heat @xmath267 , and the heights and positions of the peaks are found dependent on f couplings @xmath1 , @xmath36 and af coupling @xmath2 . for the spin-@xmath0 non - frustrated heisenberg diamond chains with af couplings @xmath1 , @xmath2 and f coupling @xmath36 , the magnetization plateau at @xmath12 in the ground state coincides with a perfect fixed sequence of the averaged local magnetic moment such as @xmath310 with @xmath313 , which could be understood by trimerized states . the static structure factor @xmath9 is observed to exhibit the peaks at wave vectors @xmath13 and @xmath14 ( @xmath15 ) for different external fields and different af couplings @xmath1 , @xmath2 and f coupling @xmath36 , which could be experimentally detected in the related diamond - type compound . in addition , it is found that the zero - field spin correlation function @xmath127 is similar to that of the @xmath29 heisenberg af chain . at finite temperatures , the magnetization @xmath273 , susceptibility @xmath189 and specific heat @xmath267 are found to reveal different behaviors for different couplings , i.e. , the magnetization plateau at @xmath12 is obtained , whose width is found to depend on the couplings ; the double - peak structure is observed for the temperature dependence of the susceptibility @xmath189 and specific heat @xmath267 , where the heights and positions of the peaks depend on different af couplings @xmath1 , @xmath2 and f coupling @xmath314 . the effect of the anisotropy of the af and f interactions on the physical properties of the non - frustrated heisenberg diamond chain is also investigated . for the case of the couplings satisfying @xmath269 , when the anisotropic ratio @xmath315 , it is found that the width of the plateau at @xmath12 , the saturation field , and the susceptibility @xmath231 show the same tendency , but quantitatively different , under the external field @xmath60 along the @xmath271 and @xmath275 directions , while the specific heat @xmath267 for @xmath60 along the @xmath271 direction coincides with that along the @xmath275 direction . for the case of the couplings satisfying @xmath277 , when the anisotropic ratio @xmath316 , it is seen that the width of the plateau at @xmath12 , the saturation field , and the susceptibility @xmath231 exhibit the opposite trends for @xmath60 along the @xmath271 and @xmath275 directions , while the specific heat @xmath267 for @xmath60 along the @xmath271 direction also coincides with that along the @xmath275 direction . for all the three cases , plateau states of @xmath12 are observed during the magnetization , whose static structure factor @xmath9 shows peaks at wavevectors @xmath13 , @xmath14 and @xmath15 . but in absence of the magnetic field , the static structure factor @xmath9 in the ground state displays peaks at @xmath13 , @xmath153 , @xmath14 , @xmath137 , @xmath15 , and @xmath124 for the frustrated case with @xmath1 , @xmath2 , @xmath31 ; peaks at @xmath13 , @xmath153 , @xmath128 , and @xmath124 for the frustrated case with @xmath1 , @xmath4 , @xmath48 ; and a peak at @xmath125 for the non - frustrated case with @xmath1 , @xmath48 , @xmath4 . in addition , the dmrg results of the zero - field static structure factor can be nicely fitted by a linear superposition of six modes , where the fitting equation is proposed . at finite temperatures , the double - peak structure of the susceptibility and specific heat against temperature can be obtained for all the three cases . it is found that the susceptibility shows ferrimagnetic characteristics for the two frustrated cases with some couplings , while no ferrimagnetic behaviors are observed for the non - frustrated case . the compound cu@xmath16(co@xmath16)@xmath17(oh)@xmath17 is regarded as a model substance for the spin-@xmath0 heisenberg diamond chain . the @xmath30 magnetization plateau and the two broad peaks both in the magnetic susceptibility and the specific heat have been observed experimentallycucooh1 . our tmrg calculations with @xmath282 and @xmath283 capture well the main characteristics of the experimental susceptibility , specific heat and magnetization , indicating that the compound cu@xmath16(co@xmath16)@xmath17(oh)@xmath17 may not be a spin frustrated magnet@xcite . in this appendix , the low - lying excitations of the spin-1/2 frustrated heisenberg diamond chain are investigated by means of the jordan - wigner ( jw ) transformation . the hamiltonian of the system reads @xmath317 where @xmath318 is the total number of spins in the diamond chain , @xmath46 ( @xmath42 ) represent the af coupling while @xmath47 the f interaction , and @xmath60 is the external magnetic field along the @xmath271 direction . in accordance with the spin configuration of the diamond chain , we start from the jordan - wigner ( jw ) transformation with spinless fermions @xmath319 , \notag \\ s_{j}^{z}&=&a_{j}^{+}a_{j}-\frac{1}{2},\end{aligned}\ ] ] where @xmath320 . because the period of the present system is 3 , three kinds of fermions in moment space can be introduced through the fourier transformations @xmath321 ignoring the interactions between fermions , the hamiltonian takes the form of @xmath322,\end{aligned}\ ] ] where @xmath323 , @xmath324=@xmath325 , @xmath326=@xmath327 , @xmath328=@xmath329 , @xmath330=@xmath331 , @xmath332=@xmath333 , and @xmath334=@xmath335 . via the bogoliubov transformation @xmath336 the hamiltonian can be diagonalized as @xmath337 the coefficients of the bogoliubov transformation can be found through equations of motion @xmath338 $ ] : @xmath339 for a given @xmath340 , the eigenvalues @xmath341 and eigenvectors @xmath342 can be numerically calculated by the driver @xmath343 of the @xmath344 , which is available on the website@xcite . figs.[aaaexcita ] show the zero - field low - lying fermionic excitation @xmath181 for the frustrated diamond chain with different af coupling , while figs.[fafexcita ] present the zero - field low - lying fermionic excitation @xmath345 for the frustrated diamond chain with @xmath1 , @xmath4 and @xmath5 . we are grateful to prof . d. p. arovas for useful communication . this work is supported in part by the national science fund for distinguished young scholars of china ( grant no . 10625419 ) , the national science foundation of china ( grant nos . 90403036 , 20490210 , 10247002 ) , and by the most of china ( grant no . 2006cb601102 ) . m. drillon , e. coronado , m. belaiche and r. l. carlin , j. appl . phys . * 63 * , 3551 ( 1988 ) ; m. drillon , m. belaiche , p. legoll , j. aride , a. boukhari and a. moqine , j. magn . magn . mater . * 128 * , 83 ( 1993 ) . the system under interest involves possibly ferrimagnetic , dimerized and spin liquid phases , leading to the spin - spin correlation functions exhibit different behaviors including exponential or power - law decaying . the constant @xmath180 in eq . ( [ sixeq ] ) characterizes the long - rang order in ferrimagnetic phase .

the recently introduced sommerfeld enhancement of the dark matter annihilation cross section has important implications for the detection of dark matter annihilation in subhalos in the galactic halo . in addition to the boost to the dark matter annihilation cross section from the high densities of these subhalos with respect to the main halo , an additional boost caused by the sommerfeld enhancement results from the fact that they are kinematically colder than the galactic halo . if we further believe the generic prediction of cdmthat in each subhalo there is an abundance of substructure which is approximately self - similar to that of the galactic halo , then i show that additional boosts coming from the density enhancements of these small substructures and their small velocity dispersions enhance the dark matter annihilation cross section even further . i find that very large boost factors ( to ) are obtained in a large class of models . the implications of these boost factors for the detection of dark matter annihilation from dwarf spheroidal galaxies in the galactic halo are such that , generically , they outshine the background gamma - ray flux and are detectable by the fermi gamma - ray space telescope .

0903.0413

in the standard flat , gaussian , adiabatic , and scale - invariant @xmath2cdmparadigm @xmath3 % of the matter density of the universe is in the form of dark matter ( dm ) @xcite . the properties of dark matter are largely unconstrained , although a weakly interacting massive particle ( wimp ) with a mass @xmath4 gev 1 tev has many attractive features , including that it leads to a relic density of dark matter that is remarkably close to the measured value @xcite . no interaction of dark matter either with itself or with standard model ( sm ) particles other than gravity has ever been observed . however , at the present day we can not exclude that the dark sector is brimming with a rich phenomenology at weak - interaction scales or lower . one exciting possibility is that dark matter self - annihilates at a level observable in the galactic halo today . detecting this self - annihilation seems unlikely considering the standard self - annihilation cross section ; @xmath5 @xmath6 s@xmath7 for consistency with the observed relic density . however , there are various mechanisms to boost the annihilation signal to a value larger by a few orders of magnitude , e.g. , clumpiness of the dark halo or a combination of coannihilations and sommerfeld enhancement @xcite . the possibility of detecting the dm annihilation signal has attracted much attention recently because of the anomalous excesses of high energy electrons and positrons reported by two different experimental groups : the payload for antimatter exploration and light - nuclei astrophysics ( pamela ) satellite , which saw a sharp upturn in the positron fraction @xmath8 from 10 to 100 gev @xcite , and the advanced thin ionization calorimeter ( atic ) group , which reported an excess over the expected background in the combined number of electrons and positrons at energies 500 to 800 gev @xcite . the pamela result on the positron fraction confirms earlier results @xcite and can be explained by dm annihilation @xcite . there is still a good chance that these anomalies can be obviated by conventional astrophysical effects , e.g. , pulsars . nevertheless , a plethora of models has been proposed in the last few months explaining the pamela and atic results in terms of dm annihilation into electron - positron pairs ( e.g. , @xcite ) , a description that also naturally accounts for other experimental anomalies , such as the `` wmap haze '' @xcite , egret observations of an excess of high - energy gamma - rays in the galactic center @xcite , and recent hess measurements of the flux of very high - energy electrons @xcite . model building is complicated by the fact that pamela did not observe a similar excess in the ratio of anti - protons to protons @xcite , which means that the coupling of the dmparticles to quarks must somehow be suppressed . the largest obstacle to modeling the pamela / atic signal with dmannihilation is the very large boost factor ( @xmath9 to @xmath0 ) with respect to the fiducial @xmath10 @xmath6 s@xmath7 required to produce an observable annihilation flux . the boost factor from the clumpiness of the dark matter halo is expected to be small , typically of the order 1 to 10 . therefore , a different mechanism must be responsible for the large boosts necessary . one such mechanism that has received much attention lately is the sommerfeld enhancement @xcite . the sommerfeld enhancement is a generic effect present whenever there is an attractive force acting between the dark matter particles , e.g. , a yukawa interaction or a gauge interaction through vector bosons @xcite . the main effect of this attractive force is to enhance the annihilation cross section @xmath11 with a factor proportional to @xmath12 ( @xmath13 ) ; this is known as `` @xmath14 '' enhancement . there are regions in the parameter space made up of the dark matter mass @xmath15 , the force carrier mass @xmath16 , and the force `` fine - structure constant '' @xmath17 coupling@xmath18/@xmath19 in which the enhancement is near a resonance , where it approximately behaves as @xmath20 . however , the enhancement saturates at a certain @xmath21 because of the finite range of the attractive force large ( @xmath22 to @xmath23 ) attractive forces with infinite range lead to a burst of dm annihilation in the first dm halos formed at @xmath24 @xmath25 100 200 which would lead to observable effects incompatible with measurements of the diffuse extragalactic gamma - ray background today and the cosmic microwave background @xcite . since the typical velocities of dark matter particles in the galactic halo today are of the order of the velocity dispersion of the halo , which is non - relativistic ( @xmath26 ) , the cross section of dark matter particles annihilating in the halo could be significantly enhanced by this sommerfeld effect . both observations @xcite as well as numerical simulations @xcite have firmly established that dark matter halos in cdm are not smooth structures , but that they are filled with a large amount of substructure . these substructures are both in the form of subhalos , self - bound objects orbiting in the galactic halo , and in the form of streams , relics of tidally disrupted subhalos . since these substructures are overdense regions in the halo , the dark matter annihilation cross section will be correspondingly higher : annihilation cross sections scale as the density squared . the effects of these kinds of substructure boosts have been studied extensively during the last decade . however , these subhalos are not only denser than the galactic halo in which they orbit , they are also kinematically colder than the galactic halo , with velocity dispersions that are typically two orders of magnitude smaller . this means that the sommerfeld enhancement in these subhalos can be many orders of magnitude larger than the sommerfeld enhancement acting in the galactic halo taken as a whole . in addition to this larger boost caused by the small velocity dispersion in the subhalos , these subhalos themselves are filled with an abundance of substructure which is approximately self - similar to that of the main halo . each of these even smaller substructure has a larger density than the subhalo in which they are orbiting ( quantified by the well established concentration - mass relationships for cdmhalos @xcite ) as well as a smaller velocity dispersion . this opens up the possibility that the dark matter annihilation signal in these already cold subhalos can be boosted even further by contribution of their substructure . in this paper we quantify the magnitude of this new substructure boost and discuss its implications for the detection of dark matter annihilation from subhalos in the galactic halos . as we will show , the total combined boost from sommerfeld enhancement and substructure can reach very large values ( 10@xmath27 to 10@xmath28 ) , and these boosts are only weakly dependent on the assumptions that we have to make about the abundance of substructure and the dm density profile . we use these total boosts together with published dm density profiles of some of the dwarf spheroidal satellite galaxies of the milky way to predict the total flux of gamma - rays from dm annihilation that would be detectable by the fermi gamma - ray space telescope ( _ fermi _ ) and find that in most of the models that posit the sommerfeld enhancement the dm annihilation signal from the dwarf spheroidals shines above the diffuse extragalactic gamma - ray background , which is the only background signal at the high galactic latitudes of these satellite galaxies . finally , we also consider the prospects for detecting smaller subhalos of the galaxy by computing annihilation fluxes of subhalos in the via lactea ii simulation . whether these subhalos will be detectable in the near future depends on the exact details of the sommerfeld enhancement . the simplest example of the sommerfeld effect that contains all of the generic features described in the introduction is that from an attractive yukawa type force mediated by a spin-0 boson @xmath29 between the dm particles , and we adopt that description here . the sommerfeld effect is a non - perturbative effect in the sense that in the language of quantum field theory it comes about from feynman diagrams in which the force carrier is exchanged many times before the annihilation actually happens , giving rise to so called `` ladder diagrams '' . however , we can study the sommerfeld effect in simple , non - relativistic quantum mechanics ( see the appendix in ref . we are instructed to solve the radial schrdinger equation ( in natural units ) @xmath30 subject to the boundary condition @xmath31 as @xmath32 . here @xmath33 and @xmath34 are the mass of the dark matter and the force carrier , respectively . using the substitution @xmath35 this becomes @xmath36 with boundary condition @xmath37 the sommerfeld enhancement @xmath38 is then given by @xmath39 equation ( [ eq : sommdiff2 ] ) shows that the sommerfeld enhancement is only a function of two variables , which can be chosen as @xmath40 and @xmath41 . the qualitative features of equation ( [ eq : sommdiff2 ] ) have been widely discussed and , therefore , we will merely mention them here without proof ( see , e.g. , refs . when the kinetic energy of the collision is much larger than the product of the force carrier mass and the coupling , then the yukawa interaction can be approximated as an attractive coulomb interaction , for which an analytical solution in terms of hypergeometric functions exists . the sommerfeld enhancement is simply given by @xmath42 from this we see why the sommerfeld enhancement is often referred to as a `` @xmath14 '' enhancement for small velocities @xmath43 . in the limit of a vanishing interaction ( @xmath44 ) the sommerfeld enhancement disappears as @xmath45 . the @xmath14 behavior , which diverges for small @xmath21 and leads to an overproduction of high - energy photons in the early universe @xcite , saturates as a consequence of the finite range of the yukawa interaction at a velocity @xmath46 and settles on a value @xmath47 . however , at some parameter combinations the yukawa potential has bound states , at which there are large boosts to the annihilation cross section which go approximately as @xmath48 . these resonance regions , however , are cut - off at small @xmath21 by the finite lifetime of the bound state . we expect these three different behaviors ( @xmath14 , saturation at low @xmath49 , and resonance effects from the existence of bound states ) to be generic among different models of the sommerfeld enhancement ( see also the discussion in the appendix of ref . @xcite ) . . because of the scaling properties of equation([eq : sommdiff2 ] ) the behavior of the sommerfeld enhancement as a function of @xmath50 can actually be read off along diagonals in the plane.,title="fig : " ] + . because of the scaling properties of equation([eq : sommdiff2 ] ) the behavior of the sommerfeld enhancement as a function of @xmath50 can actually be read off along diagonals in the plane.,title="fig : " ] equation ( [ eq : sommdiff2 ] ) can also be numerically integrated efficiently , by integrating the solution @xmath51 inwards from a radius satisfying @xmath52 in the coulomb regime ( @xmath53 ) this means that the initial value for the integration should be such that @xmath54 ; otherwise we should start at @xmath55 . figure [ fig : sommb34 ] shows the sommerfeld enhancement in the parameter plane made of @xmath56 and @xmath57 . all of the qualitative features we discussed above are present : in the coulomb regime ( lower part of the two panels ) we see that the enhancement scales as @xmath56 , saturating at @xmath58 when @xmath56 becomes small ; the fact that the upper left corner of both panels is approximately the same shows that the sommerfeld enhancement saturates at @xmath46 ; the sharp diagonal lines in the upper right of both panels show the resonance region , in which the enhancement is sharply peaked . because of the scaling properties of equation ( [ eq : sommdiff2 ] ) , shifting @xmath59 is equivalent to shifting both @xmath60 and @xmath61 by the same amount . therefore , the @xmath21 behavior of the enhancement can also be read off from these figures along diagonals . inspection of the resonance diagonals in the second panel of figure [ fig : sommb34 ] shows that the enhancement grows as @xmath20 in these regions . of course , when considering the sommerfeld enhancement of a halo of a certain velocity dispersion @xmath62 we need to average the sommerfeld enhancement over the distribution of relative velocities of that halo . we will approximate the one - particle velocity distribution of the halo as a single truncated maxwell - boltzmann distribution with velocity dispersion @xmath62 @xmath63 this approximation of a single maxwell - boltzmann distribution for the velocity distribution means that we assume that the halo has a constant velocity dispersion over the halo , i.e. , that the halo is isothermal . this is not necessarily the case indeed , it is not if we believe that the density profiles of dm halos are well fit by einasto or nfw profiles but it is not a bad assumption for intermediate distances from the halo s center , i.e. , between the innermost and outermost parts of the halo @xcite . a more detailed treatment of the sommerfeld enhancement in subhalos should take the non - uniform velocity dispersion in subhalos into account . note that in the pure @xmath14 regime the averaging will tend to sustain the @xmath14 behavior up to a factor @xmath64 , as we have that @xmath65 since the truncation does not significantly influence the result of these integrals . similarly , in the resonance region the @xmath66 behavior is conserved by the averaging . when the enhancement has leveled off the averaging has no effect . + in figure [ fig : sommavg ] we show this averaged sommerfeld enhancement for four different decades of velocity dispersion . the overall behavior of the sommerfeld enhancement seen in figure [ fig : sommb34 ] is indeed conserved well by the averaging : this has to do with the fact that the most probable relative velocity in a maxwell - boltzmann relative velocity distribution is the velocity dispersion . the sommerfeld enhancement attains very large values for subhalos with very small velocity dispersions , mostly , but not exclusively , in the resonance regions . this means that the smallest subhalos that exist in cdm will get large boosts to their dmannihilation from sommerfeld enhancement . the photon flux from dm annihilation from a solid angle @xmath67 along a given line - of - sight ( los ) is given by @xmath68 where @xmath69 is the threshold energy of the detector , @xmath70 denotes ( possibly ) different final states of the dm annihilation , @xmath71 is the spectrum of the annihilation to that state , @xmath72 is the annihilation cross section to state @xmath70 ( without the sommerfeld enhancement ) , @xmath73 is the sommerfeld enhancement which we have factored out of the cross section , and @xmath74 is the dm density along the line - of - sight . we emphasize that this framework also allows us to consider different final annihilation products , such as the electrons and positrons observed by pamela / atic . we will focus our analysis on the `` structure quantity '' @xmath75(@xmath76 ) , which we define here as @xmath77 this structure quantity does not only depend on the mass and internal properties of the dm halos in which we are interested , it depends on the particle physics parameters as well . we will keep all of the other factors in equation ( [ eq : flux ] ) fixed . however , in order to derive actual numbers for the photon flux from dm annihilation , we briefly describe the other quantities appearing in equation([eq : flux ] ) . for the annihilation cross section prior to any sommerfeld enhancement we will adopt the fiducial value @xmath78 @xmath6 s@xmath7 , and we choose a fiducial mass for the dm particle of @xmath15 = 700 gev . the energy spectrum of the dm annihilation depends on the model used to describe the dm and its interactions ; it is outside of the scope of this paper to provide a detailed treatment of this for different dm models . therefore , we will adopt an annihilation spectrum for dm annihilation into photons appropriate for the newer models that were proposed to explain the pamela / atic results . specifically , we use a gamma ray spectrum from a cascade annihilation model . in these models the dm annihilates into a new light degree of freedom which then in turn decays into leptons and photons . the injection spectrum per dm annihilation is given by @xcite @xmath79 in which @xmath80 is the electromagnetic coupling constant and @xmath81 is the mass of the leptons created in the process @xmath82 . integrating this spectrum using an energy range appropriate for _ fermi _ we find that @xmath83 we find a similar number if we integrate the spectrum from one cascade step . it is interesting to note that these photon spectra have fewer photons in them , about an order of magnitude less , than the annihilation spectra from neutralino annihilation which are often used @xcite . of course , since the energy spectrum contributes a multiplicative factor to the dmannihilation flux , the effect of different annihilation spectra can be incorporated by simply multiplying our results with the appropriate factor . since the photon flux from dm annihilations depends on the density squared of the dm the exact form of the inner density profile has important ramifications for the detectability of dmannihilation . the density profile of dark matter halos has been discussed extensively in the last few years . while the shape of the outer density profile seems to have converged in the various numerical simulations of dm halos and a consensus has been reached , the detailed shape of the inner density profile is still a matter of debate . the main point of contention is whether the density profile is cusped near the center of the dm halo , and if so , how steep this cusp is , or whether the density asymptotes to a constant value at the center . since this discussion has a non - negligible impact on the prospects for dm annihilation detection , we will briefly discuss the various plausible density profiles . the inner structure of dm halos is one of the key predictions of the cdm paradigm , and over a decade ago it was shown that the spherically averaged density profiles of dm halos are well described by a universal profile , the so - called nfw profile @xcite , depending only on the virial mass of the halo and its concentration . moreover , the concentration was shown to depend systematically on the halo mass @xcite . this profile was found to provide a good fit to many numerical simulations ( e.g. , @xcite ) . while there seems to be a consensus between different numerical simulations on the shape of the outer density profile , no agreement was reached on the inner density structure of the dm halos found in the simulations . the nfw profile has an inner slope of -1 , other studies , however , found evidence for steeper inner slopes @xcite , shallower inner slopes @xcite , or did nt find any convergence toward a power - law behavior @xcite . even the universality of the density profile has been brought into question @xcite . more recently , high - resolution numerical simulations have essentially ruled out the steepest cusps , such as moore s profile , which has an inner cusp a with logarithmic slope of -1.5 , however steeper inner cusps than the original nfw profile are still observed in numerical simulations @xcite . other recent simulations find inner slopes that are shallower than an nfw profile @xcite . it also has become apparent in recent years that the inner density profile of dm halos might not be described by a power - law divergence , but that , on the contrary , the logarithmic slope becomes ever shallower , leading to a cored density profile @xcite . therefore , density profiles such as the einasto profile have been shown to provide an excellent fit to the numerical simulations with the highest resolution yet @xcite . the implications of the exact form of the inner density profile have been widely discussed before in the context of various dmannihilation models ( e.g. , @xcite ) . roughly speaking , the annihilation flux is smaller for inner density profiles which are cored , and larger for inner density profiles which show a cuspy behavior . the steeper the slope of the inner cusp , the larger the dm annihilation flux ( for obvious reasons ) . since the core vs. cusp debate is still undecided we will consider dm distribution models of both types . the cuspy dm profile we will use is a generalized nfw profile ( gnfw ) : @xmath84 in which @xmath85 is a characteristic density and @xmath86 a characteristic ( scale-)radius . the original nfw profile is recovered for @xmath87 = 1 @xcite , which we will use as the fiducial model in what follows . varying the value of the inner logarithmic slope @xmath87 will allow us to describe the effect of steeper and shallower inner slopes than the nfw model . note that this functional form does _ not _ include the exact form of moore s profile ; this need not concern us since inner cusps as steep as moore s profile are excluded by the latest numerical simulations . the contribution of the dm density profile on the structure quantity @xmath75 can be simply evaluated in the context of any density model . as in ref . @xcite we assume that the distance to the subhalo @xmath88 is much larger than the scale - radius of the dmhalo . for an nfw profile 90% of the flux originates from the region within one scale - radius , and we will , conservatively , only take contributions to the dm annihilation flux coming from this region for all of the different profiles . therefore , we find that @xmath89 the normalization of this relationship depends on the logarithmic slope of the inner cusp and is given by @xmath90 the normalization for the nfw profile is @xmath91 and it depends strongly on the value of the logarithmic slope of the inner density profile , as can be seen in figure [ fig : normgammanfw ] . the bottom panel of fig . [ fig : normgammanfw ] shows the fraction of the dmannihilation flux coming from region within one scale - radius of the halo . for reasonable values of the inner slope this fraction is in the range 80 to 95% . relation of equation ( [ eq : lumrhosrsnfw ] ) for the gnfw profile . the functional dependence of the normalization on the exponent @xmath87 in the profile given in equation ( [ eq : nfw ] ) is given in equation ( [ eq : normgammanfw ] ) . bottom : fraction of the emission that comes from the region within @xmath86 . ] the einasto profile is characterized by an ever - decreasing logarithmic inner slope . it s functional dependence on the radius is given by @xmath92\ , .\ ] ] here , @xmath93 is the density at the radius @xmath94 where the local slope is -2 and @xmath56 is a shape parameter often fixed at a value @xmath95 @xcite . as above , assuming @xmath96 , calculating the flux originating from within the radius @xmath94 , we find that @xmath97 the normalization depends on the value of @xmath56 and is given by @xmath98\ , , \ ] ] where @xmath99 $ ] is the lower incomplete gamma function . the dependence of this normalization on @xmath56 is shown in figure [ fig : normgammaeinasto ] . in order to compare the einasto profile as described here to the gnfw profile , we can relate the parameters @xmath100 of the gnfw profile to the parameters @xmath101 of the einasto model by calculating the radius at which the logarithmic slope of the density versus radius profile equals -2 . thus , we find that @xmath102 note that the dm annihilation flux from the einasto profile , although it seems larger from figs . [ fig : normgammanfw ] and [ fig : normgammaeinasto ] , is actually smaller in general , since there is a factor of 2@xmath103 involved in going from one to the other . relation of equation([eq : lumrhosrseinasto ] ) for the einasto profile . the functional dependence of the normalization on @xmath56 in the profile given in equation ( [ eq : einasto ] ) is given in equation([eq : normgammaeinasto ] ) . bottom : fraction of the emission that comes from the region within @xmath94 . ] further on it will sometimes be useful to describe the dm density profiles by different parameters than a characteristic density and radius . another set of parameters is given by a concentration parameter and a mass . the virial radius @xmath104 of a halo is defined to be the radius within which the mean density of the halo is @xmath105 times the critical density @xmath106 , where @xmath107 @xcite . the virial mass is simply the mass contained within a sphere of radius @xmath104@xmath108 the concentration for both the gnfw profile as for the einasto profile is then defined by @xmath109 a characteristic mass of a halo is given by the virial mass defined above . the virial mass of the gnfw profile in terms of the characteristic density @xmath85 and the characteristic radius @xmath86 is @xmath110 in which the function @xmath111 is given by @xmath112 the function @xmath113 is the gauss hypergeometric function . in the @xmath114 case @xmath111 is given by the familiar form @xmath115 . the virial mass for the einasto profile in terms of the characteristic density @xmath93 and the characteristic radius @xmath94 is @xmath116\ , .\ ] ] in this last expression @xmath117 $ ] is the lower incomplete gamma function . the density can then be expressed in terms of the concentration by using the definition of the virial mass given in equation ( [ eq : virialmass ] ) . another pair of parameters that one can use to characterize the dmdensity profile is the maximum circular velocity @xmath118 and the radius at which this maximum occurs , @xmath119 . these parameters are especially well - suited when dealing with numerical simulations , since they can be readily read off from the raw simulation data . the circular velocity is given by @xmath120 in which @xmath121 and @xmath122 is defined in equation([eq : hypergeometric ] ) in the case of a gnfw profile and is given by @xmath123 $ ] in the case of an einasto profile . one can find @xmath119 by ( numerically ) solving the equation @xmath124 = 0\ , , \ ] ] and @xmath118 follows then immediately . as is customary , we will write the total structure quantity @xmath75(@xmath76 ) as a sum of a smooth component and a substructure component : @xmath125 the smooth contribution to the structure quantity is the product of the sommerfeld enhancement @xmath38 and the line - of - sight integral over the dm density squared , given for the different dm density profiles in equations ( [ eq : lumrhosrsnfw ] ) and ( [ eq : lumrhosrseinasto ] ) . therefore , @xmath126 in which @xmath127 is the luminosity that would come from the smooth component of the halo in the absence of sommerfeld enhancement . we can now write @xmath128 { \ensuremath{{\ensuremath{\mathcal{l}}}_{\mbox{{\footnotesize sm}},v = c}}}(m)\ , . \end{split}\ ] ] we will from now on write the substructure boost as @xmath129(m ) . then we have the following @xmath130 { \ensuremath{{\ensuremath{\mathcal{l}}}_{\mbox{{\footnotesize sm}},v = c}}}(m)\ , . \end{split}\ ] ] here we have introduced the subhalo mass function @xmath131 , which gives the abundance of subhalos of mass @xmath132 . before proceeding we should specify the upper and lower limits of the integral appearing in equation ( [ eq : boost ] ) . the integration over ever smaller substructures is cut - off at the size of the smallest subhalos . the mass scale at which this cut - off occurs is the thermal free - streaming scale , a dm model - dependent quantity , which in many dm models lies in the range 10@xmath133 to 10@xmath134 @xmath135@xcite . the upper limit to this integral is some fraction @xmath136 of the halo mass @xmath76 , since the abundance of substructure does not continue up to this mass scale . in the following we will set @xmath136 equal to 0.1 . in order to calculate the total boost from substructure to the annihilation cross section , we will write all of the quantities appearing in equation ( [ eq : boost ] ) as a function of the mass @xmath132 . for instance , from numerical simulations , the subhalo mass function is well constrained to follow a power - law type behavior : @xmath137 the value of the exponent @xmath138 is still under debate . the two highest resolution @xmath139-body simulations yet , the aquarius and via lacteaprojects , both find values for @xmath138 that put roughly equal amounts of mass per subhalo mass decade . while the via lactea simulation seems to prefer a value @xmath140 @xcite , which would lead to a logarithmically diverging total mass in substructures if the subhalo abundance were not cut off at the thermal free - streaming limit of the dark matter , the aquarius project prefers a value @xmath141 @xcite . as the aquarius project has made detailed convergence studies that support their claim for a value of @xmath138 that is shallower than -1 , we will adopt the value @xmath141 as our fiducial value , although even shallower slopes have also been observed @xcite . the same slope has also been observed in other simulations @xcite . with the exponent @xmath138 fixed , we turn to the normalization of the relation given in equation ( [ eq : massfraction ] ) . again , different high - resolution numerical simulations give different answers to this question as it relates to the abundance of subhalos of galactic subhalos . the problem of the normalization is complicated by the fact that numerical simulations have only recently begun to be able to resolve substructure in subhalos in a galaxy - size halo . while the simulations generally agree that the resolved substructure abundance of subhalos in the host halo adds up to @xmath142% of the total mass of the halo @xcite , the question of whether the substructure abundance in subhalos is just a scaled - down version of the substructure abundance in the host halo remains unresolved . one would not expect the abundance to be self - similar , since , while substructure in both the main host halo and subhalos gets diminished by tidal disruptions , substructure in the subhalos does not get replenished by the infall of halos from the field , as is the case for the host halo . nevertheless , many numerical simulations do find approximately self - similar substructure abundances @xcite . recently , the aquarius simulation has concluded from its simulation that the substructure abundance is not self - similar , although the deviation from the self - similar relation is small @xcite . therefore , we will adopt a normalization here which places 10% of the mass of the subhalo in the form of `` resolved '' substructure , where we adopt the definition that resolved substructure is in halos of mass @xmath143 . in the previous section we calculated @xmath144 for the different dm density profiles and found in all cases that @xmath145 for a characteristic density @xmath146 and radius @xmath147 , the exact definition of which depends on the specific dm density profile used . in order to find the mass dependence of @xmath144 , we use the description of the dm density profile in terms of virial mass @xmath148 and concentration @xmath149 . we then make use of the fact that many simulations have found a relation between the concentration of halos and their mass @xcite , such that everything depends only on the mass . this relation is well established over a large range in halo masses , and we will adopt the particular relationship found in ref . @xcite @xmath150 in good agreement with the results of other , similar analyses @xcite , although the normalization of the concentration - mass relation is about 15% lower than the original relation . however , the concentration of subhalos is generally larger than the concentration of field halos of the same mass as a result of tidal mass - loss @xcite . this is a small effect , of order 10% , and it does not play a large role in the results we obtain below . this concentration - halo mass concentration has not just been established in numerical simulations : statistical weak lensing analysis of luminous red galaxies and clusters from the sloan digital sky survey ( sdss ) has shown that the halo profiles of these objects are consistent with the nfw profile and mass - concentration relation given above @xcite . the only remaining quantity in equation ( [ eq : boost ] ) that we havent related to the mass of the subhalo yet , is the sommerfeld enhancement factor @xmath38 . the sommerfeld enhancement depends on the mass of the subhalo through the velocity dispersion . therefore , we need to relate the velocity dispersion of a dm subhalo to its mass . it is important to note that all we really need is an approximate relationship between the velocity dispersion and the halo mass which fits well the range of galactic size halos to a scale not much smaller than the mass scale of the smallest dwarf spheroidals known today , since in most cases the sommerfeld enhancement will saturate at velocity dispersions of the order of the velocity dispersion of the dwarf spheroidals . this means that the specific form that we adopt below for the velocity dispersion - mass scaling relation does not matter greatly . + + + on the scale of clusters there is a well - defined relationship between the mass of a dm halo and its one - dimensional velocity dispersion which one can derive from adiabatic scaling arguments @xcite given by @xmath151 this scaling relation can be easily derived in the case of an isothermal distribution function @xcite , since then @xmath152 from which equation ( [ eq : vdispscaling ] ) immediately follows . this scaling relation has been found to describe numerical simulations of cluster - sized halos extremely well @xcite . from recent numerical simulations it seems that we can use this relation even for subhalos of subhalos in a galaxy sized dark matter halo . in ref . @xcite it is shown that the relation between the maximum circular velocity of a subhalo and its mass is well fit by a power - law of the type @xmath153 using the circular velocity at @xmath94 as a measure of the halo s velocity dispersion gives @xmath154 . alternatively , we could use the circular velocity at the virial radius as a measure of the velocity dispersion . in the case of an nfw profile we have that @xmath155 in which @xmath156 in the relevant concentration range . using equation ( [ eq : concscaling ] ) , this gives @xmath157 therefore , we will use ( as a rough approximation ) the scaling relation given in equation ( [ eq : vdispscaling ] ) . we calibrate this relationship at the mass scale of the dwarf spheroidals by using the velocity and mass of canes venatici i from ref . @xcite . we define the total boost for a given subhalo mass as ( see equation ) @xmath158 in which @xmath159 is the sommerfeld enhancement for a halo of mass @xmath76 and @xmath160 is the substructure boost for a halo of that mass , as defined in eq . ( [ eq : boost ] ) . this total boost for subhalos of mass @xmath76 = @xmath161 @xmath135 and @xmath76 = @xmath162 @xmath135 calculated by following the procedure outlined in the previous subsection is shown in figure [ fig : subboost ] ( _ top panels _ ) . it is immediately obvious that the total boost factors can get very large in certain regions of the parameter space spanned by @xmath16/@xmath15and @xmath56 : boosts vary between @xmath163 to @xmath1 . these masses correspond to the typical masses of the classical and newly - discovered dwarf spheroidals ( dsphs ) and as such their velocity dispersions are of the order @xmath164 . the pattern of resonance regions and non - resonance regions that appears for the pure sommerfeld enhancement ( as shown in figure [ fig : sommavg ] ) is also apparent in these plots of the total boosts , however , it is noteworthy that the resonances seems to be more highly peaked than they were in the pure sommerfeld enhancement case , and that the boost the lower half of the parameter space , i.e. , for smaller values of @xmath16/@xmath15 , are higher than one would expect from the pure sommerfeld enhancement combined with a boost factor from pure density enhancements . this is confirmed by looking at the `` reduced boost '' defined as @xmath165 in which @xmath166 is the boost we would find if we turned off the sommerfeld enhancement in the calculation of the total boost , i.e. , the boost we would get if we set @xmath167 equal to one in equation([eq : boost ] ) . this @xmath168 corresponds to the boost factor that has been used before to estimate the total boost from density enhancements in dsphs and it is generally @xmath169(10 ) . this reduced boost therefore divides out the contribution of the sommerfeld enhancement of the top - level subhalo and the boost from the overdensities in lower - level halos , and it would be equal to one if there were no extra boost coming from the fact that the substructures are kinematically colder than the main subhalo . as we can see in figure [ fig : subboost ] ( _ bottom panels _ ) , in various parts of the yukawa coupling parameter space this reduced boost is equal to one such that there is no extra boosts caused by the lower velocity dispersion in the substructures . this is mainly the case for large values of the parameter @xmath16/@xmath15 , since the sommerfeld enhancement , in the 1/@xmath49 regime , levels off around @xmath170 , and since the top - level subhalo has a velocity dispersion of @xmath171 this means that the sommerfeld enhancement has mostly saturated at these large values of @xmath16/@xmath15 , even in regions of parameter space in which @xmath56 is small . indeed , in the 1/@xmath49 regime the sommerfeld enhancement is @xmath172 @xmath173 @xmath174 for @xmath175 , such that the sommerfeld enhancement has fully saturated over the whole @xmath56 range for large values of @xmath16/@xmath15 ( @xmath176 ) . however , the sommerfeld enhancement has not saturated in all regions of parameter space at this mass scale , and for smaller values of @xmath16/@xmath15and especially in resonance regions , the reduced boost can reach high values , signaling a large extra boost caused by the kinematically coldness of the substructure . for small values of @xmath16/@xmath15 we see that we can get boosts that are an order of magnitude to two orders of magnitude larger than what we would expect from naively combining the sommerfeld enhancement from the velocity dispersion of the subhalo with the boost from density enhancements alone . this can be easily understood as being a consequence of the fact that the sommerfeld enhancement has not saturated yet at the mass scale of the top - level subhalo , such that it can be boosted up to it saturation point by the contribution from the kinematically colder substructure . the largest values of the reduced boost are found in the resonance regions . in these regions the sommerfeld enhancement both grows faster with decreasing relative velocity as well as saturates at a much larger value of the enhancement . both of these effects conspire to give very large extra boosts . the resonance regions are clearly distinguishable in this plot of the reduced boost as 45@xmath177 lines . because the resonance regions are so narrow the apparent extra boosts in different resonances seems to vary greatly , but this is mostly caused by the finite resolution of this figure . extra boosts up to @xmath178 are obtained in these resonances , and these extra boosts increase along the resonances when decreasing @xmath56 , again caused by the higher saturation level for smaller @xmath57 values . the total boost factors obtained by assuming an nfw dark matter density profile do not depend much on the inner slope of the density profile . this is mainly because the calculation of the boost only considers _ relative _ luminosities , for which the difference in normalization of the @xmath179 relation for different inner slopes is irrelevant , and because the mass of the subhalo is only weakly affected by the value of the inner slope the function @xmath111 is only a weak function of the inner slope . this is confirmed by calculating the total boost for a @xmath180 @xmath135subhalo for a subhalo with a very shallow inner cusp as well as for a subhalo with a very steep inner slope . these are shown in figure [ fig : subboostprofiles ] ( _ top panels _ ) , and , from comparing this with the total boost shown in figure [ fig : subboost ] , it is clear that the inner slope does not affect the total boost caused by substructure , neither the overall magnitude of the effect nor the structure in the ( @xmath16/@xmath15,@xmath56 ) plane . the main difference in dm annihilation flux between the different dark matter density profiles is therefore given by the normalization of the @xmath181 relation , shown in figure [ fig : normgammanfw ] . if the dm density profiles of halos is better described by an einasto profile then the substructure boost is affected more strongly by the shape parameter @xmath56 which basically describes how cored the dm density profile is . changing @xmath56 has a greater impact on changing the overall dm density profile and therefore also plays a larger role in the relation between the characteristic density , the concentration , and the mass of the subhalo . the total boosts for a @xmath76 = @xmath161 @xmath135 subhalo is shown in figure [ fig : subboostprofiles ] ( _ bottom panels _ ) for two different values of the shape parameter @xmath56 which span the range found in numerical simulations ( for values of @xmath56greater than 0.2 the behavior basically saturates at that of @xmath56 = 0.2 ) . for @xmath56 = 0.1 the total boost is about the same as for the gnfw profile shape , which is understandable given that the einasto profile is more cuspy for smaller values of @xmath56 . the same dependence on the parameter @xmath16/@xmath15 as is seen for the gnfw profile is displayed for the einasto profile . in the case that @xmath56 is larger the substructure boost reaches slightly larger values than for the gnfwdensity profile . thus , as was the case for the gnfw profile , the impact of the different normalization of the @xmath182 for different values of @xmath56 will play a leading role in the difference between the different profile shapes . [ cols="<,>,>,<,>,<,>,<,>,<,>,<,>,<,<,^ " , ] + + in figure [ fig : subboostslope ] the effect of varying two more of the many parameters introduced in the previous section is shown . the slope of the subhalo mass function @xmath138 was set to -0.9 before , corresponding to the results from the latest high - resolution numerical simulations . in figure [ fig : subboostslope ] ( _ top panels _ ) we show the effect of both a shallower as well as a steeper subhalo mass function . the value of @xmath183 corresponds to the shallow slope found in ref . we see that the effect of such a shallow slope is to seriously reduce the substructure boost , which is simply a consequence of the fact that for a shallower slope less substructure is found in smaller and more concentrated clumps . a steeper slope such as @xmath140 , as was found in the recent via lacteasimulation @xcite leads to a substructure boost which is about an order of magnitude larger than for the fiducial @xmath184 value . the effect of the thermal free - streaming scale of dm on the substructure boost is shown in figure [ fig : subboostslope ] ( _ bottom panels _ ) . this effect is relatively large since the free - streaming scale sets the scale of the smallest and most concentrated dm halos and raising or lowering this scale therefore changes an important part of the contribution to the substructure boost . for very large values of the free - streaming scale such as @xmath185 @xmath135 , the substructure boost becomes about an order of magnitude smaller . when the free - streaming scale becomes very small , much of the substructure is in very small , cold clumps , such that the total boost becomes about an order of magnitude larger . besides detection of dm annihilation from the galactic center or the diffuse component from the galactic halo , there idea exists that the best candidates for dm annihilation detection might be the dwarf spheroidal satellite galaxies ( dsphs ) of our galaxy , since the dwarf galaxies are believed to be the most dark matter dominated structures in the universe , and the dm annihilation signal coming from them would be much easier to distinguish from background sources than , e.g. , would be the case for dm annihilation photons coming from the galactic center . that dsphs make good candidates for dmannihilation detection has been much discussed before , especially for the `` classical '' , or pre - sdss , dsphs@xcite . the annihilation signal coming from draco has been the subject of much research in recent years because of its relative proximity to the earth @xcite . the prospects for dm annihilation detection from dsphs has in general been found to be rather poor , with detection with _ fermi _ being unlikely @xcite . some of the dsphs have already been the subject of observational searches for dm annihilation and some limits on the dm annihilation signal have already been set @xcite . the dsphs discovered by sdss have also been suggested to make good candidates for dm annihilation detection , because of their very large mass - to - light ratios ( @xmath186 ) @xcite . in the context of the sommerfeld enhancement , these recently discovered dsphs are especially interesting , since they also have the smallest velocity dispersion of all of the known satellites of our galaxy . predicting the dm annihilation signal from these low - surface brightness galaxies is complicated by the dearth of kinematical tracers of the dm distribution and their contamination with background stars @xcite . nevertheless , detailed studies have constrained the dm density and velocity dispersion profiles of some of these dsphs@xcite . we will use some of the results on the dm density and velocity profiles found in these studies to show the extra effect of the sommerfeld enhancement on the predicted dm annihilation signal from some of these dsphs . we consider the dsphs given in table [ tab : dsphs ] here . eight of the dsphs are classical dwarf galaxies and we use the dm density profiles fitted to them in ref . @xcite . in this the dm density profiles of the dsphs were not completely determined by the available data a degeneracy between the parameters of the nfw profile used to describe the dm density and the king profile used to fit the luminous component of the galaxies was found and the concentration - mass relation from cdm@xcite was used to break the degeneracy . since these authors expressed their results in the form of the @xmath119 and @xmath118 parameters of the dm density profiles , we use the procedure outlined in section [ sec : dmprofile ] to convert these to @xmath85 and @xmath86 values . as these @xmath118 and @xmath119 parameters were fit to an nfw dm profile , we will only consider an nfw profile for the dsphs here , but the general conclusions about the importance of the exact profile shape also hold in these particular cases . the velocity dispersions of these dsphs are taken from ref . and we again model these subhalos as having a constant velocity dispersion . we also consider some of the newly discovered dsphs . in particular we discuss the prospects for dm annihilation detection from coma berenices and ursa major ii , and we use the best fit values to the available data from ref . since these authors directly quote @xmath85 and @xmath86 values we do not include @xmath119and @xmath118 values for these dsphs . we use the velocity dispersions determined for these galaxies in ref . @xcite . in order to calculate the substructure boost to the dm annihilation cross section we need the mass of the dsphs . many definitions of the mass of the galaxy could be used here , among the more appropriate ones is the mass contained inside the tidal radius , i.e. , the radius inside of which material in the dsph is safe from being tidally stripped by the galaxy . however , since we are considering the dm annihilation flux coming from within the radius @xmath94 we will use the mass contained within this radius to calculate the substructure boost . this is more appropriate since ( 1 ) it can be calculated straightforwardly from the dm density profile of the dsph and ( 2 ) the mass contained within @xmath94 is really the more relevant mass to calculate the substructure boost , since we are only interested in contributions to the dm annihilation flux coming from within the radius @xmath94 . because of the similarities between some of the dm density profiles of the dsphs in table [ tab : dsphs ] we do not show the predicted annihilation fluxes for all of the dsphs in figure [ fig : dsph ] , instead focusing on a representative sample of them . inspection of table [ tab : dsphs ] shows that annihilation flux of the dsphs not shown here will be similar to that of the ones shown in figure [ fig : dsph ] . at this point we can also make a connection to another quantity that is often used to describe the astrophysical contribution to the dmannihilation signal , @xmath187 . this quantity is generally defined as @xmath188 and is therefore equal to the line - of - sight integral over the density squared averaged over solid angle . we can define a slight variant of this which takes into account the sommerfeld enhancement @xmath189 in which @xmath38 is the sommerfeld enhancement averaged over the distribution of relative velocities . if we work this out for the dwarf spheroidals , using the expression from equation([eq : lumrhosrsnfw ] ) for the line - of - sight integral averaged over the solid angle spanned by the scale radius @xmath86 we find that this @xmath190 is approximately equal to the total boost @xmath191 , with a pre - factor @xmath169(1 ) . therefore in the case of the dsphs studied here the @xmath190 can be approximately read off from the top panels of figure [ fig : subboost ] . it is clear from figure [ fig : dsph ] that the dm annihilation flux predicted to be coming from the dsphs is very large in a large class of models . we have to stress here that our predicted fluxes are all for the particle physics specific parameters given in section [ sec : partphys ] , i.e. , @xmath192 = @xmath193 @xmath6 s@xmath7 , @xmath15 = 700 gev and a particular form for the dmannihilation spectrum . to predict the gamma - ray flux for a particular model with different values of these particle physics parameters one is required to calculate the particle physics parameter specific factor to the dm annihilation flux and appropriately rescale the values given here . in order to compare these signals to the expected background flux we will only consider contributions to the background flux of extragalactic gamma rays . although there are also contributions to the background flux from pion - decay , inverse compton scattering , and bremsstrahlung , at the high galactic latitudes at which the dwarf spheroidals are found these are all of the same order or smaller than the extragalactic background @xcite . we have integrated the extragalactic gamma ray flux as measured by egret over the energy range between 4 gev to 250 gev and multiplied it with the solid angle that the dsphs span on the sky . the extragalactic gamma ray flux is given by a power law @xcite @xmath194 where @xmath195 @xmath196 s@xmath7 sr@xmath7 gev@xmath7 , @xmath56 = 2.10 , and @xmath197 gev . we have included the background flux of extragalactic photons in figure [ fig : dsph ] . this background flux is in all cases of the order of 10@xmath198 to 10@xmath199 photons @xmath196 s@xmath7 , such that the dmannihilation boosted by substructure and sommerfeld enhancement outshine this background in most of the parameter space , and by a few orders of magnitude in most models . given that the flux above 100 mev over a 55 days time - span needed to separate a point source of dmannihilation from the background by _ fermi _ to give a 5@xmath200 detection is about 10@xmath201 photons cm@xmath7 s@xmath7 and that the necessary fluxes scale roughly as @xmath203 @xcite , the dm annihilation of these dsphs should be detectable by _ fermi _ in many of the models parametrized by ( @xmath204 ) . the ursa major ii dsph is the only dsph for which the dm annihilation dominates the background in all the regions of parameter space . this shows that the newly discovered , most - dark - matter - dominated satellite galaxies are possibly the best targets for the indirect detection of dm . we conclude that the prospects for detecting dm annihilation from the dsphs with _ fermi _ are very good in models that include a significant sommerfeld enhancement . /@xmath15 = 10@xmath205 and @xmath206 . bottom : histogram of the total boosts of the subhalos for the same values of @xmath16/@xmath15 and @xmath56 . ] /@xmath15 = 10@xmath205 and @xmath206 . for comparison we show the total gamma - ray flux from subhalos in the absence of sommerfeld enhancement as the dashed histogram and the top horizontal axis . ] but for @xmath16/@xmath15 = 10@xmath207 and @xmath208 . ] but for @xmath16/@xmath15 = 10@xmath207 and @xmath208 . ] besides the satellite galaxies that have been observed in the galactic halo cdm predicts that the halo is filled with hundreds of individual subhalos a fact which gives rise to the so - called `` missing satellites problem '' @xcite . each of these subhalos has a mass of the order of the masses of the observed dsphs or smaller , and thus will be kinematically cold with respect to the main halo . thus , the total dm annihilation signal coming from all of the substructure will be boosted by the sommerfeld enhancement described above , such that an appreciable dm annihilation signal could be expected from all of the substructure . in this section we get a feeling for the expected signal from dm substructures in a typical galaxy - sized halo . if the dm annihilation signal from subhalos without star formation could be detected this could give a clear signal that the dm exists and would open up a whole new arena for testing cdm . but for a resonance : @xmath16/@xmath15 = 10@xmath209 and @xmath210 . ] but for a resonance : @xmath16/@xmath15 = 10@xmath209 and @xmath210 . ] we use the distribution of dm subhalos from the via lactea ii numerical @xmath139-body simulation @xcite . via lactea ii is a state - of - the - art @xmath139-body simulation of a galaxy - sized halo since redshift 104.3 . it sampled the region of the main halo with 1.1 @xmath211 10@xmath28 particles , each with a mass of 4,100 @xmath135 . over 40,000 subhalos can be resolved within 402 kpc of the center of the main halo and 20,047 subhalos are detected which had a peak circular velocity larger than 4 km s@xmath7 at some time . the properties of the subhalos in this latter sample can be accurately established , i.e. , they are unaffected by resolution effects . we select from this sample the 9,830 subhalos which lie within the tidal radius of the main halo , which is approximately 462 kpc . we show the distribution of these subhalos in mass and distance from the sun in figure [ fig : vl23_1 ] . we have placed the sun at random locations at a distance of 8.5 kpc of the center of the main halo , corresponding to an earth - based - observer - like vantage point . the distribution of the distances in figure [ fig : vl23_1 ] is given for three of these random observers and the distribution is very similar for all three observers . we see that most of the subhalos have masses in the range 10@xmath212 to 10@xmath213 @xmath135 , with subhalos with masses of the order of the dsphs being relatively rare . the lower limit is of course set by the finite numerical resolution of the simulation and does nt correspond to a physical cut - off indeed , the physical cut - off lies at the thermal free - streaming scale . the distribution of the distance to the subhalos shows that the distribution of the subhalos is strongly radially anti - biased , with a sharp cut - off caused by our cut at the tidal radius of the main halo . this radial anti - bias is a consequence of the fact that there is much more space far away from the center of the galaxy , and because of the strong tidal forces near the center of the galaxy , which strongly affects halos which pass near the center of the galaxy . for each halo we can estimate the velocity dispersion using the mass - velocity dispersion relation given in equation([eq : vdispscaling ] ) and we can calculate the characteristic density @xmath85 and the characteristic length - scale @xmath86 from the values of @xmath118 and @xmath119 given for each subhalo , by the procedure outlined at the end of sec . [ sec : dmprofile ] . for a given set of particle physics parameters @xmath16/@xmath15 and @xmath56 this allows us to calculate the total boost to the dm annihilation cross section for each of the subhalos , and by using the assumptions about the particle physics given in sec . [ sec : partphys ] we can also calculate the photon flux from dm annihilation coming from each subhalo , and the total gamma - ray flux coming from all of the subhalos combined . we only cover some typical cases for the sommerfeld enhancement here which suffices to show the general trend . we consider the particle physics parameters @xmath16/@xmath15 and @xmath56 which give rise to the smallest sommerfeld enhancements in figures [ fig : vl23_2 ] and [ fig : vl23_3 ] , i.e. , @xmath16/@xmath15 = 10@xmath205 and @xmath56 = 0.001 . the top panel of figure [ fig : vl23_2 ] shows the distribution of gamma - ray fluxes from dm annihilation for the subhalos , while the bottom panel shows the distribution of the total boost factors . as can be seen , the total boost factors are quite small : the typical boost factor is only about 30 with the highest boost factors being about 100 . the total gamma - ray flux coming from subhalos is then also quite small in this case , as is shown in figure [ fig : vl23_3 ] , in which the total gamma - ray flux is shown for 1,000 random observers at a distance of 8.5 kpc from the galactic center . at most observation locations this total flux is about 10@xmath214 cm@xmath18 s@xmath7 , with the high - end tail not extending further than about half an order of magnitude from this . for comparison the dashed histogram in figure [ fig : vl23_3 ] shows the same distribution of total gamma - ray flux for different random observers in the absence of any sommerfeld enhancement . this confirms that the total gamma - ray flux with sommerfeld enhancement is about an order of magnitude larger than without sommerfeld enhancement . figures [ fig : vl51_1 ] and [ fig : vl51_2 ] show the same for a different extremum of the sommerfeld enhancement : the case in which @xmath16/@xmath15 = 10@xmath207 and @xmath56 = 0.1 . as we saw before , in this case the sommerfeld enhancement can reach very high values , and thus we see that the typical dm annihilation signal coming from dmsubhalos is large : the typical flux for one subhalo is equal to the total flux in the scenario described in the previous paragraph . the total boost factors for most of the halos are very large , of the order of 10@xmath216 , which gives rise to this very large photon - flux . the total predicted flux in this case is many orders of magnitude larger than the total flux in the other extremum of the sommerfeld enhancement and about seven orders of magnitude larger than the total gamma - ray flux in the absence of any sommerfeld enhancement . thus , the exact value of the parameters @xmath16/@xmath15 and @xmath56 , upon which the sommerfeld enhancement depends , matters a great deal for the predicted gamma - ray flux . finally , we look at the expected signal coming from subhalos when the sommerfeld enhancement is at a resonance , the same distribution as before are shown in figures [ fig : vlresonance1 ] and [ fig : vlresonance2 ] . again we find large boost factors and large individual and total gamma - ray fluxes , although they are smaller than the values found for the @xmath16/@xmath15 = 10@xmath207 and @xmath56 = 0.1 parameter set . however , since these values are near a resonance in a region of parameter space around which the predicted signal looks more like that for the parameter set @xmath16/@xmath15 = 10@xmath205 and @xmath56 = 0.001 , this again shows that whether dark halos , i.e. , subhalos without star formation , can be detected through dm annihilation products depends very sensitively on the particle physics parameters upon which the sommerfeld enhancement depends . in conclusion we can say that in a scenario in which there is sommerfeld enhancement there is much more hope to detect dark subhalos than in more conventional scenarios , but that the details of the prediction depend strongly on the underlying particle physics . the recently proposed sommerfeld enhancement to the dm annihilation cross section forces us to reconsider statements made in the past about the expected signal of dm annihilation from cold substructures in the halo of the galaxy . since the sommerfeld enhancement depends critically on the velocity dispersion of the substructure in question , as well as on the details of the particle physics model used to describe the attractive ( or , possibly , repulsive ) force that gives rise to the sommerfeld enhancement , the contributions from the specific particle physics model and the details of the dmdistribution in the substructure can no longer be as cleanly separated as they have been before . this means that in a scenario in which there is a significant sommerfeld enhancement , previous claims of `` particle physics independent '' conclusions about the dm annihilation signal from substructure ( e.g. , @xcite ) need to be reconsidered in this new framework . this complicates the analysis but it opens up new possibilities for detecting dm annihilation and learning about the properties of the dark sector . as we showed above , the combination of the sommerfeld enhancement and the substructure that is believed to exist down to small scales leads to very large boost factors because of the combined effect of the higher densities of small subhalos as well as the fact that they are kinematically colder than the larger halos . this leads to predictions for the dmannihilation gamma - ray flux that are much larger than was previously believed . as we showed in sec . [ sec : dsphs ] , in models in which there is a significant sommerfeld enhancement at the level of the galactic halo and its subhalos , the dm annihilation signal from the dwarf spheroidals outshines the extragalactic background flux and the predicted gamma - ray fluxes are so large that they should be detectable by _ fermi_. given a detailed knowledge of the density structure and velocity dispersion profile of the dwarf spheroidal galaxies , much can then be learned if we are able to detect the dm annihilation signal coming from these dsphs , by considering the relative signal and using the results of sec . [ sec : dsphs ] , or by using the detailed velocity dispersion profile of the dsphs @xcite . we have focused on the consequences of the sommerfeld enhancement for dm subhalos in a @xmath2cdm cosmology , using the latest results from high - resolution numerical @xmath139-body simulations to inform our assumptions about the abundance and properties of the substructure . under these assumptions we were able to show that there are very large boosts to the dm annihilation cross section from the combined effect of substructure boosts and sommerfeld enhancement in large parts of the parameter space consisting of the parameters of the attractive yukawa type force between the dm particles which is responsible for the sommerfeld enhancement . these regions of parameter space are in no way unrealistic . indeed , in one of the most straightforward interpretations of the atic / pamela results , the sharp cut - off around 1 tev seen in the atic spectrum is interpreted as being a consequence of the fact that the dark matter mass is about 700 gev . the absence of an excess in anti - protons similar to the excess in positrons seen by pamela can be explained by positing that the dark matter particles annihilate to an intermediate boson @xmath29 , which is lighter than a proton ( or even , lighter than the effective quark mass @xmath217 ) in order to kinematically forbid the annihilation into protons @xcite . this naturally leads to a @xmath16/@xmath15@xmath218 @xmath23 . the coupling constant of the new force in the dark sector is also generally @xmath219 @xcite . thus , from figure [ fig : subboost ] , we expect there to be a rather large boost to the dm annihilation cross section in these models . this leads to particularly large dm annihilation signals predicted for the dsphs ( see figure [ fig : dsph ] ) . we have not discussed the substructure boost to the dm annihilation signal from unresolved substructure , such as very small subhalos and dm tidal debris from disrupted subhalos , in the galactic halo . this is , however , interesting in the light of the sommerfeld enhancement , since it might lead to a different prediction about the relative magnitudes of the annihilation signal coming from the resolved galactic halo and unresolved halo , as well from the resolved substructures in the galactic halos . it is already clear that in the generic 1/@xmath49 sommerfeld enhancement the main halo , with its typical velocity dispersion of @xmath220 km s@xmath7 , will receive a much smaller enhancement than the dsphs , with their typical velocity dispersions of 10 km s@xmath7 . this , however , ignores the effects of saturation and resonances , which might be important in the specific particle physics model adopted for the sommerfeld enhancement . additionally , while the velocity dispersion of the galactic halo might be of the order of 100 km s@xmath7 , the actual velocity dispersion relevant to the sommerfeld enhancement might be much smaller for halo dm particles . if a large fraction of the dmin the galactic halo is part of a cold stream and various numerical simulations predict that large parts of the halo , if not all of the halo , are made up of such cold streams with typical velocity dispersions of the order of 1 km s@xmath7 @xcite then , in addition to any boost caused by density enhancements in the stream @xcite , the boost from the sommerfeld enhancement could also play a large role , as the relevant relative velocity of dm particles is set by the velocity dispersion of particles in the stream , and not by the velocity dispersion of the streams in the halo , which is what the 100 km s@xmath7 galactic halo velocity dispersion corresponds to in these scenarios . thus , the expected dmannihilation signal from the main halo could be much larger than that naively expected for a halo with a velocity dispersion of 100 km s@xmath7 .

we compute the energy dependence of the-integrated cross section of directly produced quarkonia in collisions at next - to - leading order ( nlo ) , namely up to , within nonrelativistic qcd ( nrqcd ) . our analysis is based on the idea that the-_integrated _ and the-_differential _ cross sections can be treated as two different observables . the colour - octet nrqcd parameters needed to predict the-integrated yield can thus be extracted from the fits of the-differential cross sections at mid and large . for the first time , the total cross section is evaluated in nrqcd at full nlo accuracy using the recent nlo fits of the-differential yields at rhic , the tevatron and the lhc . both the normalisation and the energy dependence of the , and , we obtained , are in disagreement with the data irrespective of the fit method . the same is true if one uses cem - like colour - octet nrqcd parameters . if , on the contrary , one disregards the colour - octet contribution , the existing data in the tev range are well described by the contribution in the colour - singlet model which , at , however shows an unphysical energy dependence . a similar observation is made for . this calls for a full nnlo or for a resummation of the initial - state radiation in this channel . in any case , past claims that colour - octet transitions are dominantly responsible for low- quarkonium production are not supported by our results . this may impact the interpretation of quarkonium suppression in high - energy proton - nucleus and nucleus - nucleus collisions .

1504.00317

understanding the production mechanism of low-@xmath0 quarkonia in nucleon - nucleon collisions is of fundamental importance to properly use them as probes of deconfinement or collectivity in heavy ion collisions . indeed , most of the analysis of quarkonium production in nucleus - nucleus collisions are carried out on the bulk of the cross section , namely at low @xmath0 . recently , comparisons between alice data @xcite without @xmath0 cut and cms data @xcite with @xmath0 cut in pbpb collisions at @xmath8 tev showed an unexpected suppression pattern of the charmonia , at variance with the simple picture of quarkonium melting in deconfined quark matter @xcite . however , to properly interpret this observation , it is essential to rule out the possibility that a part of the effect observed could be due to a difference in the production mechanism in individual nucleon - nucleon collisions at low and at larger @xmath0 . the propagation of a colour - octet pair in a deconfined medium certainly differs from that of a colour - singlet pair ; this can result into a different nuclear suppression ( see _ e.g. _ @xcite ) . on the contrary , as regards the bottomonia , the observation of the expected sequential - suppression pattern has been claimed by cms @xcite . further , the effect of normal nuclear matter may also significantly depend on how the pair is produced : the recently revived fractional energy loss @xcite would for instance act on long - lived colour - octet states and probably differently if the heavy quark state is already produced colourless at short distance , as postulated in the csm @xcite . saturation effects in @xmath9 collisions also do depend on the colour state of the perturbatively produced heavy - quark pair @xcite despite the possibility that nrqcd factorisation would not hold at low @xmath0 , several nrqcd analyses have thus been carried earlier to evaluate the impact of the colour - octet channels to the @xmath0-integrated @xmath3 yields @xcite . a first study of the impact of initial state radiations ( isr ) on the very low @xmath0 @xmath3 s and @xmath10 s was recently carried out successfully in nrqcd @xcite yet at the cost of introducing additional non - perturbative parameters . whereas , based on an analysis of the sole early rhic data , cooper _ et al . _ argued @xcite that the universality of nrqcd was safe and that colour - singlet contributions to the @xmath0-integrated @xmath3 yields were negligible , the global analysis of maltoni _ et al . _ at nlo showed @xcite that the colour - octet long - distance matrix elements ( ldmes ) required to describe the total prompt @xmath3 yield from fixed - target energies to rhic were _ one tenth _ of that expected from the leading - order fit of the @xmath0-differential cross sections at tevatron energies . such fits of the @xmath0-differential @xmath3 cross sections have recently been extended to nlo _i.e . _ one - loop accuracy on the prompt @xmath3 yields some of them focusing on the larger @xmath0 data and explicitly including the feed - down contributions @xcite , some enlarging the analysis beyond hadroproduction and including rather low-@xmath0 data @xcite and on the @xmath11 yields @xcite . thanks to these studies , we can significantly extend the existing nrqcd studies of the @xmath0-integrated cross section by combining in a coherent manner , the hard parts or wilson coefficients up to @xmath2 , first derived by @xcite and which we have systematically checked with fdc @xcite , with the nrqcd matrix elements fitted at nlo on the @xmath0 dependence of the yields . one can indeed consider the _ @xmath0-integrated _ and the _ @xmath0-differential _ cross sections as two different observables their born contributions involve different diagrams and such a procedure is not a all trivial physics - wise . as we detail later , our results show that the data do not allow for a global description of both the @xmath0-integrated and @xmath0-differential quarkonium yields . as a point of comparison , we also had a look at colour - evaporation - model - like ( cem ) predictions derived from nrqcd following the work of @xcite and we found out that it can not reproduce @xmath0-integrated yields using the ldmes obtained following the relations of @xcite after identifying the minimal singlet transition to that of the csm . _ a contrario _ , results obtained from the traditional cem implementation at one loop do not show a similar issue . the inability of colour - octet dominance within nrqcd to provide a global description of both low and large @xmath0 data is in line with the recent findings @xcite that the sole lo colour - singlet contributions are sufficient to account for the magnitude of the total cross section and its dependence in rapidity , @xmath12 , from rhic , tevatron all the way to lhc energies . any additional contribution in this energy range creates a surplus as compared to data . however , as we also study in a dedicated section , the total nlo csm cross section shows a weird energy dependence at lhc energies . the problem is striking for the @xmath3 , less for the @xmath10 . in any case , one should be very careful in interpreting these results . in particular , such nlo results can not be considered as a improvement of the lo ones . we also observe the same issue for @xmath13 and @xmath14 production for which there is no final - state - gluon radiation at born order . we are therefore tempted to attribute this behaviour to large loop contributions which become negative at low @xmath0 , rather than to specific effects related to the @xmath15 production _ per se_. a quick inspection of the rather concise one - loop results @xcite for @xmath13 and @xmath14 production in the tmd factorisation formalism unfortunately does not reveal any obvious negative contributions and does not help in the understanding of this rather general issue of quarkonium production in collinear factorisation . the same problems appear with some co channels as well and may therefore cast doubts on the reliability of our results in the @xmath16 region where some contributions shows a strange behaviour in particular at lhc energies . at this stage , we are not able to conclude from our observations whether these problems are indicative of the break down of nrqcd factorisation at low @xmath0 and low @xmath17 or at low @xmath17 only . however , for sure , none of the above observations can reasonably support the idea that co transitions are dominant at low @xmath0 . such a conclusion would be for the least premature . the structure of the paper is as follows . in section [ sec : x - section ] , we detail the procedure to evaluate the @xmath0-integrated yield at one - loop accuracy in nrqcd and we explain the idea underlying this first complete one - loop analysis . in section [ sec : ldmes ] , we explain our selection of ldmes determined at nlo . in section [ sec : data ] , we briefly comment on the existing world data sets for @xmath3 , @xmath18 and @xmath5 analysis of gong _ et al . _ @xcite treats the @xmath5 , @xmath19 and @xmath20 , the lack of knowledge on the @xmath21 and @xmath22 yields and their corresponding feed down to @xmath19 and @xmath20 makes the analysis of their _ direct _ yield delicate ; we have thus decided not to consider these in the present study . our choice has been confirmed by the recent lhcb result @xcite that a large fraction of the @xmath19 and @xmath20 yield actually comes from @xmath21 and @xmath22 decays up to 40% in the @xmath20 case . ] . we also explain how we estimate the direct yields . in section [ sec : results_nrqcd ] , we show and discuss our results for the first full one - loop nrqcd analysis of quarkonium hadroproduction . to go further in the interpretation of some of our results , we discuss in section [ sec : results_cem ] the prediction of nrqcd using cem - like ldmes . this is also compared with the conventional approach based on quark - hadron duality . section [ sec : results_csm ] focuses on csm results both for the @xmath15 states considered here and for the @xmath23 states for which analytical results exist . our conclusions are presented in section [ sec : conclusion ] . following the nrqcd factorisation , the cross - section for quarkonium hadroproduction can be expressed from the parton densities in the colliding hadrons , @xmath24 , a hard - part the partonic cross section for the production of a heavy - quark pair with zero relative velocity , @xmath25 , in a definite angular - momentum , spin and colour state , and a ldme connected to the hadronisation probabilities of the intermediate state into the quarkonium . namely , one has for the production of a quarkonium @xmath26 along with some unidentified set of particles @xmath27 , @xmath28\langle{\cal o}_{{{\cal q}}}(n)\rangle \label{sum - section } \end{split}\ ] ] where the indices @xmath29 run over all partonic species and @xmath30 denotes the colour , spin and angular momentum states of the intermediate @xmath31 pair . for the @xmath15 quarkonium states , the first co states which appear in the @xmath25 expansion are the @xmath32}_{0},^{3}s^{[8]}_{1}$ ] and @xmath33}_{j}$ ] states , in addition to the leading @xmath25 contribution @xmath34}_{1}$ ] from a cs transition . one however has to note that , for hadroproduction , whereas the co contributions already appear at @xmath35 ( fig . [ diagram - a ] & [ diagram - b ] ) , the cs one only appear at @xmath2 ( fig . [ diagram - c ] ) . these @xmath35 co graphs nevertheless do not contribute to the production of quarkonia with a _ @xmath0 since they would be produced alone without any other hard particle to recoil on . the born contributions from cs and co transitions are indeed different in nature : the former is the production of a quarkonium in association with a recoiling gluon , which could form a jet , while the latter is the production of a quarkonium essentially alone at low @xmath0 . let us now have a look at the @xmath2 co contributions ( fig . [ diagram - d]-[diagram - f ] ) which are then nlo or one loop corrections to quarkonium production and which are potentially plagued by the typical divergences of radiative corrections . yet , the real - emission @xmath2 corrections to co contributions ( fig . [ diagram - d ] & [ diagram - e ] ) can also be seen as born - order contributions to the production of a quarkonium + a jet or , equally speaking , of a quarkonium with @xmath36 . as such , they do not show any soft divergences for @xmath37 . these are supposed to be the leading contribution to the @xmath0-differential cross section in most of the data set taken at hadron collider ( tevatron , rhic and lhc ) . these are now known up to one - loop accuracy , namely up to @xmath6 ( see _ e.g. _ @xcite ) ( fig . [ diagram - g ] & [ diagram - h ] ) . it is important to note that one can not avoid dealing with the divergences appearing at @xmath2 if one study the @xmath0-integrated cross section . at @xmath35 , the co partonic processes are : @xmath39}_{1 } ] & \hbox { ( { { fig.~\ref{diagram - a } } } ) } \\ & g+g\rightarrow q\bar q[^{1}\!s^{[8]}_{0 } , ^{3}\!p^{[8]}_{j=0,2 } ] & \hbox { ( { { fig.~\ref{diagram - b } } } ) } \end{split}\ ] ] where @xmath40 denotes @xmath41 . at @xmath2 , the qcd corrections to the aforementioned channels include real ( fig . [ diagram - d ] & [ diagram - e ] ) and virtual ( fig . [ diagram - f ] ) corrections . one encounters uv , ir and coulomb singularities in the calculation of the virtual corrections . the uv - divergences from the self - energy and triangle diagrams are removed by the renormalisation procedure . since we follow the same lines as @xcite where all the procedure is described , we do not repeat its description . as regards the real - emission corrections , they arise from 3 kinds of processes ( not all drawn ) : @xmath42}_{0},^{3}\!s^{[8]}_{1},^{3}\!p^{[8]}_{j=0,2}]+g , \\ & g+q(\overline{q})\rightarrow q \bar q[^{1}\!s^{[0]}_{8},^{3}\!s^{[8]}_{1},^{3}\!p^{[8]}_{j=0,2}]+q(\overline{q } ) , \\ & q+\overline{q}\rightarrow q \bar q[^{1}\!s^{[8]}_{0},^{3}\!s^{[8]}_{1},^{3}\!p^{[8]}_{j=0,1,2}]+g.\end{split}\ ] ] as usual , the phase - space integrations generate ir singularities , which are either soft or collinear and can be conveniently isolated by slicing the phase space into different regions . here we adopt the two - cutoff phase space slicing method to deal with the problem @xcite . as we previously alluded to , the @xmath2 cs contribution is particular since , in the limit @xmath43 , it would be strictly speaking born order for both the production of a quarkonium and of a quarkonium + a jet . it arises from the well - known process : @xmath44}_{1}]+g & \hbox { ( { { fig.~\ref{diagram - c}}})}\end{split}\ ] ] our calculations is equivalent to a previous work by maltoni _ et al . _ @xcite and we have checked that we reproduce their results for all the relevant channels . as announced in the introduction , one of the novelty in our study resides in the use of the ldmes fitted at the same order , _ i.e. _ one loop , to the @xmath0-differential cross sections . as such , this is the first global nlo analysis of hadroproduction . since we also look at data at rather low energies , we also included a cs channel via @xmath45 exchange . indeed , as noted in a different context in @xcite , the qed cs contributions via @xmath45 are naturally as large as the co @xmath46}$ ] transition via @xmath47 the @xmath48 suppression being compensated by the small relative size of the @xmath46}$ ] co ldme ( @xmath49 ) as compared to the @xmath50}$ ] cs ldme ( @xmath51 ) . the real - emission contributions arise from @xmath52}_{1}]+g , \\ & g+q(\overline{q})\rightarrow q \bar q[^{3}\!s^{[1]}_{1 } ] + q(\overline{q}),\end{split}\ ] ] whereas the loop contributions are only from @xmath52}_{1}].\end{split}\ ] ] fig . [ diagram - a ] ( fig . [ diagram - f ] ) with the @xmath53-channel gluon replaced by a @xmath45 would depict the born ( a one - loop ) contribution . we have however found that they do not matter in the regions which we considered . the cs ldmes can either be extracted from the leptonic decay width at nlo or can be estimated by using a potential model result , which gives for the buchmuller - tye potential @xcite @xmath54 gev@xmath55 , @xmath56 gev@xmath55 and @xmath57 gev@xmath55 . as regards the co ldmes , they can only be extracted from data . as we discussed above , our aim is to analyse the _ @xmath0-integrated _ yield using the constraints from the _ @xmath0 dependence _ of the yields . in the @xmath3 case , we will use the results of five fits of this dependence @xcite . the first two were limited to @xmath1 data but explicitly took into account the effect of the feed - down yield and polarisation data from cdf(run ii ) . in @xcite , et al . _ chose to fit the cdf and lhcb experimental data for the yield only ( no polarisation data ) . ] the latter fit was based on a wider set of data including @xmath58 and @xmath59 systems but the feed - down effects were only implicitly included through constant fractions for these systems . the fourth one includes the recent @xmath13 measurement at @xmath60 gev by lhcb @xcite by relying on the heavy - quark spin symmetry of nrqcd which relates colour - octet matrix elements of spin - singlet and triplet quarkonia with the same principal quantum number @xmath30 . the fifth one incorporates the leading - power fragmentation corrections together with the usual nlo corrections , which results in a different short - distance coefficient and allows for different ldmes . .values of @xmath61}_{0})\rangle$ ] , @xmath62}_{0})\rangle$ ] and @xmath63}_{1})\rangle$ ] from 5 nlo ( _ i.e. _ at one loop ) fits of the @xmath0 dependence of the yields , which we will use to compute [ the energy dependence of ] the @xmath0-integrated yields . [ cols="^,^,^,^ " , ] as regards the @xmath5 , the data set is surprisingly wider than that of @xmath18 despite a significantly smaller production cross section . it is certainly due to the larger energy of the decay leptons and to the smaller background . for a long time , it was considered that only half of the ( low-@xmath0 ) @xmath5 were directly produced ( @xmath64 ) based on an early cdf measurement @xcite . recent lhcb studies of @xmath65 production @xcite along with @xmath66 cross section measurements @xcite , rather indicate that two thirds of the @xmath5 are directly produced , we will therefore opt for @xmath67 . yet , a number of experiments could not resolve the 3 @xmath10 states . in this case , one should apply @xcite a slightly smaller direct fraction which we take to be @xmath68 . as we take this fraction to be energy independent , we chose a conservative estimate of their uncertainty . table [ tab : jpsi - data ] shows the @xmath3 data set , table [ tab : psi2s - data ] that of @xmath18 and table [ tab : upsi - data ] that of @xmath10 . in the numerical computation at nlo , the cteq6 m pdf @xcite , and the corresponding two - loop qcd coupling constant @xmath69 are used at two loop . this remark does not concern the real - emission radiative corrections of a given channel which are treated as usual , _ i.e. _ with nlo pdfs . ] . the charm quark mass , @xmath70 , is set by default to 1.5 gev and the bottom quark one , @xmath71 , to 4.5 gev . our default choices for the renormalisation , factorisation , and nrqcd scales are @xmath72 with @xmath73 and @xmath74 , respectively . when other choices are made , in particular to estimate the theoretical uncertainty due to the lack of knowledge of corrections beyond nlo , they are indicated on the corresponding plots . we have taken @xmath75 and @xmath76 for the two phase space cutoffs the insensitivity of the result on the chosen values for these cut - off has been checked . our results for direct @xmath3 , @xmath18 and @xmath5 are shown on respectively fig . [ fig : energy_dependence ] ( a ) , ( b ) and ( c ) . we first discuss the comparison between the five fits and the @xmath3 data ( fig . [ fig : energy_dependence ] ( a ) ) . without a surprise , our study shows that the global fits including rather low @xmath77 data , that is the one of butenschoen _ et al . _ @xcite provides the only acceptable description of the total cross section . we however note that the latter fit does not provide a good description of the @xmath3 polarisation data and , as recently noted @xcite , it yields to negative cross section for @xmath78 at large @xmath0 . finally , it does not allow @xcite to describe the @xmath13 data . the fits of gong _ et al . _ @xcite , and ma _ et al . _ @xcite greatly overshoot the data in the energy range between rhic and the tevatron , whereas these fits a priori provide a good description of the @xmath0-differential cross section at these energies . the fit of bodwin _ et al . _ @xcite gives the worse account of the @xmath0-integrated @xmath3 data in the whole energy range . indeed , the new ingredient of @xcite allows one to describe high-@xmath0 data with a large @xmath79}$ ] ldme ( see table [ tab : ldme - jpsi ] ) as for @xcite but without negative ldmes for the other octet ldmes which results in too large a yield at low @xmath0 . in addition , we also note the strange energy dependence of at least the @xmath80-wave octet channel . we postpone its discussion to section [ section:1s08-nlo ] where this is analysed in more detail for the @xmath79}$ ] transition and , in section [ section : csm - nlo ] , where we discuss a similar observation for the csm yield at nlo . as regards the @xmath18 ( fig . [ fig : energy_dependence ] ( b ) ) , our nlo nrqcd results do not reproduce the data at all at rhic energies and , since both fits as dominated by the @xmath80-wave octet channel , shows a nearly unphysical behaviour at lhc energies . the comparison for the @xmath5 ( fig . [ fig : energy_dependence ] ( c ) ) , is more encouraging . at rhic energies and below , the agreement is even quite good , while at tevatron and lhc energies , the nlo nrqcd curves _ only _ overshoot the data by a factor of 2 . we finally note that from rhic to lhc energies , the lo csm contributions ( the blue in all the plots ) accounts well for the data . the agreement is a bit less good for @xmath18 if we stick only to the default / central value . this is not at all a surprise and is in line with the previous conclusions made in @xcite . in fact , strangely enough , it seems that it is only at low energies ( below @xmath81 gev ) that the co contributions would be needed to describe the data . the more recent data from the lhc and the tevatron tend to agree more with the lo csm . overall , this shows unless the resummation of isr modifies our predictions by a factor of ten that it would be difficult to achieve a global description of the total and @xmath0-differential yield and its polarisation at least for the charmonia . as we discussed in the introduction , a first resummation study has recently been performed within nrqcd @xcite . when combined with the results of @xcite , this resummation yields @xcite to a good description of the low-@xmath0 data . it should however be stressed that this study introduces 3 new parameters @xmath82 to parametrise the so - called @xmath83 function used the css resummation procedure . moreover , we stress that such values of the co ldmes would result in a negative nlo @xmath0-differential cross section for @xmath78 at large @xmath0 where nrqcd factorisation should normally hold . to close the discussion of the theory - data comparison , let us note that the @xmath84}$ ] channel alone would provide a decent energy dependence . if we were to refit the low-@xmath0 data and thus obtain a dominance of the @xmath84}$ ] channel , the yield at large @xmath0 would nevertheless dominantly be transversely polarised in disagreement with existing data ( _ e.g. _ @xcite ) . yet , the better energy dependence of @xmath84}$ ] at nlo with the respect to the other octet channels , which shows a flat energy dependence in the tev region , is probably a fortunate `` accident '' . indeed , most of the @xmath84}$ ] yield up @xmath85 is in fact not at one loop , but for the suppressed @xmath86 contribution , since the @xmath87}$ ] is zero . the curve for @xmath84}$ ] as well as @xmath88}$ ] shown on fig . [ fig : energy_dependence ] ( a ) is effectively a born - order one . clearly , these behave better than the channels where the loop contributions are allowed . in view of the observations just made in the previous paragraph , we have found it useful to analyse more carefully the behaviour of one specific channel . we have decided to look more carefully at the simplest one , that is from the @xmath79}$ ] transition in particular at how the scale choices influence the behaviour of the yield at large @xmath16 . analytical results for the hard - scattering partonic amplitude squared can be found in appendix c of @xcite . we have used them in a small numerical code to convolve it with pdfs and checked them with the results of fdc . the advantage of using fdc is that we can easily cut on @xmath0 and @xmath89 . as one could have anticipated from the band of the lower panel of fig . [ fig : energy_dependence ] ( a ) , one observes that for a wide range of scale choices , the energy dependence remains extremely flat . for some of the choice where @xmath90 , one even sees that the cross section clearly decreases and becomes negative when the yield becomes negative the curve stops . this is of course not satisfactory . at this stage , we are not able to conclude from our observations normalisation and high energy issue whether these point at the break down of nrqcd factorisation , nrqcd universality or should force us to continue questioning our understanding of the mid- and high-@xmath0 quarkonium production mechanisms . to investigate this a bit further , we look at the predictions of another approach the colour evaporation model in the next section and later in the colour - singlet model both for @xmath15 and @xmath23 quarkonia . to go further in our investigations of qcd one - loop effects on the energy dependence of quarkonium production , we have found it useful to compare our results with those of the colour - evaporation model ( cem ) which directly follows from the quark - hadron duality @xcite . the quarkonium production cross section is obtained by considering the cross section to produce a @xmath91 pair in an invariant mass region compatible with its hadronisation into a quarkonium , namely between @xmath92 and the threshold to produce open heavy flavour hadrons , @xmath93 . to this , one should multiply a phenomenological factor accounting for the probability that the pair eventually hadronises into a given quarkonium state . overall , one considers @xmath94 in a sense , the factor @xmath95 , _ i.e. _ the probability ( or fraction ) of @xmath96 pair in the relevant invariant mass region to directly hadronise into @xmath97 , plays a similar role as the ldmes in nrqcd , except that its size can be guessed . indeed , it is expected @xcite that one ninth one colour singlet @xmath96 configuration out of 9 possible of the open charm cross section in this invariant mass region eventually hadronise into a `` stable '' quarkonium . taking into account this factor 9 , in the case of @xmath3 , it was argued @xcite that a simple statistical counting , which would give : @xmath98 where the sum over @xmath99 runs over all the charmonium states below the @xmath100 threshold , could describe the existing data in the late 90 s . the solid turquoise curve computed at nlo as opposed to the analysis of @xcite on fig . [ fig : energy_dependence_cem ] ( a ) illustrates this gev , @xmath101 and @xmath102 . ] . following the fit of vogt in @xcite , @xmath103 lies between 1.5 % and 2.5 % . this indeed remarkably coincides with the simple statistical counting - waves do not work and illustrate the limit of the model . ] . for the @xmath10 , the corresponding quantity is of similar size , between 2 % and 5 % , although following the state - counting argument , one may expect a smaller number than for @xmath3 . let us nevertheless stress that a violation of eq . [ eq : fdir_cem ] can not be used to invalidate the cem since this relation completely ignores phase - space constraints . what cem predicts is that @xmath95 is process independent . in @xcite , bodwin _ et al . _ studied the connexion between the cem and nrqcd . following @xcite up to @xmath104 corrections , only 4 intermediate @xmath96 states contribute to @xmath105 quarkonium production in a cem - like implementation of nrqcd . one indeed has : @xmath106}_{1})\rangle = & 3 \times \langle{\cal o}_{^3\!s_1}(^{1}\!s^{[1]}_{0})\rangle,\\ \langle{\cal o}_{^3\!s_1}(^{1}\!s^{[8]}_{0})\rangle = & \frac{4}{3 } \times \langle{\cal o}_{^3\!s_1}(^{1}\!s^{[1]}_{0})\rangle,\\ \langle{\cal o}_{^3\!s_1}(^{3}\!s^{[8]}_{1})\rangle = & 4 \times\langle{\cal o}_{^3\!s_1}(^{1}\!s^{[1]}_{0})\rangle . \end{split}\ ] ] all these nonvanishing ldmes are then fixed if one makes the reasonable assumption that @xmath107}_{1})\rangle$ ] is indeed the usual cs ldme , _ i.e. _ @xmath108 . as compared to the results presented in the previous section , the only additional piece to perform a full one - loop analysis is the hard part for @xmath109}_{0}$ ] which normally needs not to be considered for @xmath105 production at this level of accuracy in @xmath25 . computing it with fdc @xcite does not present any difficulty . + fig . [ fig : energy_dependence_cem ] ( a ) & ( b ) show the resulting cross sections of @xmath3 and @xmath5 production for the relevant channels and their sum , to be compared to the world data set used in the previous section . by construction , the @xmath110}_{1}$ ] curve is the same as in the previous section . one notes that , as for the @xmath80-wave octet , the @xmath109}_{0}$ ] curve strangely flattens out in the @xmath3 case at high energies . we will come back to this in the next section . the total cem - like contribution greatly overshoots the data , by a factor as large as 100 . this was to be expected since ( i ) , following eq . [ eq : cem - ldme ] , all the ldmes are roughly of the same size , ( ii ) the @xmath110}_{1}$ ] is roughly compatible with the data and ( iii ) the hard part for the other transitions appear at @xmath35 , are not suppressed as @xmath111 and are thus expected to give a larger contribution than the @xmath110}_{1}$ ] if one disregards the ldme . of course , one could question our assumption that @xmath107}_{1})\rangle = \frac{2 n_c}{4 \pi}~(2j+1)~|r(0)|^2 $ ] and rather fit @xmath112}_{0})\rangle$ ] . in both the @xmath3 and @xmath5 cases , the corresponding ldmes would then approximately be 100 times smaller . in particular , the singlet transition would be 100 times less probable that what one expects from the leptonic decay . this would be an unlikely and dramatic violation of factorisation which should have implications elsewhere . in particular , a pair produced at short distances with the same quantum number as the physical state , among these the colour , would have a much larger probability to be broken up before eventually hadronising than expected . although it is not as obvious as in the nrqcd formulae of eq . [ eq : cem - ldme ] , where the hypotheses of the cem are translated into direct relations between co and cs transition probabilities , the same should happen in eq . [ eq : sigma_cem ] where all the colour configurations are summed over and then considered on the same footage . in a process where the cs configurations dominate , such as @xmath113 , csm and cem predictions necessarily differ . contrary to nrqcd which encompasses the csm , the cem does not encompass the csm . if one agrees with the data , the other can not . the matter is then how precise the predictions and the data are to rule out one approach or the other . overall , one has to acknowledge that the conventional cem central curves as simplistic as the underlying idea of the model can be give an account ( fig . [ fig : energy_dependence_cem ] ( a ) & ( b ) ) of the world data points as satisfactory as the lo csm . the latter seems to underestimate the data at low energies while the former only has trouble to account for the tev @xmath3 points ; the slope being more problematic than the normalisation which can be adjusted . all this is qualitative since the theoretical uncertainties on the cem are as large as that on the open heavy - flavour production which are admittedly large ( see @xcite for an up - to - date discussion on the @xmath114 case ) . [ section : csm - nlo ] as we just stated , the lo csm curves are providing a surprisingly good description of the @xmath3 and @xmath10 data at high energies without adjusting and even less twisting any parameters . although the central lo csm curves agree with the data , the conventional theoretical uncertainties from the arbitrary scales and the heavy - quark mass are large ( see the lower panels of fig . [ fig : energy_dependence ] ) ) . it is therefore very natural to look whether these uncertainties are reduced at one - loop accuracy . such an observation was already made for the @xmath10 case in @xcite but this study was limited to a single @xmath16 , _ i.e. _ 200 gev . contrary the co channels , the one - loop corrections to the cs channels only arise at @xmath6 ( see _ e.g. _ fig . [ diagram - j ] & [ diagram - k ] ) . nevertheless , these are know since 2007 @xcite and can also be computed with fdc as done in @xcite . in particular , there is no specific difficulty to integrate the @xmath2 and @xmath6 contributions in @xmath0 since they are finite at @xmath115 . + however , as already noted in @xcite , such nlo results tends to shows negative values at low @xmath0 which can have a non - negligible impact on the total ( _ i.e. _ @xmath0-integrated ) cross section . to our knowledge , the energy dependence of the csm at nlo has never been studied in detail . this is done below . figs . [ fig : energy_dependence_csm ] show the energy dependence of the nlo csm ( 7 curves ) . note that , if a curve is not shown until 14 tev , this indicates that the total yield got _ negative_. the 3 red curves correspond to the default scale choices ( @xmath116 ) and are indicative of the heavy - quark mass uncertainty , on the order of a factor of 4 for the @xmath3 and 2.5 for the @xmath10 . in the former case , all 3 curves end up to be negative somewhere between 500 gev and 2 tev . note also that the upper curve at low energies , _ i.e. _ for @xmath117 gev , is the first to get negative and crosses the other ones as if the negative contribution were more important for lighter systems , see section [ subsec : etaq ] . ] . in the @xmath10 case , these 3 curves do not become negative at high energies we have checked it up to @xmath81 tev . nonetheless , they start to significantly differ from the lo curves ( 3 blue curves ) above 1 tev , contrary to the good lo vs nlo convergence found at rhic energies in @xcite . one might thus be tempted to identify this weird energy behaviour to a _ low-@xmath17 effect_. going further in the @xmath3 case , one can vary the factorisation and renormalisation scales about the default choice . doing so , one obtains two classes of curves . for @xmath118 ( pink and orange ) , the yield remains positive , but it is not less unphysical for it to be practically constant as the energy increases between 1 and 10 tev ! the only way to recover a semblance of increase is to take a large value of @xmath119 and seemingly also a small value for @xmath120 . obviously , whatever the reason for this behaviour is , for large enough @xmath119 , the qcd corrections which are proportional to @xmath121 necessarily get smaller and any difference between lo and nlo results should decrease . in the @xmath10 case , as for the @xmath3 case , both curves with @xmath118 ( pink and orange ) correspond to the highest yields at high energies and the lowest at low energies . when one chooses @xmath122 ( purple and green ) , the high - energy yields become negative both for @xmath3 and @xmath10 . in many respect , these observations are very similar to those made on the @xmath79}$ ] . such a pathological behaviour may thus not be related to the nature to the final state ( see also next section ) . large nnlo corrections are expected to show up at large energies ( low @xmath17 ) as discussed in @xcite . it is not clear if they could provide a solution to this issue . another way to solve this might be to resum initial state radiation as done in the cem @xcite and for some co channels @xcite . at the light of such results , the most that one can reasonably say is that the nlo csm results may be reliable for @xmath10 up to 200 gev and for @xmath3 up to 60 gev , that is up to @xmath16 about 20 times the quarkonium mass . above these value , the best that we have is the born order results . contrary to the spin - triplet case , one can obtain analytical formulae @xcite for the spin - singlet pseudo - scalar production cross section such as that of @xmath13 and @xmath14 . this can in principle be of some help to understand the weird energy behaviour of the cs @xmath15 yield and of some co channels . indeed , the lo production occurs as for some co channels without final - state - gluon radiation . in fact the final state is simply colourless . + as can be seen on figs . [ fig : energy_dependence_csm - etaq ] , the issue is similar in many respects but for the fact that one does not obtain negative yields for the @xmath14 . for the @xmath13 , the curves for @xmath118 remains positive at high energies as for the @xmath3 . one also sees that the crossing of the central lo and nlo curves occurs at larger @xmath16 than for the @xmath15 states . however , such small quantitative differences may be due to the fact that we computed the @xmath89-integrated cross sections using the analytical expressions of @xcite instead of the @xmath89-differential cross section at @xmath123 . we have investigated this in more details by looking at the different nlo contributions ( the real emissions from @xmath124 and @xmath125 fusion as well as the virtual ( loop ) contributions ) in order to see which channels induce the negative contributions and for which scale / mass values . however , it must be stressed that the decomposition between these different contributions depend on the regularisation method used . for instance , the decomposition is drastically different when using fdc with sometimes a very large cancellation between the _ positive _ real - emission @xmath124 contribution and the _ negative _ sum of the born and loop @xmath124 contributions and the formulae of @xcite where all the @xmath124 contributions are gathered . yet , we checked that we obtain exactly the same results with both methods ; the regularisation method or numerical instabilities can not be the source of the issues observed above . as regards the @xmath125 contribution , only @xmath126 matters to tell whether it will change sign . for @xmath120 close to @xmath127 and below , it will be positive ( negative ) at small ( large ) @xmath16 . otherwise , it remains negative for any @xmath16 . the value of @xmath119 only influences the normalisation . as what concerns the @xmath124 contributions , which are expected to be dominant at high energies , both @xmath126 and @xmath128 matter . for @xmath129 , the @xmath124 contribution monotonously increase as function @xmath16 irrespective of @xmath128 . for @xmath130 , the @xmath131 one is rather small and , despite being negative , does not induce a turn over in the increase of the cross section . for @xmath132 , the @xmath124 contribution gets negative at large @xmath16 for @xmath133 . for @xmath134 , it remains always positive . yet the sum @xmath135 can still become negative since , in some cases , @xmath131 increases faster with @xmath16 . all this seems in lines , for the @xmath124 [ @xmath131 ] part , with the formulae ( c.25 ) and ( c.26 ) [ ( c.32 ) and ( c.35 ) ] in the appendix c of @xcite . both contributions indeed exhibit logarithms of @xmath126 multiplied by a factor function of @xmath137 . as we can see , the results are already difficult to interpret for the spin singlet case . for the spin triplet , we do not have similar analytical results . we can only guess that the structure is similar . what seems surprising is that when one inspects similar expressions for the @xmath138 production at nlo in the tmd factorisation @xcite , such negative terms do not appear as obvious . we are thus entitled to wonder whether such a formalism , which automatically resums logarithm of transverse momenta , may provide the solution to this issue . another possible solution may be the consideration of nnlo corrections , which may show opposite signs to that at nlo . this is obviously beyond the scope of our analysis . we have performed the first full analysis of the energy dependence of the quarkonium - production cross section at one - loop accuracy both in nrqcd and in the csm . taken at face value , our results would indicate a severe break down of nrqcd universality in line with the previous analysis of maltoni _ et al . _ @xcite unless one keeps the ldmes close to the fit of butenschoen and kniehl , which however disagrees with the @xmath3 polarisation measurements and the @xmath13 cross sections . the situation is however slightly more intricate since we have uncovered a weird sometimes unphysical behaviour at large energies where one approaches the small-@xmath17 regime where non - linear effects in the parton densities may be relevant . this certainly casts doubts on the numerical values we obtained at lhc energies since collinear factorisation on which we based our analysis could break down . yet , up to a few hundred gev , the energy dependence of the different octet channels at nlo seems well behaved and there is no reason to doubt this . in this region , the nlo yield prediction by nrqcd after fitting the mid- and high-@xmath0 quarkonium data _i.e . _ the yield and its polarisation would overshoot the data by a factor ranging from 4 to more than 10 . the same holds true at lo ( see [ appendix ] ) . to reproduce the data , the co ldmes should be much smaller than what they are found in order to be to reproduce the tevatron and lhc @xmath0-differential cross section in the case of the @xmath3 and @xmath18 . on the other hand , the lo csm provides a decent energy dependence in agreement with the existing data , except for @xmath139 gev and is therefore up to ten times below the data - overshooting co contributions . at one loop , the results are however ill - behaved for the charmonia and the cross sections can even become negative at large @xmath16 for some reasonable scale choices . the situation is a bit better for @xmath5 . the same occurs for the spin - singlet quarkonia . in this case , one - loop results exist in the tmd factorisation approach and do not seem to be prone to such an issue . in the case of double @xmath3 production , the energy dependence at one loop seems well - behaved @xcite . finally , we investigated the energy dependence of the yield in the cem where the final states are treated rather differently and we did not find any specific problems . we are therefore tempted to attribute this problem to initial - state effects . in @xcite , ma and venugopalan obtained a good description of the low-@xmath0 @xmath3 data over a wide range of energy by , on the one hand , using the ldmes from @xcite our second set and , on the other , a cgc - based computation of the low-@xmath0 dependence . in reproducing the data , they found that the cs contribution is only 10% of the total yield . this 10% is reminiscent of the factor 10 between the cs and co in our `` collinear '' study . from our viewpoint , it looks as if the specific ingredient of this cgc - based computation would correspond to an effective reduction of the two - gluon flux by a factor of 10 . it is therefore very interesting to find out new processes which would be sensitive to this physics . the negative yields obtained in the collinear case observed for @xmath79}$ ] , @xmath140}$ ] , @xmath50}$ ] , @xmath141}$ ] channels could also be cured by adding the large contributions of the one - loop amplitude squared thus positive . this may look like an _ ad - hoc _ solution which certainly questions the convergence of the perturbative series in @xmath142 . however , large nnlo corrections have already been discussed 10 years ago in @xcite . another path towards a solution may also be higher - twist contributions @xcite where two gluons come from a single proton as recently rediscussed in @xcite . in any case , whatever the explanation for this situation may be , past claims that colour - octet transitions are dominantly responsible for low-@xmath0 quarkonium production were premature in light of the results presented here . we are grateful to p. artoisenet , s.j . brodsky , k.t . chao , w. den dunnen , m.g . echevarria , b. gong , y.q . ma , j.w . qiu , c. pisano , m. schlegel , h.s . shao , l.p . sun , r. venugopalan and r. vogt for useful discussions . this work is supported in part by the lia france - china particle physics laboratory ( fcppl ) and by the sapore gravis networking of the eu i3 hadron physics 3 program . sharma and vitev recently performed @xcite a lo fit of the co ldmes using rhic , tevatron and lhc @xmath3 data . setting @xmath62}_{0})\rangle=\langle{\cal o}_{j/\psi}(^{3}\!p^{[8]}_{0})\rangle / m^2_c$ ] and accounting for the possible feed - downs , they obtained : b. abelev _ et al . _ [ alice collaboration ] , phys . * 109 * ( 2012 ) 072301 [ arxiv:1202.1383 [ hep - ex ] ] . s. chatrchyan _ et al . _ [ cms collaboration ] , jhep * 1205 * ( 2012 ) 063 [ arxiv:1201.5069 [ nucl - ex ] ] . t. matsui and h. satz , phys . b * 178 * ( 1986 ) 416 . s. chatrchyan _ et al . _ [ cms collaboration ] , phys . lett . * 107 * ( 2011 ) 052302 [ arxiv:1105.4894 [ nucl - ex ] ] . s. chatrchyan _ et al . _ [ cms collaboration ] , phys . * 109 * ( 2012 ) 222301 [ arxiv:1208.2826 [ nucl - ex ] ] . j. -w . qiu , j. p. vary and x. -f . zhang , phys . * 88 * ( 2002 ) 232301 [ hep - ph/9809442 ] . f. arleo and s. peigne , phys . rev . * 109 * ( 2012 ) 122301 [ arxiv:1204.4609 [ hep - ph ] ] . t. liou and a. h. mueller , phys . d * 89 * ( 2014 ) 074026 [ arxiv:1402.1647 [ hep - ph ] ] . chang , nucl . b * 172 * ( 1980 ) 425 . d. kharzeev , e. levin , m. nardi and k. tuchin , phys . rev . lett . * 102 * ( 2009 ) 152301 [ arxiv:0808.2954 [ hep - ph ] ] . f. dominguez , d. e. kharzeev , e. m. levin , a. h. mueller and k. tuchin , phys . b * 710 * ( 2012 ) 182 [ arxiv:1109.1250 [ hep - ph ] ] . z. b. kang , y. q. ma and r. venugopalan , jhep * 1401 * ( 2014 ) 056 [ arxiv:1309.7337 [ hep - ph ] ] . m. beneke and i. z. rothstein , phys . d * 54 * ( 1996 ) 2005 [ erratum - ibid . d * 54 * ( 1996 ) 7082 ] [ hep - ph/9603400 ] . f. cooper , m. x. liu and g. c. nayak , phys . * 93 * ( 2004 ) 171801 [ hep - ph/0402219 ] . f. maltoni , _ et al . _ , phys . b * 638 * ( 2006 ) 202 [ hep - ph/0601203 ] . p. sun , c .- p . yuan and f. yuan , phys . d * 88 * , 054008 ( 2013 ) [ arxiv:1210.3432 [ hep - ph ] ] . y. -q . ma , k. wang and k. -t . chao , phys . lett . * 106 * 042002 ( 2011 ) [ arxiv:1009.3655 [ hep - ph ] ] . b. gong , l. -p . wan , j. -x . wang and h. -f . zhang , phys . rev . lett * 110 * 042002 ( 2013 ) [ arxiv:1205.6682 [ hep - ph ] ] . a. petrelli , m. cacciari , m. greco , f. maltoni and m. l. mangano , nucl . b * 514 * ( 1998 ) 245 [ hep - ph/9707223 ] . j. -x . wang , nucl . instrum . meth . a * 534 * ( 2004 ) 241 [ hep - ph/0407058 ] . j. p. lansberg , int . j. mod . a * 21 * ( 2006 ) 3857 [ hep - ph/0602091 ] . n. brambilla , s. eidelman , b. k. heltsley , r. vogt , g. t. bodwin , e. eichten , a. d. frawley and a. b. meyer _ et al . _ , eur . j. c * 71 * ( 2011 ) 1534 [ arxiv:1010.5827 [ hep - ph ] ] . g. t. bodwin , e. braaten and j. lee , phys . d * 72 * ( 2005 ) 014004 [ hep - ph/0504014 ] . r. e. nelson , r. vogt and a. d. frawley , phys . c * 87 * ( 2013 ) 014908 [ arxiv:1210.4610 [ hep - ph ] ] . s. j. brodsky and j.p . lansberg , phys . rev . d * 81 * ( 2010 ) 051502 ; [ arxiv:0908.0754 [ hep - ph ] ] . lansberg , pos ichep * 2010 * ( 2010 ) 206 . [ arxiv:1012.2815 [ hep - ph ] ] . j. p. lansberg , nucl . phys . a * 910 - 911 * 470 ( 2012 ) . [ arxiv:1209.0331 [ hep - ph ] ] . j. p. lansberg , pos ichep * 2012 * ( 2013 ) 293 [ arxiv:1303.2858 [ hep - ph ] ] . b. gong , x. q. li and j. x. wang , phys . b * 673 * ( 2009 ) 197 [ erratum - ibid . * 693 * ( 2010 ) 612 ] [ arxiv:0805.4751 [ hep - ph ] ] . b. gong , j. -x . wang and h. -f . zhang , phys . d * 83 * ( 2011 ) 114021 [ arxiv:1009.3839 [ hep - ph ] ] . m. butenschoen and b. a. kniehl , phys . d * 84 * ( 2011 ) 051501 [ arxiv:1105.0820 [ hep - ph ] ] . h. han , y. q. ma , c. meng , h. s. shao and k. t. chao , phys . * 114 * ( 2015 ) 9 , 092005 [ arxiv:1411.7350 [ hep - ph ] ] . g. t. bodwin , h. s. chung , u. r. kim and j. lee , phys . * 113 * ( 2014 ) 2 , 022001 [ arxiv:1403.3612 [ hep - ph ] ] . r. aaij _ et al . _ [ lhcb collaboration ] , arxiv:1409.3612 [ hep - ex ] . h. f. zhang , z. sun , w. l. sang and r. li , phys . * 114 * ( 2015 ) 9 , 092006 [ arxiv:1412.0508 [ hep - ph ] ] . c. morel _ et al . _ [ ua6 collaboration ] , phys . b * 252 * ( 1990 ) 505 . a. g. clark , p. darriulat , k. eggert , v. hungerbuhler , h. r. renshall , j. strauss , a. zallo and b. aubert _ et al . _ , nucl . b * 142 * ( 1978 ) 29 . c. kourkoumelis , l. resvanis , t. a. filippas , e. fokitis , a. m. cnops , j. h. cobb , r. hogue and s. iwata _ et al . _ , phys . b * 91 * ( 1980 ) 481 . a. adare _ et al . _ [ phenix collaboration ] , phys . d * 85 * ( 2012 ) 092004 [ arxiv:1105.1966 [ hep - ex ] ] . b. abelev _ et al . _ [ alice collaboration ] , phys . b * 718 * ( 2012 ) 295 [ arxiv:1203.3641 [ hep - ex ] ] . k. aamodt _ et al . _ [ alice collaboration ] , phys . b * 704 * ( 2011 ) 442 [ erratum - ibid . b * 718 * ( 2012 ) 692 ] [ arxiv:1105.0380 [ hep - ex ] ] . d. acosta _ et al . _ [ cdf collaboration ] , phys . rev . d * 71 * ( 2005 ) 032001 [ hep - ex/0412071 ] . l. y. zhu _ et al . _ [ nusea collaboration ] , phys . * 100 * ( 2008 ) 062301 [ arxiv:0710.2344 [ hep - ex ] ] . l. adamczyk _ et al . _ [ star collaboration ] , phys . b * 735 * ( 2014 ) 127 [ arxiv:1312.3675 [ nucl - ex ] ] . c. albajar _ et al . _ [ ua1 collaboration ] , phys . b * 186 * ( 1987 ) 237 . d. acosta _ et al . _ [ cdf collaboration ] , phys . rev . lett . * 88 * ( 2002 ) 161802 . v. m. abazov _ et al . _ [ d0 collaboration ] , phys . * 94 * ( 2005 ) 232001 [ erratum - ibid . * 100 * ( 2008 ) 049902 ] [ hep - ex/0502030 ] . g. aad _ et al . _ [ atlas collaboration ] , phys . rev . d * 87 * ( 2013 ) 5 , 052004 [ arxiv:1211.7255 [ hep - ex ] ] . s. chatrchyan _ et al . _ [ cms collaboration ] , phys . b * 727 * ( 2013 ) 101 [ arxiv:1303.5900 [ hep - ex ] ] . t. affolder _ et al . _ [ cdf collaboration ] , phys . * 84 * ( 2000 ) 2094 [ hep - ex/9910025 ] . r. aaij _ et al . _ [ lhcb collaboration ] , jhep * 1211 * ( 2012 ) 031 [ arxiv:1209.0282 [ hep - ex ] ] . r. aaij _ et al . _ [ lhcb collaboration ] , eur . j. c * 74 * ( 2014 ) 10 , 3092 [ arxiv:1407.7734 [ hep - ex ] ] . r. aaij _ et al . _ [ lhcb collaboration ] , eur . j. c * 72 * ( 2012 ) 2025 [ arxiv:1202.6579 [ hep - ex ] ] . j. pumplin , d. r. stump , j. huston , h. l. lai , p. m. nadolsky and w. k. tung , jhep * 0207 * ( 2002 ) 012 [ hep - ph/0201195 ] . a. d. martin , w. j. stirling , r. s. thorne and g. watt , eur . j. c * 63 * ( 2009 ) 189 [ arxiv:0901.0002 [ hep - ph ] ] . r. li and j. x. wang , phys . d * 89 * ( 2014 ) 114018 [ arxiv:1401.6918 [ hep - ph ] ] . m. butenschoen , z. g. he and b. a. kniehl , phys . * 114 * ( 2015 ) 9 , 092004 [ arxiv:1411.5287 [ hep - ph ] ] . k. -t . chao , y. -q . ma , h. -s . shao , k. wang and y. -j . zhang , phys . * 108 * ( 2012 ) 242004 [ arxiv:1201.2675 [ hep - ph ] ] . h. fritzsch , phys . lett . b * 67 * ( 1977 ) 217 . f. halzen , phys . b * 69 * ( 1977 ) 105 . j. f. amundson , o. j. p. eboli , e. m. gregores and f. halzen , phys . b * 372 * ( 1996 ) 127 [ hep - ph/9512248 ] . j. campbell and r. k. ellis , mcfm - monte carlo for femtobarn processes , http://mcfm.fnal.gov/. j. m. campbell , f. maltoni and f. tramontano , phys . * 98 * ( 2007 ) 252002 [ hep - ph/0703113 [ hep - ph ] ] . b. gong and j. x. wang , phys . * 100 * ( 2008 ) 232001 [ arxiv:0802.3727 [ hep - ph ] ] . v. a. khoze , a. d. martin , m. g. ryskin and w. j. stirling , eur . j. c * 39 * ( 2005 ) 163 [ hep - ph/0410020 ] . e. l. berger , j. w. qiu and y. l. wang , phys . d * 71 * ( 2005 ) 034007 [ hep - ph/0404158 ] . j. h. kuhn and e. mirkes , phys . d * 48 * ( 1993 ) 179 [ hep - ph/9301204 ] . y. q. ma and r. venugopalan , phys . * 113 * ( 2014 ) 19 , 192301 [ arxiv:1408.4075 [ hep - ph ] ] . j. l. alonso , j. l. cortes and b. pire , phys . b * 228 * ( 1989 ) 425 . l. motyka and m. sadzikowski , arxiv:1501.04915 [ hep - ph ] .

we have obtained 48 high resolution echelle spectra of the pre - main sequence eclipsing binary system kh 15d ( v582 mon , p = 48.37 d , 0.6 , m = 0.6 m , m = 0.7 m ) . the eclipses are caused by a circumbinary disk seen nearly edge on , which at the epoch of these observations completely obscured the orbit of star b and a large portion of the orbit of star a. the spectra were obtained over five contiguous observing seasons from 2001/2002 to 2005/2006 while star a was fully visible , fully occulted , and during several ingress and egress events . the h line profile shows dramatic changes in these time series data over timescales ranging from days to years . a fraction of the variations are due to edge effects " and depend only on the height of star a above or below the razor sharp edge of the occulting disk . other observed variations depend on the orbital phase : the h emission line profile changes from an inverse p cygni type profile during ingress to an enhanced double - peaked profile , with both a blue and red emission component , during egress . each of these interpreted variations are complicated by the fact that there is also a chaotic , irregular component present in these profiles . we find that the complex data set can be largely understood in the context of accretion onto the stars from a circumbinary disk with gas flows as predicted by the models of eccentric t tauri binaries put forward by artymowicz & lubow , gnther & kley , and de val - borro et al . in particular , our data provide strong support for the pulsed accretion phenomenon , in which enhanced accretion occurs during and after perihelion passage . draft version :

1204.1334

kh 15d , first observed at van vleck observatory in 1995 , gained its notoriety due to the regular and rather large amplitude variability it displayed ( kearns & herbst 1998 ) . at that time , the system s light was observed to diminish by nearly 3 magnitudes every 48.4 days , and remain in this diminished or eclipsed state for approximately 16 days , with an unusual return to near - normal brightness close to mid - eclipse . by 2000 , the central reversal in the light curve had faded and the length of time that the star spent in the eclipsed state had grown slightly ( @xmath11 day ) . spectra taken of the object in and out of eclipse had shown that there was effectively no color change between states , implying that the source of obscuration was either an optically thick disk or rather large particles ( hamilton et al . several models were proposed at this time ( herbst et al . 2002 ; barge & viton 2003 ; agol et al . 2004 ) that involved some sort of warp or swarm of particles in a circumstellar disk around a single star . kh 15d was also discovered to be the source of a bipolar jet revealed in h@xmath5 and [ oi ] ( hamilton et al . 2003 ) whose possible launching mechanism has been discussed in detail by mundt et al . the system is associated with a shocked h@xmath6 emission filament ( deming , charbonneau , & harrington 2004 ; tokunaga et al . 2004 ) . during eclipse , the spectrum of kh 15d often exhibits extended wings of h@xmath5 emission , up to several hundreds of km s@xmath7 , characteristic of actively accreting classical t tauri stars ( cttss ) . out of eclipse , the ew of h@xmath5 is only a few angstroms , which would normally lead to a classification as a weak - lined t tauri star ( wtts ) ( hamilton et al . it is perhaps best described as a weakly accreting t tauri star . its age , estimated from membership in ngc 2264 , is @xmath1 3 myr . in 2004 , it was discovered that kh 15d is a binary system . johnson et al . ( 2004 ) conducted a radial velocity survey of the system over the course of two observing seasons ( 2002/2003 and 2003/2004 ) during which kh 15d underwent significant radial velocity variations consistent with a binary companion with an orbital period equal to that of the photometric period of 48 days . almost simultaneously , two models were put forward to explain the changing light curve as the result of the progressive occultation of a pre - main sequence binary orbit by a precessing circumbinary inner disk or ring ( winn et al . 2004 ; chiang & murray - clay 2004 ) . this type of model had the advantage of also being consistent with the historical light curves obtained from a study of archival photographic plates ( winn et al . 2003 ; johnson & winn 2004 ) . winn et al . ( 2006 ; hereafter w06 ) further refined this model by including a wealth of additional photometric data available from the literature ( johnson et al . 2005 ; maffei et al . 2005 ; hamilton et al . 2005 ; barsunova et al . 2005 ; kusakabe et al . 2005 ) . our current working model of the system ( w06 ) has the distinguishing aspect that the system is fortuitously observed nearly along the plane of an opaque circumbinary ring that is , itself , somewhat tilted with respect to the binary s orbital plane . to orient the reader , a schematic of the kh 15d system is shown in figure 1 . the ring is presumably the inner part ( @xmath8 5 au ) of a more extensive circumbinary disk ( cbd ) that provides gas for continuing accretion . the outer disk has recently been detected at mm wavelengths with the sma ( herbst & wilner , in preparation ) . precession of the ring has , over the past 50 years , gradually occulted the orbit of the binary as projected on the sky . various studies have shown ( e.g. , herbst et al . 2002 ; herbst et al . 2010 ) that the edge of this occulting ring is surprisingly sharp ( much less than one stellar radius ) and behaves like a knife edge . for a more general discussion of the current properties of the kh 15d light curve and its interpretation , the reader is referred to herbst et al . ( 2010 ) . when first noticed in 1995 ( kearns & herbst 1998 ) the occulting edge had just about completely covered the orbit of the more massive star ( designated star b following the nomenclature of w06 ) . from 1995 to 2009 the knife edge moved progressively across the orbit of star a , producing the observed light curve ( hamilton et al . 2005 ; herbst et al . 2010 ) . during this time the system exhibited dramatic photometric variations on the orbital cycle as star a regularly rose and set with respect to the ring horizon . throughout part of this phase of its evolution ( 2001 - 2006 ) we were able to obtain a substantial number of spectra ( n=48 ) of the system at a variety of orbital phases and heights of star a above and below the ring edge . this has allowed us to do a kind of occultation mapping " of gas flows in the magnetosphere of star a and perhaps elsewhere within the inner cavity of the cbd . by far the most useful spectral feature available to us for this work is h@xmath5 . hamilton et al . ( 2003 ) used 3 spectra obtained during one cycle in 2001 to infer the presence of a jet and a magnetospheric accretion flow . at that time , the system was believed to be only a single star with a circumstellar disk containing a warp . in this paper , we present the analysis of additional high resolution spectroscopic data of the kh 15d system , specifically , the h@xmath5 emission line profiles , obtained while in its bright state , during ingress / egress , and during eclipse over the course of a 5 year observing period ( 2001 - 2006 ) . section 2 discusses the observations and reductions . the h@xmath5 emission line profiles are presented and characterized in section 3 . profile decompositions based on gaussian fitting and the velocity behavior of the absorption component are discussed in section 4 . the high velocity emission component of the h@xmath5 line profile is presented in section 5 , and constraints on the size of the magnetosphere of star a are given in section 6 . general conclusions are drawn in section 7 . the high resolution echelle spectra analyzed here were obtained during five observing seasons from 2001/2002 through 2005/2006 at various observing facilities . table 1 gives the ut dates of the observations , telescope / instrument used , wavelength coverage and resolution . we attempted to exploit the natural coronagraph " that this system represents by focusing primarily on phases during or near ingress and egress , however , spectra were also obtained near mid - eclipse , and when star a was well out of eclipse . figure 2 shows the cousins @xmath9-band ( @xmath10 ) data from hamilton et al . ( 2005 ) or herbst et al . ( 2010 ) phased with the orbital period for each of the five observing seasons overlain with the dates on which spectra were obtained . it is clear that the width of the eclipse increased significantly between the beginning and end of the data acquisition period so a particular orbital phase that represented an out - of - eclipse " observation in 2001/2002 might be in eclipse " in 2005/2006 . this is demonstrated in figure 3 . each spectrum is characterized , therefore , by two numbers : orbital phase , ranging from 0.5 to + 0.5 with 0.0 representing the time of mid - eclipse , and the position of star a with respect to the ring edge , ranging from @xmath11x = 4 ( 4 stellar radii above the edge ) to @xmath11x = + 10 ( 10 stellar radii below the ring edge ) . the parameter @xmath11x used here comes from model 3 of w06 and is the elevation of the center of star a above ( ) or below ( + ) the edge of the occulting disk in units of the radius of star a ( r@xmath2 = 1.3 r@xmath3 = 9 x 10@xmath12 m ) . as is evident from table 1 , there were five different instruments employed to obtain the observational material analyzed here . these are : the keck 10-m telescope and high resolution echelle spectrometer ( hires ; solid red lines in fig . 2 ) , the european southern observatory s 8.2-m very large telescope and uv - visual echelle spectrograph ( uves ; long - dashed blue lines ) , the 2.1-m otto struve telescope at mcdonald observatory and sandiford cassegrain echelle spectrometer ( ce ; dash - dot green lines ) , the 8-m hobby - ebberly telescope at mcdonald observatory and high resolution spectrograph ( hrs ; dashed pink lines ) and the 6.5-m magellan ii ( clay ) telescope and magellan inamori kyocera echelle spectrograph ( mike ; dash - dot - dot - dot orange lines ) . brief descriptions of the data and reduction process are now given for each instrument . the data obtained with the vlt and uves used in this analysis were taken during the 2001/2002 and 2004/2005 observing seasons . the data from 2001/2002 were obtained in queue mode and have already been published . the reader is referred to hamilton et al . ( 2003 ) for more information regarding their reduction . the data taken during the 2004/2005 observing season were reduced with a set of custom echelle reduction routines written in idl . the data reduction procedure is described by valenti ( 1994 ) and hinkle et al . ( 2000 ) and includes bias subtraction , flat fielding by a normalized flat spectrum , scattered light subtraction , and optimal extraction of the spectrum . due to nebular h@xmath5 emission near kh 15d , it is important for this study to perform a sky subtraction when reducing the spectra . the vlt / uves spectral format contains enough room between the orders that background sky spectra are recorded both above and below the stellar spectrum in the slit . the spectral traces defining the order locations for the orders in the stellar spectrum were offset in each direction along the slit and sky spectra were extracted . the two sky spectra were averaged and used to subtract the sky background from the spectra of kh 15d . the wavelength solution was determined by fitting a two - dimensional polynomial to @xmath13 as function of pixel and order number , @xmath14 , for several hundred extracted thorium lines observed from an internal lamp assembly . some of the data that were obtained with the keck i telescope and hires for this project were also used as a part of the radial velocity study johnson et al . a full description of the reduction procedure used for the keck data obtained in 2003/2004 is given in that paper . additional keck data obtained in feb 2005 were reduced using the same custom idl echelle reduction routines referenced above in 2.1 . similar to the vlt / uves data , sky subtraction was also performed for the keck / hires data . the january 2004 data obtained with the mcdonald observatory 2.1-m otto struve telescope and ce spectrometer were also used as a part of the radial velocity survey presented by johnson et al . the reader is again directed to that paper for a full description of the reduction procedures for these spectra . the spectral format of the ce is quite compact , requiring a relatively short slit ( in this case 2.@xmath155 ) in order to keep the orders well separated on the ccd . as a result , traditional sky subtraction with this system is not possible . as mentioned above , there is an h@xmath5 component in the line profiles of kh 15d due to the nebular emission from ngc 2264 , the cluster in which kh 15d is located . in order to study the intrinsic variations of the h@xmath5 profile of kh 15d , it is helpful to remove this nebular component . to do so , we utilized a sky spectrum obtained with one of the het / hrs spectra of kh 15d ( see below ) as a proxy for sky subtracting the mcdonald 2.1-m ce data . after continuum normalizing the h@xmath5 order in the 2.1-m ce spectra , we scaled the het sky spectrum so that when it is subtracted from the 2.1-m ce spectra the nearby [ nii ] @xmath166583 emission line is entirely removed . this line is a pure nebular emission line , so the underlying assumption in this procedure is that the ratio of this line to the nebular h@xmath5 line stays constant throughout the nebula surrounding kh 15d . as a check on this assumption , we examined the spectra taken with keck / hires which has a long enough slit to separately extract sky spectra on either side of the stellar spectrum . we scaled the sky spectrum on one side of the stellar specrum so that the flux in the [ nii ] line matched that from the sky spectrum taken on the other side of the star . we then differenced the two sky spectra . in each case , no residual h@xmath5 flux is detected , and the 3@xmath17 upper limit on the h@xmath5 flux in the difference spectrum is @xmath18% of the nebular h@xmath5 emission . it thus appears that this is a fairly robust way to remove the sky@xmath19nebular emission from the 2.1-m ce spectra . the mike spectra were reduced with the mike redux " code written by s. burles , x. prochaska , and r. bernsteini ( see http://web.mit.edu/@xmath1burles/www/mike/ ) . this reduction package performs bias subtraction , flat fielding , order - edge tracing , sky subtraction , and optimal extraction . wavelength calibration is also performed as part of the package using spectra of a thorium - argon arc lamp obtained just prior to each observation of kh 15d . the wavelength solution for the mike data was performed with the custom idl software described above and referenced in valenti ( 1994 ) and hinkle et al . ( 2000 ) . the het / hrs spectra used here were all reduced with the custom idl software described above in 2.1 and referenced in valenti ( 1994 ) and hinkle et al . the hrs instrument at the het is a fiber fed spectrometer . for each observation of kh 15d , sky fibers were placed on either side of the star fiber . as a result of the spectra from these sky fibers appearing on the ccd , the optimal extraction routines in the reduction package were not used . instead , the counts normal to the dispersion direction were summed at each wavelength in order to produce the extracted spectrum . the observations used here come from two programs looking at kh 15d . as a result , about half the observations used 2@xmath20 fibers while the other half used 3@xmath20 fibers . the throughput of the fibers used to feed the stellar and sky spectra vary somewhat . as a result , the two sky spectra were summed to increase the signal to noise and the result was then scaled to match the [ nii ] @xmath166583 flux in the stellar spectrum before the sky subtraction was performed . spectra of a thorium - argon lamp were also extracted for each night and again a wavelength solution was created by fitting a two - dimensional polynomial to @xmath13 as function of pixel and order number , @xmath14 , for several hundred thorium lines . in order to interpret the wealth of complex information produced by the occultation mapping of the h@xmath5 line profile we required the height of star a above or below the occulting screen , which is taken from model 3 of w06 . table 2 lists the julian date of observation , telescope / instrument , orbital phase , the height of star a above the disk , the radial velocity ( rv ) of star a as predicted by w06 , the radial velocity of star b as predicted by w06 , the calculated or measured ( when available ) cousins @xmath21 magnitude ( @xmath22 ) , the measured @xmath10 ( from hamilton et al . 2005 or herbst et al . 2010 ) , the flux ratio used to scale the normalized spectrum , and the barycentric correction . all profiles have been continuum - fit and normalized to 1.0 . their flux was then scaled in reference to the 2002 december 10 out - of - eclipse spectrum obtained with the het at mcdonald observatory . the 2002 december 10 spectrum was chosen as the reference spectrum to which all others would be scaled because star a had the largest height above the occulting screen as predicted by the w06 model on that date and presumably represents the h@xmath5 profile most intrinsic to star a. the scale factor was calculated by computing the difference between the @xmath22 magnitude on the particular date of interest and the @xmath22 magnitude observed on 2002 december 10 . the flux on 2002 december 10 was assumed to be 1.00 . the cousins @xmath21 filter encompasses the wavelength region corresponding to h@xmath5 , making it the best available proxy for measurement of the flux in the continuum near h@xmath5 . actual observed @xmath22 magnitudes were used when available , however , the majority of the photometric monitoring during these seasons was only done in the cousins @xmath9-band . when @xmath22 magnitudes were not available in the photometric database , they were estimated from measured @xmath10 values using a color - magnitude relation from hamilton et al . ( 2005 ) based on usno 2002/2003 data . typical photometric errors are @xmath1 0.01 out of eclipse and @xmath1 0.1 in eclipse ( hamilton et al . 2005 ) . the scaled profiles are shown in figures 4 and 5 . the profiles are plotted from left to right in these figures according to distance above or below the projection of the edge of the occulting disk as calculated by w06 . they are shown , in reference to figure 2 , from out of eclipse , through ingress , during mid - eclipse , through egress , and out of eclipse again . the profiles are plotted this way to assist in identifying whether or not there are systematic differences in the profiles given the specific location in the orbit , not just the height above or below the occulting disk . all profiles have been corrected for the earth s motion , and are shown in the rest frame of the center of mass of the system . when trying to understand the complex variations in the h@xmath5 line profiles presented here , one should consider three more or less independent factors influencing the variability : ( 1 ) the edge effect " - occultation by the cbd as measured by the distance @xmath11x with respect to the occulting edge ; ( 2 ) the orbital phase effect " - potential variability due to increased accretion on both stars or other changes during certain phases of the eccentric binary orbit ( see de val - borro et al . 2011 and references therein ) ; and ( 3 ) an irregular / chaotic component that is best seen when comparing the profiles taken at similar heights and orbital phase but at different times . we have several multi - day spectral sequences ( some with coverage on consecutive days ) obtained during ingress , egress , and while the observable star is out of eclipse . these observations occurred over the course of five different observing seasons so we can assess both the short - term and long - term variability of the h@xmath5 emission line profile . in the following section , we will begin by demonstrating the changes that are seen in the h@xmath5 emission line profile as the stars move throughout their orbits , followed by changes that are produced as star a is occulted by the cbd . finally , we will discuss variations observed which can be attributed to the irregular nature of the t tauri accretion process . by looking at figures 4 and 5 , it is evident that the kh 15d system exhibits highly variable h@xmath5 profiles . we begin our analysis by taking the appropriately scaled ( see 3 ) profiles from figures 4 and 5 and averaging them according to the predicted distances of star a with respect to the edge of the occulting disk at the time of observation . the spectra were split up into two sets of 7 different bins ( @xmath11x = 4 to 2 , @xmath11x = 2 to 1 , @xmath11x = 1 to 0 , @xmath11x = 0 to + 1 , @xmath11x = + 1 to + 2 , @xmath11x = + 2 to + 4 , and @xmath11x = + 7 to + 10 ; see table 3 for details ) , with the first set of bins for the spectra taken during ingress ( approaching mid - eclipse , phases 0.5 to 0 according to figure 2 ) , and the second set of bins for those taken during egress ( following mid - eclipse , phases 0 to + 0.5 according to figure 2 ) . these sets of averaged profiles are displayed in figure 6 . to orient the reader , time " in figure 6 runs down the left hand side of panels and then up the right hand side of panels back to the top . as one examines the averaged , scaled emission line profiles presented in figure 6 , three distinct features " can be discerned and are indicated by arrows in the appropriate panels . they are : 1 . the absorption component seen mostly out of eclipse and during ingress / egress . the broad component mainly seen once star a is fully occulted . 3 . the double peak emission profile primarily observable during full eclipse . the top four panels in figure 6 ( a - d ) represent the out - of - eclipse spectra obtained when @xmath11x ranges from 4 to 1 . the next four panels ( e - h ) show the ingress and egress spectra obtained when @xmath11x ranges from 1 to + 1 . in these eight panels , a central " absorption feature near velocity = 0 is predominant , and sometimes extends well below the stellar continuum . the bottom six panels in figure 6 ( i - n ) show the averaged h@xmath5 profile after the star is completely obscured and at a distance of @xmath11x = + 1 or greater . here , we see a generally double - peaked emission line profile , with an underlying broad component . the underlying broad component ranges in velocity from approximately -300 to + 300 km s@xmath7 . to emphasize the presence of this faint broad component , which is difficult to see in the regularly scaled plots , the profiles plotted in the bottom 6 panels of figure 6 have been multiplied by a factor of 10 and over - plotted on the same graph . these modified profiles are shown as a dash - dot - dot - dot line in red . note that the bottom two panels in figure 6 ( m and n ) are identical . the largest variation in the profiles examined here naturally occurs during the eclipse itself as the star and its magnetosphere sink below , and alternatively rise from behind , the occulting screen . to focus on these phases , we show in figures 7 and 8 , two ingress sequences ; one from 2003 february and one from 2004 december . the ingresses are plotted two ways : 1 ) the top panel shows the fluxed profile with an offset , so that each individual profile is visible . the zero point for each spectrum is indicated by the small horizontal line along the y - axis . 2 ) the bottom panel shows each spectrum properly fluxed with its continuum adjusted back up to 1.0 so that they can be plotted on top of one another allowing one to see the systematic decline of the flux in the far wings . figures 7 and 8 demonstrate that a typical " ingress spectrum may be characterized as an inverse p cygni type profile , consisting of red - shifted h@xmath5 absorption , superimposed upon blue - shifted h@xmath5 emission . we note that the absorption feature starts out extending well below the stellar continuum . as star a sinks below the occulting edge during ingress , the overall photospheric spectrum and all of the absorption that is seen in the h@xmath5 profile drops to near zero flux levels . the profile that is left once the star is fully occulted and several stellar radii below the occulting edge primarily consists of a rather narrow double - peaked emission feature with peaks at @xmath1 @xmath23 30 km s@xmath7 , as well as an underlying faint broad component ranging in velocity from approximately -300 to 300 km s@xmath7 , as mentioned earlier . this double - peaked emission feature was first noted by hamilton et al . ( 2003 ) , and its interpretation in terms of a bipolar jet has been discussed in great detail by mundt et al . ( 2010 ) , and will not be further discussed here . after mid - eclipse , or periastron , star a emerges again , rising from behind the occulting screen during egress . during this phase , we see a more generally enhanced , roughly symmetric , double - peaked h@xmath5 emission profile with a slightly blue - shifted absorption feature , as shown in figure 9 . additionally , figure 9 shows that during this particular egress , there is a reduction of peak emission flux as more of the star is exposed ( i.e. , 2005 february 28 vs. 2005 march 01 ) . this seems to be contrary to what one would expect ; as more of the star is exposed , more of the h@xmath5 emitting region , i.e. , the magnetosphere , should be exposed , and therefore a greater amount of h@xmath5 flux would be expected . we return to this issue later . in summary , the disappearance / reappearance of both the central " absorption feature as well as the overall photospheric spectrum observed in the kh 15d system can be attributed to the effect that the edge of the cbd and its occultation of star a and its close stellar environment has on the observed h@xmath5 emission line spectrum . additionally , the variation from inverse p cygni profile to double - peaked profile must be dependent on where star a is in its orbit , since the inverse p cygni profile is mostly observed out of eclipse and into ingress . the significance of these observations will be discussed in 5 . t tauri stars ( ttss ) are generally characterized by their irregular variability due to accretion and other effects . as a way of assessing the baseline variability of the visible tts in this system , high - resolution spectra were obtained during the out - of - eclipse phases with the het during the 2002/2003 ( hessman , pi ) and 2005/2006 ( herbst , pi ) observing seasons . these data have been scaled by the appropriate flux ratio , had their continuua adjusted back up to 1 for ease of comparison , and are shown in figure 10 . the left - hand side of figure 10 shows the h@xmath5 emission line profiles obtained out - of - eclipse in 2002 december and 2003 january / february . the bottom panel shows the 2002/2003 light curve and the vertical lines indicate when the spectra were obtained . the phases are listed in each plot as are the model predicted heights above the edge of the cbd on that given date . on the right - hand side of figure 10 , we show the out - of - eclipse h@xmath5 emission line profiles obtained during 2005 december and 2006 february . the bottom panel shows where the spectra were obtained in reference to the 2005/2006 light curve . focusing our attention first on the 2002/2003 season , we note that in this season , the spectra were obtained when star a was located anywhere between @xmath243 and 4 stellar radii above the edge of the cbd ( see figure 3 and table 3 ) . the phases also range from + 0.42 to 0.37 , passing through phase @xmath25 0.5 . we remind the reader that positive phases are those following mid - eclipse , representing the egress side of the orbit , while negative phases are those preceding mid - eclipse , representing the ingress side of the orbit . a phase of @xmath250.5 is exactly opposite mid - eclipse , which is denoted as phase = 0 . a comparison of the blue - shifted peaks observed in 2002 december and 2003 jan / feb does not show a great deal of variation in the maximum flux within the line . in both cases , the relative line flux is @xmath24 2.1 - 2.2 at the peak . the maximum flux occurs when the star is at its greatest height ( @xmath11x = 4.02 , phase = 0.48 for 2002 december , and @xmath11x = 3.95 , phase = + 0.49 for 2003 january ) . what is highly variable , however , is the emission , and alternatively , the absorption , in the red - shifted portion of the h@xmath5 line profile . the interpretation of this absorption feature and the cause of its variability is discussed in 4 . when comparing the observed profiles obtained during 2005/2006 season , one can see that the maximum flux in the blue - shifted peak in 2005 december is greater than that observed in 2006 february , one full orbital cycle apart , suggesting that this may be the result of the intrinsic variability of ttss . however , it is important to note that the 2005 december spectra were taken as star a was rising up from the edge of the cbd , as denoted by its x position ( phases ranging from 0.38 to 0.46 ) . we see in this set of profiles that as the star continues to rise from behind the edge , the red - shifted emission increases in flux as the blue - shifted emission becomes somewhat narrower and lower in flux . during this time , we also see an increase in the depth of the central " absorption feature , eventually extending below the continuum . comparatively , the profiles obtained in 2006 february ( phases ranging from 0.39 to 0.5 and then to 0.47 ) are similar in peak blue - emission flux to the profiles shown on the left - hand side of figure 10 from 2002/2003 . the maximum flux occurs around phase 0.5 when the star is at its greatest predicted height above the occulting edge ( 2006 february 12 ) . here we see both blue and red emission with a narrow central " absorption that extends below the stellar continuum for the first three profiles ( 2006 february 9 , 10 , and 12 ) . as the star begins its descent " back toward the obscuring edge ( phases have gone from + to ) , the red - shifted emission feature disappears and is replaced by a broad absorption feature , apparently separate from the central " absorption , and centered around + 100 km s@xmath7 . this type of change in profile shape is also seen in the top left panel of figure 10 ( 2002 december ) . as star a moves through phase 0.5 and begins its descent to ingress , the profile loses its red - ward emission as it is replaced by a broad red - shifted absorption feature . note also that the peak blue - emission flux in 2006 february is similar in nature to the peak blue - emission flux seen in the out - of - eclipse spectra observed in 2002/2003 , and shown on the left - hand side of figure 10 . in summary , the observations show that the h@xmath5 emission line profile in the kh 15d system varies systematically with the position of star a in its orbit and as a result of obscuration of the h@xmath5 emitting region by the cbd . we see a characteristic inverse p cygni profile during ingress and a double - peaked , enhanced emission profile during egress . our observations also reveal that the central " absorption feature is primarily red - shifted with respect to the systemic velocity during ingress and blue - shifted during egress . additionally , it appears as if the profile makes its transition from double - peak to inverse p cygni just as star a passes through phase = @xmath25 0.5 , or apastron . in the next section , we explore the various components that make up the h@xmath5 emission line profile . it is clear from figs . 4 and 5 that there is a substantial , but highly variable absorption component to the h@xmath5 profiles when star a is above the occulting edge . figure 6 shows that this feature depends on the orbital phase of star a. during ingress the central absorption is broad and red - shifted relative to the systemic velocity ( see panels a , c , e , and g of figure 6 ) . during egress , the absorption feature tends to be narrower and more central or slightly blue - shifted ( figure 6 , panels b , d , f , and h ) . whenever star a is visible above the cbd , the absorption feature extends below the continuum , indicating that the gas responsible for it is at least partly projected onto the stellar photosphere . to quantify the velocity behavior of the absorption component we have fit the profiles when the star is not fully eclipsed with three gaussians . normally these were chosen to fit the red and blue emission wings and the central absorption . in one or two cases the red - shifted absorption was so strong that there was no blue emission wing visible and we fit two absorption components . note that the intention here is to simply help quantify the mean velocity of the absorbing material clearly visible in the profiles . examples of this fitting are shown in figure 11 , with the combined fit shown as the solid curve in red . note that in this figure , the left hand panels represents the ingress portion of the orbit , while the right hand panels represents the egress portion of the orbit . additionally , we draw attention to the fact that the top panels show out of eclipse spectra ( @xmath11x values @xmath1 + 4 ) , while the bottom panels represent spectra when the star was mostly eclipsed ( @xmath11x @xmath1 + 0.5 ) . plotting the profiles in this way helps to identify the differences between ingress and egress . there is a substantial amount of red - shifted absorption that comes and goes on the orbital cycle , being strongest during ingress and weaker or perhaps even absent during egress . it is not clear whether there is a separate narrow absorption component centered near the systemic velocity or the velocity of star a. in at least one case shown in figure 11 , where we used two absorption components to model the line , this appears to be the case . the phase - dependent red - shifted component seen during ingress is , however , always present . figure 12 shows the velocity of the red - shifted absorption component plotted against binary phase . the top panel shows the velocity of the gas relative to the center of mass of the system , while the bottom panel shows it in the frame of reference of star a. again , the feature is only visible when at least some portion of star a is above the occulting edge , so there is a gap during perihelion passage . it is quite clear from the figure that as star a becomes visible on the far side of the disk hole from us , the central absorption is redshifted by only a small amount ( @xmath110 km s@xmath7 ) with respect to star a. but once the star turns the corner " and starts to move toward the observer on its way to the near side of the gap where it will soon begin to be occulted , the highly red - shifted absorption component appears , with velocities reaching up to + 100 km s@xmath7 or more . we believe that this relatively high velocity absorption component is a regular feature of the system that varies with orbital phase , and is not merely a stochastic feature attributable to unsteady accretion . the evidence is that the feature is present in every spectrum during the appropriate orbital phase , even those taken many cycles ( and years ) apart . it is not always precisely the same strength or velocity , but it is always identifiable as a feature of the h@xmath5 profile . this suggests that it is a stable accretion feature ( or stream " ) associated with star a. it is interesting that models of accretion in eccentric binary systems ( e.g. , de val - borro et al . 2011 ) predict precisely the sort of stable accretion streams that these observations require . in these models , the disk and binary are co - planar , however one can expect the general flow pattern to be rather similar for a slightly inclined system ( 10 - 20@xmath26 ) . furthermore , one can expect stronger accretion streams in the inclined system , which may help to understand why this wtts has such strong accretion signatures . a schematic representation of how such an accretion stream would need to be configured to explain our observations was introduced in figure 1 . detailed modeling will be required to determine whether such a model satisfies the observations , but an accretion stream does seem promising as a way to understand this absorption component and its variation with orbital phase . note that the maximum velocity observed , 100 km s@xmath7 , is close to the free - fall velocity for gas falling from the inner edge of the cbd near 0.6 au to @xmath110 stellar radii . in this interpretation , then , we would identify the high velocity gas observed during ingress to be material in the accreting stream just outside the magnetosphere . it is quite possible that the gas accretes onto a disk at this location where it awaits a more rapid transfer to the star ( or ejection into a jet ) during perihelion passage ; see next section ) . the lower velocity , but still accreting , gas seen during egress and perhaps at all times , may be part of a remnant disk of gas that did not accrete during perihelion passage or the outer parts of the accretion stream . additionally , since the streams are in co - rotation , when star a is at apastron , gas will be seen approaching with much larger velocities than the star , which would produce a blue - shifted wing . again , a detailed analysis of these events requires a dynamical model tuned to the properties of the kh 15d system and that is beyond the scope of this work . figure 6 shows clearly that the high velocity wings of the h - alpha profiles vary substantially and systematically with orbital phase . for single cttss , this emission is believed to arise within the magnetospheres of stars as they accrete material from a circumstellar disk and its strength is a proxy for the accretion rate . during phases when star a is only partly occulted or entirely visible the high velocity component is associated with its magnetosphere . at other phases there may be a contribution from star b or from the combined magnetospheres . during these fully occulted phases much , if not all , of the light reaches us by reflection from the back wall of the cbd ( herbst et al . 2008 ) , allowing us to observe phenomena occurring during perihelion passage that would otherwise be invisible . the data show an interesting and clear trend that repeats in all cycles observed . namely , we observe that the high velocity emission is more intense following perihelion passage ( i.e. during egress ) than it is prior to perihelion passage ( i.e. during ingress ) . this may be seen by comparing the left hand side of figure 6 , corresponding to ingress , and the right hand side , corresponding to egress . the effects are particularly noticeable in panels near the bottom that apply to the fully occulted system just before and just after perihelion passage . on the left hand panels , one clearly sees the high velocity emission but it is nowhere near as intense as it is on the right hand panels . to make this clearer we have multiplied the relevant profiles by a factor of ten in the bottom six panels . another way to demonstrate this point is to compare individual profiles obtained at the same locations with respect to the occulting disk and determine whether the ingress profiles look different from the egress profiles . this direct comparison is shown in figure 13 where the spectra have been properly fluxed and had their continuum adjusted back up to 1.0 , and for two different values of @xmath11x where we happened to have both ingress and egress spectra . it is clear that at both locations relative to the occulting disk ( partly visible , @xmath11x @xmath1 + 0.5 , top panel , and fully occulted , @xmath11x = + 1.09 , bottom panel ) the magnetospheric emission is more intense during egress than ingress . this will be quantified further in the next section where we discuss the equivalent widths of the blue emission wing in more detail . we note that in addition to the phase - dependent variations discussed above there is clearly a time - dependent variation as well . it is not the case that every egress spectrum looks identical in its profile to others obtained at the same orbital phase . this time variability is illustrated in figure 14 where we show some egress spectra obtained near the same orbital phase and at times when star a was fully occulted , or nearly so . these spectra are of somewhat lower resolution than many but serve to make the main point : there is substantial variation in the emission line wings that is independent of orbital phase or location relative to the occulting disk edge . the 2001 dec 20 spectrum is particularly notable , and was discussed in hamilton et al . ( 2003 ) in the context of magnetospheric accretion . here we see a very extended and substantial wing indicating that during this perihelion passage there was a substantial amount of activity . in figure 15 we show a sequence of spectra taken during a single egress while the system was fully occulted . here , also , we see substantial variability in the high velocity emission , now on a daily timescale . there is some evidence for a high velocity feature " that shows up as a small bump in the blue wing on 2005 dec 12 and migrates to lower velocity and higher intensity by 2005 dec 12 . whether this is related to the changing aspect of the star caused by stellar rotation and orbital motion or to some feature of the accretion is unknown . it is clear , however , that there is variability in the magnetospheric emission even on timescales as short as a day . our observations appear to confirm a general prediction of models of the binary accretion process in eccentric systems ( artymowicz & lubow 1996 ; gnther & kley 2002 ; and de val - borro et al . 2011 ) that accretion onto the stars occurs predominantly during periastron . it is during egress , i.e. after perihelion passage , that we see the most activity in the high velocity wings of the line and the greatest intensity overall . similar results have been reported for dq tau ( whose binary parameters are very similar to those for kh 15d ; see table 4 ) by mathieu et al . ( 1997 ) and basri et al . ( 1997 ) , for v4046 sgr by stempels & gahm ( 2004 ) , for uz tau e by jensen et al . ( 2007 ) , and for v773 tau a by boden et al . ( 2007 ) . in some cases mentioned above , the evidence for pulsed accretion includes phase - dependent continuum variations detected by broad band photometric studies . it has long been known that kh 15d shows a brightening near perihelion passage ( see hamilton et al . 2001 ; hamilton et al . 2005 ; herbst et al . 2010 ) at @xmath10 and some variations in color . the general phenomenon has been interpreted as the influence of star b , which is inferred to be the brighter of the pair , reaching its greatest extension toward the edge of the cbd at perihelion . at these orbital phases it , therefore , becomes the dominant source of the reflected light . it could , of course , be that some of this variation is caused by an increased brightness of one or both stars during perihelion passage . if the accretion is primarily onto the lower mass star ( star a ) , we may not be able to witness the more dramatic brightness enhancements that occur , because star a is at its lowest point with respect to the disk edge then and perhaps completely invisible to us , even by reflected light . this would , however , be in contrast to the current models , which predict that mass is preferentially channeled to the primary . although the photometry can not directly confirm the case for increased accretion during perihelion passage in the case of kh 15d , they are certainly not inconsistent with the spectroscopic evidence reported here . our data taken during eclipses can , at most , be used to constrain the size of the magnetosphere of star a in the vertical direction . as outlined in previous papers ( e.g. , herbst et al . 2002 ) the occulting edge of the circumbinary disk / ring is extremely sharp , " i.e. , the transition region from optically thin to optically thick is less than @xmath1 0.1 r@xmath2 ( and perhaps much less ) so that the eclipses can be modeled with a knife - edge . " the magnetospheric emission model profiles of hartmann and collaborators ( hartmann et al . 1994 ; muzerolle et al . 1998 ; muzerolle et al . 2001 ) show that the blue wing of these profiles is an emission component , while the red wing has contributions from both emission and absorption . we have calculated the emission flux and equivalent width of the blue - shifted wing of the h@xmath5 profile between 280 km s@xmath7 and 85 km s@xmath7 , ignoring the velocity range from 0 to 85 km s@xmath7 because it is likely affected by emission from the jet . for velocities less than -280 km s@xmath7 , there is very little emission . in figure 16 we show the fluxes ( top panel ) and equivalent width ( lower panel ) for two eclipse sequences obtained during ingress ( 2003 feb and 2004 dec ; red squares and blue triangles , respectively ) and two eclipse sequences obtained during egress ( 2005 feb and 2005 dec ; green stars and pink circles , respectively ) . each ingress or egress is indicated in the upper left corner of the figure , and the change in position from the center of mass of the binary of star a ( r ) is indicated for each ingress / egress shown . for reference , the binary semi - major axis is 0.25 au = 41.4 r@xmath2 . note that for the ingress observations , time increases from left to right , while for the egress observations it is just the opposite . to interpret the data displayed in figure 16 , we make use of the knife - edge " model , and assume that the emission region is spherically symmetric and homogeneously emitting as an initial thought experiment . we begin with the top panel of figure 16 , which shows the blue wing emission flux . within the knife - edge model , the somewhat rapid decrease of the relative line fluxes during ingress can be most easily explained by a decreasing geometrical size of the emission region ( i.e. , less of the emission region is exposed as star a sets " below the edge of the occulting disk ) , provided this region is not highly variable on time scales of a few days . in principal , most of the flux data for the two egress phases can be explained in the same way . however , for the 2005 december egress , there is also a strong time variable component superimposed , which causes a strong decline in the line flux for the end of the egress phase ( recall that for egress , increasing time runs from right to left ) . as already outlined elsewhere in this paper ( see 5 ) the line fluxes during egress are typically a factor 3 higher than during ingress , which has been attributed to increased accretion from the circumbinary disk / ring onto the two stars at the time near or after periastron passage . to constrain the size of the h@xmath5 emission region _ relative to the size of the stellar disk _ , we use the equivalent width ( ew ) of the emission contained within the velocity range of 85 to 280 km s@xmath7 , because the ew is a measure of the _ ratio _ between line flux and continuum flux . to interpret the ew measurements within the context of the above mentioned knife - edge " model , assuming a spherically symmetric emission line region , we consider two extreme cases : 1 ) the case of an emission region that is at least a few times larger than the stellar disk , and 2 ) the other extreme where the emission region is not extended and is roughly the same size as the stellar disk . in the first case , the ew should increase between @xmath11x = 0 and @xmath11x = + 1 , because at @xmath11x = + 1 , all of the stellar disk is effectively covered and the only continuum emission visible would be that due to the scattered component . if the emission region were indeed larger than the stellar disk , say with a radius of 2r@xmath2 , only a fraction of that region would be covered . therefore , the ratio between emission line flux and continuum flux should be highly increased . we do , in fact , see a slight increase in the ew between @xmath11x = 0 and @xmath11x = + 1 for both ingresses , however , it is more pronounced for the 2003 feb ingress . if instead , we examine the second case where the emitting region is roughly the same size as the stellar disk , we would not expect a change in the ew at all . it is difficult to interpret the observed ew measurements in exactly this way , though the relatively small change in ew suggests a relatively small magnetosphere . the egress data displayed in figure 16 show that during egress the ew of h@xmath5 is about a factor of 2 - 4 greater than during ingress . this means that within the simplified spherically symmetric model considered here either the radius or the surface brightness of the line emission region has increased . unfortunately , the egress data are not well sampled between @xmath11x = 0 and @xmath11x = + 1 in order to check whether the radius of the emission line region has indeed increased . the 2005 feb egress data suggest that there is a factor of 2 decrease in ew from @xmath11x = + 1 to + 0.5 , but this rests on only one data point . in 2005 dec , the same sort of trend is seen ( decreasing ew as the star moves closer to the occulting edge ) but the trend is not as steep for that egress , and the ew is on average greater . the ew data in figure 16 can only constrain the extended emission perpendicular to the knife edge since it is believed that star a moves roughly up and down with respect to the occulting edge . we can not exclude a disk - like emission region around the star with a disk of e.g. several stellar radii in size . in fact , in the models of hartmann et al . ( 1994 ) , the vertical extent of their magnetospheric accretion flow is barely bigger than the star , however , the magnetosphere extends out to 2 - 3 r@xmath27 . at most , we can argue that this disk like structure could be roughly 10 - 15% thicker than the stellar diameter ( in the direction perpendicular to the knife edge ) . if the h@xmath5 emitting region were uniformly bright , we would expect that the ew measurements for both ingress and egress to be the same . we show in figure 17 , the flux ( top panel ) and ew ( bottom panel ) of the h@xmath5 emission contained within the velocity range of 85 to 280 km s@xmath7 for all observed profiles . open symbols represent ingress measurements , while filled symbols represent egress measurements . here we see a generally increasing ew for ingress from @xmath11x = 0 to @xmath11x = + 1 , and that all egress measurements are a factor of 2 - 4 times greater than ingress measurements . if we take the ew of the h@xmath5 emitting region as a proxy for accretion rate , we can conclude that the accretion rate onto star a has increased significantly after periastron ( phase = 0 ) . this has been modeled for eccentric binaries and is shown in figure 2 of artymowicz & lubow ( 1996 ) , figure 12 of gnther & kley ( 2002 ) , and de val - borro et al . both gnther & kley ( 2002 ) and de val - borro et al . ( 2011 ) model the dq tau system , which can be used as a proxy for the kh 15d system ( see table 4 ) . all models demonstrate that the accretion rates are greater during periastron passage , however , both artymowicz & lubow ( 1996 ) and gnther & kley ( 2002 ) show that the peak accretion occurs just prior to periastron passage . to compare our results to those presented in figure 2 of artymowicz & lubow ( 1996 ) , we plot our ew measurements as a function of phase and approximated their model for gas accretion rate as a simple gaussian . figure 18 shows that if we scale the gaussian to match our data on ingress ( solid red line ) , we can not match the egress accretion rate . if instead , we scale the gaussian to match our egress data , we can not match the ingress observations ( dashed blue line ) . this suggests that the simulation does not quite accurately predict the timing of the increase in accretion , or there may be something significant within the kh 15d system that is not included within the model , like a differential occultation of the magnetosphere . the early calculations by artymowicz & lubow ( 1996 ) used here were based on smooth particle hydrodynamics and rather coarse . a new theoretical approach to this problem is warranted . our spectroscopic studies have revealed intrinsic variability and amazing structure in the h@xmath5 line profiles caused by high velocity gas flows in the vicinity of the stars within the kh 15d system . the profiles observed during ingress and egress are distinctly different , changing from an inverse p - cygni profile during ingress to an enhanced double - peaked profile with broad extended wings during egress . the differences in these profiles can be understood in terms of models of accretion flows within an eccentric binary ( artymowicz & lubow 1996 ; gnther & kley 2002 ; and de val - borro et al . measurements of the flux and ew of the blue - shifted emission located in the wings of h@xmath5 ( velocity ranging from 280 km s@xmath7 to 85 km s@xmath7 ) give us an indication that the h@xmath5 emitting region is compact and variable in brightness on the time scale of the orbital period . there is potentially a great deal more information locked in these spectra than we have been able to extract . the present study will hopefully serve as an incentive and guide to more detailed modeling . the dynamics of the binary system are very well understood ( w06 ) and it should be quite possible to model with definiteness the expected gas flows and accretion dynamics using codes that have already been applied to similar systems . these data could potentially serve as a critical test of such models . some subtleties of the profile variations have undoubtedly escaped us and comparison with models of this particular system may reveal them . we would like to thank the anonymous referee for the suggested improvements to this paper . we also thank dr . frederic v. hessman for graciously allowing us to use the data he collected with the het . c.m.h would like to thank dr . eric mamajek and dr . david james for helpful advice and discussions regarding the creation of figure 1 . c.m.h . acknowledges partial support by the american association of university women through an american fellowship . .- k . acknowledges partial support by the nasa origins of solar systems program through the following grants to rice university : nnx08ah86 g and nnx10ai53 g . w.h . acknowledges partial support by the nasa origins of solar systems program . some of the 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 . the authors wish to recognize and acknowledge the very significant cultural role and reverence that the summit of mauna kea has always had within the indigenous hawaiian community . we are most fortunate to have the opportunity to conduct observations from this mountain . the hobby - eberly telescope ( het ) is a joint project of the university of texas at austin , the pennsylvania state university , stanford university , ludwig - maximilians - universitt mnchen , and georg - august - universitt gttingen . the het is named in honor of its principal benefactors , william p. hobby and robert e. eberly . this paper is based in part on observations collected at the european southern observatory ( program 074.c- 0604a ) . c.m.h . would also like to thank darla and steve mckee , as well as the masci family , for graciously supporting her and her daughter throughout the beginning , and final production , of this work , respectively . this paper is dedicated to robert masci . hinkle , k. , wallace , l. , valenti , j. , & harmer , d. 2000 , visible and near infrared atlas of the arcturus spectrum 3727 - 9300 ed . k. hinkle , l. wallace , j. valenti , & d. harmer . ( san francisco : asp ) isbn : 1 - 58381 - 037 - 4 2001 nov 29 & vlt / uves & 4800 - 6800 & 44,000 + 2001 dec 14 & vlt / uves & 4800 - 6800 & 44,000 + 2001 dec 20 & vlt / uves & 4800 - 6800 & 44,000 + 2002 dec 06 & het / hrs & 6380 - 7330 & 15,000 + 2002 dec 10 & het / hrs & 6380 - 7330 & 15,000 + 2002 dec 13 & het / hrs & 6380 - 7330 & 15,000 + 2003 jan 23 & het / hrs & 6380 - 7330 & 15,000 + 2003 jan 26 & het / hrs & 6380 - 7330 & 15,000 + 2003 feb 02 & het / hrs & 6380 - 7330 & 15,000 + 2003 feb 05 & het / hrs & 6380 - 7330 & 15,000 + 2003 feb 08 & keck / hires & 4200 - 6600 & 70,000 + 2003 feb 09 & keck / hires & 4200 - 6600 & 70,000 + 2003 feb 10 & keck / hires & 4200 - 6600 & 70,000 + 2003 mar 06 & het / hrs & 6380 - 7330 & 15,000 + 2003 mar 23 & het / hrs & 6380 - 7330 & 15,000 + 2003 dec 16 & keck / hires & 4400 - 6800 & 40,000 + 2003 dec 18 & keck / hires & 4400 - 6800 & 40,000 + 2004 jan 04 & mcd / ce & 5600 - 6900 & 30,000 + 2004 jan 05 & mcd / ce & 5600 - 6900 & 30,000 + 2004 jan 10 & mcd / ce & 5600 - 6900 & 30,000 + 2004 feb 05 & keck / hires & 4400 - 6800 & 39,000 + 2004 mar 10 & keck / hires & 4700 - 7100 & 40,000 + 2004 mar 12 & keck / hires & 4800 - 7100 & 40,000 + 2004 dec 10 & mcd / ce & 5600 - 6900 & 30,000 + 2004 dec 13 & vlt / uves & 5800 - 7300 & 44,000 + 2004 dec 14 & vlt / uves & 5900 - 7300 & 44,000 + 2004 dec 15 & vlt / uves & 5800 - 7300 & 44,000 + 2004 dec 16 & vlt / uves & 5800 - 7300 & 55,000 + 2004 dec 17 & vlt / uves & 5800 - 7300 & 55,000 + 2004 dec 18 & vlt / uves & 5800 - 7300 & 55,000 + 2005 feb 27 & keck / hires & 4800 - 7100 & 42,000 + 2005 feb 28 & keck / hires & 4800 - 7100 & 42,000 + 2005 mar 01 & keck / hires & 4800 - 7100 & 42,000 + 2005 dec 12 & mag / mike & 5800 - 6800 & 25,000 + 2005 dec 13 & mag / mike & 5800 - 6800 & 25,000 + 2005 dec 14 & mag / mike & 5800 - 6800 & 25,000 + 2005 dec 15 & mag / mike & 5800 - 6800 & 25,000 + 2005 dec 20 & het / hrs & 5880 - 6770 & 30,000 + 2005 dec 21 & het / hrs & 5880 - 6770 & 30,000 + 2005 dec 22 & het / hrs & 5880 - 6770 & 30,000 + 2005 dec 23 & het / hrs & 5880 - 6770 & 30,000 + 2005 dec 24 & het / hrs & 5880 - 6770 & 30,000 + 2006 feb 07 & het / hrs & 5880 - 6770 & 30,000 + 2006 feb 09 & het / hrs & 5880 - 6770 & 30,000 + 2006 feb 10 & het / hrs & 5880 - 6770 & 30,000 + 2006 feb 12 & het / hrs & 5880 - 6770 & 30,000 + 2006 feb 13 & het / hrs & 5880 - 6770 & 30,000 + 2006 feb 14 & het / hrs & 5880 - 6770 & 30,000 + 2452242.7446 & vlt / uves & -0.26 & -1.46 & -9.95 & 4.52 & 15.28 & 14.49 & 0.990 & 16.23 + 2452257.8271 & vlt / uves & 0.05 & 8.74 & 59.86 & -27.21 & 18.86 & 18.18 & 0.036 & 9.10 + 2452263.7026 & vlt / uves & 0.17 & 1.95 & -2.19 & 0.99 & 18.31 & 17.61 & 0.060 & 6.35 + 2452614.9333 & het / hrs & 0.44 & -3.90 & -18.89 & 8.58 & 15.25 & 14.48 & 1.020 & 12.80 + 2452618.9292 & het / hrs & -0.48 & -4.02 & -18.43 & 8.61 & 15.27 & 14.50 & 1.000 & 10.89 + 2452621.7431 & het / hrs & -0.43 & -3.65 & -17.34 & 7.88 & 15.26 & 14.49 & 1.010 & 9.90 + 2452662.6451 & het / hrs & 0.42 & -3.68 & -18.83 & 8.55 & 15.25 & 14.48 & 1.020 & -10.68 + 2452665.8014 & het / hrs & 0.49 & -3.95 & -18.76 & 8.52 & 15.25 & 14.48 & 1.020 & -12.51 + 2452672.6090 & het / hrs & -0.37 & -2.91 & -15.82 & 7.18 & 15.29 & 14.52 & 0.980 & -15.23 + 2452675.7688 & het / hrs & -0.31 & -1.71 & -12.99 & 5.90 & 15.29 & 14.52 & 0.980 & -16.92 + 2452678.8193 & keck / hires & -0.26 & -0.12 & -8.94 & 4.06 & 15.57 & 14.80 & 0.760 & -18.29 + 2452679.8276 & keck / hires & -0.24 & 0.50 & -7.20 & 3.27 & 16.20 & 15.46 & 0.420 & -18.71 + 2452680.8356 & keck / hires & -0.22 & 1.17 & -5.15 & 2.42 & 18.09 & 17.39 & 0.070 & -19.13 + 2452704.6771 & het / hrs & 0.29 & -1.39 & -15.35 & 6.97 & 15.33 & 14.56 & 0.950 & -26.23 + 2452721.6361 & het / hrs & -0.36 & -2.59 & -15.33 & 6.96 & 15.28 & 14.46 & 0.990 & -28.70 + 2452990.0431 & keck / hires & 0.18 & 2.72 & -4.83 & 2.19 & 18.29@xmath28 & 17.62 & 0.060 & 7.84 + 2452992.0413 & keck / hires & 0.22 & 1.09 & -10.63 & 4.83 & 18.11@xmath28 & 17.42 & 0.070 & 6.84 + 2453008.7960 & mcd / ce & -0.42 & -2.71 & -17.33 & 7.87 & 15.25 & 14.47 & 1.020 & -1.33 + 2453009.8729 & mcd / ce & -0.40 & -2.48 & -16.74 & 7.60 & 15.22 & 14.44 & 1.050 & -2.06 + 2453014.7920 & mcd / ce & -0.30 & -0.77 & -12.58 & 5.71 & 15.34 & 14.57 & 0.940 & -4.42 + 2453040.8025 & keck / hires & 0.22 & 0.92 & -11.45 & 5.20 & 18.24 & 17.56 & 0.060 & -16.88 + 2453074.7882 & keck / hires & -0.07 & 7.61 & 23.23 & -10.56 & 18.85@xmath28 & 18.25 & 0.040 & -27.41 + 2453076.8125 & keck / hires & -0.03 & 9.36 & 42.73 & -19.43 & 18.51@xmath28 & 17.81 & 0.050 & -27.79 + 2453349.8684 & mcd / ce & -0.38 & -1.31 & -15.86 & 7.20 & 15.28 & 14.51 & 0.990 & 10.85 + 2453352.7976 & vlt / uves & -0.31 & -0.28 & -13.29 & 6.03 & 15.35 & 14.59 & 0.930 & 9.56 + 2453353.6774 & vlt / uves & -0.29 & 0.11 & -12.31 & 5.59 & 15.51 & 14.74 & 0.800 & 9.28 + 2453354.7199 & vlt / uves & -0.27 & 0.60 & -11.01 & 5.00 & 16.35 & 15.60 & 0.370 & 8.73 + 2453355.6510 & vlt / uves & -0.25 & 1.09 & -9.69 & 4.40 & 17.55 & 16.85 & 0.120 & 8.32 + 2453356.6724 & vlt / uves & -0.23 & 1.66 & -8.04 & 3.65 & 17.89 & 17.20 & 0.090 & 7.80 + 2453357.6852 & vlt / uves & -0.21 & 2.28 & -6.16 & 2.79 & 18.20 & 17.50 & 0.067 & 7.28 + 2453428.8268 & keck / hires & 0.26 & 1.09 & -13.25 & 6.02 & 17.31 & 16.60 & 0.150 & -24.91 + 2453429.8250 & keck / hires & 0.28 & 0.52 & -14.70 & 6.67 & 15.86 & 15.10 & 0.580 & -25.18 + 2453430.8201 & keck / hires & 0.30 & 0.02 & -15.84 & 7.19 & 15.47 & 14.70 & 0.830 & -25.45 + 2453716.7708 & mag / mike & 0.21 & 3.31 & -8.20 & 3.72 & 18.80 & 18.13 & 0.040 & 9.98 + 2453717.7708 & mag / mike & 0.23 & 2.58 & -10.71 & 4.86 & 18.27 & 17.59 & 0.060 & 9.49 + 2453718.7708 & mag / mike & 0.25 & 1.93 & -12.69 & 5.76 & 18.14 & 17.45 & 0.070 & 9.00 + 2453719.7708 & mag / mike & 0.27 & 1.35 & -14.25 & 6.47 & 17.60 & 16.90 & 0.120 & 8.51 + 2453724.7264 & het / hrs & 0.38 & -0.63 & -18.26 & 8.30 & 15.38 & 14.61 & 0.900 & 6.35 + 2453725.7340 & het / hrs & 0.40 & -0.87 & -18.59 & 8.44 & 15.31 & 14.52 & 0.960 & 5.83 + 2453726.7222 & het / hrs & 0.42 & -1.05 & -18.79 & 8.53 & 15.28 & 14.51 & 0.990 & 5.35 + 2453727.7181 & het / hrs & 0.44 & -1.19 & -18.89 & 8.58 & 15.27 & 14.50 & 1.000 & 4.84 + 2453728.7090 & het / hrs & 0.46 & -1.28 & -18.90 & 8.58 & 15.22 & 14.45 & 1.050 & 4.35 + 2453773.7556 & het / hrs & 0.39 & -0.67 & -18.49 & 8.40 & 15.49 & 14.73 & 0.820 & -17.81 + 2453775.7458 & het / hrs & 0.43 & -1.03 & -18.86 & 8.57 & 15.34 & 14.57 & 0.940 & -18.59 + 2453776.5862 & het / hrs & 0.45 & -1.12 & -18.91 & 8.59 & 15.28 & 14.50 & 0.990 & -18.59 + 2453778.7409 & het / hrs & 0.50 & -1.21 & -18.72 & 8.51 & 15.23 & 14.46 & 1.040 & -19.74 + 2453779.5889 & het / hrs & -0.49 & -1.19 & -18.55 & 8.43 & 15.25 & 14.48 & 1.020 & -19.74 + 2453780.7305 & het / hrs & -0.47 & -1.12 & -18.22 & 8.28 & 15.26 & 14.49 & 1.010 & -20.45 + @xmath11x = -4 to -2 & 2002 dec 10 & -4.02 & het / hrs + ( ingress ) & 2002 dec 13 & -3.65 & het / hrs + & 2003 feb 02 & -2.91 & het / hrs + & 2004 jan 04 & -2.71 & mcd / ce + & 2003 mar 23 & -2.59 & het / hrs + & 2004 jan 05 & -2.48 & mcd / ce + @xmath11x = 2 to 1 & 2001 dec 20 & 1.95 & vlt / uves + ( egress ) & 2005 dec 14 & 1.93 & magellan / mike + & 2005 dec 15 & 1.35 & magellan / mike + & 2005 feb 27 & 1.10 & keck / hires + & 2003 dec 18 & 1.09 & keck / hires + @xmath11x = -1 to -2 & 2006 feb 09 & -1.03 & het / hrs + ( egress ) & 2005 dec 22 & -1.05 & het / hrs + & 2006 feb 10 & -1.12 & het / hrs + & 2005 dec 23 & -1.19 & het / hrs + & 2006 feb 12 & -1.21 & het / hrs + & 2005 dec 24 & -1.28 & het / hrs + & 2003 mar 06 & -1.39 & het / hrs +

the harmonic lanczos bidiagonalization method can be used to compute the smallest singular triplets of a large matrix . we prove that for good enough projection subspaces harmonic ritz values converge if the columns of are strongly linearly independent . on the other hand , harmonic ritz values may miss some desired singular values when the columns of almost linearly dependent . furthermore , harmonic ritz vectors may converge irregularly and even may fail to converge . based on the refined projection principle for large matrix eigenproblems due to the first author , we propose a refined harmonic lanczos bidiagonalization method that takes the rayleigh quotients of the harmonic ritz vectors as approximate singular values and extracts the best approximate singular vectors , called the refined harmonic ritz approximations , from the given subspaces in the sense of residual minimizations . the refined approximations are shown to converge to the desired singular vectors once the subspaces are sufficiently good and the rayleigh quotients converge . an implicitly restarted refined harmonic lanczos bidiagonalization algorithm ( irrhlb ) is developed . we study how to select the best possible shifts , and suggest refined harmonic shifts that are theoretically better than the harmonic shifts used within the implicitly restarted lanczos bidiagonalization algorithm ( irhlb ) . we propose a novel procedure that can numerically compute the refined harmonic shifts efficiently and accurately . numerical experiments are reported that compare irrhlb with five other algorithms based on the lanczos bidiagonalization process . it appears that irrhlb is at least competitive with them and can be considerably more efficient when computing the smallest singular triplets . singular values , singular vectors , svd , lanczos bidiagonalization , refined projection , harmonic , refined harmonic , implicit restart , harmonic shifts , refined harmonic shifts . 65f15 , 15a18

0906.2242

we assume that a large sparse matrix @xmath1 has full column rank and let @xmath2 be its singular value decomposition ( svd ) @xcite , where @xmath3 and @xmath4 are @xmath5 and @xmath6 orthogonal matrices , @xmath7 and @xmath8 is diagonal . @xmath9 , are called the singular values of @xmath0 , @xmath10 s and @xmath11 s are the associated left and right singular vectors , respectively , and @xmath12 s are called singular triplets . in this paper , slightly different from the convention , the singular values are labeled as @xmath13 . we are concerned with the following problem . * problem 1*. _ compute numerically the @xmath14 smallest singular triplets @xmath12 of @xmath0 , @xmath15 , where @xmath16 . _ there are many applications of problem 1 , including determination of numerical rank and of spectral condition number , least squares problems , total least squares problems , regression analysis , image and signal processing , pattern recognition and information retrieval , to name a few . consider the @xmath17 augmented matrix @xmath18 then , the eigenvalues of @xmath19 are just @xmath20 and @xmath21 zeros , the associated eigenvectors of @xmath22 and @xmath23 are @xmath24 and @xmath25 , respectively , and the eigenvectors associated with zero eigenvalues have the form @xmath26 , where @xmath27 s are orthogonal to all @xmath28 . therefore , we obtain the following formulation of problem 1 . * problem 2*. _ compute numerically the @xmath14 smallest positive eigenvalues and the associated eigenvectors of @xmath29 . _ for the @xmath14 smallest eigenpairs of @xmath19 , problem 2 is a symmetric interior eigenvalue problem . since @xmath30 and @xmath31 are assumed to be large , we can only resort to projection methods . a typical method is the symmetric lanczos method @xcite . it and other standard projection methods usually favor the extreme eigenvalues and the associated eigenvectors but are generally very inefficient for computing interior eigenpairs @xcite . another drawback is that in finite precision the computed eigenvalues do not come in plus - and - minus pairs and the computed eigenvectors do not respect the special structures that the true eigenvectors have . because of the mentioned drawbacks , we should not work on @xmath19 explicitly for computing the smallest singular triplets of @xmath0 . instead we attempt to solve problem 1 directly by working on @xmath29 implicitly . it appears that lanczos bidiagonalization type methods @xcite and jacobi - davidson svd type methods @xcite can solve the mentioned problems elegantly . the lanczos bidiagonalization type methods available have in common that they are all based on the lanczos bidiagonalization process to build up orthonormal bases of certain krylov subspaces . however , their mathematical backgrounds can be fundamentally different . basically , there are three kinds of projection principles that extract different approximate singular triplets with respect to the subspaces . some methods use the standard projection principle @xcite to extract ritz approximations @xcite , some methods use the harmonic projection principle @xcite to extract harmonic ritz approximations @xcite and some methods use the refined projection principle @xcite to extract refined singular vector approximations @xcite . jacobi - davidson type svd methods for problem 1 have several versions that are based on the three projection principles as well as their generalizations , respectively . as observed and claimed in @xcite , the refined extraction version appears to give the best accuracy in general . for problem 1 , due to the storage requirement and computational cost , all the lanczos bidiagonalization type methods as well as jacobi - davidson type methods have to be restarted generally in order to make them converge . that is , for given projection subspaces , if the methods do not converge , then one repeatedly chooses new better starting vectors , constructs better subspaces and computes new approximate singular triplets until they converge . the implicit restarting technique due to sorensen @xcite is a powerful tool for restarting krylov subspace algorithms in various contexts including large svd problems @xcite . the success of an implicitly restarted algorithm heavily depends on both the underlying method itself and a proper selection of the shifts involved ; see , e.g. , @xcite . based on the lanczos bidiagonalization method and one of its harmonic versions , jia and niu @xcite and larsen @xcite have developed an implicitly restarted lanczos bidiagonalization algorithm ( irlb ) , and kokiopoulou _ et al . _ @xcite have proposed an implicitly restarted harmonic lanczos bidiagonalization algorithm ( irlanb ) for computing the smallest singular triplets . irlb uses the unwanted ritz values and irlanb uses the unwanted ritz or harmonic ritz values as shifts , respectively . these shifts are called exact shifts and harmonic shifts . baglama and reichel @xcite propose a thick restarting technique that explicitly augments small subspaces with certain ritz or harmonic ritz vectors , leading to augmented restarted lanczos bidiagonalization algorithms ( irlba ) . @xcite analyze a parallel implementation of this algorithm . based on stewart s work for large eigenproblems @xcite , stoll @xcite presents a krylov - schur type algorithm that is restarted explicitly and is easily implemented . it is shown in @xcite that the lanczos bidiagonalization method may fail to compute singular vectors though it converges for computing singular values for sufficiently good subspaces . to correct this deficiency , applying the refined projection principle proposed by the first author @xcite ( see also @xcite ) , we have proposed a refined lanczos bidiagonalization method , analyzed its convergence and developed an implicitly restarted refined lanczos bidiagonalization algorithm ( irrlb ) @xcite . based on the refined approximations to singular vectors , we have proposed refined shifts that are theoretically better than the exact shifts used within irlb . numerical experiments have demonstrated that irrlb often outperforms irlb @xcite considerably and is more efficient than several other available schemes : propack @xcite , lanso @xcite , the matlab internal function svds and some others when computing the largest and smallest singular triplets . hochstenbach @xcite shows that for nested subspaces ritz values approach the largest singular values monotonically but approach the smallest ones irregularly . so the lanczos bidiagonalization method is more suitable for computing the largest singular triplets and may exhibit irregular convergence behavior when computing the smallest singular triplets . in contrast , the smallest harmonic ritz values converge to the smallest singular values monotonically from above and may be better approximations . we continue to study how to compute the smallest singular triplets more efficiently in this paper . based on the lanczos bidiagonalization process , we propose a harmonic lanczos bidiagonalization method by combining it with the harmonic projection principle . our derivation is different from that in @xcite . the method is the same as that in @xcite but different from the one in @xcite . we prove that for good enough projection subspaces harmonic ritz values converge if the columns of @xmath0 are strongly linearly independent . on the other hand , harmonic ritz values may miss some desired singular values when the columns of @xmath0 are almost linearly dependent . so harmonic ritz values may not be reliable . furthermore , harmonic ritz vectors may converge irregularly and even may fail to converge . these results imply that either implicitly or explicitly restarted algorithms may converge very slowly , converge irregularly or fail to converge . to circumvent these drawbacks , combining the harmonic projection principle with the harmonic projection principle , we propose a refined harmonic lanczos bidiagonalization method that takes the rayleigh quotients of harmonic ritz vectors as more accurate and reliable approximate singular values and extracts the best approximations to the desired singular vectors from the given subspaces that minimize the residuals formed with the rayleigh quotients . we prove that refined harmonic ritz approximations converge once the krylov subspaces are good enough and the rayleigh quotients converge . we then develop an implicitly restarted refined harmonic lanczos bidiagonalization algorithm ( irrhlb ) . based on the refined harmonic ritz approximations to the desired singular vectors , in the spirit of jia s work @xcite , we propose a new shifts scheme , called the refined harmonic shifts , that we show to be theoretically better than the harmonic shifts used within the implicitly restarted harmonic lanczos bidiagonalization algorithm ( irhlb ) and irlanb . motivated by @xcite , we propose an efficient procedure to compute the refined harmonic shifts accurately . it is worth noting that kokiopoulou _ _ @xcite also use the refined projection principle to compute the refined harmonic ritz approximations . they exploit the lower lanczos bidiagonalization process , use the ritz or harmonic ritz values as shifts in the algorithm and compute the smallest singular triplets one by one by exploiting deflation . they only use the refined projection principle as refinement postprocessing at the end of each restart . the authors demonstrate that computing refined ( harmonic ) ritz vectors and thus refined ritz values benefits the overall convergence process . in particular , they show that while convergence is not apparent in terms of harmonic residual norms , monitoring refined residuals predicts convergence more accurately and safely . in contrast , based on the upper lanczos bidiagonalization process and the refined projection principle , we propose a truly new method the refined harmonic lanczos bidiagonalization method that computes refined harmonic ritz vectors as new approximations . we then develop irrhlb with use of the new better shifts , called refined harmonic shifts , based on refined harmonic ritz approximations . irrhlb computes all the desired smallest singular triplets simultaneously . the paper is organized as follows . in 2 , based on the lanczos bidiagonalization process , we derive the harmonic lanczos bidiagonalization method and then present some basic and important properties of approximate singular vectors to be used later . exploiting jia s results in @xcite , we then make a convergence analysis . in 3 we propose the refined harmonic lanczos bidiagonalization method . we prove that the refined harmonic ritz approximations converge for good enough subspaces once the rayleigh quotients converge . in 4 , we consider selection of the shifts involved . for irhlb , similar to what is done in @xcite , we use the harmonic ritz values . for irrhlb , by exploiting the available refined harmonic ritz approximations , we propose the refined harmonic shifts that are proved to be theoretically better than the harmonic shifts . we then present an efficient procedure to compute them . we show that in finite precision the refined harmonic shifts can be computed accurately . meanwhile , we extend the adaptive shifting strategy proposed by larsen @xcite and modified by jia and niu @xcite to irhlb and irrhlb . in 5 , we report numerical results and compare irrhlb with the five other state of art algorithms : irhlb , irrlb , irlb , irlanb and irlba , indicating that irrhlb is at least competitive with the five other algorithms and can be considerably more efficient when computing the smallest singular triplets . finally , we conclude the paper with some remarks in 6 . we introduce some notations to be used . denote by @xmath32 the spectral norm of a matrix and the vector 2-norm , by @xmath33 , by @xmath34 the @xmath35-dimensional krylov subspace generated by the matrix @xmath36 and the starting @xmath37 , by the superscript ` t ' the transpose of a matrix or vector , by @xmath38 the identity matrix with the order clear from the context and by @xmath39 the @xmath35-th coordinate vector of dimension @xmath35 . golub _ et al . _ @xcite propose a lanczos bidiagonalization method that can compute either the largest or the smallest singular triplets of @xmath0 . the method is equivalent to the symmetric lanczos method for the eigenproblem of @xmath19 starting with a special vector @xcite and is based on the lanczos bidiagonalization process @xcite , which satisfies the following relations if it does not break down before step @xmath35 : @xmath40 where the @xmath41 matrix @xmath42 and the columns of @xmath43 and @xmath44 form orthonormal bases of the krylov subspaces @xmath45 and @xmath46 , respectively . so we have @xmath47 the lanczos bidiagonalization method computes the singular triplets @xmath48 @xmath49 of @xmath50 and then uses some of @xmath51 , called the ritz approximations , to approximate the largest and/or smallest singular triplets of @xmath0 . in finite precision , the columns of @xmath52 and of @xmath53 may rapidly lose orthogonality . a partial reorthogonalization strategy @xcite is an effective technique for maintaining numerical orthogonality . however , simon and zha @xcite show that it generally suffices to partially reorthogonalize only @xmath52 or @xmath53 rather than reorthogonalizing them simultaneously . this may reduce the computational cost considerably when only reorthogonalizing the columns of @xmath53 for @xmath54 . in our codes , we adopt the strategy from @xcite which is based on @xcite . given the subspace @xmath55 the harmonic projection method of @xmath56 computes @xmath57 satisfying the requirements @xmath58 and uses them as approximations to some eigenpairs of @xmath19 @xcite . making use of ( [ projmatrix ] ) , we see that ( [ proj ] ) is equivalent to the generalized eigenproblem @xmath59 @xmath50 is nonsingular as @xmath0 has full column rank and its singular values interlace those of @xmath0 @xcite . this is a symmetric positive definite generalized eigenproblem , so its eigenvalues are all real and nonzero @xcite . furthermore , we present the following result . [ th1 ] if @xmath60 satisfies ( [ harmbl ] ) , then @xmath61 or equivalently @xmath62 satisfies ( [ harmbl ] ) too , that is , the eigenvalues of ( [ harmbl ] ) come in plus - and - minus pairs and the eigenvectors have a special structure . _ equation ( [ harmbl ] ) gives rise to @xmath63 so we can readily see that the assertion holds . assume that the nonnegative eigenvalues of ( [ harmbl ] ) are ordered as @xmath64 where @xmath65 . then we use @xmath66 as approximations to the @xmath14 smallest singular triplets @xmath12 . this method is called the harmonic lanczos bidiagonalization method . @xmath67 s are called the harmonic ritz values , @xmath68 s and @xmath69 s the ( left and right ) harmonic ritz vectors , and @xmath70 s the harmonic ritz approximations . it is proved in @xcite that @xmath71 we have the following basic and important properties , which will play a key role in 4.2 . [ th2 ] for @xmath72 it holds that @xmath73 and @xmath74 _ proof . _ since ( [ harmbl ] ) is a symmetric positive definite generalized eigenproblem , for @xmath75 we have @xmath76 from which it follows that ( [ borth ] ) holds . ( [ borth2 ] ) follows from @xmath77 and ( [ projmatrix ] ) . we now discuss efficient computation of harmonic ritz approximations . it is disappointing that ( [ harmbl ] ) is a @xmath78 generalized eigenproblem . fortunately , we can reduce ( [ harmbl ] ) to a half sized svd problem that can be solved more cheaply and accurately , as shown below . we get from ( [ harmbl ] ) @xmath79 from which it follows that @xmath80 so , the @xmath81 s are the singular values and right singular vectors of the @xmath82 matrix @xmath83 and the left singular vectors @xmath84 therefore , we can get @xmath81 more accurately and efficiently by computing the svd of the half sized ( [ svd1 ] ) and obtain all the @xmath85 s by solving the bidiagonal linear systems @xmath86 at a total cost of @xmath87 flops . we comment that , based on the harmonic projection of @xmath88 onto @xmath45 , baglama and reichel @xcite also derive ( [ reduction])([bidiag ] ) . our method is the same as that in @xcite but is different from the one in @xcite , which is based on the lower lanczos bidiagonalization process . from ( [ bid11 ] ) and ( [ bid21 ] ) , we have @xmath89 and @xmath90 therefore , if @xmath91 where @xmath92 is a user prescribed accuracy , then the method is accepted as converged for @xmath92 . we now analyze the convergence . jia @xcite establishes a general convergence theory of harmonic projection methods for large eigenproblems . the theory can be adapted here . from ( [ harmbl ] ) , set the matrices @xmath93 and @xmath94 recall from theorem [ th1 ] that @xmath95 s are the eigenvalues of @xmath96 . the following result is direct from theorem 2.1 and corollary 2.2 of @xcite . [ theo1 ] assume that @xmath97 is a singular triplet of @xmath0 and define @xmath98 to be the distance between the vector @xmath99 and the subspace @xmath100 . then there exists a perturbation matrix @xmath101 satisfying @xmath102 such that the exact singular value @xmath103 of @xmath0 is an eigenvalue of @xmath104 . furthermore , there exists an eigenvalue @xmath105 of @xmath106 satisfying @xmath107 this theorem shows that if @xmath108 tends to zero and @xmath109 is uniformly bounded then there always exists one harmonic ritz value @xmath105 that converges to the desired singular value @xmath103 . the interlacing theorem of singular values @xcite tells us that @xmath110 is uniformly bounded . as a result , if @xmath111 , we should have @xmath112 . however , the situation is by no means so simple , and it is instructive to see what will happen when only speaking of @xmath108 small . note that @xmath113 so if @xmath114 , that is , the columns of @xmath0 are almost linearly dependent , then @xmath115 may not be near zero , so that @xmath116 may not be small . actually , @xmath117 once the smallest singular value ( ritz value ) of @xmath50 converges to @xmath118 . in this case , the harmonic lanczos bidiagonalization method may miss @xmath103 . so the method may not be reliable , and @xmath105 is only guaranteed to be a good approximation to @xmath103 only if @xmath108 is very small and the columns of @xmath0 are strongly linearly independent . to improve convergence and reliability of the method , we recommend the rayleigh quotient @xmath119 as a new approximation to @xmath22 , as was also done in @xcite . @xmath120 is more accurate and reliable than @xmath67 . correspondingly , @xmath67 in ( [ harmstop ] ) is replaced by @xmath120 . we refer to @xcite for more theoretical results and arguments on such a replacement . the following result is a direct application of theorem 3.2 of @xcite . [ theo2 ] let @xmath121 be an eigenpair of @xmath96 , and assume that @xmath122 is such that the square matrix @xmath123 is orthogonal and transforms @xmath96 into @xmath124 then if @xmath125 it holds that @xmath126 theorems [ theo2 ] states that if the separation @xmath127 of @xmath105 and the other harmonic ritz values is uniformly bounded below by a positive constant then the harmonic ritz approximations @xmath128 converge . unfortunately , however , for a general @xmath0 , @xmath129 can be arbitrarily near zero , i.e. , @xmath105 can be arbitrarily close to the other harmonic ritz values . as a result , the upper bounds ( [ vector2 ] ) and ( [ sep ] ) can converge to zero very slowly and irregularly and even fail to do so as @xmath130 . this means that the approximate singular vectors may converge very slowly and irregularly and even may fail to converge . the previous analysis shows that the harmonic ritz approximations may converge slowly and irregularly and even may fail to converge . to overcome this intrinsic drawback , we now combine the harmonic lanczos bidiagonalization method with the refined projection principle and derive a refined harmonic lanczos bidiagonalization method . recall that @xmath131 . we use @xmath132 satisfying @xmath133 to replace @xmath134 as a new approximation to an eigenpair of @xmath19 . the method first uses the harmonic lanczos bidiagonalization method to compute @xmath57 and then forms the rayleigh quotient @xmath135 . with each @xmath136 , it computes @xmath137 by solving the minimization problem . the following results adapted from @xcite can be used to compute @xmath138 efficiently and accurately . [ theo4 ] let @xmath139 be the right singular vector of the matrix @xmath140 associated with its smallest singular value @xmath141 . then @xmath142 @xmath143 with @xmath138 at hand , we define the new left and right approximate singular vectors as @xmath144 and use @xmath145 s to approximate the @xmath14 smallest singular triplets of @xmath0 . we call @xmath146 a refined harmonic ritz triplet and @xmath147 a refined harmonic ritz approximation . similar to ( [ harmstop ] ) , @xmath145 is accepted as converged if @xmath148 the following result is taken directly from theorem 4.1 of @xcite . [ theo5 ] let @xmath97 be a singular triplet of @xmath0 . then there exist @xmath149 and @xmath150 such that @xmath151 and @xmath152 are orthogonal and @xmath153 where @xmath154 . set @xmath155 and assume that @xmath156 is approximating @xmath97 . then if @xmath157 we have @xmath158 note that @xmath159 is the gap of @xmath103 and the other singular values of @xmath0 and is a fixed constant . theorems [ theo5 ] shows that the refined harmonic ritz approximations converge once @xmath160 and @xmath161 . therefore , the refined harmonic lanczos bidiagonalization method overcomes , to great extent , the possible non - convergence of the harmonic ritz approximations . due to the storage requirements and computational cost , in practice , the number of steps can not be large and must be limited . for a relatively small @xmath35 , however , the @xmath35-dimensional subspaces @xmath45 and @xmath162 , in general , do not contain enough information on the desired right and left singular vectors , so that both the harmonic and refined harmonic lanczos bidiagonalization methods do not converge . therefore , it is necessary to restart the methods . the idea is to repeatedly update new starting vectors based on the information available and construct increasingly better krylov subspaces until the methods converge . implicit restarting is usually preferable not only because of efficiency of the restart procedure , but also because the implicit procedure is more effective at locking in desired directions and purging unwanted ones . we briefly review the implicit restarting technique for the lanczos bidiagonalization process @xcite . note that @xmath163 . after running @xmath164 implicit qr iteration steps on @xmath50 using the shifts @xmath165 , we get @xmath166 where @xmath167 and @xmath168 are the accumulations of givens rotations applied to @xmath50 from the left and right , respectively . define @xmath169 and @xmath170 . this process is achieved implicitly from @xmath171 to @xmath172 by working on @xmath50 directly . performing the above @xmath164 implicit qr iteration steps gives the following relations @xcite : @xmath173 where @xmath174 is the entry of @xmath167 in position @xmath175 and the updated starting vector has the form @xmath176 with @xmath177 a factor making @xmath178 . since @xmath179 is orthogonal to @xmath180 , we have obtained a @xmath14-step lanczos bidiagonalization process starting with @xmath181 . it is then extended to a @xmath35-step lanczos bidiagonalization process in a standard way . so we avoid restarting the process from scratch and do it from step @xmath182 upwards . this saves the computational cost of the first @xmath14 steps of the process . applying the implicit restarting technique to the harmonic lanczos bidiagonalization method and its refined version in such a way , we have formally sketched an implicitly restarted harmonic lanczos bidiagonalization algorithm ( irhlb ) and an implicitly restarted refined harmonic lanczos bidiagonalization algorithm ( irrhlb ) . we can run irhlb and irrhlb once the shifts @xmath183 are given . however , in order to make them work as efficiently as possible , we should select the best possible shifts in some sense for each algorithm . in the same spirit of @xcite , it has been shown in @xcite that if the shifts are more accurate approximations to some of the unwanted singular values of @xmath0 then the resulting subspaces contain more information on the desired singular vectors . the better the subspaces are , the faster irhlb and irrhlb may converge . for eigenproblems and svd problems , morgan @xcite and kokiopoulou _ et al_. @xcite suggest using unwanted harmonic ritz values as shifts , called the harmonic shifts here . these shifts are natural choices as they are the best approximations available to some of the unwanted eigenvalues and the unwanted singular values , respectively . so , for our irhlb we also use the @xmath164 unwanted harmonic ritz values @xmath184 as shifts . since the refined harmonic approximations @xmath185 are optimal in the sense of residual minimizations , they are generally more accurate than the harmonic ritz approximations @xmath186 . therefore , based on @xmath187 , it should be possible to find better possible shifts than the harmonic shifts . the following important result on the harmonic shifts is crucial for us to introduce and understand new better shifts for use within irrhlb . [ theo6 ] define @xmath188 then the harmonic shifts @xmath184 are the absolute values of the @xmath164 harmonic ritz values of @xmath19 with respect to the subspace @xmath189 . _ from definition ( [ proj ] ) of the harmonic projection as well as the relationship between ( [ proj ] ) and ( [ harmbl ] ) , it is easily verified that for @xmath190 , if the @xmath191-th column of @xmath192 and that of @xmath193 have the same or opposite @xmath194 sign , then @xmath19 has @xmath67 or @xmath195 as one harmonic ritz value with respect to the subspace @xmath196 . we see from ( [ borth2 ] ) that @xmath197 and @xmath198 define @xmath199 and @xmath200 , and let @xmath201 be matrices with @xmath202 columns satisfying @xmath203 and @xmath204 where @xmath205 denotes the direct sum . jia @xcite derives a number of theoretical results that compare refined ritz vectors and ritz vectors . at this moment , we temporarily regard @xmath19 as a general matrix , and @xmath206 are a ritz and the corresponding refined ritz vector of @xmath19 with respect to a general subspace @xmath100 , respectively . one of jia s results says that we always have @xmath207 if the left - hand side is not zero and @xmath208 may occur if @xmath120 is close to some @xmath209 for @xmath210 . by standard perturbation theory in terms of residual norms , these two results demonstrate that @xmath211 is more accurate and can be much more accurate than @xmath212 . here we should point out that these claims hold without requiring that @xmath100 is sufficiently good . jia @xcite constructs a number of symmetric matrices having well separated simple eigenvalues and accurate subspaces to illustrate this . more precisely , assuming that @xmath120 and @xmath213 are used to approximate the eigenvalue @xmath214 and the eigenvector @xmath215 of @xmath19 and @xmath108 is the distance between @xmath215 and the subspace @xmath100 , jia s examples show that we can indeed have @xmath216 the above results are easily adapted to the harmonic and refined harmonic ritz vectors . some similar symmetric matrices are constructed by jia in @xcite for which the harmonic ritz vectors have no accuracy at all but refined harmonic ones have accuracy @xmath217 . coming back to our svd context , the above results and analysis indicate that @xmath218 and @xmath219 are more accurate and can be much more accurate than @xmath68 and @xmath69 without the assumption that projection subspaces are sufficiently good . based on the above , it is evident that the subspaces @xmath220 and @xmath221 contain ( possibly much ) more accurate approximations to @xmath222 and @xmath223 , @xmath224 than the subspaces @xmath225 and @xmath226 do . this , in turn , means that the subspace @xmath227 contains ( possibly much ) more accurate approximations to the eigenvectors @xmath228 associated with the eigenvalues @xmath229 , @xmath224 , of @xmath19 than the subspace @xmath189 does . recall from theorem [ theo1 ] that a better subspace should generally produce more accurate harmonic ritz values . hence , combining with theorem [ theo6 ] , we have come to the following key result . [ theo9 ] as approximations to some of @xmath230 , the absolute values of the harmonic ritz values @xmath231 of @xmath19 with respect to the subspace @xmath232 are more accurate and can be much more accurate than the harmonic shifts @xmath184 . this theorem holds without assuming that @xmath233 and @xmath234 are sufficiently good . it suggests that we use better @xmath235 as shifts for use within irrhlb . we call them the refined harmonic shifts . computationally , at first glance , it seems quite complicated and expensive to get the refined harmonic shifts as it involves constructing @xmath201 that are related with the large @xmath19 . inspired by the tricks in @xcite , however , we can exploit ( [ projmatrix ] ) to propose an efficient procedure for computing them accurately , as shown below . recall ( [ refinevector ] ) and define @xmath236 and @xmath237 . we use householder transformations to compute the full qr decompositions @xmath238 which costs @xmath239 flops . partition @xmath240 where @xmath241 and @xmath242 are the first @xmath14 columns of @xmath243 and @xmath244 , respectively , and let @xmath245 then it can be readily verified that @xmath246 so @xmath247 and @xmath248 defined in this way meet conditions ( [ aorth ] ) and ( [ sumdecomp ] ) and are just what we need . by ( [ sumdecomp ] ) , we have @xmath249 the harmonic ritz values @xmath231 of @xmath19 with respect to @xmath250 satisfy @xmath251 exploiting ( [ projmatrix ] ) , we get a @xmath252 symmetric positive definite generalized eigenvalue problem @xmath253 the @xmath254 s are computed by the qz algorithm @xcite using @xmath255 flops . so the total cost of computing the refined harmonic shifts is @xmath239 flops , negligible compared with the harmonic lanczos bidiagonalization method . we give more details on computation of the refined harmonic shifts . denote by @xmath256 and @xmath257 the matrices of the left and right - hand sides in ( [ shift ] ) , respectively , and observe that @xmath258 noting that the two matrices in @xmath256 are transposes each other , we only need to form @xmath259 by computing @xmath260 or @xmath261 , and @xmath262 or @xmath263 is then used to form @xmath257 . since @xmath257 is symmetric , we only need to compute its upper triangular part . the total cost of forming @xmath256 and @xmath257 is @xmath239 flops . we then compute the eigenvalues @xmath264 s of the symmetric positive definite matrix pencil @xmath265 . now we show that in finite precision the above procedure is numerically stable and can compute the refined harmonic shifts accurately . there are three major steps in the procedure : the qr decompositions in ( [ qr ] ) , computation of @xmath256 and @xmath257 and the solution of the eigenvalue problem of @xmath266 by the qz algorithm . note that the qr decompositions can be computed using householder transformations in a numerically stable way ( we use the matlab built - in code qr in our implementation ) and the qz algorithm are numerically stable . therefore , omitting details on roundoff errors , we finally compute the eigenvalues @xmath267 s of a perturbed matrix pencil @xmath268 , where @xmath269 and @xmath270 are the matrices of roundoff error accumulations and satisfy @xmath271 with @xmath272 being the machine precision and @xmath273 the frobenius norm . for an eigenpair @xmath274 of the pencil @xmath265 with @xmath275 , let @xmath276 and @xmath277 , so that @xmath278 is a projective representation of the eigenvalue @xmath279 @xcite . then it is known @xcite that there is an eigenvalue @xmath280 of the matrix pencil @xmath281 such that the chordal distance @xmath282 it is important to point out that for not too small @xmath283 the choral distance behaves like the ordinary distance @xmath284 ; see a remark in @xcite . so , how accurate @xmath285 is depends on the @xmath279 s condition number @xmath286 if one of @xmath287 and @xmath288 is not small , @xmath289 is not large and thus by ( [ backerror ] ) the relative error of @xmath285 is @xmath290 if @xmath291 is not very small . we look at the smallest @xmath291 . by theorem [ theo9 ] , it is known that the absolute values @xmath291 s better approximate some of @xmath230 than @xmath292 . furthermore , recall from ( [ mono ] ) that @xmath293 . so the smallest @xmath291 is approximately bounded below by @xmath294 . in the following , we establish lower bounds for @xmath295 and @xmath296 and an upper bound for @xmath289 rigorously and prove when our proposed procedure can numerically compute the refined harmonic ritz shifts accurately . [ theo10 ] for any refined harmonic ritz shift @xmath283 , we have @xmath297 and the @xmath279 s condition number is bounded from above : @xmath298 if @xmath294 is not very small , then numerically the procedure described can compute the refined harmonic shifts @xmath299 s with relative accuracy @xmath290 . _ proof_. we prove ( [ nubound ] ) and ( [ xibound ] ) in turn . to prove ( [ nubound ] ) , we first estimate @xmath295 and then @xmath296 . from the definition of @xmath256 and @xmath257 , we have @xmath300 where @xmath301 denotes the smallest singular value of a matrix @xmath36 . note that @xmath302 and @xmath303 form orthogonal bases of the left subspace @xmath304 and the right subspace @xmath305 . exploiting ( [ projmatrix ] ) , ( [ partition ] ) , ( [ perp ] ) and ( [ newbasis ] ) and keeping in mind that @xmath306 and @xmath307 , we obtain from ( [ aorth ] ) that the projection matrix of @xmath0 with respect to @xmath302 and @xmath303 is @xmath308 whose singular values @xmath309 , labeled in increasing order , are the union of the singular values of @xmath310 and @xmath259 and are just the ritz values of @xmath0 with respect to the left and right subspaces @xmath234 and @xmath233 . by the singular value interlacing property , we have @xmath311 . furthermore , note that @xmath310 and @xmath312 are the projection matrices of @xmath0 with respect to the left subspaces @xmath313 and @xmath314 and the right subspaces @xmath315 and @xmath316 , respectively . therefore , the singular values of @xmath317 are @xmath318 and approximate the @xmath14 desired smallest singular values @xmath22 s from above , while the singular values of @xmath319 are @xmath320 and approximate @xmath321 from above too . in particular , we have @xmath322 so it holds that @xmath323 next we estimate @xmath288 . we obtain from ( [ gm ] ) , ( [ perp ] ) and ( [ projmatrix ] ) @xmath324 observe that @xmath325 and @xmath326 are the projection matrices of @xmath327 and @xmath88 with respect to @xmath328 and @xmath329 , respectively . so their eigenvalues approximate some of the eigenvalues @xmath330 of @xmath327 and @xmath88 . furthermore , since @xmath331 is symmetric nonnegative definite , the smallest eigenvalue of @xmath332 is no less than the smallest eigenvalue @xmath333 of @xmath334 . as a consequence , it follows from @xmath335 that the smallest eigenvalue of @xmath336 is bounded below by @xmath337 . similarly , we have @xmath338 which is symmetric nonnegative definite . as a result , the smallest eigenvalue of @xmath325 is no less than the smallest eigenvalue @xmath333 of @xmath339 and is bounded below by @xmath337 too . therefore , we get @xmath340 so it follows from ( [ alphabound ] ) that the @xmath279 s condition number @xmath289 in ( [ condnum ] ) satisfies @xmath341 which is ( [ nubound ] ) . if @xmath294 is not very small , then @xmath289 is not large and all @xmath283 s are not too small . keep in mind the comments on ( [ backerror ] ) and ( [ chi ] ) . it is then clear that numerically the proposed procedure can compute the refined harmonic shifts @xmath299 s with relative accuracy @xmath290 . we should point out that theorem [ theo10 ] holds without any assumption on @xmath233 and @xmath234 , as is clearly seen from the proof . it has been observed @xcite that if the @xmath14-th desired @xmath342 is very near a shift then irlb with the exact shifts ( the unwanted ritz values ) converges very slowly and even stagnates . this is also the case for irrhlb with the refined harmonic shifts and irhlb with the harmonic shifts . the reason is that if some shift @xmath343 is very near @xmath342 then the new starting vector @xmath181 will nearly annihilate the component of the desired @xmath344 , so that the new subspace @xmath345 contains very little information on @xmath344 and @xmath346 converges to @xmath342 very slowly or not at all . in order to overcome this problem , for irlb with the exact shifts , larsen @xcite proposes an adaptive shifting strategy for computing the largest singular triplets . he simply replaces a bad shift to be defined below by a zero shift . jia and niu @xcite adapt it to irrlb for computing the largest singular triplets but modify it for computing the smallest singular triplets . their strategy works for irhlb and irrhlb : define the relative gaps of @xmath346 and all the shifts @xmath347 by @xmath348 where @xmath349 is the residual norm ( [ harmstop ] ) or ( [ refinedstop ] ) . we should note that @xmath350 is an approximation to @xmath342 . if @xmath351 , @xmath343 is a bad shift and should be replaced by a suitable quantity . expand @xmath352 as a linear combination of the right singular vectors @xmath353 : @xmath354 then for the harmonic shifts @xmath355 we have from ( [ update ] ) @xmath356 so if @xmath357 is very near @xmath342 , which is the case that @xmath294 is very near @xmath342 , then @xmath181 has a very small component in the direction of @xmath344 . a good strategy is to replace @xmath357 by the largest one among all the shifts , as this strategy amplifies the components of @xmath181 in @xmath358 and meanwhile dampens those in @xmath359 . the above strategy applies to the refined harmonic shifts as well . we now present irhlb with the harmonic shifts and irrhlb with the refined harmonic shifts , respectively . * algorithm 1 . irhlb with the harmonic shifts * 1 . given a unit length starting vector @xmath352 of dimension @xmath31 , the steps @xmath35 , the number @xmath14 of the desired singular triplets and the convergence tolerance @xmath92 . 2 . run the @xmath35-step lanczos bidiagonalization process and construct @xmath360 and @xmath361 . 3 . calculate the triplets @xmath362 , by computing the singular values and right singular vectors of ( [ svd1 ] ) and by solving ( [ bidiag ] ) and normalizing the solutions , and use the rayleigh quotients @xmath363 as approximations to @xmath364 . 4 . replace @xmath67 by @xmath120 in ( [ harmstop ] ) . for @xmath15 , test if ( [ harmstop ] ) is satisfied . if yes , compute @xmath68 and @xmath69 explicitly and stop . 5 . implicitly restart the lanczos bidiagonalization process using the harmonic shifts and the adaptive shifting strategy . * algorithm 2 . irrhlb with the refined harmonic shifts * 1 . given a unit length starting vector @xmath352 of dimension @xmath31 , the steps @xmath35 , the number @xmath14 of the desired singular triplets and the convergence tolerance @xmath92 . 2 . run the @xmath35-step lanczos bidiagonalization process and construct @xmath360 and @xmath361 . 3 . calculate the triplets @xmath362 by computing the singular values and right singular vectors of ( [ svd1 ] ) and by solving ( [ bidiag ] ) and normalizing the solutions , and use the rayleigh quotients @xmath365 as approximations to @xmath366 . 4 . for each @xmath367 , compute @xmath368 and @xmath369 in theorem [ theo4 ] . 5 . for @xmath15 , test if ( [ refinedstop ] ) is satisfied . if yes , compute @xmath218 and @xmath219 by ( [ refinevector ] ) explicitly and stop . 6 . implicitly restart the lanczos bidiagonalization process using the refined harmonic shifts and the adaptive shifting strategy . we have developed the experimental matlab codes of irrhlb , irhlb , irrlb and irlb . the latter two were named irrbl and irbl in @xcite and were originally developed based on the lower lanczos bidigonalization process . here we have developed their upper lanczos bidiagonalization versions . these four codes call the upper lanczos bidiagonalization process in baglama and reichel s code irlba , and some parameters and defaults are the same as those used in irlba . we compare irrhlb with irrhlb , irrlb , irlb , irlba and irlanb in this section and report numerical results . numerical experiments were run on an intel core 2 e6320 with cpu 1.86ghz and ram 2 gb under the window xp operating system using matlab 7.1 with @xmath370 . the stopping criteria are @xmath371 and @xmath372 if @xmath373 then stop . similar criteria apply to irlb and irrlb as well . in ( [ stop ] ) , @xmath374 is replaced by the maximum of the current largest ( harmonic ) ritz value and the old one obtained at last restart . some parameters in irrhlb , irhlb , irrlb and irlb are described in table [ description ] . [ htp ] .parameters in irrhlb , irrlb , irhlb and irlb [ cols= " < , < " , ] clearly , for fidap4 and lshp3205 , irrhlb worked well and solved the problem successfully while irrlb only performed well in some cases and was less efficient than irrhlb . in contrast , irhlb , irlb , irlba and irlanb performed more poorly and they all failed to converge for fidap4 and lshp3205 . for jagmesh8 , irrhlb still worked robustly and efficiently , but irrlb succeeded only in a few cases and irhlb , irlb , irlba and irlanb behaved more badly . they all were considerably less efficient than irrhlb if they worked . we found that , generally , the bigger @xmath14 was , the more restarts irrhlb and irrlb used for the same @xmath35 . this should not be surprising as the problem for a bigger @xmath14 is generally more difficult to solve than that for a smaller @xmath14 . we also observed that all the smallest singular values were computed with relative errors no more than a modest multiple of @xmath375 . we had more observations on the behavior of irhlb , irlb , irlba and irlanb on these three difficult problems . for example , the residual norms obtained by them may oscillated but decreased very slowly ; they may have first decreased to some stage and then oscillated ; they might have first decreased , then stabilized and did not decrease further ; they might have decreased to some stage and then increased . therefore , irrhlb is not only the best but also the unique choice for fidap4 , jagmesh8 and lshp3025 for most of the given @xmath14 s and @xmath35 s . we tested well1850 , pde2961 , dw2048 and plat1919 for @xmath376 , respectively . tables [ tablewell6][tableplat6 ] report the results for @xmath377 . we do not list the corresponding results for @xmath378 , as will be explained shortly . @xmath35 & & & + algorithms & @xmath379 & @xmath380 & @xmath381 & @xmath379 & @xmath380 & @xmath381 & @xmath379 & @xmath380 & @xmath381 + irrhlb & 71&2.67&785&43&2.98&692&35&4.18&739 + irrlb & 69&2.51&763&62&4.48&996&35&4.16&739 + irhlb & 168&5.32&1852&83&4.99&1332&51&5.55&1075 + irlb & 183&5.99&2017&91&5.63&1460&55&6.01&1159 + irlba & 191&1.90&2105&93&1.21&1492&57&1.00&1201 + irlanb & 279&7.88&2795&133&6.61&2000&82&6.40&1645 + + @xmath35 & & & + algorithms & @xmath379 & @xmath380 & @xmath381 & @xmath379 & @xmath380 & @xmath381 & @xmath379 & @xmath380 & @xmath381 + irrhlb & 80&3.10&726&51&3.53&720&35&4.08&671 + irrlb & 103&4.15&933&63&4.43&888&41&5.16&785 + irhlb & 171&5.85&1545&76&4.51&1070&46&4.47&880 + irlb & 184&4.96&1662&82&4.84&1154&49&4.80&937 + irlba & 189&1.70&1707&83&1.01&1168&50&0.86&956 + irlanb & 259&6.11&2079&109&4.79&1424&63&4.10&1141 + + @xmath35 & & & + algorithms & @xmath379 & @xmath380 & @xmath381 & @xmath379 & @xmath380 & @xmath381 & @xmath379 & @xmath380 & @xmath381 + irrhlb & 105&3.18&743&60&4.00&728&43&4.99&739 + irrlb & 161&5.53&1135&70&4.51&848&50&4.64&688 + irhlb & 248&6.43&1744&94&5.14&1136&53&4.06&909 + irlb & 275&6.08&1933&103&4.70&1244&58&4.69&994 + irlba & 292&2.23&2052&108&1.26&1304&60&0.97&1028 + irlanb & 388&7.30&2337&128&4.58&1417&69&3.41&1113 + + @xmath35 & & & + algorithms & @xmath379 & @xmath380 & @xmath381 & @xmath379 & @xmath380 & @xmath381 & @xmath379 & @xmath380 & @xmath381 + irrhlb & 114&6.81&811&63&6.56&769&40&6.54&693 + irrlb & 171&9.47&1210&69&7.18&841&42&6.76&727 + irhlb & 194&5.81&1371&77&4.34&937&45&3.94&778 + irlb & 202&5.50&1427&82&4.77&997&47&4.05&812 + irlba & 170&1.61&1196&72&0.98&871&43&0.81&739 + irlanb & 282&5.73&1706&99&3.89&1103&56&3.43&910 + @xmath35 & & & + algorithms & @xmath379 & @xmath380 & @xmath381 & @xmath379 & @xmath380 & @xmath381 & @xmath379 & @xmath380 & @xmath381 + irrhlb & 93&24.8&2242&77&38.1&2776&64&37.7&2948 + irrlb & 156&36.9&4060&52&23.3&1876&46&30.0&2120 + irhlb & 236&59.6&6140&128&57.2&4612&82&47.9&3776 + irlb & 266&67.5&6920&145&62.9&5224&93&57.5&4282 + irlba & 276&9.71&7180&148&7.54&5332&94&6.64&4328 + irlanb & 406&64.2&10155&219&70.0&7670&142&60.4&6395 + + @xmath35 & & & + algorithms & @xmath379 & @xmath380 & @xmath381 & @xmath379 & @xmath380 & @xmath381 & @xmath379 & @xmath380 & @xmath381 + irrhlb & 100&22.7&2406&81&34.8&2760&67&40.9&2954 + irrlb & 117&26.2&2814&72&31.5&2454&46&32.8&2030 + irhlb & 209&40.8&5022&110&41.8&3746&71&42.7&3130 + irlb & 230&44.0&5526&120&45.2&4086&78&44.9&3438 + irlba & 238&7.84&5718&124&6.21&4222&80&5.55&3526 + irlanb & 280&29.3&6447&142&42.7&4693&92&35.5&3963 + + @xmath35 & & & + algorithms & @xmath379 & @xmath380 & @xmath381 & @xmath379 & @xmath380 & @xmath381 & @xmath379 & @xmath380 & @xmath381 + irrhlb & 107&22.6&2362&80&30.9&2568&69&49.8&2906 + irrlb & 146&32.7&6220&75&35.2&2408&49&33.6&2066 + irhlb & 186&31.5&4100&95&32.7&3048&63&35.1&2654 + irlb & 266&47.0&5860&134&43.0&4296&85&47.8&3578 + irlba & 287&9.27&6321&142&6.79&4552&89&6.06&3746 + irlanb & 224&28.9&4713&114&23.4&3543&73&28.0&3002 + + @xmath35 & & & + algorithms & @xmath379 & @xmath380 & @xmath381 & @xmath379 & @xmath380 & @xmath381 & @xmath379 & @xmath380 & @xmath381 + irrhlb & 190&38.9&3243&110&46.2&2983&89&67.7&3306 + irrlb & 246&51.4&4195&136&65.2&3685&68&54.1&2529 + irhlb & 443&63.5&7544&187&59.3&5062&110&56.8&4083 + irlb & 483&69.0&8224&203&63.6&5494&117&60.0&4342 + irlba & 335&9.99&5636&159&7.21&4250&90&5.85&3322 + irlanb & 546&54.7&8750&222&43.7&5786&126&44.2&4550 + @xmath35 & & & + algorithms & @xmath379 & @xmath380 & @xmath381 & @xmath379 & @xmath380 & @xmath381 & @xmath379 & @xmath380 & @xmath381 + irrhlb & 127&44.1&3306&92&41.3&3316&85&78.9&3914 + irrlb & 230&77.7&5984&147&80.3&5296&66&59.7&3040 + irhlb & 371&116&9650&193&106&6952&122&105&5616 + irlb & 425&136&11054&226&133&8140&142&122&6536 + irlba & 463&21.6&12042&238&16.6&8572&148&14.5&6812 + irlanb & 490&100&12255&252&103&8825&157&99.0&7070 + + @xmath35 & & & + algorithms & @xmath379 & @xmath380 & @xmath381 & @xmath379 & @xmath380 & @xmath381 & @xmath379 & @xmath380 & @xmath381 + irrhlb & 149&44.7&3582&113&62.2&3848&101&83.1&4450 + irrlb & 254&83.6&6102&149&90.3&5072&85&63.3&3746 + irhlb & 475&137&11406&239&131&8132&146&113&6430 + irlb & 537&143&12894&272&134&9254&167&139&7354 + irlba & 581&27.9&13950&284&19.3&9662&172&16.3&7574 + irlanb & 686&133&15785&339&111&11194&204&97.3&8779 + + @xmath35 & & & + algorithms & @xmath379 & @xmath380 & @xmath381 & @xmath379 & @xmath380 & @xmath381 & @xmath379 & @xmath380 & @xmath381 + irrhlb & 161&52.0&3550&127&78.6&4082&100&97.6&4208 + irrlb & 239&65.6&5266&110&66.2&3528&103&99.2&4334 + irhlb & 518&122&11404&248&114&7944&149&111&6266 + irlb & 575&113&12658&278&126&8904&165&116&6938 + irlba & 579&25.1&12745&273&18.2&8743&164&15.2&6895 + irlanb & 604&99.7&12693&284&84.6&8813&169&71.5&6938 + + @xmath35 & & & + algorithms & @xmath379 & @xmath380 & @xmath381 & @xmath379 & @xmath380 & @xmath381 & @xmath379 & @xmath380 & @xmath381 + irrhlb & 290&83.7&4943&176&80.7&4765&139&145&5156 + irrlb & 571&145&9720&189&108&5116&144&132&5341 + irhlb & 925&212&15738&373&163&10084&205&150&7598 + irlb & 1004&193&17081&403&157&10894&223&140&8264 + irlba & 601&23.3&10157&258&15.9&6948&145&12.8&5355 + irlanb & 1302&138&20846&502&119&13066&269&106&9698 + @xmath35 & & & + algorithms & @xmath379 & @xmath380 & @xmath381 & @xmath379 & @xmath380 & @xmath381 & @xmath379 & @xmath380 & @xmath381 + irrhlb & 146&8.20&1610&129&15.0&2068&74&12.0&1558 + irrlb & 351&22.7&3865&200&23.0&3204&105&19.0&2209 + irhlb & 605&37.5&6659&286&27.0&4580&163&28.3&3427 + irlb & 671&41.6&7385&313&32.9&5012&183&29.3&3847 + irlba & 727&12.7&8001&337&8.08&5396&191&6.15&4015 + irlanb & 943&33.2&9435&425&27.9&6380&234&29.2&4685 + + @xmath35 & & & + algorithms & @xmath379 & @xmath380 & @xmath381 & @xmath379 & @xmath380 & @xmath381 & @xmath379 & @xmath380 & @xmath381 + irrhlb & 214&10.9&1932&209&17.7&2890&153&21.4&2913 + irrlb & 605&30.5&5451&179&15.3&2512&123&16.4&2343 + irhlb & 923&37.2&8313&379&31.9&5312&214&28.4&4072 + irlb & 1089&48.7&9807&451&37.8&6320&253&28.8&4813 + irlba & 1277&17.7&11499&511&10.5&7160&273&8.44&5193 + irlanb & 1428&46.4&11431&545&32.7&7092&286&25.6&5155 + + @xmath35 & & & + algorithms & @xmath379 & @xmath380 & @xmath381 & @xmath379 & @xmath380 & @xmath381 & @xmath379 & @xmath380 & @xmath381 + irrhlb & 385&15.3&2703&237&19.7&2582&171&22.2&2915 + irrlb & 1983&86.0&13889&454&40.8&5456&266&35.8&4530 + irhlb & @xmath382&-&-&793&48.2&9524&387&43.7&6587 + irlb & @xmath382&-&-&869&64.0&10436&432&46.7&7352 + irlba & @xmath382&-&-&722&13.5&8647&352&9.77&5983 + irlanb & @xmath382&-&-&1150&54.5&12659&524&45.9&8393 + + @xmath35 & & & + algorithms & @xmath379 & @xmath380 & @xmath381 & @xmath379 & @xmath380 & @xmath381 & @xmath379 & @xmath380 & @xmath381 + irrhlb & 685&45.1&4808&368&45.5&4429&395&94.9&6728 + irrlb & @xmath382.&-&-&1283&165&15409&490&116&8343 + irhlb & @xmath382.&-&-&1809&192&21721&891&155&15160 + irlb & @xmath382.&-&-&@xmath382.&-&-&1428&209&24289 + irlba & @xmath382.&-&-&1317&26.8&12412&562&17.2&8330 + irlanb & @xmath382.&-&-&@xmath382.&-&-&1039&92.7&16638 + we found that all the algorithms computed the desired smallest singular values correctly once they converged . the computed smallest singular values had relative errors no more than a very modest multiple of @xmath383 . we observed that , in terms of restarts and matrix - vector products , irrhlb was often considerably more efficient and several times faster than the others except irrlb . irrlb was nearly as efficient as irrhlb in many cases , and it was slightly better than irrhlb in a few cases ; see , e.g. , table [ tabledw6][tablepde6 ] for the results on dw2048 for @xmath384 and pde2961 for @xmath385 , @xmath384 . however , for the relatively difficult plat1919 , it was less robust than irrhlb and failed to converge for some @xmath14 and @xmath35 ; see table [ tableplat6 ] . irhlb , irlb , irlba and irlanb were less robust and efficient than irrlb , as the tables indicate . in addition , the results demonstrate that the bigger @xmath14 was , the more restarts the algorithms used generally . for @xmath378 , we observed similar phenomena and had similar findings . the only essential exception is that for plat1919 and @xmath386 , @xmath387 , irrlb did not converge after 2000 restarts were used . furthermore , we found that for the four test matrices all the algorithms used more restarts for @xmath378 than those for @xmath377 and they continued converging very smoothly from @xmath377 to @xmath378 , provided they converged . hence we do not list the results anymore . to be more illustrative , we draw some typical curves that feature general convergence processes of the six algorithms . figures [ fig3][fig4 ] depict absolute residual norms versus restarts for well1850 when @xmath385 and @xmath388 , respectively . the figures clearly demonstrate that irrhlb is the fastest , irrlb is the second best , irhlb is faster than irlb while irhlb , irlb , irlba and irlanb are comparable and competitive though irlanb may be slightly slower . the tables tell us that irhlb was faster than irlb . we see from the figures that after some stages the algorithms started converging quite smoothly and they used more but not too more restarts for the smaller @xmath378 . besides , for irlanb , we see that they computed the smallest singular triplet after many restarts then found the second and third smallest singular triplets very quickly . this is because after the previous singular triplet(s ) was ( were ) computed the available subspaces had contained rich information on the later desired singular vectors . we have done more experiments and have similar findings . based on them , we may conclude that irrhlb is the best and the most robust for general purpose and irrlb is the second best . as far as overall efficiency is concerned , in terms of restarts and matrix - vector products , irrhlb is the fastest and irrlb is the second best while irhlb , irlb , irlba and irlanb are all comparable each other and no one is considerably superior to the others . a further observation tells us that irhlb is faster than irlb . that irrhlb is superior to irrlb and irhlb is better than irlb sheds light on the fact argued in the introduction : the refined harmonic projection and the harmonic projection are more suitable for computing the smallest singular triplets than the refined standard projection and the standard projection , respectively . meanwhile , we find that , as far as cpu timings are concerned , irrhlb can be inferior to irlba . this may be partly because @xmath0 is not very large or @xmath0 too sparse , so that the savings of the first @xmath389 steps of the lanczos bidiagonalization process can not compensate implicit restarting with @xmath390 shifts , and partly because our code on implicit restarting is not far from optimized . in any event , as a whole , we can draw an overall conclusion that irrhlb is at least competitive with and can be much more efficient than the five other state of the art algorithms in both robustness and efficiency . the advantages of irrhlb are twofold : it extracts the best left and right approximate singular vectors from the given subspaces in the sense of residual minimizations ; it uses the better refined harmonic shifts to construct better subspaces at each restart . each of these two advantages alone may not gain much , but , as restarts proceed , the cumulative effect of their combination may be very striking , so that irrhlb can be much more efficient than the other algorithms , as also noticed and commented on the refined algorithms for the large eigenproblem in @xcite . we have presented the refined harmonic lanczos bidiagonalization method for computing the smallest singular triplets of large matrices . we have developed a practical implicitly restarted algorithm with the refined harmonic shifts scheme suggested . we have done many numerical experiments and have compared the new algorithm with the five other state of the art algorithms . the results show that the new algorithm is at least competitive with and can be much more efficient than the five other algorithms in both robustness and efficiency . we have reported the numerical results of computation of the smallest singular triplets . we have also done many numerical experiments on computation of the largest singular triplets . as indicated in @xcite , irrlb is generally preferable to irlb @xcite , propack @xcite , lanso @xcite and svds as well as some others ; it is the most robust among the restarted algorithms . note that irlanb is designed to only compute the smallest singular triplets . for computation of the largest singular triplets , we have found that irrlb is at least competitive with the four other algorithms , in which irlba uses ritz approximations . however , more observations reveal that irrhlb and irhlb are considerably inferior to irrlb and irlb , respectively . this suggests that irrlb and irlb are suitable for computing both the largest singular triplets and the smallest ones but irrhlb and irhlb are more suitable for computing the smallest singular triplets . the matlab codes of irrhlb , irrlb , irhlb and irlb can be obtained from the authors upon request . we thank two referees very much for their very valuable and helpful suggestions and comments , which made us improve on the presentation considerably . many thanks also go to kokiopoulou , bekas , gallopoulos and baglama and reichel for generously providing us their irlanb and irlba codes , which made our numerical experiments and comparisons possible . , _ test matrix collection for non - hermitian eigenvalue problems _ , technical report cs-97 - 355 , university of tennessee , knoxville , 1997 . lapack note # 123 . data available at http://math.nist.gov/marketmatrix . , _ lanczos bidiagonalization with partial reorthogonalization _ , chapter of ph.d . thesis , department of computer science , university of aarhus , danmark , 1998 . available online from http://soi.standford.edu/@xmath391rmunk .

we simulate the neutrino and-ray emissions of the galaxy which are originated from the hadronic scattering of cosmic rays ( cr ) with the interstellar medium ( ism ) . rather than assuming a uniform cr density , we estimate the spatial distribution of cr nuclei by means of numerical simulations . we consider several models of the galactic magnetic field and of the ism distribution finding only a weak dependence of our results on their choice . we found that by extrapolating the predicted-ray spectra down to few gev we get a good agreement with egret measurements . then , we can reliably compare our predictions with available observations above the tev both for the-rays and the neutrinos . we confirm that the excesses observed by milagro in the cygnus region and by hess in the galactic centre ridge can not be explained without invoking significant cr over - densities in those regions . finally , we discuss the perspectives that a km neutrino telescope based in the north hemisphere has to measure the diffuse emission from the inner galaxy .

astro-ph0701856

it is well known that the interaction of cosmic ray ( cr ) nuclei with the interstellar medium ( ism ) should give rise to diffuse @xmath0-ray and neutrino emissions @xcite . several satellites ( see @xcite for a review ) , especially egret @xcite , already observed the @xmath0-ray diffuse emission of the galaxy up to @xmath2 . around the gev , the hadronic origin of a large fraction of that emission is testified by its good correlation with the distribution of the interstellar hydrogen . glast observatory @xcite should soon provide a deeper insight into the nature of that emission as it will probe the @xmath0-ray sky with much better sensitivity and angular resolution than egret and extend the explored energy window up to 300 gev . above that energy , however , satellite observatories can hardly provide significant data . ground based experiments are divided in two main groups : extensive air shower arrays and atmospheric cherenkov telescopes . in the former class , experiments like milagro @xcite and tibet @xcite can probe large regions of the sky at energies larger than few tev s with an angular resolution of several degrees . those experiments already provided interesting constraints on the diffuse emission from the galactic disk @xcite and , in the case of milagro , a diffuse emission from the cygnus region has been recently observed @xcite . unfortunately , none of those experiments covers a sky window encompassing the galactic centre ( gc ) . atmospheric cherenkov telescopes , like hess @xcite and magic @xcite , are sensitive to @xmath0-rays with energy in the range @xmath3 and have a very good angular resolution ( better than @xmath4 ) . however , although these instruments can search for a diffuse emission in some limited regions of the sky , they are mainly dedicated to observe quite localised sources . neutrino telescopes ( nts ) may soon provide a new valuable probe of high energy cr physics which can be complementary to @xmath0-ray observatories . the large detection volume , large field of view , available observation time , and relatively good angular resolution ( which should be better than @xmath5 for water cherenkov nts looking for up - going muon neutrinos ) make , at least in principle , these instruments well suited to search for the diffuse neutrino emission of the galaxy . + the amanda telescope , which is operating at the south pole , put already an upper limit on the @xmath6 s flux from the gp @xcite ( see ) and icecube @xcite should significantly improve that limit . however , in this context the most promising nts are those based in the north hemisphere ( antares @xcite , nemo @xcite and nestor @xcite and especially their planned @xmath7 upgrade , the km3net project @xcite ) as they will be sensitive to up - going @xmath6 s coming from the gc region . furthermore , since these instruments will be deployed in water , their angular resolution should be better than nt at the south pole improving their chances to disentangle the signal from the background . from the theoretical point of view , several calculations of the @xmath0-ray and neutrino diffuse fluxes above the tev have been already performed . in @xcite the cr distribution in the galactic disk was assumed to be homogeneous so that the secondary @xmath8 and @xmath0-ray fluxes come out to be proportional to the gas column density ( in @xcite a uniform gas distribution was assumed ) . while that was a very reasonable approximation to start with , a more accurate analysis should account for the inhomogeneous cr distribution which arises due to the not uniform scatter of sources . it should be noted that , since cr sources are likely to be related to supernova remnants ( snr ) and those objects are most abundant in the gas ( target ) rich regions , the secondary @xmath0-ray and neutrino emissions may be significantly enhanced with respect to the case in which a uniform cr distribution is assumed . the cr homogeneity assumption was released in more recent works where the high energy nuclei distribution was modelled by solving the spatial diffusion equation @xcite . those papers , however , were mainly addressed to model the diffuse @xmath0-ray emission below the tev and did not consider neutrinos . in @xcite a different , and promising , approach was considered to solve the diffusion equation and to model the neutrino emission up to the pev . in that paper , however , several approximations were done which did not allow to reach an angular accuracy better than @xmath9 . the aim of this work is to model both the diffuse @xmath0-ray and neutrino emissions of the galaxy above the tev with a better accuracy . that is necessary in order to be able to establish the detection perspectives of air shower arrays and neutrino telescopes and to interpret correctly forthcoming observations . our approach is comprehensive as we carefully account both for the spatial distribution of cr nuclei and for that of the ism . in we start by discussing the main properties of the ism . we pay special attention to the distribution of snr ( see ) and to that of the atomic and molecular hydrogen ( ) . in both cases we adopt models which we apply for the first time in this context . details of gas models and their comparison with those adopted in previous works are given in the [ sec : appendix_gas ] . in we model the distribution of the cr nuclei in the galaxy by solving numerically the diffusion equation . our approach is similar to that sketched in @xcite that we fully exploit here . differently from @xcite we use expressions for the diffusion coefficients as determined from montecarlo simulations of charged particle propagation in turbulent magnetic fields @xcite . this allows us to get more detailed cr distributions above the tev and to test how much the large uncertainties in the knowledge of the turbulent component of the gmf ( see ) affect our predictions . in we combine the simulated cr distributions and the gas models to map the expected @xmath6 and @xmath0-ray emissions . then , in we compare our predictions with available experimental results . in we briefly discuss the perspectives that a @xmath10 nt to be built in the mediterranean sea has to detect the muon neutrino emission from the inner galaxy . finally , in we summarise our conclusions . in order to assess the problem of the propagation of crs and their interaction with the ism we need the knowledge of three basic physical inputs , namely : 1 . the distribution of supernova remnants ( snr ) which we assume to trace that of cr sources ; 2 . the properties of the galactic magnetic field ( gmf ) in which the propagation occurs ; 3 . the distribution of the diffuse gas providing the target for the production of @xmath0-rays and neutrinos through hadronic interactions . since all these inputs are affected by large uncertainties and systematics , we need to estimate how their poor knowledge is transferred in the resulting @xmath0-ray and @xmath8 emission . therefore , we compare the results obtained using several models . in the following we assume cylindrical symmetry and adopt the sun galactocentric distance @xmath11 . the rate of galactic sn explosions , as inferred from observations of sn in external galaxies similar to the milky way @xcite , is roughly @xmath12 respectively for type i and type ii sn . the total rate is @xmath13 which is in reasonably good agreement with records of historical sn . since we will normalise the cr injection rate by requiring the simulated cr flux to meet the observed value at earth position , we are only interested in the @xmath14 and @xmath15 ratio . what is even more crucial to our analysis is the _ spatial distribution _ of type i and ii sn . since type - ia and core - collapsed sn originate from different stellar populations their spatial distribution , as well as their rate , are different . several methods to determined the snr distribution in the galaxy are discussed in the literature . one of the most commonly adopted methods is that to estimate the snr distances on the basis of the surface brightness - distance ( @xmath16 ) relation @xcite . such an analysis does not cover the gc region . furthermore , several doubts have been risen on the accuracy of that method as it is plagued by a number of systematics concerning the completeness of the available snr catalogue and the proper handling of selection effects @xcite . here we adopt a snr distribution a distribution which is inferred from observations of related objects , such as pulsars or progenitor stars , as done e.g. in @xcite which , in our opinion , is a safer approach . in @xcite the spatial distribution of type - ia snr adopted was assumed to follow that of old disk stars which have an exponential scale length @xmath17 along @xmath18 and an exponential scale height @xmath19 . therefore @xmath20 where @xmath21 is a normalisation factor . although type - ia sn are globally less frequent than core - collapsed sn , their rate is dominating in the inner few kpc s of the galaxy . to trace core - collapsed originated snr one may use either hii regions , which are produced by their luminous progenitors , or pulsars , which are a likely left - over of the collapse . the pulsar distribution at birth was estimated to be @xcite @xmath22 \ , & r < 3.7 \ { \rm kpc } \nonumber \\ \noalign{\medskip } \exp \left [ - { \displaystyle r^2 - r_\odot^2 \over \displaystyle ( 6.8 \ { \rm kpc})^2 } \right ] \ , & r > 3.7 \ { \rm kpc } \end{array } \right . \label{snr2}\ ] ] where @xmath23 + 0.21 \ \exp \left [ - \left ( { z \over 0.6 \ { { { \rm kpc } } } } \right)^2 \right]~.\ ] ] while the absolute values of @xmath24 and @xmath25 are irrelevant here , their ratio is needed to normalise the relative weights of type - i and type - ii snr distributions . by requiring that @xmath26 , as it follows from ( [ i - ii - rates ] ) , we find @xmath27 . in we show the total snr radial distribution as obtained in that way and compare it with that given in @xcite . both distributions have been normalised so to take the same value at the solar circle where observations are most reliable . it is evident that in the inner galaxy the snr rate distribution which we adopt in this work , is significantly higher than that given in @xcite . rather , we have a relative good agreement with the distribution adopted in @xcite ( see also @xcite ) which is based on pulsars , and with independent observations of the @xmath28 line of @xmath29al which is thought to be a reliable tracer of sns @xcite . is shown in arbitrary units versus the distance from the gc . the distribution is such that , in cylindrical symmetry , @xmath30 represents the total sn rate in the whole galaxy . the upper line ( red , continue ) is that derived in @xcite which we adopt here , while the lower one ( blue , dashed ) is from @xcite . they are both normalised to unit at @xmath31 . ] we assume the spatial distribution of the cr sources to coincide with that of snrs and the cr energy spectrum at source to be everywhere a power - law with exponent @xmath32 ( @xmath32 , which refers to the injection spectrum , should not be confused with the slope of the observed cr spectrum ) . therefore , the source term in our diffusion equation will be @xmath33 its absolute normalisation @xmath34 and spectral slope @xmath32 will be fixed so to reproduce the observed spectrum at earth below the knee . for protons @xcite @xmath35 the milky way , as well as other spiral galaxies , is known to be permeated by large - scale , so called _ regular _ , magnetic fields . the orientation and strength of those fields is measured mainly by means of faraday rotation measurements ( rms ) of polarised radio sources . from those observations it is known that the regular field in the disk of the galaxy is prevalently oriented along the disk plane and it seems to follow the galactic arms as observed in other spiral galaxies . according to @xcite , its strength at the sun position is @xmath36 while at smaller radii @xmath37 where @xmath38 . a @xmath39 profile seems to give a worst fit of data . unfortunately , observations are not significant for @xmath40 . most likely @xcite the regular field in the disk has a bi - symmetric structure ( bbs ) with a counterclockwise field in the spiral arms and clockwise in the interarm regions . concerning its vertical behaviour , it is generally assumed that @xmath41 decreases exponentially for increasing values of @xmath42 with a scale height of few hundred parsecs . there are increasing evidences that the field is symmetric for @xmath43 ( bbs - s ) @xcite . superimposed to the regular field a random , or turbulent , component of the gmf is known to be present . in the disk , this component is comparable to , or even larger than , the regular one . indeed , the locally observed rms value of the total field is about @xmath44 . which is @xmath45 times larger than @xmath46 . from polarimetric measurements of stellar light and rms of close pulsars it has been inferred that the gmf is chaotic on all scales below @xmath47 . the power spectrum of the gmf fluctuations is poorly known . observational data , obtained from rm of pairs of close pulsars , are compatible with a kolmogorov spectrum , i.e. @xmath48 , though with a very large uncertainty ( see e.g. @xcite and ref.s therein ) . what is most relevant here is the galactic magnetic halo ( mh ) since most of cr propagation takes place outside the disk . the regular component of the mh has been studied by means of rms of polarised extra - galactic radio sources . those observations showed that @xmath49 is mainly azimutal , the same as @xmath41 , and that its vertical scale height is @xmath50 @xcite . the radial behaviour of @xmath49 is poorly known and it is generally assumed that it traces that of @xmath51 . here we share the same attitude . it is worth noticing that the vertical symmetry of the mh might be opposite with respect to that in the disk : this fact may have relevant consequences for the propagation of crs with @xmath52 @xcite . in the following we assume a symmetric structure so that the regular component of the mh can be considered as an extension of @xmath41 and we can combine both in a unique simple structure : @xmath53 with @xmath54 and @xmath55 left as a free parameter . similarly to @xcite we use a @xmath56 function to regularise the exponential vertical profile of the regular and turbulent components of the gmf at @xmath57 . concerning the properties of the random mf component in the halo , very little is known from observations . both the vertical extension of the radio halo and the isotopic relative abundances of cr suggest that the vertical scale height of the random fields is significantly larger than that of the regular one , @xmath58 or larger . it is also likely that the turbulence strength increases in the regions with the highest star - forming activity . we account for this possible radial dependence of @xmath59 by means of the parameter @xmath60 . therefore , we write @xmath61 due to the relative small number of extra - galactic radio sources , the maximal scale of magnetic fluctuations @xmath62 and their power spectrum in the mh are practically unknown . in the following we assume that @xmath62 is the same as in the disk and consider only kolmogorov and kraichnan @xmath63 power spectra . the diffuse gas accounts for about 10 - 15% of the total mass of the galactic disk ( not including dark matter ) and its chemical composition is dominated by hydrogen ( about @xmath64 by number and @xmath65 by mass ) and helium ( @xmath66 by number and @xmath67 by mass ) . hydrogen is shared in three main components @xcite : ionised ( hii , total mass @xmath68 ) , atomic ( hi , @xmath69 ) and molecular ( @xmath70 , @xmath71 ) . while all these components have comparable masses , their spatial distributions are quite different . the much hotter hii has a scale height along the vertical axis considerably larger than the other hydrogen components ( @xmath72 ) . therefore its contribution to the @xmath0-ray and neutrino emissivity from the galactic plane is subdominant and can be neglected in the following . the construction of galacto - centric radial density profiles for the hi and @xmath70 intrinsically needs model dependent assumptions ( see [ sec : appendix_gas ] for further details ) . therefore large uncertainties are involved in this operation . here we consider two models , which we call gas models a and b , both for the atomic and molecular hydrogen , that will be discussed in details in [ sec : appendix_gas ] . model a : : it has been developed by nakanishi and sofue ( ns ) in @xcite for the hi and by the same authors in @xcite for the @xmath70 . model b : : we construct model b by suitably combining the results of different analyses which have been separately performed for the disk and the galactic bulge . for the @xmath70 and hi distributions in the bulge we use a detailed 3d model recently developed by ferriere et al . @xcite on the basis of several observations . for the molecular hydrogen in the disk we use the well known bronfman s et al . model @xcite . for the hi distribution in the disk , we adopt wolfire et al . @xcite 2-dimensional model . although ferriere et al . model is not cylindrically symmetric , we verified that by fitting their 3d distribution with a cylindrically symmetric one and assuming a gaussian vertical profile peaked on the gp , we get @xmath0-ray and neutrino fluxes that , when integrated over windows larger than 1 degrees squared , differ very little from those obtained using the complete 3d models . for this reason , in the following we work only with averaged 2d distributions . in and we show the volume density radial and vertical profiles of the hi and the @xmath70 obtained with models a and b together with the continuos fits we use in our analysis without reporting observational errors as they are typically much smaller than systematic ones ( the difference between model a and b will provide a glimpse of the amount of those uncertainties ) . volume density along the galactic plane ( @xmath57 ) is plotted versus the radius @xmath18 . green crosses display the binned distribution as given in @xcite while and the continuos line ( green ) is our fit of those points ( model a ) . blue crosses are derived from @xcite . the dashed ( red ) line is our fit of those points with the distribution given in @xcite . for @xmath73 ( model b ) . right panel : the radial profile of the hwhm scale height ( @xmath74 ) for the same models . ] volume density along the galactic plane as a function of @xmath18 . green crosses display the binned distribution as given in @xcite and the continuos ( green ) line is our continuos fit of those points ( model a ) . red and blue crosses are derived ( see text ) from @xcite and @xcite respectively . the dashed ( red ) line is our combined fit of all those points ( model b ) . right panel : the radial profile of the hwhm scale height ( @xmath74 ) for the same models . ] while the gas densities in model a and b differ relatively little close to the solar circle ( what is most relevant here is the @xmath70 ) the main discrepancies arise in the galactic bulge . indeed , that is the region where the uncertainties on the gas velocity are the largest . that discrepancy , however , has little consequences on the gas column density ( see ) as it should since this quantity is almost directly related to the observed co emission . in the following we will use model b as our reference model since , in the central region of the ge , it provides a better fit of the @xmath75co emission survey @xcite . in we also compare the hydrogen column density distributions obtained with models a and b with that used in @xcite finding that the former are more narrowly peaked along the gp . ) column density are shown along the galactic plane ( left panel ) and along the @xmath76 line ( right panel ) . the three diagrams correspond to model a ( dashed , blue line ) , model b ( continuos , red line ) and to the hydrogen distribution adopted in @xcite . column densities are averaged over @xmath77 angular bins . ] in the following we will assume that helium is distributed in the same way as all hydrogen nuclei . the ism is a quite turbulent magneto - hydro - dynamic ( mhd ) environment . since the larmor radius of high energy nuclei , @xmath78 is much smaller than the maximal scale of the magnetic field fluctuations @xmath47 , the propagation of those particles takes place in the spatial diffusion regime . the diffusion equation which describes such a propagation is ( see e.g. @xcite ) @xmath79 where @xmath80 is the differential cr density averaged over a scale larger than @xmath81 , @xmath82 is the cr source term ( [ eq : source_term ] ) and @xmath83 are the spatial components of the diffusion tensor . we assume cylindrical symmetry and use einstein notation for repeated indexes . does not contain energy loss / gain terms . this is justified by the fact that in the high energy range relevant for our problem ( @xmath84 i.e. @xmath85 ) nuclei energy losses and re - acceleration are expected to be negligible ( see e.g. @xcite ) . under the assumption of cylindrical symmetry the diffusion equation takes the simpler form @xcite @xmath86 - \partial_z \left [ d_\perp \partial_z \right ] + u_r \partial_r + u_z \partial_z \right\ } n ( e , r , z ) = q ( e , r , z)\;,\ ] ] where @xmath87 and @xmath88 are respectively the diffusion coefficient in the direction perpendicular to @xmath89 and the antisymmetric ( hall ) coefficient , while @xmath90 and @xmath91 are the drift velocities as defined in [ sec : appendix_num ] . will be derived in [ sec : appendix_num ] . several simulations of the diffusion coefficients in high turbulence mhd media have been performed @xcite . we adopt here the expressions for these coefficients which have been derived in @xcite and used in @xcite . those expressions provide @xmath92 and @xmath88 as functions of the regular magnetic field and of the _ turbulence level _ @xmath93 in that point . respect to other works , where the cr density has been simulated by using a mean value of the diffusion coefficients as derived from the observed secondary / primary ratio of cr nuclear species , our approach offers the advantage to provide the diffusion coefficients _ point - by - point_. this allows to test how the expected neutrino and @xmath0-ray emissions depend on the properties of the turbulent components of the gmf . we verified , by means of dedicated runs , that the simulated escape time of nuclei found with our numerical code is compatible with that estimated from observations of the b / c ratio measured at energies around the gev ( see e.g. @xcite ) . all the advantages of getting a realistic spatial distribution of cr by solving the diffusion equation , however , were not fully exploited in @xcite . in that work , in fact , the author solved equation ( [ eq : diff2 ] ) analytically , which was possible only considering an over - simplified spatial dependence of the diffusion coefficients . for example , @xmath92 was assumed to be spatially homogeneous . + in this work we solve numerically[multiblock footnote omitted ] ( [ eq : diff2 ] ) under more general conditions . we impose the boundary conditions @xmath94 on a the cylindrical surface @xmath95 ) where the turbulent halo is supposed to end and crs escape to infinity without further diffusing . for any given value of the particle rigidity the code provides a bi - dimensional histogram @xmath96 mapping the particle spatial distribution in the stationary limit . in order to verify how much the large uncertainties in the knowledge of the gmf properties may affect our predictions for the cr distribution , we consider and test several models . our first result of these tests is that the cr spatial distribution and spectrum are almost independent of the radial and vertical lenght - scales of the regular magnetic field . therefore the large observational uncertainties in @xmath55 and @xmath97 do not affect our final results . in the following we will always assume @xmath98 and @xmath54 . more significant is the effect of changing the turbulent halo scale height @xmath99 as that parameter determines the length over which cr have to diffuse before escaping to infinity ( here we always assume that @xmath100 , as suggested form observations ; see ) . indeed we found that in the inner galaxy , i.e. where sources are most abundant , the cr density increases with @xmath99 . we considered several values of that parameter finding that @xmath101 provide the best matching of egret data ( see ) . in we summarise the main features of all models that will be discussed in this paper . @ccccc model # & snr & @xmath102 & turbulence & @xmath103-symmetry + 0 & cb @xcite & 1 & kolmogorov & s + 1 & ferriere @xcite & 1 & kolmogorov & s + 2 & ferriere @xcite & like snr & kolmogorov & s + 3 & ferriere @xcite & 1 & kraichnan & s + 4 & ferriere @xcite & 1 & kolmogorov & a + in and [ fig : crprof_he ] we show the high energy proton fluxes along two significant sections of the galactic halo : the galactic plane and a plane perpendicular to it at @xmath104 . the effect of changing the snr distribution is evident when comparing models 0 and 1 as both have the same gmf turbulent spectrum ( kolmogorov ) and strength ( @xmath105 everywhere on the gp ) . it is evident that the adoption of the snr distribution given in @xcite gives rise to a @xmath106 increase in the cr density in the gc region with respect to that derived following @xcite . we also investigate the effect of changing the turbulence level uniformly in the magnetic halo finding a marginal effect at energies below the pev . more interesting is the effect of assuming a radially dependent turbulence strength @xmath102 . in model 2 we assume that quantity to follow the same radial profile of snr as shown in the upper curve of . that choice is justified by the well known argument according to which mhd turbulence in the ism is powered by sn ejecta . by comparing proton flux profiles in , [ fig : crprof_he ] for the models 1 and 2 ( as both are obtained by using the same value of @xmath105 at the solar circle ) we find that a radially dependent @xmath93 gives rise to a smoothing of the cr density distribution with respect to that of sources . that is to be expected as regions which are poor of sources are more easily filled by cr coming from more active regions if the turbulence strength is locally smaller . the effect is less evident at higher energies since in that case cr escape more rapidly along the @xmath103 axis . in the above we assumed a kolmogorov power spectrum ( i.e. @xmath107 ) for the turbulent component of the gmf . it is interesting to investigate how the cr spatial distribution is affected by adopting a different spectrum . a reasonable possibility is to assume a kraichnan spectrum ( model 3 ) which is characterised by @xmath108 . noticeably , in this case @xmath109 so that the value of the spectral index at the sources , which is required to fit the observed proton spectrum at the earth , is @xmath110 . this is in good agreement with the value measured for the @xmath0-ray spectrum of several snrs ( see e.g. @xcite and ref.s therein ) . since the kraichnan spectrum is harder than kolmogorov s , it gives rise to a more tight confinement of cr in the nearby of sources explaining the higher density in the inner galaxy which we observe for that model in , [ fig : crprof_he ] . the deformation of the vertical profiles observed for all models at high energy is due to the hall diffusion which gives rise to a drift in a direction perpendicular to both @xmath89 ( along @xmath111 ) and the radial component of @xmath112 . are plotted along the galactic midplane ( left panel ) and along the @xmath103-axis at @xmath104 ( right panel ) . the dash - dotted line ( orange ) , continuos ( red ) , dotted ( green ) , and dashed ( blue ) correspond to the models 0,2,1,3 respectively ( see text ) . all fluxes have been normalised to the observed value at the sun position . ] . ] finally , we also considered the effect of changing the symmetry of the regular magnetic field with respect to the galactic plane ( model 4 ) . in all previous cases a symmetric ( s ) configuration ( i.e. @xmath113 does not reverse its sign for @xmath114 ) was considered . since observations did not settle what the actual symmetry is yet , it is interesting to consider the possible implications of its change . in @xcite it was already shown that if the magnetic halo is a - symmetric , hall diffusion gives rise to a significant increase of the cr density in the gc region at energies above the pev . we confirm the existence of that effect . the consequences for the neutrino spectrum where investigated in @xcite . however , as they are negligible in the energy range considered in this work we will disregard the a - symmetric models in the following . as we mentioned , the diagrams in this sections refer to protons only . the corresponding flux profiles for composite nuclei can be obtained by a simple energy rescaling @xmath115 and by using the proper normalisation given by the observed flux of each nuclide . therefore , @xmath116 where @xmath117 and @xmath118 are respectively the locally observed fraction at 1 tev of the species with charge @xmath119 with respect to protons and their spectral slope . recent compilations of measurements of both quantities can be found in @xcite . we conclude this section by noticing that although the cr energy spectrum is not exactly the same over the whole galaxy , this has negligible implications for the following results . we start by considering @xmath0-ray production by the decay of neutral pions which are generated by the interaction of the nucleonic component of cr with the ism ( mainly hydrogen and helium ) . here we do not consider a possible contribution to the diffuse @xmath0-ray flux which may be originated by inverse - compton ( ic ) scattering of relativistic electrons with the background radiation ( see e.g. @xcite ) . bremsstrahlung is negligible at the energies considered in this work . due to the low density of the ism practically all mesons decay before interacting with matter . since at those high energies constituent nucleons interact independently one from the other , here we need only to consider elementary inelastic nucleon - nucleon scattering . furthermore , we can safely assume proton - neutron invariance . in the energy range considered in this work @xmath120 , the attenuation of the @xmath0-ray flux due to pair - production scattering onto the cmb radiation can be neglected ( see e.g. @xcite ) . high energy neutrinos @xmath121 are prevalently originated by the decay of charged pions and kaons which are generated by the same hadronic scattering process considered for the production of @xmath0-rays . at the source only electron and muon neutrinos are produced . during the propagation over galactic distances neutrino oscillations redistribute equally the neutrino budget among all lepton families . in the following we will be interested only in the @xmath122 flux reaching the earth . under the assumption that the primary proton spectrum is a power - law and that the differential cross - section follows a scaling behaviour ( which is well justified at the energies considered in this paper ) , the @xmath0-ray ( muon neutrino ) emissivity , can be written as ( see e.g. @xcite ) @xmath123 here @xmath124 is the primary cr spectral index and @xmath125 is the @xmath126 inelastic cross - section at @xmath127 @xcite . the @xmath0-ray ( muon neutrino ) yield @xmath128 , as determined in @xcite , is shown in . contribution to the photon emissivity coming from @xmath129 decay , which was not considered in @xcite , has been included here . ] those values are in agreement with previous results ( see e.g. @xcite ) within 20% . -ray and @xmath130 yields are represented as a function of the primary nuclei spectral index @xmath124 . neutrino oscillations are accounted for . ] above few gev s , yields are practically independent of energy . for @xmath131 the @xmath132 flux ratio is @xmath133 . the factor @xmath134 in ( [ qnu ] ) represents the contribution from all other nuclear species both in the cr and the ism . we find @xmath135 which is in good agreement with the value found in other works ( see e.g. @xcite ) . here we used experimental values of @xmath117 and @xmath118 as given in @xcite and the helium / hydrogen ratio in the ism @xmath136 @xcite . all other nuclear components in the ism give a negligible contribution . the differential @xmath0-ray ( neutrino ) flux reaching the earth is given by the line integral @xmath137 where @xmath138 is the distance from the earth and @xmath139 are related to @xmath140 through @xmath141 in cylindrical symmetry . finally , the integrated fluxes are determined by integrating the power law spectrum over the energy up to 1 pev . for the sake of clarity , in the following we will show flux diagrams as obtained only with our model 3b ( model 3 for the cr distribution and b for the gas ) which is our preferred model . as we mentioned in , the gas model b gives the best matching of co surveys , while the cr model 3 has to be preferred because , by adopting gas model b , it best reproduces egret observations above few gev s ( see ) . at the end of this section we will briefly discuss how our predictions would change adopting different models . in we show two representative sections of the neutrino flux profile above 1 tev . in order to show how the expected signal may depend on the experimental angular resolution , in the same figure we draw the flux averaged over angular bins of different sizes . it is evident that due to the narrowly peaked behaviour of the gas density along the gp ( @xmath142 ) , the averaged flux which may be measured from these regions should change significantly by varying the angular resolution . flux is shown along the galactic equator ( @xmath142 , left panel ) and along @xmath76 ( right panel ) . fluxes are averaged over @xmath77 ( dot - dashed , orange line ) , @xmath143 ( continue , red line ) , @xmath144 ( dashed , brown line ) angular bins . the corresponding @xmath0-ray flux can be obtained by multiplying this diagram by 3.1 . ] in we compare our results with those obtained in @xcite and @xcite which have been derived assuming a uniform cr density . we also show the flux as obtained by using the same gas model b but a uniform cr density . by comparing those profiles the reader can see as the flux that we expect from the central region of the gp is significantly larger than in @xcite . that discrepancy is due , for a large fraction , to the snr overdensity in the molecular ring region conversion factor is @xmath145 while we have @xmath146 . ] . flux for @xmath147 , averaged over @xmath77 angular bins , is shown . the continuos ( red ) , dot - dashed ( blue ) , dashed ( green ) , and dotted ( orange ) curves correspond respectively to : our model 3b , a model with the same gas density distribution ( model b ) but a uniform cr density as observed at the earth position ; the model considered in @xcite ; the model considered in @xcite . the corresponding @xmath0-ray flux can be obtained by multiplying this diagram by 3.1 . ] finally in we represent a full sky map of @xmath148 obtained using model 3b . this figure has been done with the healpix package @xcite . is shown in galactic coordinates ( the gc is in the center ) . this map has been obtained with healpix @xcite . the flux distribution is projected in such a way that all pixels in this map correspond to the same solid angle . the corresponding @xmath0-ray flux can be obtained by multiplying this diagram by 3.1 . ] we conclude this section by discussing how our predictions would change if different models for the gas and cr distributions are adopted . we find that by using the gas model a rather than b , the @xmath0-ray and neutrino fluxes along the gp grow at most by a factor 2 . we find comparable displacements by adopting any other of the cr models considered in . therefore , we conclude that , under the hypothesis on which this analysis is based , our results are very little model dependent . first of all , we would like to compare our results with egret observations . that is required to put our results onto more solid grounds and to verify that our choice of the preferred model is consistent with those observations . here we consider only egret measurements of the diffuse emission along the gp between 4 and 10 gev @xcite . such a comparison requires a considerable extrapolation of our previous findings . we are allowed to do that since , already for @xmath149 , nuclei propagation takes place deep into the spatial diffusion regime so that the energy dependence of the diffusion coefficients do not change going to lower energies ( see e.g. @xcite ) . as a consequence , in the stationary regime the cr distribution above 10 gev is given by a uniform @xmath150 rescaling of that shown in . we neglect small corrections due to the variations of the @xmath126 scattering cross section with the energy . in and in we compare respectively theoretical mean fluxes and flux profiles along the gp with egret measurements @xcite . considering the large uncertainties involved in the modelling of the cr and gas distribution , the agreement is rather good . it should be noticed that in the energy range @xmath151 the ic contribution to the photon flux from the gp is expected to be subdominat ( see e.g. @xcite ) . -ray flux along the galactic plane , between 4 and 10 tev , as obtained with our reference model 3b ( blue , dashed line ) is compared with that measured by egret ( red , continuos line ) . fluxes are averaged over @xmath152 angular bins . ] in we also show our prediction for the @xmath0-ray differential spectrum between 1 gev and 100 tev in the region @xmath153 and we compared it with egret @xcite and milagro measurments @xcite in the same region @xcite . that is compatible with that adopted in our paper within errors . ] . all the theoretical fluxes shown in this section have been derived using the cr model 3 and gas model b. as we mentioned in the previous section , while we prefer gas model b because it provides the best description of co surveys , once made that choice , models 3 is that which gives the best fit of egret data . anyhow , passing from one model to another , our predictions change by a factor 2 at most . -ray spectrum multiplied by @xmath154 as obtained with our reference model 3b in the region @xmath153 . green ( red ) bars represent egret ( milagro ) measurements in the same window . ] reassured by these findings we are now ready to compare our results with measurements performed at higher energies . several air shower array experiments looked for the @xmath0-ray diffuse emission of the galaxy above the tev . the most interesting results are those of tibet @xcite and milagro @xcite . both experiments probed different regions of the gp . in all those regions only upper limits were found but in cygnus where milagro found a significant excess on the background @xcite . in we compare our predictions , as obtained with our preferred 3b model ( see ) , with those measurements . with the exception of cygnus ( see also the discussion about the gc ridge at the the of this subsection ) in all other regions we predict fluxes which are significantly below the experimental limits . therefore , there is still room for a large ic contribution . @cccc sky window & @xmath155 & + & & our model & measurements + @xmath156 & @xmath157 & @xmath158 & @xmath159 @xcite + @xmath160 & @xmath161 & @xmath162 & @xmath163 @xcite + & @xmath157 & @xmath164 & @xmath165 @xcite + @xmath153 & @xmath166 & @xmath167 & @xmath168 @xcite + & @xmath157 & @xmath169 & @xmath170 @xcite + @xmath171 & @xmath172 & @xmath173 & @xmath174 @xcite + & @xmath157 & @xmath175 & @xmath176 @xcite + concerning the excess in the cygnus region observed by milagro , we confirm the conclusion @xcite that it can not be explained by the interaction of the diffuse component of galactic cr with the gas in that region ( see ) . indeed , we find that a cr local over - density of about 20 is required to explain that signal in terms of hadronic emission . a similar effect has to be invoked to explain hess measurement of the diffuse @xmath0-ray tev emission from the gc ridge region @xcite . in fact , even accounting for the detailed 3-dimensional gas model recently developed in @xcite for the gc region , we find that in the region @xmath177 the expected @xmath0-ray flux for @xmath147 is @xmath178 , which is almost 10 times smaller than hess measurement . it is worth to notice that the slope of the spectrum measured by hess is close to @xmath179 which is quite different from that of galactic cr . that probably means that primary particles in the gcr have a local origin . it is evident that the analysis performed in this work , as well as in related papers @xcite , can not account for those kind of local emissions . the only available upper limit on the neutrino flux from the galaxy has been obtained by the amanda - ii experiment @xcite . being located at the south pole , amanda can not probe the emission from the gc . in the region @xmath180 , and assuming a spectral index @xmath131 , their present constraint is @xmath181 which implies @xmath182 . according to our model the expected flux in the same region is @xmath183 . in the energy range @xmath184 water ( or ice ) cherenkov neutrino telescopes are best suited to look for up - going muons produced by charged - current interactions of muon neutrinos in the earth . while the earth offers almost a complete shielding from up - going atmospheric muons below the horizon , an unavoidable background is given by atmospheric neutrinos . several estimates have been made of the atmospheric neutrino flux , based on different assumptions modelling hadronic interactions ( see e.g. @xcite ) . above the tev all calculations almost agree predicting an averaged flux @xmath185 though a @xmath186 uncertainty remains due to the experimental error on the primary cr spectrum and the theoretical error modelling strange particle production . since , as it is evident from our previous results , the expected neutrino flux from the galaxy is significantly smaller than such background , a suitable procedure has to be adopted to disentangle the signal . one possible approach is to search only for neutrinos with @xmath187 @xcite . that however may be hard to do due to the very low flux at that high energies and to the earth opacity ( only earth skimming neutrinos or shower - like events produced by down - going neutrinos can be detected in that case ) . in our opinion a more promising strategy is to search for up - going muon neutrinos above @xmath188 tev . their arrival direction can be reliably reconstructed with an angular resolution as good as @xmath189 in water and that information may be used to identify the galactic emission as a localised excess of events . while it is quite evident that the most natural search window ( _ on - source region _ ) is a narrow band along the galactic equator ( see ) , its optimal sizes have to be chosen by taking into account the angular extension of the source . in the previous section we showed that most of the neutrino flux from the galaxy should be concentrated in the region @xmath190 . since the gaussian width of the signal @xmath191 is comparable to the expected angular resolution of neutrino telescopes , some attention has to be paid when choosing the latitude width of the search window . by assuming that both the line spread function of the experiment and the signal profile along @xmath192 are gaussian having widths @xmath193 and @xmath194 , respectively , the optimal search window width is approximatively given by ( see e.g. @xcite ) @xmath195 . since for a km@xmath1 water nt the expected _ line spread function _ is @xmath196 , we think that @xmath197 is a reasonable value to adopt . although to perform a detailed calculation of the expected signal in a given nt is beyond the purposes of this work , here we perform a simplified estimate which is only intended to give the reader a feeling of the chances that forthcoming experiments have to achieve a positive detection . the expected muon detection rate of neutrinos coming from that window is @xmath198 where @xmath199 is the effective area of the neutrino telescope and @xmath200 is the visibility function ( i.e. the fraction of time that a point in a sky with galactic coordinates @xmath201 spends above the visibility horizon ) . for example , @xmath202 at antares geographical position . since the effective area is relatively weakly dependent on the arrival direction of the neutrinos , we adopt here a mean value obtained by averaging over all possible nadir angles . as a reference we use @xmath199 as provided by the antares collaboration ( see e.g.@xcite ) and assume that for a @xmath7 experiment it will be 40 times larger . this is a rapidly growing function of the energy for @xmath203 . it is interesting to observe that the product @xmath204 , which is related to the detection efficiency , is peaked at about 1 tev for a @xmath131 power law spectrum . in we show the integrated muon neutrino detection rate as a function of the minimum energy to be expected in a @xmath7 nt placed at the same geographical position of antares . the upper curve delimiting the signal band corresponds to model 3a , while the lower to model 2b . we compare that rate with the atmospheric neutrino detection rate expected in antares @xcite multiplied by a factor 40 to account for the larger effective area . our calculation accounts for the angular dependence of the expected atmospheric neutrino detection rate . according to , the background is about 50 times larger than the signal above 1 tev in the @xmath205 region , and about 5 times above 10 tev . therefore , an excess may be detectable for @xmath206 only after a considerable number of years . detection rate ( red band , between continuos line ) from the regions @xmath207 ( left ) , @xmath208 ( centre ) , @xmath209 ( right ) , is compared with the background due to atmospheric neutrinos from the same region ( ligth blue band , between dashed lines ) . an effective area 40 times larger than antares has been adopted . e is the neutrino energy . ] we have performed a comprehensive calculation of the neutrino and @xmath0-ray flux distributions which are expected to be originated by the interaction of cr nuclei with the interstellar gas . our computation improves and updates previous analyses under several aspects . + modelling the spatial distribution of cr in the galaxy , we accounted for a spatial dependence of the diffusion coefficients . that approach allowed us to test how different models of the turbulent component of the gmf affect the cr distribution and the secondary @xmath0-ray flux and to choose our preferred model as that which best reproduces egret @xmath0-ray observations above few gevs . interestingly , we found that such a model gives also the best agreement with the hard @xmath0-ray spectra observed for several snrs . + having assumed that cr sources are distributed like snrs , we adopted a detailed model of the spatial distribution of those objects . + concerning the gas distribution , we considered two models , both for the hi and @xmath70 , which we applied for the first time in this framework . although these models have significant differences in some regions , we showed that they give rise to neutrino and @xmath0-ray fluxes which , when averaged over windows larger than few degrees squared , coincide within a factor of a 2 . we also found that the column density profile across the gp is almost the same for those two models , while it differs significantly from that adopted in previous works @xcite . as a consequence , we predict the neutrino and @xmath0-ray emissions to be more narrowly peaked along the gp . this improves significantly the perspectives to disentangle the signal from the almost isotropic background if the experimental angular resolution will be better than @xmath210 . we compared our predictions with the available observations . + concerning the @xmath0-rays , we found that above the tev our predicted fluxes are below the experimental limits by less than an order of magnitude . since our analysis does not account for a possible ic contribution , which may be not negligible above the tev , it is possible that those experiments will get a positive signal in the next few years . the comparison of the experimental results with our predictions , as well as with those of previous works ( see e.g. @xcite ) , will help to understand which fraction of that emission is of leptonic origin . + concerning the signal observed by milagro in the cygnus region @xcite , we confirm that it can not be explained without invoking a significant cr overdensity , or a large ic contribution , which may be due to a local concentration of sources . a similar effect takes place in the gc region . we found that in order to explain the excess observed by hess from the gc ridge @xcite the cr flux must exceed our prediction by a factor of about 10 . this is not unexpected since the distribution of star forming regions , hence of cr sources , is known to be quite clumpy . it is understood that the kind of simulation performed in this work , as well as in previous ones @xcite , can only model the mean cr distribution smoothed on scale of several hundred parsecs . we also extrapolated our results down to few gevs , where the ic scattering emission is expected to be subdominant . we found that in the most dense regions the flux distribution agrees reasonably well with egret observations . this result encourages us to improve the accuracy of our analysis by using , for example , more detailed models of the gas distribution , so to be able to simulate what glast may observe above the gev at least in some limited regions of the sky . going back to neutrinos , we compared our predictions for the muon neutrino flux from the gp with the experimental limit recently established by amanda - ii @xcite . we found that our predicted flux is almost two orders of magnitude below that limit . unfortunately , also icecube @xcite will hardly be able to get a positive signal . + since a neutrino telescope placed in the north hemisphere may have better changes , we investigated this possibility in some details . assuming that such an instrument will be placed at the same position of antares @xcite and have a 40 times larger effective area we estimated the expected signal and the background along the galactic equator . we found that the detection of the smooth component of the diffuse emission may require more than 10 years of data taking . it should be noted , however , that our analysis provides only a lower limit to the expected emission from the galactic plane . as we mentioned in the above , several observations suggest that the cr and gas distributions may be more clumpy than what considered in this work . furthermore , fluctuations of those quantities are likely to be spatially correlated . this may lead to a significant enhancement of the neutrino flux from some regions as may be the case for the gc ridge @xcite and the cygnus region @xcite . in a forthcoming paper we will try to develop our analysis in order to model at least some of these effects . we thank j. candia for valuable discussions and for providing us with the fortran code which was used in @xcite to solve the cosmic ray diffusion equation . we are grateful to h. nakanishi for sending to us unpublished results of his analysis on the hi density distribution and thank d. gaggero for comparing our results with those obtained using a 3d gas distribution in the galactic bulge region . we thank v. cavasinni and v. flaminio for their encouragement and for reading this manuscript before its submission . we also thank , r. aloisio , c. baccigalupi , r. bandiera , p. blasi , a. celotti , s. deglinnocenti , k. ferriere , f. halzen , j. kelley , j. kirk , t. montaruli , for useful discussions and comments . finally we thank the referee for constructive criticisms . d. g. and l.m . are members of the antares collaborations . is also affiliated to the glast collaboration . most of the work by c.e . was done at the universit di pisa and infn , pisa in the antares group . some of the results presented in this paper have been obtained using the healpix @xcite package . the hi and @xmath70 density distributions are generally inferred from radio observations of the @xmath211 line and @xmath75co rotational line emissions respectively . the galacto - centric radial density profiles are estimated by converting the line - of - sight velocity ( which is determined from the doppler shift ) into heliocentric distance by means of the galactic rotation curve ( see e.g. @xcite ) . this operation is impossible for the whole galaxy and it always involves large uncertainties so that some phenomenological models have to be invoked . we are going here to describe in detail the models sketched in . model a : : it has been developed by nakanishi and sofue ( ns ) in @xcite for the hi and by the same authors in @xcite for the @xmath70 . for both gas components ns developed 3-d models ( @xmath212 ) which , however , do not cover the entire galaxy . by averaging over the azimuth they get 2d - distributions which are approximatively symmetric with respect to the galactic plane . ns assumed that the vertical profiles of the hi and @xmath70 densities have the form @xmath213 which is almost coincident with a gaussian peaked at the galactic plane half - width @xmath214 . the quantity @xmath215 is usually called the full width to half maximum ( fwhm ) scale height . then , they derived binned radial distributions for @xmath216 and the gas density @xmath217 along the galactic plane . model b : : it is the product of the combination of results of different analyses which have been separately performed for the disk and the galactic bulge . for the @xmath70 and hi distributions in the bulge we use a detailed 3d model recently developed by ferriere et al . @xcite on the basis of several observations . in that model both the @xmath70 and the hi are concentrated in two main structures . the central molecular zone ( cmz ) appears as a quite dense @xmath218 wide ( sizes at half maximum density ) ellipse on the plane of the sky ( for the hi the vertical extension is 90 pc ) containing almost @xmath219 in @xmath70 and @xmath220 in hi . it gives rise to a pronounced peak in the gas column density in the direction of the gc . the holed galactic bulge disk ( gbd ) is 3 kpc long and 1 kpc wide toroidal structure . its mass is comparable to that in the cmz ( @xmath221 in @xmath70 and @xmath222 in hi ) but it is spread over a much larger volume so that the gas density in that region is much smaller than in the cmz . furthermore , the gbd is significantly inclined with respect to the plane of the sky so that its contribution to the gas column density along the line of sight is quite small . + for the molecular hydrogen in the disk we use the bronfman s et al . model @xcite . although such a model is almost two decades old , it was shown @xcite that it still provides a good description of recent co surveys @xcite . once averaged over the azimuth , bronfman s et al . distribution is well approximated by a disk with a gaussian vertical profile symmetric with respect to the galactic plane . since in @xcite @xmath223 was adopted , we correct the gas densities and the scale heights given in that paper to make them compatible with the value @xmath11 we use in this work . + furthermore , we accounted for the different values of the @xmath70-co conversion factor which is @xmath224 in @xcite and @xmath225 in @xcite . even by taking into account a possible increasing of @xmath226 with @xmath18 ( see e.g. eq.(7 ) in @xcite ) , these values are too different to be compatible . we assume that for @xmath227 @xmath228 as this is the mean value which one gets in that region by taking @xmath229 at @xmath230 and assuming that it grows like @xmath231 as argued in @xcite . hence , we correct bronfman s et al . density by multiplying it by the factor @xmath232 . we will derive here equation ( [ eq : diff2 ] ) . the diffusion equation which describes the propagation of crs in a turbulent medium is , @xmath233 cylindrical symmetry will be assumed . the diffusion tensor can be conveniently decomposed into @xmath234 where @xmath235 are the components of the regular magnetic field versor . the symmetric components @xmath236 and @xmath92 are the diffusion coefficients along and perpendicularly to @xmath89 , while @xmath88 is the antisymmetric ( hall ) diffusion coefficient . all those coefficients are functions of the energy though with different behaviours . for a fixed value of @xmath81 and for @xmath237 , @xmath236 and @xmath92 are proportional to @xmath238 while @xmath239 ( see e.g. @xcite ) , where @xmath0 is the power - law index of the gmf turbulent fluctuations as defined in . the hall diffusion becomes dominant only for very high values of the rigidity , i.e. for @xmath240 . as motivated in and similarly to what done in other works @xcite , we can approximate the regular component of the magnetic field to be azimutally oriented , i.e. @xmath241 and @xmath242 . under such an assumption @xmath236 becomes not physically relevant and the diffusion equation takes the simpler form @xcite @xmath243 - \partial_z \left [ d_\perp \partial_z \right ] + u_r \partial_r + u_z \partial_z \right\ } n ( e , r , z ) = q ( e , r , z)\;,\ ] ] where we defined @xmath244 9 cillis a n and hartman r c 2005 , _ astrophys . j. _ * 621 * 291 gehrels n and michelson p [ for the glast collaboration ] 1999 _ astropart . phys . _ * 11 * 277 ( experiment website : http://www-glast.stanford.edu ) atkins r w _ et al . _ [ milagro collaboration ] 2005 _ phys . lett . _ * 95 * 251103 amenomori m _ et al . _ [ tibet collaboration ] 2005 ( _ preprint astro - ph/0511514 _ ) abdo a a _ et al . _ [ milagro collaboration ] 2006 , ( _ preprint astro - ph/0611691 _ ) hofmann w [ for the hess collaboration ] 2003 _ proceeding of the 28th international cosmic ray conference " , tskukuba _ 2811 ; + ( experiment website : www.mpi-hd.mpg.de/hfm/hess/hess.html ) kelley j l _ et al . _ [ icecube collaboration ] 2005 _ proceeding of the 29th international cosmic ray conference " , pune _ ( _ preprint astro - ph/0509546 _ ) desiati p [ for the icecube collaboration ] , ( _ preprint astro - ph/0611603 _ ) ( experiment website : icecube.wisc.edu ) aslanides e _ et al . _ [ antares collaboration ] 1999 _ preprint astro - ph/9907432 _ ; ( experiment website : antares.in2p3.fr ) de marzo c n [ for the nemo collaboration ] 2001 _ nucl . * b 100 * 344 ; ( experiment website : nemoweb.lns.infn.it/project.htm ) berezinsky v s , gaisser t k , halzen f and stanev t 1993 _ astropart . _ * 1 * 281 ingelman g and thunman m 1996 ( _ preprint hep - ph/9604286 _ ) strong a w and moskalenko i w 1998 _ astrophys . j. _ * 509 * 212 strong a w , moskalenko i v and reimer o 2004 _ astrophys . j. _ * 613 * 962 candia j 2005 jcap * 0511 * 002 casse f , lemoine m and pelletier g 2002 _ phys . rev . _ * d 65 * 023002 candia j , roulet and epele l n 2002 jhep * 0212 * 033 cappellaro e _ et al . _ 1997 _ astron . _ * 322 * 431 lorimer d r 2004 _ in young neutron stars and their environments , iau symp . _ * 218 * ( _ preprint astro - ph/0308501 _ ) strong a w et al . 2004 _ astron . _ * 422 * l47 knodlseder j _ et al . _ 1999 _ astron . _ * 345 * 813 plueschke s _ et al . _ 2004 _ aip conf . proc . _ * 510 * 35 ( _ preprint _ astro - ph/0104047 ) hrandel j r 2003 _ astropart . phys . _ * 19 * 193 han j l , manchester r n , lyne a g , qiao g j , van straten w 2006 _ astrophys . j. _ * 642 * 868 beck r 2001 _ space science rev . _ * 99 * 243 han j l , ferriere k and manchester r n 2004 _ astrophys . j. _ * 610 * 820 han j l and qiao g j 1994 , _ astron . astrophys . _ * 288 * 759 nakanishi h and sofue y 2003 _ publ . astron . _ * 55 * 191 ( _ preprint astro - ph/0304338 _ ) nakanishi h and sofue y 2006 _ publ . _ * 58 * 847 ( _ preprint astro - ph/0610769 _ ) ferriere k , gillard w and jean p 2007 ( _ preprint astro - ph/0702532 _ ) bronfman l _ et al . _ 1988 _ astrophys . j. _ * 324 * 248 cavasinni v , grasso d , maccione l 2006 _ astropart . _ * 26 * 41 costantini m l and vissani f 2005 _ astropart . _ * 23 * 477 lipari p 2006 _ nucl . instrum . _ * a 567 * 405 kelner s r , aharonian f a and bugayov v v 2004 _ phys . rev . _ * d 74 * 034018 dermer c 1986 _ astro . _ * 157 * 223 montaruli t [ antares collaboration ] 2005 _ acta phys . _ * b 36 * ( 2005 ) 509 ( _ preprint hep - ex/0410079 _ ) zornoza j d 2005 , phd thesis , universitat de valencia + ( http://www.slac.stanford.edu/spires/find/hep/www?key=6077153 ) beacom j f and kistler m d 2007 _ phys . rev . _ * d 75 * 083001 anchordoqui l , halzen f , montaruli t and omurchadha a 2006 ( _ preprint astro - ph/0612699 _ ) arimoto n , sofue y and tsujimoto t 1996 _ publ . astron . soc . _ * 48 * 275

we report on an ultraviolet spectroscopic survey of red giants observed by the _ hubble space telescope _ , focusing on spectra of the mg ii h & k lines near 2800 in order to study stellar chromospheric emission , winds , and astrospheric absorption . we focus on spectral types between k2 iii and m5 iii , a spectral type range with stars that are noncoronal , but possessing strong , chromospheric winds . we find a very tight relation between mg ii surface flux and photospheric temperature , supporting the notion that all k2-m5 iii stars are emitting at a basal flux level . wind velocities ( ) are generally found to decrease with spectral type , with decreasing from km s at k2 iii to km s at m5 iii . we find two new detections of astrospheric absorption , for pup ( k5 iii ) and eri ( m1 iii ) . this absorption signature had previously only been detected for tau ( k5 iii ) . for the three astrospheric detections the temperature of the wind after the termination shock correlates with , but is lower than predicted by the rankine - hugoniot shock jump conditions , consistent with the idea that red giant termination shocks are radiative shocks rather than simple hydrodynamic shocks . a full hydrodynamic simulation of the eri astrosphere is provided to explore this further .

1607.07732

ultraviolet and x - ray observations have demonstrated that among main sequence stars stellar coronae are a ubiquitous phenomenon , with coronal emissions being detectable for all stars with spectral types later than a5 v ( e.g. , * ? ? ? this is not the case , however , above the main sequence . for giant stars , f and g type giants are coronal , but this coronal emission either disappears entirely or at least dramatically weakens for k and m type giants . for the k and m giants , the coronae seem to be largely replaced by strong winds ( @xmath7 m@xmath8 yr@xmath2 ) at chromospheric temperatures ( @xmath9 k ) . the first recognition of the existence of this coronal dividing line among evolved stars between coronal and `` windy '' stars was by @xcite . if attention is limited to giant stars , i.e. luminosity class iii stars , the dividing line lies at a spectral type of about k2 iii @xcite . the nature of the transition across this dividing line is a topic of particular interest . some stars to the right of the dividing line , referred to as `` hybrid chromosphere '' stars , seem to maintain some level of coronal emission , with @xmath5 dra ( k5 iii ) being perhaps the best - studied luminosity class iii example @xcite . this has led to the proposition that magnetic loops with coronal temperatures still persist beyond the dividing line , but the coronal emissions from these loops are completely or mostly hidden by the dense , thick chromospheres that have developed for such stars , from which emanate the strong chromospheric winds @xcite . in order to further investigate the transition from coronae to chromospheric winds , we here present a new survey of red giant stars with spectral types later than k2 iii , using uv spectra taken by the _ hubble space telescope _ ( hst ) . we are not only interested in the winds of these stars , but also in the interactions of these winds with the interstellar medium ( ism ) . ultraviolet spectroscopy from hst has provided the first spectroscopic detections of the wind - ism interaction regions , or `` astrospheres , '' of many cool stars , including the heliospheric structure surrounding our own sun . hydrogen lyman-@xmath6 absorption has been observed from the `` hydrogen wall '' region outside the heliopause @xcite , and from the heliotail @xcite . hydrogen wall absorption has also been observed around other sun - like stars , representing the first method by which solar - like coronal winds can be detected around other stars @xcite . however , we are here interested in the astrospheres of red giant stars . observations of the mg ii h & k lines near 2800 have been used to detect absorption from the astrosphere of @xmath6 tau ( k5 iii ) @xcite . models of the @xmath6 tau astrosphere were computed using codes designed to study the heliosphere . analogous to the solar wind , the red giant wind expands radially from the star until it reaches the termination shock ( ts ) , where it is shocked to subsonic speeds . the post - ts wind is slower , hotter , and more dense . it is this post - ts region that provides mg ii column densities sufficiently high to yield the observed mg ii astrospheric absorption . due to its high positive stellar radial velocity ( @xmath10 km s@xmath2 ) , our line of sight towards @xmath6 tau is estimated to be at an angle of @xmath11 from the upwind direction of the ism flow vector in the rest frame of the star . this line of sight is therefore very much through the `` astrotail , '' and this may be essential for detecting the astrospheric absorption . a downwind line of sight provides a much longer path length through the post - ts material than an upwind or sidewind line of sight , leading to much higher column densities through the post - ts material @xcite . nevertheless , the hydrodynamic models of the @xmath6 tau astrosphere were unable to reproduce the observed astrospheric absorption very well , predicting far too little absorption , and placing the absorption farther from the stellar rest frame than observed . the proposed explanation was that the @xmath6 tau ts is a radiative shock , with radiative cooling from h lyman lines yielding further compression and deceleration behind the ts , resulting in a stronger mg ii absorption feature closer to the rest frame of the star , in better agreement with the observed absorption @xcite . the @xmath6 tau observation has long been the only example of a detected red giant astrosphere . our new hst survey of red giant stars was not only designed to study the red giant winds , but also designed to try to find new detections of astrospheric absorption around such stars . we are interested in the warm , partially ionized winds and resulting astrospheres of giants with spectral types between k2 iii and m5 iii . we do not consider stars later than m5 , because such giants contain miras and other long period variables , which have cooler and more massive winds of a different character , and we do not consider giants earlier than k2 , which lie on the wrong side of the coronal dividing line mentioned in section 1 , exhibiting high - temperature coronal emission instead of strong chromospheric winds ( e.g. , * ? ? ? based on the idea that a downwind line of sight through the astrotail is essential for detecting the mg ii astrospheric absorption signature , all the selected targets have @xmath12 km s@xmath2 . this choice has the added benefit of shifting the astrospheric absorption and stellar wind absorption away from the expected locations of narrow interstellar mg ii absorption , which can confuse analysis of the line profile . lcccccccccccc star & alternate & spect . & v@xmath13 & v - k@xmath14 & t@xmath15 & r@xmath16 & d@xmath17 & @xmath18@xmath19 & log f@xmath20@xmath21 & @xmath0@xmath22 & @xmath23@xmath24 & @xmath25 + & name & type & & & ( k)&(r@xmath8 ) & ( pc ) & ( km / s ) & & ( km / s ) & ( km / s ) & ( deg ) + + hd 66141 & hr 3145 & k2 iii & 4.38 & 2.97 & 4274 & 23.9 & 77.9 & 71.6 & 5.08 & ... & 76 & 142 + hd 211416 & @xmath6 tuc & k3 iii & 2.82 & 2.91 & 4310 & 37.3 & 61.2 & 42.2 & 5.02 & ... & 52 & 140 + hd 87837 & hr 3980 & k4 iii & 4.38 & 3.34 & 4086 & 33.6 & 90.6 & 39.8 & 4.69 & ... & 45 & 143 + hd 50778 & @xmath26 cma & k4 iii & 4.08 & 3.42 & 4052 & 35.4 & 79.9 & 96.2 & 4.96 & @xmath27 & 95 & 150 + hd 59717 & @xmath4 pup & k5 iii & 3.25 & 3.66 & 3959 & 43.7 & 59.4 & 87.3 & 4.84 & @xmath28 & 88 & 147 + hd 25025 & @xmath5 eri & m1 iii & 2.94 & 3.87 & 3889 & 58.9 & 62.3 & 60.8 & 4.67 & @xmath29 & 57 & 146 + hd 44478 & @xmath30 gem & m3 iii & 2.87 & 4.73 & 3675 & 107.7 & 71.0 & 54.4 & 4.50 & @xmath31 & 55 & 144 + hd 20720 & @xmath32 eri & m3.5 iii & 3.70 & 4.85 & 3652 & 102.9 & 93.4 & 41.7 & 4.49 & @xmath33 & 34 & 149 + hd 120323 & hr 5192 & m4.5 iii & 4.19 & 5.85 & 3500 & 82.4 & 56.1 & 40.7 & 4.20 & @xmath31 & 43 & 172 + + hd 124897 & @xmath6 boo & k2 iii & @xmath34&2.86 & 4341 & 25.2 & 11.3 & @xmath35 & 5.23 & @xmath36&113 & 92 + hd 164058 & @xmath5 dra & k5 iii & 2.23 & 3.58 & 3989 & 53.4 & 47.3&@xmath37 & 4.79 & ( 74 ) & 18 & 40 + hd 29139 & @xmath6 tau & k5 iii & 0.86 & 3.90 & 3880 & 51.0 & 20.4 & 54.3 & 4.77 & @xmath38 & 44 & 170 + hd 108903 & @xmath5 cru & m3.5 iii & 1.64 & 4.76 & 3669 & 73.9 & 27.2 & 21.0 & 4.49 & @xmath39 & 37 & 116 + table 1 lists properties of the nine red giants that constitute our new red giant survey sample . within the hst archives , there are high resolution mg ii spectra of four other k2-m5 iii stars . these stars are also listed in table 1 , as we will consider their mg ii spectra as well , although only @xmath6 tau has the high @xmath18 that we presume to be advantageous for astrospheric detection . the properties listed in table 1 are spectral type , v band magnitude , v - k color , temperature , radius , distance , radial velocity ( @xmath18 ) , logarithmic mg ii surface flux ( f@xmath20 , in ergs @xmath40 s@xmath2 units ) , stellar wind terminal velocity ( @xmath0 ) , the ism velocity seen by the star ( @xmath23 ) , and the orientation of our line of sight to the star with respect to the upwind direction ( @xmath26 ) . the simbad database was utilized to find the v magnitudes , v - k colors , distances @xcite , and heliocentric radial velocities . the v - k colors are used in the estimation of both the photospheric effective temperatures and stellar radii , based on the prescriptions of @xcite and @xcite , respectively . the radii computed in this manner should have uncertainties of 11.7% @xcite . five of the stars have interferometric radii measured by @xcite . the only one initially found to be in poor agreement was @xmath30 gem . the discrepancy was resolved by replacing the initially assumed magnitude with the 2mass catalog s @xmath41 value @xcite , changing the v - k color and inferred radius significantly . in computing the last two quantities of table 1 regarding the ism flow direction and magnitude at each star , which are relevant for astrospheric studies , we note that our target stars are all close enough to be within the local bubble ( lb ) , a region of generally low ism density extending roughly 100 pc from the sun in most directions @xcite . we verify that the stars are within the lb using the lb maps of @xcite . most of the volume of the lb is believed to be very hot , ionized plasma , which produces much of the soft x - ray background emission @xcite . the flow vector of the lb material is uncertain , but a plausible assumption is that it is roughly at rest relative to the local standard of rest ( lsr ) ( e.g. , * ? ? ? * ) . combining this assumption with the known proper motion and radial velocity of our target stars , we can estimate the ism flow vector seen in the rest frame of the star and its orientation relative to our line of sight to the star , characterized by the @xmath23 and @xmath26 parameters in table 1 . all the new survey stars have @xmath42 , confirming that the lines of sight to these stars are expected to be very downwind , which was one of the desired selection criteria to maximize the likelihood of detecting astrospheric absorption . the nine newly observed stars were observed by the space telescope imaging spectrograph ( stis ) instrument on hst @xcite , with the observations occurring between 2013 october and 2015 january . for each of our targets , the hst visit consisted of a spectrum of the 25742851 wavelength range taken with stis s high resolution ( @xmath43 ) e230h grating , followed by a longer exposure of the 11501700 wavelength range taken with the moderate resolution ( @xmath44 ) e140 m grating . our focus here is on the mg ii h & k lines within the e230h spectrum , with rest wavelengths of 2803.5315 and 2796.3543 , respectively @xcite . the @xmath6 tau spectrum considered is the same as that studied by @xcite , which is from 1994 april 8 , and is a product of the ech - b grating on the uv spectrometer that preceded stis on hst , the goddard high resolution spectrograph ( ghrs ) . the other three archival mg ii spectra considered are all stis / e230h observations , from 1998 august 24 , 2011 august 11 , and 2011 august 4 for @xmath6 boo , @xmath5 dra , and @xmath5 cru , respectively . figure 1 displays the mg ii spectra of all the k2-m5 iii stars listed in table 1 , in spectral type order . we also include in the figure an archival hst / ghrs mg ii spectrum of the k0 iii star @xmath45 gem @xcite . being on the coronal side of the dividing line mentioned in section 1 , the @xmath45 gem spectrum serves to illustrate what a red giant chromospheric mg ii line profile should look like in the absence of an opaque stellar wind . the stellar rest frame is indicated in figure 1 , but for two stars in our sample that are spectroscopic binaries the figure instead indicates the systemic center - of - mass radial velocity and the range of velocities covered by the red giants in their elliptical orbits . these two systems are @xmath6 tuc ( k3 iii+ ? ) and @xmath4 pup ( k5 iii+g5 v ) , with orbital periods of 4197.7 and 257.8 days , respectively @xcite . the actual radial velocities of the red giants at the times of observation are best estimated from narrow h@xmath46 lines in the fuv e140 m spectrum , which suggest radial velocities of 42 and 103 km s@xmath2 for @xmath6 tuc and @xmath4 pup , respectively . the implications of @xmath4 pup s binarity will be discussed in detail in section 6 . the mg ii profiles are complicated by the presence of intervening absorbers in between the stellar chromosphere and the observer . these absorbers include the stellar wind , the ism , and the stellar astrosphere . given that all our stars are relatively nearby , we expect the ism absorption to be relatively close to the velocity predicted by the local interstellar cloud ( lic ) vector measured for the ism cloud immediately surrounding the sun ( e.g. , * ? ? ? * ; * ? ? ? multiple ism velocity components are often observed even for very nearby stars , but not typically at velocities separated by more than @xmath47 km s@xmath2 from the lic velocity @xcite . the lic velocity is shown explicitly in all the figure 1 panels . there is always absorption observed at that location . although the figure identifies the absorption as lic , there could be a blend of ism velocity components , particularly in cases like hd 87837 where the absorption feature is particularly broad and not well centered on the lic velocity . in several cases ( @xmath6 tuc , @xmath30 gem , and @xmath32 eri ) there is a separate , distinct narrow absorption feature adjacent to the lic absorption that we suspect is interstellar , and it is identified as such in figure 1 . our selection of stars with high , positive radial velocities shifts the stellar wind absorption redward of the ism absorption in all cases , generally far enough to adequately separate the two . among our new survey targets , only for @xmath32 eri have we ended up with a case where the lic absorption is clearly within the saturated core of the wind absorption , making it impossible to separate the two and complicating analysis of the absorption . we compute integrated mg ii h and k line fluxes for all the red giant target stars . we correct for ism absorption by interpolating a continuum over the ism features identified in figure 1 . figure 2a shows the k / h flux ratio plotted versus photospheric temperature . the ratio clearly increases towards later spectral types . the ratio would be 2 in an optically thin plasma , but stellar chromospheres are highly optically thick to mg ii emission , so values less than 2 are always observed @xcite . the spectral type dependence of k / h seen in figure 2a , which is not seen for main sequence stars , is indicative of systematic changes in the chromospheric opacity towards the later type giants , which will have lower surface gravities due to larger radii ( see table 1 ) . past solar and stellar chromospheric models have discussed the diagnostic potential of the k / h ratio ( e.g. , * ? ? ? * ; * ? ? ? after summing up the flux of both the h and k lines , we can compare the hst fluxes with those measured previously by the _ international ultraviolet explorer _ for the eight stars we have in common with the @xcite compilation of evolved star iue mg ii fluxes we find reasonably good agreement , with no systematic flux discrepancies , and an average difference of only 20.8% . we convert the fluxes at earth to surface fluxes ( f@xmath20 ) in ergs @xmath40 s@xmath2 units , which are listed logarithmically in table 1 . the biggest uncertainty in the surface fluxes lies in the stellar radii . the 11.7% angular diameter uncertainty from @xcite translates into a 23.4% error in f@xmath20 , which leads to 0.09 dex uncertainties for the @xmath48 f@xmath20 values in table 1 . in figure 2b , the mg ii fluxes are plotted versus temperature . there is a strong correlation , and a simple line fits the data quite well , suggesting @xmath49 the correlation is apparent whether f@xmath20 is plotted versus spectral type , v - k , or temperature . the 1@xmath4 scatter about the fitted line is only 0.054 dex , or about 13% , which is less than the 23.4% f@xmath20 uncertainty estimated above . the remarkably tight correlation between chromospheric activity and temperature ( or spectral type ) seen for red giants is very different from what is found for main sequence stars . for main sequence stars , the mg ii fluxes span a range of about a factor of 30 at a given spectral type , due to chromospheric and coronal emission being largely determined by rotation rate , which can vary greatly ( e.g. , * ? ? ? * ; * ? ? ? * ; * ? ? ? the homogeneity of k2-m5 iii mg ii fluxes at a given spectral type is consistent with the idea that on the noncoronal side of the dividing line giant stars are all rotating very slowly , and at a given spectral type the stars are therefore producing chromospheric emission at basically the same basal flux level @xcite . we found rotational velocity measurements for eight of our thirteen stars @xcite , and all are slow rotators . only @xmath4 pup has a rotational velocity above 2 km s@xmath2 , with @xmath50 km s@xmath2 . in figure 2b , our mg ii fluxes are compared with the `` v - k model '' basal flux line of @xcite , based on iue measurements of many more stars over a wider range of spectral types , albeit with many coronally active stars not emitting at a basal level . there is reasonably good agreement , though our measurements suggest a slightly steeper relation , @xmath51 instead of @xmath52 , at least in this spectral type range . one paradigm for the basal chromosphere phenomenon is that it represents the chromospheric heating level due to the dissipation of acoustic waves generated from photospheric oscillations , and in the absence of significant magnetic activity this is the dominant chromospheric heating mode @xcite . acoustic heating models appear to be able to successfully reproduce the observed dependence of basal flux on photospheric temperature , a dependence that is similar on both giants and main sequence stars . however , alternative views exist , which suggest that a base level of magnetic activity is instead responsible ( e.g. , * ? ? ? support for this view comes from the existence of a basal level of coronal x - ray emission for main sequence stars , which it is widely accepted requires a magnetic origin @xcite . it is less clear that a basal x - ray emission level exists for k2-m5 iii stars , as so few have reliable x - ray detections @xcite . nevertheless , the detection of magnetic fields on the supergiant betelgeuse @xcite and the mira @xmath53 cyg @xcite , as well as the observation of an x - ray outburst on mira itself @xcite , demonstrates that even very cool , evolved , and slowly rotating stars still have some magnetic activity , and many models of red giant chromospheres and wind acceleration rely on the presence of magnetic fields and alfvn waves @xcite . even in the absence of the ism absorption , wind scattering , and astrospheric absorption , the mg ii line profiles are already complex due to the high opacity of the chromospheric mg ii lines . the @xmath45 gem line profile in figure 1 is double - peaked , with a stronger blue peak . such `` self - reversals '' are typical for chromospheric mg ii lines , including ones observed from the sun , and the blue peak is commonly found to be the stronger one @xcite . none of the k2-m5 iii mg ii profiles are like that of @xmath45 gem . the closest are the spectra of hd 66141 ( k2 iii ) and @xmath6 tuc ( k3 iii ) , which might be interpreted as being chromospheric lines that happen to have stronger red peaks , but it is far more likely that this shift in flux from blue peak to red peak is an indication of a chromospheric wind scattering flux from the blue side of the line to the red . the wind will naturally be most opaque just bluewards of the stellar rest frame , extending to the wind s terminal velocity , and even beyond it to an extent determined by the turbulent velocities in the wind at its terminal speed . the wind will be a photon - scattering environment in which mg ii photons absorbed and reemitted by mg@xmath54 ions in the wind will preferentially escape by frequency scattering away from the velocities at which the wind is most opaque ( e.g. , * ? ? ? * ) . for most of the stars in figure 1 , the result is a broad , deep , saturated wind absorption feature blueward of the stellar rest frame . however , for three of the stars close to the dividing line at k2-k4 iii ( hd 66141 , @xmath6 tuc , and hd 87837 ) the wind is apparently weaker and less opaque , and the effect on the line profile is more subtle , yielding merely a line asymmetry with more flux emerging redward of the stellar rest frame . the exception is the well - studied k2 iii star @xmath6 boo , which shows deep wind absorption despite being very close to the coronal dividing line , implying a high mass loss rate of @xmath55 m@xmath8 yr@xmath2 @xcite . for the stars that have deep wind absorption , we can estimate the terminal wind speed ( @xmath0 ) empirically , based on the following prescription . first , we determine the maximum flux level within the mg ii k line blueward of the stellar rest frame ( e.g. , @xmath56 ergs @xmath40 s@xmath2 @xmath2 for @xmath26 cma ) . second , we divide this value by an admittedly arbitrary factor of 20 , and we then determine at what velocity that flux level is reached on the blue side of the wind absorption ( @xmath57 ) . in the stellar rest frame , this represents an upper bound for @xmath0 . next , we determine at what velocity that flux level is reached on the red side of the wind absorption ( @xmath58 ) . the average of @xmath57 and @xmath58 represents an estimate of the center of the wind absorption , which in the stellar rest frame provides an estimate of the lower bound of @xmath0 . the shaded regions in figure 1 indicate the range of @xmath0 implied by these lower and upper bounds for all the stars . if microturbulent velocities in the stellar wind at the terminal velocity are low ( e.g. , a few km s@xmath2 ) then the upper bound ( i.e. , the left edge of the shaded region ) should be close to @xmath0 ; but if turbulent velocities are high then the lower bound ( i.e. , the right edge of the shaded region ) would be a better estimate @xcite . in any case , with the bounds on @xmath0 now defined , we can now quote an estimate of @xmath0 with error bars , which are listed in table 1 . mathematically , if v@xmath59 and v@xmath60 are in a heliocentric rest frame , as in figure 1 , then @xmath61 $ ] , with uncertainty @xmath62 . we can not use this prescription for the first three giants listed in table 1 , which do not have the deep , saturated wind absorption features ( see figure 1 ) . the k5 iii giant @xmath5 dra is a special case . it does not possess a broad , deep stellar wind absorption feature , but there is a rather narrow dip at a heliocentric velocity of @xmath63 km s@xmath2 in figure 1 , which is likely indicative of the wind terminal velocity . this implies a very high wind speed of @xmath64 km s@xmath2 . an older hst / ghrs spectrum of @xmath5 dra studied by @xcite shows a similar profile , but with the absorption feature less blueshifted , suggesting @xmath65 km s@xmath2 . however , @xcite find it difficult to interpret the mg ii profile with a single wind component , and propose the existence of a slower wind component ( with @xmath66 km s@xmath2 ) in addition to the fast one producing the narrow absorption dip . the speed of this slower wind component is more consistent with that observed for the other k giants . the mg ii profiles of @xmath5 dra are in most respects similar to those of hd 87837 ( k4 iii ) , but hd 87837 does not have the absorption dip far from the stellar rest frame that would suggest that there is necessarily opaque stellar wind material at those high speeds . it is tempting to associate @xmath5 dra s unusually fast wind with the star also being one of the most well - known hybrid - chromosphere giants , with weak but detectable coronal x - ray emission @xcite , and detectable transition region uv emission from lines such as c iv @xmath671548 @xcite . @xcite found a very high wind speed of @xmath68 km s@xmath2 for @xmath6 tra ( k4 ii ) , which is another canonical hybrid chromosphere star . before the hst era , @xcite previously noted a possible correlation between the hybrid chromosphere stars and high speed mg ii wind absorption features in iue spectra . thus , both the hst and iue databases of red giant spectra seem to be consistent with the notion that red giants with unusually strong transition region and coronal emissions also have high speed wind components . the wind velocities of the m giants are clearly lower than those of the k giants in our sample , with @xmath69 km s@xmath2 for k2 iii stars near the coronal dividing line , and @xmath70 km s@xmath2 at m5 iii . the correlation is shown explicitly in figure 2c , which plots @xmath0 versus photospheric temperature . excluding the clearly discrepant @xmath5 dra , the linear fit is @xmath71 this is the clearest demonstration of such a correlation for red giants , though past mg ii observations from iue have also suggested a trend of this nature @xcite . there is some ambiguity with regards to whether the optical spectra of ca ii h & k wind absorption support a wind velocity correlation with spectral type . the wind velocities quoted by @xcite would seem to suggest lower velocities for m giants , but inferences of terminal velocities from the ca ii data are open to interpretation . at least within the m giants alone , which are the only class iii stars with reliably deep , saturated wind absorption , the centroid velocity of the ca ii absorption decreases towards later spectral types but the blueward extent of the absorption does not , leading to the early interpretation that these variations are indicative of changes in line formation depth and turbulent velocity with spectral type rather than any change in wind velocity @xcite . the very presence of ca ii wind absorption implies that the stellar wind must be relatively cold close to the star . we can further explore the comparison of ca ii and mg ii wind diagnostics using more recent ca ii observations , specifically spectra taken at a very high resolution of @xmath72 with the coude cross - dispersed spectrometer ( cs21 ) on the 2.7-meter harlan j. smith telecope at mcdonald observatory . in 2006 january , eight k0-m5 evolved stars were observed : @xmath45 gem ( k0 iii ) , @xmath6 boo ( k2 iii ) , @xmath45 umi ( k4 iii ) , @xmath6 tau ( k5 iii ) , @xmath45 and ( m0 iii ) , @xmath6 cet ( m1.5 iii ) , @xmath45 peg ( m2.5 ii - iii ) , and @xmath6 her a ( m5 ii ) . the spectral type range covered resembles our hst red giant sample , although the last two stars drift out of luminosity class iii . figure 3 shows the resulting spectra of the ca ii k line . analogous to figure 1 , @xmath45 gem is used in figure 3 to indicate what the ca ii line profile looks like for a coronal star without a strong chromospheric wind , and as for mg ii , the line profile is double - peaked , with a stronger blue peak . all of the other ca ii k lines in figure 3 have stronger red peaks , suggesting the presence of a wind . however , the latest type stars , @xmath45 peg and @xmath6 her a , are the only ones that exhibit deep , saturated ca ii wind absorption profiles . the wind signatures are more subtle for the earlier type stars . the @xmath6 boo and @xmath45 umi ca ii profiles are very similar to the mg ii profiles of hd 66141 and @xmath6 tuc ( see figure 1 ) . it is interesting that @xmath6 boo , which has strong and deep mg ii wind absorption in figure 1 , does not exhibit analogously strong ca ii absorption in figure 3 . several of the stars ( @xmath6 tau , @xmath45 and , @xmath6 cet , and perhaps @xmath45 umi ) show relatively narrow , weak ca ii absorption between @xmath73 and @xmath74 km s@xmath2 that should be indicative of the stellar wind s terminal speed . these features resemble the narrow mg ii absorption seen in figure 1 for @xmath5 dra , albeit not at nearly as high a velocity . the @xmath73 to @xmath74 km s@xmath2 range of velocities is comparable to the wind speeds inferred from mg ii . however , it is worth noting that for @xmath6 tau , we inferred @xmath75 km s@xmath2 from mg ii , while the @xmath6 tau ca ii absorption in figure 3 would suggest a lower value of @xmath76 km s@xmath2 . this is likely due to actual variability in the wind of @xmath6 tau . the variability of the ca ii wind absorption seen from @xmath6 tau has long been established @xcite . more recently , @xcite found the wind velocity inferred from ca ii wind absorption to increase from 20.9 to 28.0 km s@xmath2 between 2005 september and 2007 march . based on the @xmath6 tau example , astrospheric mg ii absorption should be weak , narrow , and slightly blueshifted from the stellar rest frame . only two of the other stars in our sample , @xmath4 pup and @xmath5 eri , show absorption of this nature ( see figure 1 ) . we not only look for astrospheric absorption in the mg ii lines , but also in other lines within our hst spectra in which it might be observed , particularly the o i line at 1302.2 , c ii at 1334.5 , and fe ii at 2600.2 , which all have line profiles similar to mg ii , with the deep wind absorption trough . for @xmath5 eri we find a marginal detection of fe ii absorption , while for @xmath4 pup we find more convincing detections for both fe ii and c ii . given that for @xmath6 tau the absorption signature was only seen in mg ii , these are the first detections of the astrospheric absorption signature in any line but mg ii . figure 4 displays all astrospheric absorption lines detected to date , for all three red giants with detected astrospheres . .astrospheric absorption line measurements [ cols="<,^,^,^,^,^,^ " , ] we have fitted these absorption features with single absorption lines , analyzing them in the same fashion as ism absorption lines ( e.g. , * ? ? ? * ; * ? ? ? required oscillator absorption strengths are taken from @xcite . each absorption component is defined by three parameters : central velocity ( @xmath77 ) , doppler broadening parameter ( @xmath78 ) , and column density ( @xmath79 ) . to constrain the fit as much as possible , the two mg ii lines are fitted simultaneously , with self - consistent fit parameters . the resulting fits to the data are shown in figure 4 , taking into account instrumental broadening using line spread functions from @xcite . the fit parameters are listed in table 2 . the column densities , with units of @xmath40 , are listed in logarithmic form . the fe ii and c ii data are noisy , and the c ii line is not well resolved in the e140 m spectrum . as a consequence , uncertainties in the fe ii and c ii fit parameters are large . the velocities of the astrospheric absorption listed in table 2 indicate the post - ts velocity , after being subtracted from @xmath18 to place them in the proper stellar rest frame . the @xmath0 velocities in table 1 are the pre - ts velocities . the ratio of the two is a measure of the compression ratio ( @xmath80 ) of the ts . mathematically , @xmath81 $ ] . the @xmath80 values of the three detected astrospheres are listed in table 2 . assuming the absorption line widths are dominated by thermal broadening , the doppler parameters ( in km s@xmath2 ) will be related to temperature by @xmath82 , where @xmath83 is the atomic weight of the species in question . the last column of table 2 lists temperatures computed from this equation using the mg ii doppler parameters ( with @xmath84 ) . if the ts is radiative , some of the astrospheric absorption may be coming from the intermediate radiative relaxation zone of the shock , rather than being entirely from post - ts material that has reached its final flow velocity and temperature . the effects of velocity gradients in this region would be to broaden the absorption line and shift the line centroid blueward of the true terminal post - ts velocity . this would mean that @xmath80 in table 2 could actually be an underestimate and t an overestimate . however , velocity gradients in the relaxation region should also induce asymmetries in the absorption profiles , for which we do not see any evidence , which is an argument in favor of the absorption being dominated by post - relaxation zone material . at the ts , the kinetic energy of the stellar wind is partly converted to thermal energy in the post - ts material . thus , there should be a correlation between @xmath0 and the temperatures listed in table 2 . we can be more specific using the conservation of momentum equation of the rankine - hugoniot shock jump relations , which says @xmath85 where @xmath86 , @xmath87 , and @xmath88 are the pressure , density , and velocity upstream of the shock ; and @xmath89 , @xmath90 , and @xmath91 are the corresponding quantities downstream . an mhd version of equation ( 3 ) would include @xmath92 terms , but such terms should not be significant here . in the strong shock limit appropriate here , @xmath86 is negligible and @xmath93 for a simple hydrodynamic shock ( e.g. , * ? ? ? thus , equation ( 3 ) reduces to @xmath94 the post - ts pressure and temperature are related by @xmath95 , where @xmath96 is the boltzmann constant and @xmath97 the number density . for a pure hydrogen fluid , @xmath98 , where @xmath99 is the proton mass . plugging these relations into equation ( 4 ) and noting that @xmath100 leads to @xmath101 if @xmath88 is in km s@xmath2 units . in figure 5 , we plot this expected relation between stellar wind speed and post - ts temperature , which is compared with the actual measurements for the three detected astrospheres . we do see an increase in temperature with @xmath0 , but the measured temperatures are lower than predicted . the explanation for the discrepancy lies in the assumption of a simple hydrodynamic shock in deriving equation ( 5 ) , which requires @xmath102 in the strong shock limit . however , from table 2 we already know @xmath103 for all three detected astrospheres . this is evidence for a dissipative shock , as suggested before for @xmath6 tau , where the astrospheric absorption could only be reproduced when the ts was modeled as a radiative shock @xcite . the dashed line is representative of the initial post - ts temperature , but radiative cooling decreases this temperature , resulting in additional compression . it is worth noting that the star with the lowest @xmath80 , @xmath5 eri , is not surprisingly closest to the expected temperature in figure 5 . for @xmath5 eri , less radiative cooling and associated compression is necessary to reach the observed temperature . this is explored further with a full hydrodynamic model of the @xmath5 eri astrosphere in the next section . a final question to address in this section concerns the issue of why only three of the red giant targets in table 1 show the astrospheric absorption signature . the @xmath26 values of the three archival nondetections ( @xmath6 boo , @xmath5 dra , and @xmath5 cru ) demonstrate that we are not looking through the astrotail , and therefore we may not have sufficient mg ii column densities along the lines of sight to these stars to detect astrospheric absorption . this , however , is not true for the 7 new targets that did not yield detections . one problem for the three m3-m5 giants in that list ( @xmath30 gem , @xmath32 eri , and hd 120323 ) is that there is very little flux near the stellar rest frame to provide a suitable background for the astrospheric signature , making it harder to detect . this is presumably due to a combination of low wind speed and high mass loss rate . as discussed above with regards to @xmath5 eri , the low wind speed also reduces the heating at the ts , thereby reducing radiative cooling and post - ts compression , which also reduces astrospheric detectability . three of the k2-k4 giants among the nondetections ( hd 66141 , @xmath6 tuc , and hd 87837 ) have winds weak enough that there is no deep , broad wind absorption feature in the mg ii lines . perhaps for these stars the winds are not strong enough to carve out astrospheres that are sufficiently large to yield enough post - ts mg ii column density for an astrospheric absorption detection . this does not explain the @xmath26 cma nondetection , though . the @xmath26 cma nondetection is the hardest to explain , as there is strong wind absorption in the mg ii profiles , there is plenty of flux near the stellar rest frame to provide background for the astrospheric signature , and the @xmath0 and @xmath26 values for the star are both large enough to expect a detection . in any case , given that only 2 of the 9 stars in our new survey yielded astrospheric detections , despite an effort to choose targets likely to provide detections , it is clear that detectable astrospheric absorption signatures are the exception and not the rule for red giants , even in high quality spectra . we compute a hydrodynamic model of the @xmath5 eri astrosphere in much the same manner as we modeled the @xmath6 tau astrosphere @xcite . this is a two - fluid model of a type used to model the heliosphere , in which the plasma and neutrals are treated as two separate , distinct fluids that are allowed to interact through charge exchange processes @xcite . a first step in the modeling process is to define the boundary conditions for the model , specifically the stellar wind boundary conditions at 1 au from the star , and the parameters for the surrounding ism . for the stellar wind , we model the mg ii wind absorption feature using techniques used in the past @xcite , leading to a mass loss rate estimate of @xmath104 m@xmath8 yr@xmath2 and a terminal velocity of @xmath105 km s@xmath2 . these values can be used to infer densities at 1 au from the star , representing the inner boundary condition of the hydrodynamic model . the model only considers hydrogen , expected to be the dominant constituent by number and mass . the ionization state of hydrogen is very uncertain . it is expected to be mostly neutral at the @xmath9 k temperature of the wind , but some degree of ionization may be present . we assume a 10% ionization . we assume a temperature of 7500 k for the wind at 1 au , though this is actually an unimportant parameter given that the flow is very supersonic . turning to the interstellar boundary conditions , the ism flow velocity and direction represented by the @xmath23 and @xmath26 values from table 1 are necessary constraints . as for density and temperature appropriate for hot lb plasma , we assume a temperature and proton density of @xmath106 k and @xmath107 @xmath108 , respectively , as assumed for model 8 of the @xmath6 tau astrosphere from @xcite . there are small , partly neutral clouds within the lb , including the lic surrounding the sun , but we assume @xmath5 eri will be surrounded by the more typical hot lb plasma . figure 6 shows the resulting hydrodynamic model of @xmath5 eri . the figure shows maps of temperature and density . the stellar wind expands radially from the star until it hits the roughly circular ts , at a distance of about 350 au in the upwind direction and about 550 au in the downwind direction . the ts is seen most clearly in the density plot in figure 6 . the heated , compressed , and decelerated post - ts wind material is partly deflected downwind towards the astrotail , separated from the plasma flow of the ism by a parabola - shaped `` astropause '' ( analogous to the heliopause ) , although the large mean free path of the neutrals allows them to cross this boundary and create the halo of neutrals around it , in green in the density plot of figure 6 . unlike most heliospheric models , the ism flow does not pass through a bow shock , as the ism flow is not supersonic due to the high temperature of the assumed lb plasma . figure 7 shows the traces of h i density , temperature , and velocity along our @xmath109 line of sight to @xmath5 eri . the locations of the ts and astropause in this direction are noted . it is the post - ts material in between the ts and astropause that will be responsible for the astrospheric mg ii absorption feature , as that is where the density is highest . computing mg ii absorption from the hydrogen parameters in figure 7 requires that we assume that mg ii is the dominant ionization state of mg , and we assume a solar mg abundance @xcite . the mg ii absorption predicted by the model is compared with the data in figure 8a . the model significantly underpredicts the amount of absorption , to the extent that it is difficult to see the difference between the predicted absorption and the assumed background flux above the absorption . thus , figure 8b shows a different version of this plot , showing the opacity across the line profile . the model is quite successful in predicting the location and width of the absorption , and if the line opacity is arbitrarily increased by a factor of 8 , the predicted absorption actually fits the data reasonably well . the model in figure 6 is actually more successful in reproducing the observed astrospheric absorption than the hydrodynamic @xmath6 tau models , most of which underpredicted the amount of absorption even more severely , and all of which placed the absorption much too far from the stellar rest frame . treating the @xmath6 tau ts as a radiative shock was necessary to increase the amount of absorption and to shift it closer to the stellar rest frame , although even with the radiative shock treatment there was still difficulty predicting a sufficient amount of absorption @xcite . for @xmath5 eri , the predicted absorption is slightly blueshifted from its observed location , but not by much . this is consistent with the @xmath110 value reported in table 2 , which indicates a compression ratio slightly higher that the @xmath102 value of a strong , hydrodynamic shock , but not by very much , unlike @xmath6 tau . the need to include radiative cooling is therefore much reduced for @xmath5 eri . including radiative cooling in the @xmath5 eri model , which we will not try here , would presumably improve agreement with the data somewhat , but it is unlikely to be the main reason the model is greatly underestimating the amount of absorption , considering that @xmath80 is not too much higher than 4 for @xmath5 eri . two other possible reasons for the absorption underestimate that are worth considering are : 1 . our assumed mg abundance in the stellar wind is too low , and 2 . the ism pressure that we are assuming may be too low . as for the first explanation , we are not aware of any precise measurement of a photospheric mg abundance for @xmath5 eri , but even if such a measurement did exist , the mg abundance in the stellar wind could be different . even in the solar wind , the abundances of elements with low first ionization potential can be higher than photospheric abundances by a factor of 4 @xcite . thus , it is plausible that simply assuming a solar photospheric mg abundance as we have could be a significant underestimate . as for the second suggestion , the effects of a high ism pressure on astrospheric absorption were explored at length in the analysis of the @xmath6 tau astrosphere @xcite . increasing the ism density pushes the ts closer to the star , where wind densities are higher , leading to the post - ts densities being higher , thereby producing more absorption . in contrast , the amount of astrospheric absorption was unaffected by increases in stellar wind density , because the wind density increase was offset by the effect of pushing the ts farther from the star , where densities are lower . the astrospheric absorption features are therefore potentially useful as ism pressure diagnostics . the ism pressure that we are assuming in the @xmath5 eri model ( @xmath111 @xmath108 k ) is consistent with measurements of pressures of neutral clouds within the lb @xcite , but it may be a little low based on recent assessments of hot plasma within the lb , which suggest @xmath112 @xmath108 k @xcite . thus , it is possible that an underestimate of the ism pressure could be part of the explanation for the underestimate of astrospheric absorption . interpretations of @xmath4 pup s mg ii lines and attempts to model its wind and astrosphere are complicated by its binarity . as noted in section 2 , the k5 iii giant has a g5 v companion , with an orbital period of 257.8 days . the half - amplitude of the orbital motion is @xmath113 km s@xmath2 @xcite . this is comparable to the wind speed , so the orbital motion will impart a significant increase in this speed in the direction of orbital motion , and a corresponding decrease in speed in the opposite direction . during the course of one 257.8 day orbit the @xmath114 km s@xmath2 stellar wind would travel about 6.5 au , or about 32 stellar radii . the region around the binary on these distance scales will involve a complex merger of high speed and slow speed streams , with uncertain consequences for the effective final terminal wind speed , although it is expected that the high and slow speed streams will have merged into a more homogeneous outflow before the ts is reached at a distance of @xmath115 au . thus , the systemic center - of - mass velocity quoted as @xmath18 in table 1 ( @xmath116 km s@xmath2 ) is definitely a better reference velocity to use when computing the post - ts velocity and @xmath80 than the actual stellar radial velocity at the time of the observation , which inspection of narrow h@xmath117 lines in the e140 m spectrum suggests is @xmath118 km s@xmath2 . the observed astrospheric absorption is certainly close to the former and not the latter . more questionable is our use of @xmath116 km s@xmath2 as the reference velocity for computing @xmath0 . which @xmath18 value is appropriate for estimating @xmath0 depends on whether the wind absorption feature is entirely from the part of the wind close to the star and moving with the star , or whether there is absorption from material farther from the star emitted when the star had a different radial velocity . by assuming @xmath116 km s@xmath2 we are implicitly assuming the latter , but the former is quite possible . if @xmath118 km s@xmath2 is more appropriate , our @xmath0 estimate would increase to @xmath119 , which would make @xmath4 pup look discrepant in figure 2c , albeit still not as discrepant as @xmath5 dra . if @xmath4 pup s mg ii lines could be monitored over the course of an orbit , we could see if the stellar wind absorption precisely follows the star or not , thereby resolving this particular issue . such monitoring has been done in the optical for the ca ii lines of the spectroscopic binary @xmath30 uma ( m0 iii+ ? ) , with a 230 day period , where it is found that the wind absorption does not seem to follow the radial velocity shifts of the red giant primary @xcite . this supports our use of the systemic @xmath116 km s@xmath2 value as the reference velocity in computing @xmath0 , instead of @xmath118 km s@xmath2 . a final complicating factor that should be mentioned concerns the presence of the g5 v companion star , which will be embedded within the wind emitted by the red giant . gravitational effects on the relatively slow red giant wind could possibly be significant , as could interactions between the winds of the two stars . if the companion star s wind is like that of the sun , it may be too weak compared with the massive red giant wind to have any effect on the red giant wind , but this is far from certain . in any case , with all these complications in mind , we choose not to compute a @xmath4 pup astrospheric model at this time . we have analyzed the chromospheric mg ii h & k lines of k2-m5 iii stars observed by hst , consisting of 9 observations obtained as part of a new hst red giant survey , and 4 archival targets . our findings are summarized as follows : 1 . the mg ii line profiles of all 13 stars in our sample show evidence for stellar wind absorption , but for three of the k2-k4 iii stars the effect is only an induced asymmetry in the line profile rather than a deep wind absorption feature . 2 . measured mg ii surface fluxes are very tightly correlated with spectral type and photospheric temperature , consistent with the idea that k2-m5 iii stars redward of the coronal dividing line are all emitting at a basal flux level . the mg ii k / h flux ratio increases towards later spectral types . wind speeds estimated empirically from the mg ii spectra correlate with spectral type and photospheric temperature , with @xmath0 decreasing from @xmath120 km s@xmath2 at k2 iii to @xmath121 km s@xmath2 at m5 iii . 4 . there are 2 new detections of astrospheric absorption among the recently observed stars ( @xmath5 eri and @xmath4 pup ) , for a total of 3 in our sample , including the previous detection towards @xmath6 tau . however , the limited number of new detections indicates that detectable astrospheric absorption in uv lines is not a common phenomenon . for both @xmath5 eri and @xmath4 pup we detect astrospheric fe ii @xmath672600.2 absorption in addition to the mg ii signature , and for @xmath4 pup , astrospheric absorption is also observed in c ii @xmath671334.5 . these are the first detections of red giant astrospheres in lines other than mg ii . analysis of the astrospheric absorption leads to measurements of ts compression ratio and post - ts temperature for the three detected astrospheres . the temperatures are correlated with @xmath0 . however , the @xmath122 k post - ts temperatures are too low and the @xmath123 compression ratios too high according to the rankine - hugoniot shock jump conditions , providing further evidence that red giant termination shocks are radiative shocks rather than simple hydrodynamic shocks . we compute a hydrodynamic model of the @xmath5 eri astrosphere , which is the one with the ts compression ratio ( @xmath110 ) closest to the strong hydrodynamic shock value of @xmath102 . not surprisingly , this model has less difficulty reproducing the observed absorption than was the case for a past study of the @xmath6 tau astrosphere @xcite , for which @xmath124 . the @xmath5 eri model places the absorption at about the right velocity and with about the correct width , but it underpredicts the mg ii opacity by a factor of 8 . this might be due to either an underestimate of the mg abundance in the stellar wind , or an underestimate of the ism pressure surrounding the star . we would like to thank drs . seth redfield and alexander brown for contributing to the acquisition and analysis of the ca ii spectra . support for hst program go-13462 was provided by nasa through an award from 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 has made use of the simbad database , operated at cds , strasbourg , france . airapetian , v. , carpenter , k. g. , & ofman , l. 2010 , apj , 723 , 1210 aurire , m. , donati , j. -f . , konstantinova - antova , r. , et al . 2010 , a&a , 516 , l2 ayres , t. r. , brown , a. , & harper , g. m. 2003 , apj , 598 , 610 ayres , t. r. , brown , a. , & harper , g. m. 2006 , apj , 651 , 1126 ayres , t. r. , brown , a. , harper , g. m. , et al . 1997 , apj , 491 , 876 basri , g. s. , & linsky , j. l. 1979 , apj , 234 , 1023 blanco , c. , bruca , l. , catalano , s. , & marilli , e. 1982 , a&a , 115 , 280 boesgaard , a. m. , & hagen , w. 1979 , apj , 231 , 128 buchholz , b. , ulmschneider , p. , & cuntz , m. 1998 , apj , 494 , 700 buzzoni , a. , patelli , l. , bellazzini , m. , fusi pecci , f. , & oliva , e. 2010 , mnras , 403 , 1592 cerruti - sola , m. , cheng , c .- c . , & pallavicini , r. 1992 , a&a , 256 , 185 cuntz , m. , rammacher , w. , & musielak , z. e. 2007 , apj , 657 , l57 cutri , r. m. , skrutskie , m. f. , van dyk , s. , et al . 2003 , ycat , 2246 , 0 de medeiros , j. r. , alves , s. , udry , s. , et al . 2014 , a&a , 561 , a126 de medeiros , j. r. , & mayor , m. 1999 , a&as , 139 , 433 dehnen , w. , & binney , j. j. 1998 , mnras , 298 , 387 deutsch , a. j. 1960 , in stellar atmospheres , ed . j. l. greenstein ( chicago : university of chicago press ) , 543 . donnelly , r. f. , white , o. r. , & livingston , w. c. 1994 , sol . phys . , 152 , 69 drake , s. a. 1986 , in new insights in astrophysics : eight years of uv astronomy with iue ( esa sp-263 ) , 193 drake , s. a. , & linsky , j. l. 1986 , aj , 91 , 602 dring , a. r. , linsky , j. , murthy , j. , et al . 1997 , apj , 488 , 760 grevesse , n. , & sauval , a. j. 1998 , space sci . , 85 , 161 haisch , b. , schmitt , j. h. m. m. , & rosso , c. 1991 , apj , 383 , l15 harper , g. m. , brown , a. , & redfield , s. 2011 , in 16th cambridge workshop on cool stars , stellar systems , and the sun , ed . c. m. johns - krull , m. k. browning , & a. a. west ( san francisco : asp ) , 1145 harper , g. m. , wood , b. e. , linsky , j. l. , et al . 1995 , apj , 452 , 407 hernandez , s. , aloisi , a. , bohlin , r. , et al . 2012 , stis instrument handbook , version 12.0 ( baltimore : stsci ) jenkins , e. b. 2002 , apj , 580 , 938 judge , p. g. , & carpenter , k. g. 1998 , apj , 494 , 828 judge , p. g. , & stencel , r. e. 1991 , apj , 371 , 357 karovska , m. , schlegel , e. , hack , w. , raymond , j. c. , & wood , b. e. 2005 , apj , 623 , l137 kelch , w. l. , linsky , j. l. , basri , g. s. , et al . 1978 , apj , 220 , 962 kimble , r. a. , woodgate , b. e. , bowers , c. w. , et al . 1998 , apj , 492 , l83 lallement , r. , welsh , b. y. , vergely , j. l. , crifo , f. , & sfeir , d. m. , 2003 , a&a , 411 , 447 lbre , a. , aurire , m. , fabas , n. , et al . 2014 , a&a , 561 , a85 leenaarts , j. , pereira , t. m. d. , carlsson , m. , uitenbroek , h. , & de pontieu , b. 2013 , apj , 772 , 89 linsky , j. l. , & avrett , e. h. 1970 , pasp , 82 , 169 linsky , j. l. , & haisch , b. m. 1979 , apj , 229 , l27 linsky , j. l. , & wood , b. e. 1996 , apj , 463 , 254 mathioudakis , m. , fruscione , a. , drake , j. j. , et al . 1995 , a&a , 300 , 775 morton , d. c. 2003 , apjs , 149 , 205 mozurkewich , d. , armstrong , j. t. , hindsley , r. b. , et al . 2003 , aj , 126 , 2502 ogorman , e. , harper , g. m. , brown , a. , drake , s. , & richards , a. m. s. 2013 , apj , 146 , 98 pauls , h. l. , zank , g. p. , & williams , l. l. 1995 , j. geophys . res . , 100 , 21,595 prez martnez , m. i. , schrder , k. -p . , & cuntz , m. 2011 , mnras , 414 , 418 pourbaix , d. , tokovinin , a. a. , batten , a. h. , et al . 2004 , a&a , 424 , 727 rammacher , w. , & cuntz , m. 2003 , apj , 594 , l51 redfield , s. , & linsky , j. l. 2002 , apjs , 139 , 439 redfield , s. , & linsky , j. l. 2004 , apj , 602 , 776 redfield , s. , & linsky , j. l. 2008 , apj , 673 , 283 reimers , d. 1975 , in problems in stellar atmospheres and envelopes ed . b. baschek , w. h. kegel , & g. traving ( new york : springer - verlag ) , 229 reimers , d. 1977 , a&a , 57 , 395 reimers , d. 1982 , a&a , 107 , 292 ribas , i. , guinan , e. f. , gdel , m. , & audard , m. 2005 , apj , 622 , 680 robinson , r. d. , carpenter , k. g. , & brown , a. 1998 , apj , 503 , 396 schmitt , j. h. m. m. 1997 , a&a , 318 , 215 schrijver , c. j. 1987 , a&a , 172 , 111 schrder , k. -p . , mittag , m. , prez martnez , m. i. , cuntz , m. , & schmitt , j. h. m. m. 2012 , a&a , 540 , a130 sfeir , d. m. , lallement , r. , crifo , f. , & welsh , b. y. 1999 , a&a , 346 , 785 shu , f. h. 1992 , the physics of astrophysics , vol . ii : gas dynamics ( mill valley : university science books ) snowden , s. l. , chiao , m. , collier , m. r. , et al . 2014 , apj , 791 , l14 snowden , s. l. , egger , r. , finkbeiner , d. p. , freyberg , m. j. , & plucinsky , p. p. 1998 , apj , 493 , 715 stepien , k. , & ulmschneider , p. 1989 , a&a , 216 , 139 suzuki , t. k. 2007 , apj , 659 , 1592 ulmschneider , p. , rammacher , w. , musielak , z. e. , & kalkofen , w. 2005 , apj , 631 , l155 van belle , g. t. 1999 , pasp , 111 , 1515 van leeuwen , f. 2007 , a&a , 474 , 653 von steiger , r. , wimmer - schweingruber , r. f. , geiss , j. , & gloeckler , g. 1995 , adspr , 15 , 3 wood , b. e. , harper , g. m. , mller , h. -r . , heerikhuisen , j. , & zank , g. p. 2007 , apj , 655 , 946 wood , b. e. , izmodenov , v. v. , alexashov , d. b. , redfield , s. , & edelman , e. 2014a , apj , 780 , 108 wood , b. e. , mller , h. -r . , redfield , s. , & edelman , e. 2014b , apj , 781 , l33 wood , b. e. , mller , h. -r . , & witte , m. 2015 , apj , 801 , 62 wood , b. e. , mller , h. -r . , zank , g. p. , linsky , j. l. , & redfield , s. 2005a , apj , 628 , l143 wood , b. e. , redfield , s. , linsky , j. l. , mller , h. -r . , & zank , g. p. 2005b , apjs , 159 , 118 woodgate , b. e. , kimble , r. a. , bowers , c. w. , et al . 1998 , pasp , 110 , 1183

we present a relativistic formulation of pion loop corrections to the coupling of photons with nucleons on the light - front . vertex and wave function renormalization constants are computed to lowest order in the pion field , including their nonanalytic behavior in the chiral limit , and studied numerically as a function of the ultraviolet cutoff . particular care is taken to explicitly verify gauge invariance and ward - takahashi identity constraints to all orders in the expansion . the results are used to compute the chiral corrections to matrix elements of local operators , related to moments of deep - inelastic structure functions . finally , comparison of results for pseudovector and pseudoscalar coupling allows the resolution of a long - standing puzzle in the computation of pion cloud corrections to structure function moments .

1306.6073

the importance of chiral symmetry in hadron physics has been understood for more than 50 years . as an explicit lagrangian representation of the approximate chiral symmetry ( pcac ) that had been recognized in low energy pion nucleon interactions , gell - mann and levy @xcite constructed the extremely successful linear sigma model . in that model the pion couples to the nucleon through pseudoscalar coupling , while an additional scalar ( @xmath1 ) field is also coupled linearly . at a more formal level , gell - mann @xcite proposed su(2)@xmath2su(2 ) as an exact algebra for the charges associated with the hamiltonian governing the strong interaction , even though chiral symmetry was not an exact symmetry of that hamiltonian . on the basis of current algebra one can show very generally that the amplitude for pion scattering or production must vanish as the four - momentum of the pion vanishes @xcite . within the linear sigma model , this important result for low energy pion - nucleon scattering , for example , is only possible through a subtle cancellation of two large contributions , the first involving pion emission and absorption through the pseudoscalar ( ps ) coupling , and the second involving @xmath1 exchange in the @xmath3-channel . keeping track of the necessary cancellations between such large terms in the linear sigma model is tedious and for that reason modern formulations of chiral effective field theory tend to prefer a lagrangian formulation based on a nonlinear realization of chiral symmetry @xcite . in such a formulation the natural @xmath4 vertex involves pseudovector ( pv ) coupling and the vanishing of pion nucleon scattering amplitudes as the pion four - momentum vanishes emerges trivially . in this work , motivated by the phenomenological simplicity of enforcing soft - pion theorems , we focus on the case of pv coupling . however , since the linear realization of chiral symmetry is still used in the literature , for completeness we also compare our results with those for ps coupling . more recently , chiral symmetry and the pion cloud of the nucleon have been shown to play a central role in understanding various flavor and spin asymmetries in quark distribution functions measured at high energies . most prominent of these has been the su(2 ) flavor asymmetry in the proton sea , with the large excess of @xmath5 quarks over @xmath6 being predicted in ref . @xcite and found in deep - inelastic scattering @xcite and drell - yan experiments @xcite . while a nonperturbative pionic component of the nucleon wave function provides a natural explanation for the sign of the observed asymmetry , calculations of the magnitude of the @xmath7 difference have typically been made in models without a direct connection to qcd . furthermore , while the most convenient framework for describing high - energy reactions is the light - front , the realization of chiral symmetry on the light - front is yet to be fully understood ( for a recent discussion see , _ e.g. _ , ref . @xcite ) . in ref . @xcite we examined the framework dependence of pion loop effects for the simple case of the nucleon self - energy . we showed that results for the model - independent , nonanalytical behavior associated with the long - range part of the pion cloud @xcite are in fact independent of whether the calculation is performed using light - front , instant form ( in the rest frame or infinite momentum frame ) , or covariant perturbation theory . on the other hand , important differences were observed for the nonanalytic structure of the self - energy when comparing the pv and ps couplings . applying the methodologies developed in refs . @xcite , we consider here the more physically relevant case of the electromagnetic coupling of the nucleon dressed by pion loops . this represents the necessary next step towards the computation of the chiral corrections to quark distribution functions of the nucleon , whose moments are given by matrix elements of twist-2 operators . the twist-2 matrix elements were studied previously by chen and ji @xcite and arndt and savage @xcite , who computed the most important pion loop contributions to the leading nonanalytic behavior within heavy baryon chiral perturbation theory . in the present analysis we compute the pion loop corrections to the vertex renormalization factors using a fully relativistic framework , which includes higher order corrections in the pion mass @xmath0 . furthermore , we demonstrate explicitly that gauge invariance and the ward - takahashi identities hold to all orders @xmath0 , provided the full set of one - loop diagrams is considered , including rainbow , tadpole and kroll - ruderman contact terms . we verify this for both pv and ps theories . using the results for the vertex corrections , we then derive the pion loop corrections to the matrix elements of the twist-2 operators for both the proton and neutron , and verify the nonanalytic behavior of the isoscalar and isovector contributions . note that for the lowest moment of the nonsinglet distribution the chiral corrections are essentially those that appear for the nucleon electromagnetic form factors at zero four - momentum transfer squared , @xmath8 . these have been computed in a relativistic formalism to one loop order in refs . in contrast to form factors , for high - energy observables such as quark distribution functions the natural framework is the light - front , to which we specialize in this work . while we also focus on the lowest moment of the quark distributions , for the reconstruction of the distributions themselves @xcite , higher moments of the distributions will of course be necessary ( sec . [ sec : me ] ) . in sec . [ sec : pin ] we review the basics of the pion nucleon interaction in terms of the chiral lagrangian evaluated to lowest order in derivatives of the pion field . the electromagnetic nucleon vertex corrections arising from pion loops are computed in sec . [ sec : z1 ] , and their nonanalytic properties studied as a function of the pion mass . to illustrate the role of the various contributions to the vertex renormalization explicitly , we compute the renormalization factors numerically as a function of the ultraviolet cutoff . the results for the vertex corrections and wave function renormalization are subsequently used in sec . [ sec : me ] to compute the chiral corrections to nucleon matrix elements of twist-2 operators . comparison of the results for pv and ps coupling also allows us to identify the origin of the discrepancy between the nonanalytic behaviors of the twist-2 moments computed in heavy baryon chiral perturbation theory and at the parton level in terms of the sullivan process @xcite . finally , in sec . [ sec : conc ] we summarize our findings and outline future extensions of the present work . in appendix [ app : feyn ] we collect formulas for the complete set of feynman rules needed to compute the vertex renormalization and wave function corrections . the demonstration that the results respect gauge invariance and the ward - takahashi identity is presented in appendices [ app : gi ] and [ app : wti ] , respectively , and some useful results for the nonanalytic behavior of integrals are listed in appendix [ app : lna ] . although some of the formal results which we summarize here can be found elsewhere , our aim will be to provide a pedagogical discussion of the derivations in order to clarify some conflicting claims in the literature about the computation of the analytic and nonanalytic contributions to the pion loop integrals . to lowest order in derivatives of the pion field @xmath9 , where @xmath10 , the @xmath4 lagrangian is given by @xcite @xmath11 where @xmath12 is the nucleon field , @xmath13 is the pauli matrix operator in nucleon isospin space , @xmath14 mev is the pion decay constant , and @xmath15 is the nucleon axial vector charge . our convention follows that in ref . @xcite , but differs by an overall minus sign for the @xmath16 term in eq . ( [ eq : lpin ] ) from that in ref . @xcite . for quantities where the @xmath17 field enters quadratically , such as the pion loop corrections discussed here , the overall sign on the @xmath16 term is immaterial . the @xmath16-dependent term in the lagrangian ( [ eq : lpin ] ) gives rise to the `` rainbow '' diagram in which a pion is emitted and absorbed by the nucleon at different space - time points , while the second is the weinberg - tomozawa coupling @xcite , which has two pion fields coupling to the nucleon at the same point . the latter gives the leading contribution to s - wave pion - nucleon scattering @xcite , and generates the pion tadpole or bubble diagrams . the above definitions mean that the field @xmath18 corresponds to an incoming negatively charged pion , with @xmath19 to an outgoing positively charged pion . in writing eq . ( [ eq : lpin ] ) we have also made use of the goldberger - treiman relation between @xmath16 , @xmath20 and the @xmath4 coupling constant @xmath21 , @xmath22 where @xmath23 and @xmath24 is the nucleon mass . the interaction of pions and nucleons with the electromagnetic field is introduced by minimal substitution , @xmath25 , where the charge @xmath26 for a photon coupling to an electron . this gives rise to a @xmath27 interaction lagrangian of the form @xmath28 where @xmath29 is the nucleon charge operator , defined in terms of the total isospin @xmath30 of the nucleon and its third component such that @xmath31 , and @xmath32 , and similarly for the pion charge operator one has @xmath33 . the first two terms in eq . ( [ eq : lping ] ) correspond to the photon coupling to the bare nucleon and pion , respectively , the third term is the kroll - ruderman coupling @xcite required by gauge invariance , while the fourth term gives rise to a photon coupling to a pion nucleon tadpole vertex . the lagrangians ( [ eq : lpin ] ) and ( [ eq : lping ] ) can be used to derive a set of feynman rules for computing lowest order amplitudes , which are summarized in appendix [ app : feyn ] . the transformation of the ps coupling lagrangian , such as in the linear @xmath1 model @xcite , to the pv coupling lagrangian in eq . ( [ eq : lpin ] ) can be understood as a canonical transformation of the field variables , analogous to the coordinate transformation from the cartesian coordinates @xmath34 and @xmath35 to the plane polar coordinates @xmath36 and @xmath37 @xcite . in analogy with the coordinate transformation , the pv coupling lagrangian can be derived from the ps lagrangian with the form of @xmath38 ( representing , for example , @xmath39 ) by taking the equivalence between the ( @xmath40 ) and ( @xmath41 ) coordinates , such that @xmath42 with an appropriate field redefinition for the nucleon , pion and @xmath1 fields . since the redefined scalar field turns out to be completely decoupled after the canonical transformation and becomes irrelevant to the chiral symmetry , one may completely remove it in the pv coupling lagrangian @xcite , as shown in eqs . ( [ eq : lpin ] ) and ( [ eq : lping ] ) . in the original ps coupling lagrangian it is crucial to keep the @xmath1 field to maintain the chiral symmetry . thus , the pseudoscalar lagrangian is not invariant under chiral transformations without the presence of a scalar field , as in the linear sigma model @xcite . however , historically the pseudoscalar pion nucleon interaction without a scalar field has often been discussed in the literature , with the lowest order lagrangian density given by @xmath43 as discussed by lensky and pascalutsa @xcite , this can be formally obtained by redefining the nucleon field @xmath44 , where @xmath45 $ ] , which leads to the ps lagrangian in eq . ( [ eq : lps ] ) , together with the weinberg - tomozawa contribution in ( [ eq : lpin ] ) replaced by an isoscalar term as in the @xmath1 model and a modified isovector term . to contrast a number of important features pertinent to the pv and ps calculations , we discuss here the consequences of neglecting the scalar field contribution for the ps theory . for on - shell nucleons obeying the free dirac equation , the ps and pv lagrangians ( [ eq : lpin ] ) give identical results for matrix elements , provided the couplings are related by eq . ( [ eq : gt ] ) . for off - shell nucleons , however , the ps and pv interactions lead to different results because of the strong coupling to negative energy states in the former . this can be illustrated by splitting the nucleon off - shell propagator into an on - shell part and an off - shell part , according to the identity @xcite @xmath46 where @xmath47 and @xmath48 . one observes that while the on - shell component gives equivalent results for pv and ps interactions , the contribution from the off - shell part @xmath49 differs for the pv and ps couplings @xcite . these differences were studied in detail for the case of the nucleon self - energy @xmath50 in ref . @xcite , where the leading nonanalytic behavior of @xmath50 was found to be of order @xmath51 for the ps case , in contrast to the @xmath52 behavior of the pv theory . in the nucleon self - energy , the difference in the off - shell part @xmath49 indeed appears as a pion tadpole contribution in the ps theory . since the pion tadpole involves the two - pion coupling , which corresponds to a scalar coupling , it is evident that the equivalence between the pv and ps coupling theories can not be attained without a scalar field to restore the chiral symmetry . important differences arise also for the vertex corrections , as we shall discuss in the following . beyond tree level , the interactions described by eqs . ( [ eq : lpin ] ) and ( [ eq : lping ] ) give rise to loop corrections which renormalize the electromagnetic photon nucleon vertex . these corrections are illustrated in fig . [ fig : loops ] . in this section we will derive the corrections arising from each of these diagrams explicitly and study their dependence on the high - momentum cutoff mass , as well as their nonanalytic properties as a function of the pion mass . for illustration , we estimate the contribution from each diagram numerically by introducing ultraviolet regularization cutoff parameters in this work . more quantitative numerical estimates using lorentz invariant regularization methods , such as dimensional regularization , will be presented in future work @xcite . the vertex renormalization constant @xmath53 is defined as @xmath54 where the operator @xmath55 is given by the sum of vertex correction diagrams in fig . [ fig : loops](c)(g ) , and for convenience is defined with the charge factor @xmath56 taken out . we use the convention in which the nucleon spinors are normalized according to @xmath57 . for small values of @xmath53 , one also has , to lowest order in the @xmath58 coupling , @xmath59 . to evaluate the vertex renormalization constants from the various diagrams in fig . [ fig : loops ] , we take the @xmath60 components , so that @xmath61 where the @xmath62 components of the momentum four - vector are given by @xmath63 . for the coupling of the photon to a proton dressed by a neutral @xmath64 loop ( @xmath65 ) , fig . [ fig : loops](c ) , the vertex correction is computed from the operator @xmath66 where we use the shorthand notation for the pion and nucleon propagators , @xmath67 respectively . for a neutron target , the correction from the coupling to an intermediate state proton dressed by a negatively charged pion ( @xmath68 ) is given by @xmath69 . taking the @xmath70 component and using eq . ( [ eq : lam+ ] ) , the contribution to the vertex renormalization factor @xmath71 is then given by @xmath72 { 1 \over d_\pi d_n^2 } \nonumber\\ & = & -i \left ( { g_a \over 2 f_\pi } \right)^2 \int\!\!{d^4k \over ( 2\pi)^4 } \left [ { 4 m^2 ( k^2 - 2y\ , p\cdot k ) \over d_\pi d_n^2}\ -\ { 4 m^2 y \over d_\pi d_n}\ -\ { 1 \over d_\pi } \right ] , \label{eq : z1n}\end{aligned}\ ] ] where @xmath73 is the fraction of the nucleon s @xmath74 component of momentum carried by the pion . in deriving eq . ( [ eq : z1n ] ) we have used the dirac equation , @xmath75 . since there are two more powers of momentum @xmath76 in the numerator of @xmath77 than in the denominator , the integral ( [ eq : z1n ] ) is formally divergent . one can perform the loop integration and regularize the divergence in several ways . in ref . @xcite we considered the nucleon self - energy arising from pion dressing , and computed the loop integrals using equal - time perturbation theory in the rest frame and in the infinite momentum frame , using light - front coordinates ( all with appropriate high - momentum cutoffs ) , and covariantly with dimensional regularization . each method was shown to give identical results for the model - independent , nonanalytic part of the integrals , with the ( model - dependent ) analytic contributions dependent upon the regularization prescription . in the case of the vertex renormalization , it is convenient to use the @xmath74 prescription to evaluate @xmath53 , and it will be particularly instructive to examine the integrands as a function of @xmath78 ( or @xmath35 ) and @xmath79 . performing the @xmath80 integration using the cauchy integral theorem by closing the contour in the lower half - plane , one can write the nucleon contribution to the vertex renormalization as @xmath81 ^ 2 } \ -\ { y \over k_\perp^2 + y^2 m^2 + ( 1-y ) m_\pi^2 } \ \right . \nonumber\\ & & \hspace*{3.5 cm } \left . -\ { 1 \over 4m^2 } \log\left ( { k_\perp^2+m_\pi^2 \over \mu^2 } \right ) \delta(y ) \right\ } , \label{eq : z1ny}\end{aligned}\ ] ] where the mass @xmath82 is an ultraviolet cutoff on the @xmath80 integration . in eq . ( [ eq : z1ny ] ) the first , second and third terms correspond to the terms in the integrand of eq . ( [ eq : z1n ] ) proportional to @xmath83 , @xmath84 and @xmath85 , respectively . the first term , which is logarithmically divergent in @xmath79 , gives a contribution that is equivalent to that obtained from a ps pion nucleon coupling ( [ eq : lps ] ) @xcite , @xmath86 ^ 2 } , \label{eq : z1nps}\end{aligned}\ ] ] where the couplings are related as in eq . ( [ eq : gt ] ) , and we use the tilde ( `` @xmath87 '' ) notation for quantities computed from the ps interaction . this result agrees exactly with the result from the infinite momentum frame calculation of drell , levy and yan @xcite within the ps @xmath58 theory . from terms in eq . ( [ eq : z1n ] ) proportional to @xmath83 ( dashed ) , @xmath84 ( dot - dashed ) , @xmath85 ( dotted ) , and the sum ( solid ) , as a function of the @xmath79 momentum cutoff @xmath88.,width=377 ] the second term in eq . ( [ eq : z1ny ] ) is a new contribution , associated with the momentum dependence of the pv pion nucleon vertex , and enters with the opposite sign to the ps - like component . the third term is nonzero only at @xmath89 , and arises from the @xmath85 term in eq . ( [ eq : z1n ] ) . for @xmath90 this serves to enhance the contribution from the @xmath83 ps - like term . numerically , the contributions to @xmath77 from each of the three terms are illustrated in fig . [ fig : z1n ] as a function of the cutoff mass @xmath88 used to render the @xmath79 integration finite , taking the @xmath80 integration cutoff @xmath91 gev . the results show large cancellation between the @xmath83 term ( which gives a negative contribution to @xmath92 ) and the @xmath84 term ( which gives a positive contribution ) , with the total closely following the residual @xmath85 contribution , which is smaller in magnitude than the other two pieces . this clearly illustrates that a calculation of the vertex renormalization using the ps @xmath58 interaction , apart from not having the correct chiral symmetry properties , yields very different results phenomenologically compared with the pv theory . for the photon coupling to a positively charged pion emitted from the proton , fig . [ fig : loops](d ) , the corresponding operator can be written @xmath93 where the overall isospin factor 2 accounts for the @xmath94 transition . for a neutron target , the operator for the coupling to the negatively charged pion in the transition @xmath95 would be @xmath96 . taking the @xmath70 component in eq . ( [ eq : lmupi ] ) , the resulting vertex renormalization factor for the pion coupling @xmath97 can be written as @xmath98 { 2y \over d_\pi^2 d_n } \nonumber\\ & = & -i \left ( { g_a \over 2 f_\pi } \right)^2 \int\!\!{d^4k \over ( 2\pi)^4 } \left [ { 8y m^2\ , p\cdot k \over d_\pi^2 d_n}\ + \ { 2y\ , p \cdot k \over d_\pi^2 } + \ { 4y\ , m^2 \over d_\pi^2 } \right ] , \label{eq : z1pi}\end{aligned}\ ] ] where @xmath99 here is defined with the isospin factor removed . because @xmath35 is odd in @xmath78 , while @xmath100 is even , the third term in eq . ( [ eq : z1pi ] ) proportional to @xmath101 will vanish after integration over @xmath78 . for the second term , proportional to @xmath102 , we can use the identity @xmath103 in fact , since @xmath104 one has , for any integer @xmath105 , @xmath106 using the relation ( [ eq : identity ] ) , and performing the @xmath80 integration by closing the contour in the upper half - plane , the photon pion coupling contribution to @xmath53 can be written as @xmath107 ^ 2 } + \ { 1 \over 4m^2 } \log\left ( { k_\perp^2+m_\pi^2 \over \mu^2 } \right ) \delta(y ) \right\}. \nonumber\\ & & \label{eq : z1piy}\end{aligned}\ ] ] note that the first term in eq . ( [ eq : z1piy ] ) is identical to the first term in eq . ( [ eq : z1ny ] ) for the @xmath108 contribution to the vertex renormalization . it is also the result one obtains from the `` sullivan process '' @xcite for the contribution of the pion cloud to the deep - inelastic structure function of the nucleon , where the current couples to the pion cloud , leaving an on - shell nucleon in the final state . these analyses all utilized the ps pion nucleon interaction , in which the vertex renormalization factor is given by @xmath109 ^ 2 } . \label{eq : z1piyps}\end{aligned}\ ] ] this result also coincides with the vertex renormalization computed in the infinite momentum frame in ref . @xcite in terms of nucleon and pion `` partons '' in the ps theory . comparison of eqs . ( [ eq : z1piyps ] ) and ( [ eq : z1nps ] ) also demonstrates that the ps model respects the charge conservation condition , @xmath110 which follows directly from the ward - takahashi identity ( see appendix [ app : wti ] ) . from terms in eq . ( [ eq : z1piy ] ) proportional to @xmath84 ( dot - dashed ) , @xmath85 ( dotted ) , and the sum ( solid ) , as a function of the @xmath79 momentum cutoff @xmath88.,width=377 ] in contrast , the full pv result for @xmath99 in eq . ( [ eq : z1piy ] ) contains in addition a singular , @xmath111-function term in @xmath35 , just as in eq . ( [ eq : z1ny ] ) for @xmath108 but with the opposite sign . as illustrated in fig . [ fig : z1pi ] , this term cancels some of the contribution from the ps , @xmath83 term , leaving an overall positive contribution to @xmath112 . clearly the pv @xmath99 and @xmath108 results are different for any value of the @xmath79 cutoff @xmath88 , and in order to demonstrate their equivalence requires consideration of additional terms arising from the derivative pv coupling . the momentum dependence of the pv @xmath58 interaction gives rise to an additional kroll - ruderman ( kr ) term @xcite which describes the photon coupling to the @xmath4 vertex , fig . [ fig : loops](e ) . for the case of the @xmath94 vertex , these are computed from the operator @xmath113 for a neutron target , the corresponding operator describing the photon coupling to the @xmath95 vertex is @xmath114 . note that the emission of a neutron @xmath64 does not give rise to a kr correction term . the contribution to the vertex renormalization factor is then given by @xmath115 , where @xmath116 performing the @xmath80 integration , the resulting contribution to @xmath117 is @xmath118 where the first term in the integrand arises from the @xmath84 term , while the @xmath119 term is associated with the @xmath85 contribution in eq . ( [ eq : z1kr ] ) . from terms in eq . ( [ eq : z1kry ] ) proportional to @xmath84 ( dot - dashed ) , @xmath85 ( dotted ) , and the sum ( solid ) , as a function of the @xmath79 momentum cutoff @xmath88.,width=377 ] the contributions from the individual terms to @xmath120 are shown in fig . [ fig : z1kr ] as a function of the @xmath79 momentum cutoff @xmath88 , as well as the total kr correction . note the large cancellation between the @xmath85 and @xmath84 terms for values of the cutoff @xmath121 gev . omission of the @xmath111-function @xmath85 contribution would thus lead to a significant overestimate of the kr correction . from the photon nucleon coupling @xmath77 ( solid ) , photon pion coupling @xmath99 ( dashed ) , and kroll - ruderman terms @xmath120 ( dot - dashed ) , as a function of the @xmath79 momentum cutoff @xmath88 . note that the sum of the pion tadpole and bubble contributions to @xmath53 vanishes.,width=377 ] formally , the kroll - ruderman terms are needed to ensure gauge invariance in the pv theory . indeed , from eqs . ( [ eq : z1ny ] ) , ( [ eq : z1piy ] ) and ( [ eq : z1kry ] ) one can verify explicitly that @xmath122 for the pv case . in contrast , in the ps theory , where the @xmath58 vertex is independent of momentum , there is no analogous kr contribution , and gauge invariance is reflected through the relation ( [ eq : psgi ] ) . in fig . [ fig : z1 ] we show the total contributions to @xmath53 from the nucleon and pion rainbow diagrams , @xmath77 and @xmath99 , and the total kr correction . the sum of these three terms is of course zero by eq . ( [ eq : pvgi ] ) . at lowest order the lagrangian ( [ eq : lping ] ) contains , in addition to the pv coupling of the pion to the nucleon , quadratic terms arising from the covariant derivative . for the coupling of the photon to the @xmath123 vertex in the pion tadpole diagram in fig . [ fig : loops](f ) the relevant operator is @xmath124 for the coupling to the @xmath125 vertex , the corresponding operator is @xmath126 . the contribution to the vertex renormalization from the @xmath127 tadpoles is then @xmath128 , where @xmath129 after integration over @xmath80 , this can be written @xmath130 for the bubble diagram with the photon coupling directly to a pion , fig . [ fig : loops](g ) , the relevant operator for a proton target is @xmath131 with that for a neutron target given by @xmath132 . taking the @xmath70 component on both sides of eq . ( [ eq : lmupibub ] ) , the contribution to the vertex renormalization is given by @xmath133 , where @xmath134 using the identify ( [ eq : identity ] ) this can be written , after @xmath80 integration , as @xmath135 which is equal and opposite to the pion tadpole contribution in eq . ( [ eq : z1ntady ] ) . the vanishing of the sum of the pion tadpole and bubble contributions , @xmath136 ensures therefore that these have no net effect on the vertex renormalization . in fig . [ fig : z1tad ] the pion tadpole and bubble diagrams are illustrated for a @xmath79 momentum cutoff @xmath88 . momentum cutoff @xmath88.,width=377 ] the model - independent , nonanalytic ( na ) structure of the vertex renormalization factors can be studied by expanding @xmath117 in powers of @xmath0 . the terms in the expansion that are even powers of @xmath0 are analytic in the quark mass @xmath137 ( from the gell - mann oakes renner relation , @xmath138 for small @xmath0 ) , while odd powers of @xmath0 or logarithms of @xmath0 are nonanalytic in @xmath137 . the na terms reflect the long - range structure of chiral loops , and are exactly calculable in terms of low energy constants such as @xmath16 and @xmath20 , independent of the details of short - range physics . the na behavior of the nucleon rainbow contribution is given by @xmath139 \right . \nonumber\\ & & \hspace*{1.4 cm } -\ { m_\pi^2 \over 2 m^2 } \log m_\pi^2\ + \ { \pi \over 2 } { m_\pi^3 \over m^3}\ + \ { m_\pi^4 \over 4 m^4 } \log m_\pi^2 \hspace*{2 cm } \left [ { 1 \over d_\pi d_n}\ { \rm term } \right ] \nonumber\\ & & \hspace*{1.4 cm } \left . + \ { m_\pi^2 \over 4 m^2 } \log m_\pi^2 \right\ } \hspace*{6.4 cm } \left [ { 1 \over d_\pi}\ { \rm term } \right ] \nonumber\\ & = & { 3 g_a^2 \over 4 ( 4\pi f_\pi)^2 } \left\ { m_\pi^2 \log m_\pi^2\ -\ \pi { m_\pi^3 \over m}\ -\ { 2 m_\pi^4 \over 3 m^2 } \log m_\pi^2 + \ { \cal o}(m_\pi^5 ) \right\ } , \label{eq : z1nlna}\end{aligned}\ ] ] where for completeness we have included the first three lowest order na terms , and the origins of the various powers of @xmath0 have been indicated in the brackets to the right of the equations . comparison with the ps result ( the @xmath83 term in eq . ( [ eq : z1nlna ] ) ) shows that for the leading na ( lna ) term ( order @xmath51 ) one has @xmath140 this makes clear the origin of the difference between the results in refs . @xcite , which were obtained for a ps @xmath58 coupling , and refs . @xcite and @xcite , which were obtained for the pv theory . note that this result does not depend on the details of the ultraviolet regulator , since the nonanalytic structure is determined entirely by the infrared behavior of the integrals . the na behavior of the pion rainbow contribution is given , to order in @xmath0 , by @xmath141 \right . \nonumber\\ & & \hspace*{1.25 cm } \left . -\ { m_\pi^2 \over 4 m^2 } \log m_\pi^2 \right\ } \hspace*{6.5 cm } \left [ { 1 \over d_\pi}\ { \rm term } \right ] \nonumber\\ & = & { 3 g_a^2 \over 4 ( 4\pi f_\pi)^2 } \left\ { m_\pi^2 \log m_\pi^2\ -\ { 5\pi \over 3 } { m_\pi^3 \over m}\ -\ { m_\pi^4 \over m^2 } \log m_\pi^2 + \ { \cal o}(m_\pi^5 ) \right\}. \label{eq : z1pilna}\end{aligned}\ ] ] note that the behavior arising from the @xmath142 term is identical to that from the @xmath83 term in @xmath77 , which reflects the gauge invariance of the ps theory , eq . ( [ eq : psgi ] ) . for the pv theory , the kroll - ruderman terms has the nonanalytic behavior @xmath143 \right . \nonumber\\ & & \hspace*{1.8 cm } \left . + \ { m_\pi^2 \over 2 m^2 } \log m_\pi^2 \right\ } \hspace*{5.7 cm } \left [ { 1 \over d_\pi}\ { \rm term } \right ] \nonumber\\ & = & { 3 g_a^2 \over 4 ( 4\pi f_\pi)^2 } \left\ { { 2\pi \over 3 } { m_\pi^3 \over m}\ -\ { m_\pi^4 \over 3 m^2 } \log m_\pi^2 + \ { \cal o}(m_\pi^5 ) \right\}. \label{eq : z1krlna}\end{aligned}\ ] ] here the @xmath144 terms cancel between the @xmath84 and @xmath85 terms , so that the leading na behavior of the kr term @xmath145 . for the sum of the @xmath99 and @xmath120 terms , @xmath146 the na behavior is therefore explicitly verified to be equivalent to that in @xmath147 in eq . ( [ eq : z1nlna ] ) . .leading nonanalytic contributions to the vertex renormalization @xmath117 , in units of @xmath148 . the asterisks @xmath149 in the @xmath77 and @xmath99 rows denote contributions that are present for the pseudoscalar @xmath58 coupling . note that the nonanalytic contributions from the kr terms cancel at this order , but are nonzero at @xmath150 , and are needed to ensure gauge invariance of the theory , eq . ( [ eq : pvgi ] ) . the sum of all contributions is given in the last two columns for the pv and ps theories , respectively . [ cols="<,^,^,^,^,^,^",options="header " , ] [ tab : budget ] finally , for the pion loop contributions from sec . [ ssec : z1tad ] , the na behavior is given by the @xmath51 term for both the pion tadpole ( fig . [ fig : loops](f ) ) and bubble ( fig . [ fig : loops](g ) ) diagrams , @xmath151 the results for the leading order nonanalytic contributions to the vertex renormalization factors are summarized in table [ tab : budget ] for each of the above terms , where the entries are given in units of @xmath148 . note that the contributions from the terms associated with @xmath83 and @xmath142 ( denoted by asterisks @xmath152 ) are the same as those in the ps theory , while the other contributions arise only for pv coupling . the table clearly illustrates the origin of the difference between the @xmath92 corrections in the pv and ps theories , and in particular the relative factor 3/4 found in heavy baryon chiral perturbation theory @xcite compared with calculations based on the sullivan process with ps coupling @xcite . the above results on the vertex renormalization factors can be used to compute the na behavior of moments of parton distribution functions ( pdfs ) arising from the pion cloud of the nucleon . the presence of the pion cloud induces corrections to the pdfs of a bare nucleon , whose bjorken @xmath34 dependence can be represented in terms of convolutions of pion and nucleon light - cone distribution functions @xmath153 and the corresponding parton distributions in the pion and nucleon @xcite . the light - cone distribution @xmath153 are defined such that when integrated over @xmath35 they give the appropriate vertex renormalization factors @xmath154 , @xmath155 . unlike in the ps coupling models @xcite , the dressed nucleon pdfs in the pv theory contain several additional terms @xcite , @xmath156 \otimes q_0 \right)(x ) + \ \left ( [ f_\pi + f_{\pi\ , \rm bub } ] \otimes q_\pi \right)(x ) + \ \left ( f_{\rm kr } \otimes q_{\rm kr } \right)(x ) , \nonumber\\ & & \label{eq : conv}\end{aligned}\ ] ] where @xmath157 is the bare nucleon pdf ( arising from the diagram in fig . [ fig : loops](a ) ) , @xmath158 is the pdf in the pion , and @xmath159 , with @xmath160 the spin - dependent pdf in the bare nucleon . the constant @xmath161 is the wave function renormalization constant , and the symbol @xmath162 represents the convolution integral @xmath163 , where for the @xmath20 and @xmath164 contributions the integration variable @xmath165 should be taken to be the fraction of the nucleon s @xmath74 component of momentum carried by the pion , @xmath73 , while for the @xmath166 , @xmath167 and @xmath168 terms @xmath169 . note that because of the additional pion field at the vertex in the kr diagram , fig . [ fig : loops](e ) , the contribution of the kr terms to the ( unpolarized ) pdf involves a convolution with a spin - dependent parton distribution . in the corresponding `` sullivan '' process based on the ps coupling @xcite , only the @xmath166 and @xmath20 functions contribute , and these are related by eq . ( [ eq : psgi ] ) . according to the operator product expansion in qcd , the moments of pdfs are related to matrix elements of local operators , @xmath170 where the braces @xmath171 denote symmetrization of lorentz indices , and the operators are given by the quark bilinears @xmath172 with @xmath173 the covariant derivative . the @xmath105-th moment of the pdf @xmath174 is given by @xmath175 for @xmath176 , we define @xmath177 to be the moment in the proton , with @xmath178 for the neutron @xmath179 . for the direct coupling of the photon to the nucleon , which includes the wave function renormalization ( fig . [ fig : loops](b ) ) , the nucleon rainbow diagram ( fig . [ fig : loops](c ) ) , and the pion tadpole diagram ( fig . [ fig : loops](f ) ) , the contributions to the twist-2 matrix elements are given by @xmath180 for the proton and neutron , respectively . the wave function renormalization @xmath161 factor in eq . ( [ eq : mnp ] ) is @xmath181 the contributions from the photon pion couplings , including the pion rainbow diagram ( fig . [ fig : loops](d ) ) , the kroll - ruderman term ( fig . [ fig : loops](e ) ) , and the pion bubble diagram ( fig . [ fig : loops](g ) ) , are given for the proton and neutron by @xmath182 using eq . ( [ eq : z2 ] ) , the pion cloud contributions to the isoscalar ( sum of proton and neutron ) moments from coupling involving nucleons cancel , @xmath183 leaving the charge of the nucleon ( or valence quark number ) unrenormalized from that given by the bare coupling , fig . [ fig : loops](a ) . similarly , the contributions to the isoscalar moments involving direct coupling to pions add to zero , as required by charge conservation , @xmath184 note that these results , eqs . ( [ eq : mn_is ] ) and ( [ eq : mpi_is ] ) , are true to all orders in the pion mass , not just for the lna parts that were discussed in refs . @xcite , and to which we turn to in the next section . the lna contributions from the nucleon coupling diagrams to the proton and neutron moments are given by @xmath185 taking the difference between the proton and neutron moments , the isovector contribution then becomes @xmath186 which agrees with the results obtained in heavy baryon chiral perturbation theory @xcite . similarly , the lna contributions from the pion coupling diagrams to the proton and neutron moments are given by @xmath187 so that the isovector contribution can be written @xmath188 the pion coupling contributions to the moment therefore cancel those of the nucleon coupling in eq . ( [ eq : mnlna ] ) , such that the total lowest moment of the pdf is not affected by pion loop corrections . the analysis is more straightforward for the ps theory , where neither tadpoles , bubbles nor kr terms are present , and lna behavior of the nucleon and pion coupling contributions to the isovector moments is given by @xmath189 this agrees with the results obtained in ref . @xcite using the light - cone momentum distributions computed for the sullivan process in the ps theory @xcite . as observed in ref . @xcite , the pv and ps results agree in the limit as @xmath190 , although they clearly differ in the general case for @xmath191 . for higher moments , @xmath192 , the pion coupling contributions will be suppressed by additional powers of @xmath193 , while the lna behavior of the nucleon coupling diagrams remains @xmath194 @xcite . cancellation will therefore not occur between the nucleon and pion coupling contributions for these moments , so that the shape of underlying pdfs will in general be modified by the presence of pion loops . we have verified that our lna results agree with those presented in ref . further details of the pion light - cone momentum distribution @xmath153 are discussed in ref . @xcite , and an analysis of their phenomenological consequences will be presented in a forthcoming publication @xcite . in this work we have presented a detailed analysis of pion cloud corrections to the electromagnetic coupling of the nucleon , using the lowest order effective lagrangian constrained by the chiral symmetry of qcd . we have computed the complete set of vertex corrections arising from the various one - loop diagrams , including rainbow diagrams with pion and nucleon coupling , kroll - ruderman contributions , and tadpole and bubble diagrams associated with @xmath58 contact interactions . explicit evaluation of the vertex renormalization factors allowed us to directly verify relations between the nucleon and pion coupling diagrams , and demonstrate the consistency of the theory with electromagnetic gauge invariance . the kr terms in particular are essential for ensuring gauge invariance to all orders in the pion mass , even though these do not contribute to the leading nonanalytic behavior of the vertex factors . we have also shown that the sum of the pion tadpole and bubble diagrams vanishes . we have examined the chiral expansion of all the vertex corrections as a function of the pion mass @xmath0 , computing the coefficients of the nonanalytic terms up to and including order @xmath195 . the lna terms agree with earlier calculations in heavy baryon chiral perturbation theory @xcite , although our formulation is relativistic and allows for higher order corrections in @xmath196 . comparison of the results for the pseudoscalar @xmath58 theory reveals the origin of the long - standing discrepancy between the lna behavior in the chiral effective theory and in approaches based on the sullivan process @xcite which use a @xmath197 coupling . to study the behavior of the total vertex corrections , rather than just their longest - range lna contributions , we have computed the vertex renormalization factors numerically as a function of the transverse momentum cutoff used to regularize the integrals . the pion and nucleon rainbow corrections give positive contributions to the vertex renormalization factor @xmath198 for the range of cutoffs considered here ( @xmath199 gev ) , while the contribution from the kr diagram is negative . the overall magnitude of the vertex correction is @xmath200 for @xmath201 gev and @xmath202 for @xmath203 gev . the tadpole and bubble contributions range up to @xmath204 for @xmath203 gev . although a transverse momentum cutoff breaks the lorentz invariance of the @xmath58 theory , for the purposes of the present study it is sufficient to illustrate the relative contributions of the various pion loop diagrams . for a more quantitative analysis , for example , of the corrections to the @xmath7 pdf difference , a covariant regularization scheme can be used @xcite . finally , using the results for the vertex and wave function renormalization constants we computed the pion loop corrections to the matrix elements of twist-2 operators for the proton and neutron , which in the operator product expansion are related to moments of parton distribution functions . for the lowest moment , we demonstrated explicitly that the pion loop corrections cancel for the isoscalar combination of moments for the nucleon and pion couplings separately . the isovector moments , on the other hand , were found to have the characteristic @xmath51 leading dependence for both the nucleon and pion coupling diagrams ( with the sum of course canceling , as required by charge conservation ) . again , comparison of the pv and ps results for the moments enabled us to clearly identify the source of the difference between the coefficients of the lna terms in the two theories . while the pv theory is clearly preferred by considerations of chiral symmetry , the explicit demonstration that the ps theory can be made consistent in this context with the introduction of a scalar @xmath1 field remains an interesting challenge . the results derived here can be used in the future to investigate the nonanalytic behavior of the nucleon pdfs , particularly the extrapolation of calculations in lattice qcd performed at unphysically large quark masses @xcite to the physical region . our findings will also pave the way for phenomenological studies , especially the quest for a consistent interpretation of the physics of the pion cloud at the parton level , enabling deeper studies of the origin of the @xmath7 asymmetry @xcite . in addition , this work will also guide investigation of the very important asymmetry between the @xmath205 and @xmath206 distributions , with its connection to the five - quark component of the nucleon wave function , as well as the spin - flavor asymmetry @xmath207 between the polarized @xmath6 and @xmath5 distributions . we thank m. birse , m. burkardt , k. hendricks , v. lyubovitskij , m. polyakov , and a. a. vladimirov for helpful discussions . this work was supported by the doe contract no . de - ac05 - 06or23177 , under which jefferson science associates , llc operates jefferson lab , doe contract no . de - fg02 - 03er41260 , and the australian research council through the arc centre of excellence for particle physics at the terascale and grant fl0992247 . for convenience we summarize in this appendix the complete set of feynman rules derived from the pv lagrangian @xmath208 in eq . ( [ eq : lping ] ) , needed to compute the vertex renormalization and wave function corrections from pion loops . the conventions throughout this work denote the nucleon momentum by @xmath209 and the pion momentum by @xmath210 , with @xmath211 the electric charge on the electron . isospin couplings that are not listed are identically zero . nucleon : @xmath212 pion : @xmath213 @xmath214 : @xmath215 @xmath216 : @xmath217 @xmath218 : @xmath219 @xmath220 : @xmath221 @xmath222 : @xmath223 @xmath224 : @xmath225 @xmath226 : @xmath227 @xmath228 : @xmath229 @xmath230 : @xmath231 @xmath228 : @xmath232 in this section we demonstrate explicitly that the electromagnetic coupling to the nucleon dressed by pions , illustrated in fig . [ fig : gi ] , is gauge invariant , for both the pv and ps pion nucleon theories . ) dressed by a pion ( with momentum @xmath76 ) : ( a ) and ( b ) wave function renormalization diagrams , ( c ) rainbow diagram with coupling to the proton , ( d ) rainbow diagram with coupling to the @xmath233 , ( e ) and ( f ) kroll - ruderman diagrams , ( g ) tadpole diagram with coupling to the @xmath123 vertex , ( h ) bubble diagram with coupling to the pion . the current brings in a momentum @xmath234.,width=377 ] for the @xmath235 coupling the current derived from the pv lagrangian in eq . ( [ eq : lping ] ) is a sum of the three contributions in fig . [ fig : gi](a ) , ( b ) and ( c ) , @xmath236 , \label{eq : jpi0}\end{aligned}\ ] ] where [ eq : jwf ] @xmath237 represent the `` wave function renormalization '' diagrams in fig . [ fig : gi](a ) and ( b ) , and @xmath238 corresponds to the rainbow diagram with coupling to the proton in fig . [ fig : gi](c ) . here the electromagnetic current operator brings in a finite momentum @xmath234 to the proton . contracting the currents with the photon four - vector @xmath239 and using the dirac equation @xmath240 , one finds [ eq : qjpi0 ] @xmath241 and @xmath242 where we have used @xmath243 and @xmath244 to simplify eqs . ( [ eq : qjwfr ] ) and ( [ eq : qjn ] ) , respectively . the sum of the three contributions then gives the required result , @xmath245 for the @xmath246 coupling the current has seven contributions , including the wave function renormalization diagrams in fig . [ fig : gi](a ) and ( b ) , the @xmath233 rainbow diagram in fig . [ fig : gi](d ) , the kroll - ruderman contributions in fig . [ fig : gi](e ) and ( f ) , and the @xmath233 tadpoles and bubbles in fig . [ fig : gi](g ) and ( h ) , @xmath247 \right . \nonumber\\ & & \hspace*{1.33 cm } \left . + \ { 1 \over 4 f_\pi^2 } \big [ { \cal j}_{n\ , \rm tad}^\mu + { \cal j}_{\pi\ , \rm bub}^\mu \big ] \right\}. \label{eq : jpi+tot}\end{aligned}\ ] ] the current for the @xmath233 rainbow is given by @xmath248 while the kroll - ruderman currents are [ eq : jpikr ] @xmath249 finally , for the pion tadpole and bubble diagrams the two currents are given by [ eq : jtad ] @xmath250 respectively . note that diagrams involving a pion tadpole with the photon coupling to a proton in the initial or final state directly involve loop integrations with an odd number of factors @xmath76 in the integrand ( see eq . ( [ eq : pip - pip ] ) ) and therefore vanish identically . contracting the @xmath233 currents with the photon momentum @xmath239 , one has [ eq : qjpi ] @xmath251 adding the wave function renormalization contributions in eqs . ( [ eq : qjwfl ] ) and ( [ eq : qjwfr ] ) , one can verify , after some tedious but straightforward manipulations , that @xmath252 note that the inclusion of the kr contributions is vital to cancel the contributions from the @xmath233 current with the pv coupling , without which the theory would not be gauge invariant . the tadpole and bubble contributions are each independently gauge invariant , as can be seen from the contractions [ eq : qjtad ] @xmath253 where the first expression can be verified by changing variables @xmath254 , and the second vanishes because of the dirac equation . the same of course holds true also for a neutron initial state . for the ps coupling , only the diagrams in fig . [ fig : gi](a)(d ) are present . as in the pv case , for the @xmath235 coupling the current in the ps theory has three contributions from fig . [ fig : gi](a ) , ( b ) and ( c ) , @xmath255 , \label{eq : jpi0ps}\end{aligned}\ ] ] where [ eq : jwfps ] @xmath256 contracting the currents ( [ eq : jwfps ] ) with @xmath239 and again using the dirac equation , one finds [ eq : qjpi0ps ] @xmath257 from eqs . ( [ eq : qjpi0ps ] ) it is evident that @xmath258 so that gauge invariance is satisfied explicitly for the proton dressed by a neutral @xmath64 with ps coupling , @xmath259 similarly for the ps @xmath246 coupling , the current is given by the three contributions in fig . [ fig : gi](a ) , ( b ) and ( d ) , @xmath260 , \label{eq : jpi+totps}\end{aligned}\ ] ] where @xmath261 the contraction of @xmath239 with @xmath262 gives @xmath263 where the first term in the parentheses cancels with @xmath264 in eq . ( [ eq : qjwfrps ] ) , and the second cancels with @xmath265 in ( [ eq : qjwflps ] ) after changing variables @xmath254 . therefore gauge invariance is explicitly verified also for the @xmath233 with ps coupling , @xmath266 note that since these results are obtained at the operator level , they are independent of the particular renormaliation scheme chosen to regulate the integrals . in this section we demonstrate the consistency of the vertex corrections and wave function renormalization with the ward - takahashi identity . to begin with , we consider the nucleon self - energy operator , which can in general be written in terms of the vector and scalar components as @xmath267 evaluating the matrix element of the self - energy operator between on - shell nucleon states gives @xmath268 where the sum is taken over the spin polarizations @xmath269 of the nucleon . the self - energy modifies the pole of the nucleon propagator according to @xmath270 where @xmath161 is the wave function renormalization constant , @xmath271 and the mass shift @xmath272 is given by @xmath273 alternatively , using eqs . ( [ eq : z2form ] ) and ( [ eq : deltam ] ) one can express the vector and scalar components of the self - energy as @xmath274 from eqs . ( [ eq : sigma_def ] ) and ( [ eq : sigma_v ] ) one observes that the nucleon wave function renormalization @xmath161 is related to the derivative of the nucleon self - energy operator by @xmath275 on the other hand , from eq . ( [ eq : z1def ] ) in sec . [ sec : z1 ] the vertex renormalization constant @xmath53 is defined in terms of the matrix element of the vertex correction @xmath55 by @xmath276 the ward - takahashi identity relates the vertex operator @xmath55 to the @xmath277 derivative of the self - energy operator @xmath278 , which can be expressed as the equality of the vertex and wave function renormalization factors , @xmath279 . to demonstrate that this relation is explicitly satisfied by the pv pion - nucleon theory defined by eqs . ( [ eq : lpin ] ) and ( [ eq : lping ] ) , recall that the self - energy for a nucleon dressed by a pion loop is given by @xcite @xmath280 where @xmath281 and @xmath282 . differentiating the nucleon propagator with respect to the nucleon momentum , which is equivalent to the insertion of a zero energy photon , @xmath283 leads to @xmath284 comparing the right - hand side of ( [ eq : dsigma_dp1 ] ) with the expression for @xmath55 in eq . ( [ eq : lamn ] ) , one can then identify @xmath285 now , changing variables @xmath286 in eq . ( [ eq : sigma ] ) enables the self - energy to be written equivalently as @xmath287 which when differentiated with respect to @xmath277 gives rise to three terms , @xmath288 changing variables @xmath286 once again then gives @xmath289 since the expressions for @xmath290 in eqs . ( [ eq : sigma ] ) and ( [ eq : sigma_pi ] ) are equivalent , this implies that the electromagnetic operators are related by @xmath291 taking matrix elements of both sides of eq . ( [ eq : lam_wti ] ) between proton states , one arrives at the relation @xmath292 or , in terms of the `` isoscalar '' vertex factors defined in sec . [ sec : z1 ] , the desired expression @xmath293 as in eq . ( [ eq : pvgi ] ) . the proof for the ps case follows similarly , and is in fact more straightforward since the kr terms are absent in this case . one can also check the validity of the ward - takahashi identity by performing the integrations explicitly , and comparing the integrated expressions for the vertex renormalization factors . in this appendix we summarize some useful results for the integrals appearing in the expressions for the vertex corrections in sec . [ sec : z1 ] and the self - energies appendix [ app : wti ] , using dimensional regularization to render ultraviolet divergent integrals finite . expanding the results in powers of @xmath0 , we also provide explicit expressions for their nonanalytic behavior in the chiral limit . for the integrals involving only the pion propagator ( @xmath85 ) , @xmath294 \nonumber\\ & \stackrel{\rm na}{\longrightarrow } & -i \pi^2 m_\pi^2 \log m_\pi^2 , \label{eq : dpi } \\ & & \nonumber\\ \int d^4k\ { 1 \over d_\pi^2 } & = & i \pi^2 \left [ { 1 \over { \varepsilon } } - \gamma - \log\pi - \log{m_\pi^2 \over \mu^2 } + { \cal o}({\varepsilon } ) \right ] \nonumber\\ & \stackrel{\rm na}{\longrightarrow } & -i \pi^2 \log m_\pi^2 . \label{eq : dpi2}\end{aligned}\ ] ] the expressions in eqs . ( [ eq : dpi ] ) and ( [ eq : dpi2 ] ) can also be related using @xmath295 note that there are no higher order contributions in @xmath0 for either of these integrals . the corresponding integrals involving only the nucleon propagator ( @xmath296 ) , on the other hand , do not have nonanalytic contributions , @xmath297 \nonumber\\ & \stackrel{\rm na}{\longrightarrow } & 0 , \\ & & \nonumber\\ \int d^4k\ { 1 \over d_n^2 } & = & i \pi^2 \left [ { 1 \over { \varepsilon } } - \gamma - \log\pi - \log{m^2 \over \mu^2 } + { \cal o}({\varepsilon } ) \right ] \nonumber\\ & \stackrel{\rm na}{\longrightarrow } & 0.\end{aligned}\ ] ] for the integral of one pion propagator and one nucleon propagator , one has @xmath298 \nonumber\\ & \stackrel{\rm na}{\longrightarrow } & -{i \pi^2 \over 2m^2 } \left [ m_\pi^2 \log m_\pi^2 + 2 \pi m m_\pi + { \cal o}(m_\pi^3 ) \right ] , \label{eq : dpidn}\end{aligned}\ ] ] where @xmath299 . note that this term is responsible for the leading @xmath150 behavior of the nucleon self - energy @xcite . finally , for integrals involving one pion and two nucleon propagators , or one nucleon and two pion propagators , the additional powers of the loop momentum render the results finite , @xmath300 \nonumber\\ & \stackrel{\rm na}{\longrightarrow } & - { i \pi^2 \over 2m^2 } \left [ \log m_\pi^2 + { \pi m \over m_\pi } + { \cal o}(m_\pi ) \right ] , \label{eq : dpi2dn } \\ & & \nonumber\\ \int d^4k\ { 1 \over d_\pi d_n^2 } & = & i \pi^2 { 1 \over m^2 } \left [ { 1 \over 2 } \log{m_\pi^2 \over m^2 } - { m_\pi \over r } \left ( \tan^{-1}{m_\pi \over r } + \tan^{-1}{2m^2-m_\pi^2 \over m_\pi r } \right ) \right ] \nonumber\\ & \stackrel{\rm na}{\longrightarrow } & { i \pi^2 \over 2m^2 } \left [ \log m_\pi^2 - { \pi m_\pi \over 2 m } + { \cal o}(m_\pi^3 ) \right ] . \label{eq : dpidn2}\end{aligned}\ ] ] the expressions in eqs . ( [ eq : dpidn ] ) and ( [ eq : dpi2dn ] ) can also be related using @xmath301

let be a smooth riemannian manifold which is the union of a compact part and a finite number of euclidean ends , for some , each of which carries the standard metric . our main result is that the riesz transform on is bounded from for and unbounded for if there is more than one end . it follows from known results that in such a case the riesz transform on is bounded for and unbounded for ; the result is new for . we also give some heat kernel estimates on such manifolds . we then consider the implications of boundedness of the riesz transform in for some for a more general class of manifolds . assume that is a-dimensional complete manifold satisfying the nash inequality and with an upper bound on the volume growth of geodesic balls . we show that boundedness of the riesz transform on for some implies a hodge - de rham interpretation of the cohomology in degree , and that the map from to cohomology in this degree is injective .

math0411648

let @xmath0 be a complete riemannian manifold with infinite measure . the riesz transform @xmath15 on @xmath0 is the operator @xmath16 where @xmath17 is the positive laplace operator on @xmath0 . the riesz transform is always a bounded map from @xmath18 to @xmath19 . it is of interest to figure out the range of @xmath20 for which @xmath15 extends to a bounded map @xmath3 . equivalently , we can ask whether @xmath21it has been shown in @xcite that the riesz transform is bounded on @xmath9 for @xmath22 as soon as the manifold satisfies the doubling property as well as a natural heat kernel upper bound . the situation for @xmath10 is more complicated : there is some understanding of what happens in the more restricted class of manifolds satisfying upper and lower gaussian estimates for the heat kernel ( see @xcite , @xcite ) , and it is also known that the above more general assumptions do not imply the boundedness of the riesz transform for all @xmath23 . the counterexample is simply the connected sum of two copies of @xmath24 , where one easily sees that the riesz transform is unbounded for @xmath25 ( see @xcite ) , precisely because the boundedness for such @xmath20 would imply a lower gaussian heat kernel estimate on this manifold , which is known to be false . the first aim of the present article is to find out what happens in the remaining range @xmath26 , and to treat more generally by the same token manifolds with a finite number of euclidean ends . our second aim is to give , in a somewhat more general class of manifolds , a cohomological consequence of the boundedness of the riesz transform for some @xmath23 , which explains the unboundedness in the range @xmath27 in the examples just mentioned . on @xmath24 , it is a classical result in harmonic analysis that the riesz transform is bounded on @xmath9 for all @xmath28 . in this case , the kernel of the riesz transform is given by @xmath29 ( we shall use @xmath30 for a euclidean coordinate throughout this paper , while the prime denotes the ` right variable ' of a kernel @xmath31 ; also , @xmath32 denotes @xmath33 and we write @xmath34 for @xmath35 . ) it is useful to compactify @xmath24 to a compact manifold with boundary @xmath36 by adding a sphere at infinity , and using @xmath37 as a boundary defining function and @xmath38 . ] for it . if we consider this kernel of @xmath39 near the ` right boundary ' of @xmath40 , namely @xmath41 , where @xmath42 , we see that it has an expansion given by @xmath43 the leading power in this expansion is @xmath44 and the coefficient multiplying it is @xmath13 . this kernel , by itself , is not bounded on @xmath9 for @xmath5 because the decay of the kernel puts it only in @xmath45 , @xmath46 , as a function of @xmath47 with @xmath30 fixed , so it can only be boundedly paired with elements in @xmath9 for @xmath48 . however , when we apply a @xmath30-derivative , the leading term is killed ( since @xmath49 ) and the kernel of @xmath15 decays at one order better , namely @xmath50 . this allows pairing with elements of @xmath9 for any @xmath51 . let @xmath0 be a manifold with euclidean ends , and assume the number of ends is at least two . it is now relatively easy to explain why the riesz transform on @xmath0 is not bounded for @xmath5 . for simplicity we shall assume here that @xmath0 has exactly two euclidean ends . let us compactify @xmath0 to a compact manifold @xmath52 in the analogous way to @xmath36 above . the boundary of @xmath52 is then the disjoint union of two @xmath53-spheres , which we shall denote @xmath54 and @xmath55 . it turns out that the kernel of @xmath39 has a similar expansion at @xmath56 , of the form @xmath57 but it is no longer true that the leading term @xmath58 . rather , @xmath59 is the harmonic function on @xmath52 that equals @xmath13 on @xmath54 , and zero on @xmath55 . as a consequence , applying the derivative operator in the left variable @xmath30 does not make this leading term disappear , since @xmath60 . hence the kernel of @xmath15 only decays to order @xmath53 at the right boundary of @xmath61 , and therefore can only be paired boundedly with elements of @xmath9 for @xmath48 . in this paper we shall prove [ main ] let @xmath0 be a complete @xmath62 riemannian manifold of dimension @xmath63 which is the union of a compact part and a finite number of euclidean ends . then the riesz transform is bounded from @xmath64 to @xmath65 for @xmath4 , and is unbounded on @xmath9 for all other values of @xmath20 if the number of ends is at least two . our method is to analyze the kernel @xmath39 based on the formula @xmath66 since @xmath0 is a manifold which is conic at infinity , the laplacian @xmath17 on @xmath0 lies in the class of scattering differential operators @xcite and we can use methods from the scattering calculus to analyze the kernel of @xmath39 . we shall analyze the kernel of @xmath39 rather precisely and work out the leading term in the expansion at the right boundary . from this , it will be straightforward to analyze the kernel of @xmath15 and to prove the theorem . the plan of the paper is as follows . we briefly describe the scattering calculus in section [ scatt ] . in sections [ para ] [ rt ] we prove the theorem using the analysis of the resolvent of the laplacian on asymptotically conic spaces in @xcite as a model . we give some large time asymptotics on derivatives of the heat kernel on manifolds with euclidean ends in section [ heat ] . in section [ coh ] we change point of view and consider a much more general class of manifolds , namely complete manifolds @xmath0 of dimension @xmath11 satisfying the nash inequality and with a uniform upper bound @xmath12 on the volume of geodesic balls of radius @xmath67 . we _ assume _ that the riesz transform on @xmath0 is bounded on @xmath9 for some @xmath10 and give several geometric and topological consequences : a hodge - de rham interpretation of the @xmath9 cohomology of @xmath0 ( proposition [ hdr ] ) , injectivity of the map from @xmath14 to @xmath9 cohomology ( lemma [ inj ] ) and derive a contradiction if @xmath7 and @xmath0 has at least two ends ( corollary [ nb ] ) , thus generalizing the unboundedness part of theorem [ main ] for @xmath7 to this larger class of manifolds . in the final section we discuss our results in the context of previously known examples and pose some open problems . as noted above , we shall use the scattering calculus @xcite to analyze the kernel of @xmath39 on manifolds with several euclidean ends . the scattering calculus is expressed in terms of compactifications on @xmath0 and , especially , of the double space @xmath68 which carries the kernel of the resolvent and of the operator @xmath39 . the space @xmath0 is compactified by adding a sphere at infinity @xmath69 for each euclidean end , and declaring @xmath70 and @xmath71 to be local coordinates near a boundary point ; in particular , @xmath72 is taken to be a defining function for the boundary ( which we shall sometimes refer to as ` infinity ' ) . we sometimes use @xmath73 to denote this boundary defining function and @xmath74 to denote boundary coordinates , extending to a collar neighbourhood of the boundary , as is customary when using the scattering calculus . the metric takes the form @xmath75 at each end , and is therefore a _ scattering metric _ as defined in @xcite ( of a particularly simple form , being an exact conic metric near infinity ) . we are mostly interested in the case when the number of ends is at least two . in fact , for the sake of clear exposition we shall assume from now on that the number of ends is exactly two , although all proofs in this paper generalize in an obvious way to any finite number of ends . we shall label these ends @xmath76 and @xmath77 , thus for example we shall use @xmath78 as the euclidean variable on the positive end , and @xmath79 for the variable on the negative end ; when it is not necessary to stipulate which end is being considered , we shall just use @xmath30 . it is not so obvious which compactification of @xmath68 is most appropriate for dealing with the schwartz kernels of operators such as the laplacian , or functions of the laplacian , on manifolds with euclidean ends . there are several different asymptotic regimes of interest when dealing with such kernels . one regime , the ` near - diagonal ' regime , is when the two variables @xmath80 remain a finite distance apart as they both go to infinity . another is when they both go to infinity with the ratio @xmath81 approaching a limit and with @xmath82 both approaching a limit . finally there is the case that one variable approaches a limit , while the other remains fixed . the kernel has different behaviour in each of these asymptotic regimes , so they need to be represented by distinct parts of the boundary of the compactification . the space @xmath83 is thus too ` small ' a compactification of @xmath68 for our purposes , since it only has the third regime distinguished ; the first two are squashed into the corner . it turns out that there is a space denoted @xmath84 , the ` scattering double space ' , which satisfies these criteria . it is obtained by performing two blowups on @xmath68 . the first is blowing up the corner @xmath85 , creating the so - called b - double space , and the second is blowing up the boundary of the diagonal ( which lifts to the b - double space to be transverse to the boundary , hence this blowup is well - defined ) . each asymptotic regime is represented by a boundary hypersurface of @xmath84 . the first , ` near - diagonal ' regime is represented by the boundary hypersuface created by the second blowup , denoted @xmath86 for ` scattering face ' ; the second is represented by the boundary hypersurface created by the first blowup , denoted @xmath87 for ` b - face ' ( since it is present in the b - calculus ) and the third regime is represented by the two boundary hypersurfaces @xmath88 , @xmath56 of @xmath83 , denoted @xmath89 and @xmath90 for ` left boundary ' a nd ` right boundary ' . note that when @xmath0 has @xmath91 ends , then @xmath87 has @xmath92 components and @xmath86 , @xmath89 and @xmath90 each have @xmath91 components . the structure of @xmath86 and @xmath87 is as follows . each component of @xmath86 is naturally diffeomorphic to @xmath93 , and @xmath94 and @xmath32 are coordinates on the interior of each component of @xmath86 . each component of @xmath87 is naturally isomorphic to a blowup of the space @xmath95 $ ] , with coordinates @xmath96 ; the blowup is of the submanifold @xmath97 which corresponds to the boundary of the diagonal . the _ scattering calculus _ is an algebra of pseudodifferential operators on @xmath0 which is defined by the properties of their schwartz kernels . namely , @xmath98 is a scattering psuedodifferential operator of order @xmath99 on @xmath0 iff the kernel of @xmath98 , when lifted to @xmath84 , is conormal has a singularity at the diagonal characteristic of pseudodifferential operators of order @xmath100 , and this holds smoothly up to the boundary , in the sense that it could be extended across the boundary as a conormal distribution . see @xcite , section 18.2 , for the precise definition . ] of order @xmath100 at the diagonal of @xmath84 smoothly up to the boundary @xmath86 , is smooth elsewhere at @xmath86 , and is rapidly decreasing at @xmath87 , @xmath89 and @xmath90 . the resolvent of the laplacian @xmath101 is a scattering pseudodifferential operator of order @xmath102 on @xmath0 for @xmath103 . in fact , the structure of the resolvent on the spectrum , i.e. the kernel of @xmath104 for real @xmath105 , can also be described on @xmath84 , although here the kernel is no longer rapidly decreasing at @xmath87 , @xmath89 and @xmath90 , rather it is a ` legendrian distribution ' @xcite . our approach is partly modelled on the analysis in this paper . however , we take advantage of the assumption here that @xmath0 has exact euclidean ends , which leads to great simplifications over the analysis of @xcite since we can exploit the well - known explicit formulae for the resolvent of the laplacian on @xmath24 and use these as ingredients for a parametrix of the resolvent kernel on @xmath0 , thereby avoiding the need to use legendrian distributions in this paper . _ notation . _ we write @xmath30 for a euclidean variable @xmath106 and write @xmath107 , while @xmath108 is used for @xmath37 , or , sometimes , where more convenient , for @xmath109 . for a manifold with corners @xmath110 , we write @xmath111 for the space of smooth functions which vanish to infinite order at the boundary of @xmath110 . we use notation @xmath112 $ ] to denote the blowup of @xmath110 at the submanifolds @xmath113 , @xmath114 , ( in that order ) . to analyze the operator @xmath39 we return to the formula . we first observe that the off - diagonal terms in the kernel of @xmath39 come from a neighbourhood of zero in the integral . indeed , let @xmath115 be a cutoff function equal to @xmath13 in a neighbhourhood of @xmath116 and equal to zero outside a compact set , and let @xmath117 . then we may insert the factor @xmath118 into the integral . with the factor @xmath119 inserted , the integral gives a function @xmath120 of @xmath17 , @xmath121 or @xmath13 , where @xmath122 is easily checked to be a classical symbol of order @xmath123 . by the symbolic functional calculus @xcite , this term is a scattering pseudodifferential operator of order @xmath123 , hence @xmath124 is a scattering pseudodifferential operator of order zero . it is therefore bounded on @xmath9 for all @xmath28 @xcite . so we are reduced to studying @xmath125 , given by the integral with factor @xmath115 inserted . we shall write down a fairly explicit parametrix for @xmath126 for small @xmath91 . in doing so , we need to consider the different asymptotics that this kernel takes when @xmath127 and @xmath128 . indeed , on @xmath24 the kernel decays as @xmath129 for @xmath128 and @xmath130 for @xmath131 , as @xmath132 , which ( except when @xmath133 ) is a different rate . this can be encoded geometrically by blowing up at the boundary when @xmath131 . consider the space @xmath134 ; \bfc \times \ { 0 \ } ; { \operatorname{lb}}\times \ { 0 \ } ; { \operatorname{rb}}\times \ { 0 \ } ] . \label{scsp - blowup}\ ] ] we shall denote the boundary hypersurfaces which are the lifts of @xmath135 $ ] , @xmath136 $ ] and @xmath137 $ ] to @xmath138 by @xmath87 , @xmath89 and @xmath90 , and @xmath139 $ ] by @xmath86 ; this is of course an abuse of notation , but in the context it will always be clear whether it is a boundary hypersurface of @xmath84 or @xmath138 that is referred to . we shall denote the new boundary hypersurfaces corresponding to the three blowups by @xmath140 , @xmath141 and @xmath142 , according as they arise from the first , second or third blowups in respectively , and we shall denote @xmath143 by @xmath144 , for ` zero face ' . we also define @xmath145 to be @xmath146 \subset \mmscsp$ ] . let @xmath147 be a smooth function on @xmath138 which is equal to one in a neighbourhood of @xmath145 , and whose support meets the boundary of @xmath138 only at @xmath86 and @xmath144 . we recall the well - known expression for the resolvent kernel @xmath126 on @xmath24 for @xmath63 : @xmath148 where @xmath149 is symbolic of order @xmath150 as @xmath151 , while it is @xmath152 and has a classical expansion in powers and logarithms as @xmath153 . in fact , @xmath154 is a polynomial of order @xmath150 when @xmath63 is odd . it is straightforward to check [ conormal ] let @xmath36 be the compactification of @xmath24 . then the resolvent kernel @xmath126 is such that @xmath155 is conormal at @xmath145 , and @xmath156 here , ` conormal to @xmath145 ' means that the kernel is conormal in @xmath157 which defines @xmath145 and smooth in the remaining variables @xmath158 , uniformly up to the boundary . for example , let us check the statement of the lemma near the triple intersection @xmath159 . coordinates near this codimension three corner are @xmath160 and boundary defining functions @xmath161 for @xmath142 , @xmath162 for @xmath90 and @xmath163 for @xmath140 . near this corner , @xmath164 is much larger than @xmath165 so we may expand @xmath166from which it is easy to check that holds . we also need a single space version of this space . let @xmath167 ; \partial m \times \ { 0 \}].\ ] ] denote the boundary hypersurfaces @xmath168 , @xmath144 and @xmath169 which arise from @xmath170 $ ] , @xmath171 and from the blowup , respectively , and denote corresponding boundary defining functions by @xmath172 , @xmath173 and @xmath174 . again , it will always be clear in context whether @xmath144 refers to the zero - face of @xmath175 or @xmath138 . we have [ const ] let @xmath176 . then there is a function @xmath177 , such that @xmath178 is equal to @xmath179 , and @xmath180 vanishes to infinite order at both @xmath169 and @xmath168 . we first use results from @xcite to show that we can solve @xmath181 on @xmath0 . the laplacian on an asymptotically euclidean manifold @xmath0 may be written in the form @xmath182where @xmath183 is an elliptic b - differential operator on @xmath0 . a short computation shows that near infinity , @xmath183 takes the form @xmath184where @xmath185 is the standard laplacian on the @xmath186-sphere . this is a strictly positive operator , so @xmath183 is ` totally elliptic ' , and hence is fredholm acting between the b - sobolev spaces is the @xmath14-space with respect to the b - metric @xmath187 ; thus @xmath188 . also @xmath189 is the b - sobolev space of order @xmath190 , defined as the set of functions @xmath191 such that @xmath192 for all b - differential operators of order @xmath190 on @xmath0 ( near infinity , such operators take the f orm @xmath193 , with @xmath194 ) . ] @xmath195 . thus @xmath17 itself is fredholm acting between @xmath196 . also @xmath183 is self - adjoint with respect to the measure induced by @xmath197 , so its index is equal to zero , hence it is invertible if and only if its null space is trivial . this is therefore also true for @xmath198.it is also shown in @xcite , section 5.25 , that if @xmath199 , with @xmath200 in @xmath18 , then @xmath200 has an asymptotic expansion of the form @xmath201 where @xmath202 is a spherical harmonic with eigenvalue @xmath203 . in particular , such a function tends to zero at infinity . it follows from this and from the maximum principle that there is no nontrivial solution to @xmath204 , with @xmath205 , because by @xmath200 would be a harmonic function tending to zero at infinity . hence we can solve @xmath181 , @xmath206 , where @xmath200 has an expansion . let @xmath200 be as in the previous paragraph . we first find a formal expansion for @xmath207 near the corner @xmath208 of @xmath175 . coordinates near this corner are @xmath209 and @xmath210 . let us look for an expansion for @xmath207 of the form @xmath211 where @xmath212 is given by the expansion for @xmath200 , so that @xmath213 . the operator @xmath214 may be written @xmath215 acting on the @xmath216th term this gives @xmath217 here @xmath218 indicates the derivative keeping @xmath91 fixed . when we switch to using coordinates @xmath219 , then we must replace @xmath220 by @xmath221 , getting @xmath222 acting on the @xmath216th term , we may replace @xmath185 by @xmath223 and @xmath224 by @xmath225 , getting the operator @xmath226 the equation @xmath227 has a smooth solution for every @xmath216 , with initial condition @xmath212 determined by the coefficient in . we may cut this off with a cutoff function in @xmath228 whose derivative is supported where @xmath229 $ ] . the error term is then of the form @xmath230 with @xmath231 supported in @xmath232 $ ] . we now change to variables which are smooth at the other corner , @xmath233 , namely @xmath91 and @xmath234 . the error term above may be written @xmath235 where @xmath236 is supported in @xmath237 $ ] . let us try to solve it away with a series of the form @xmath238 writing the operator in these new variables we get @xmath239 let @xmath240 . then @xmath241 satisfies the equation @xmath242 this is a regular singular ode with a solution of the form @xmath243 where @xmath244 is smooth down to @xmath245 . this gives us a formal series in powers of @xmath174 at @xmath169 in which each term is uniformly rapidly decreasing with all derivatives at @xmath168 ( i.e. , as @xmath246 ) . borel summing at @xmath169 , we get a formal solution that matches with @xmath200 to infinite order at @xmath169 . making a correction that vanishes to infinite order at @xmath144 , in order to make @xmath207 agree exactly with @xmath200 at @xmath144 , we get a function @xmath207 which satisfies all conditions of the lemma . we now use this lemma to define a harmonic function on @xmath0 which will be key to the parametrix construction . we begin by choosing a smooth function @xmath247 which is equal to @xmath13 for @xmath165 large and is supported in @xmath248 . using this we define functions @xmath249 on @xmath0 , with @xmath249 supported on the @xmath250 end of @xmath0 , in the obvious way . then let @xmath251 be the function given by lemma [ const ] from the function @xmath252 . it follows that @xmath253 moreover , @xmath254 we now define our parametrix . it is based on the resolvent kernel for @xmath24 , but there is a crucial additional term ( @xmath255 below ) which corrects the leading order coefficient of the kernel at the face @xmath142 ( see the discussion of this coefficient in the introduction ) . we now write @xmath249 to denote this function of the left variable o @xmath68 and @xmath256 denote this function of the right variable on @xmath68 . let @xmath257 be a parametrix , modulo smoothing operators , for @xmath258 in the interior of @xmath0 . we may assume that it is localized sufficiently close to the diagonal . we recall that the resolvent of the laplacian on @xmath24 has the form . using this notation we define @xmath259 in this section , we correct the parametrix to the exact resolvent . the main point is that we obtain complete information about the regularity of the kernel of the error term on @xmath138 , and therefore of the resolvent itself on this space . this allows us to determine the regularity of @xmath39 and @xmath260 and compute its behaviour to leading order at the boundary hypersurfaces of @xmath84 . applying @xmath214 to our parametrix ( on the left ) , we get @xmath261 where this equation defines @xmath262 . we may think of @xmath262 either as a kernel on @xmath138 or , by restricting to @xmath263 $ ] which is the interior of @xmath138 together with the interior of @xmath144 , a family of kernels parametrized by @xmath91 acting on functions on @xmath0 . by construction , the complete symbol of the diagonal singularity of @xmath264 is the inverse , modulo symbols of order @xmath265 , of the complete symbol of @xmath214 . thus @xmath262 is smooth at the diagonal . also , we see due to the properties of @xmath266 that @xmath262 vanishes to infinite order at @xmath267 , @xmath90 and @xmath141 . the crucial property of @xmath262 is the order of vanishing at @xmath142 . to calculate this , we need to determine the leading coefficient of the expansion of @xmath264 at @xmath142 . these terms come from @xmath268 and @xmath269 . since @xmath270 at @xmath142 , we have @xmath271 ; note that @xmath272 vanishes on @xmath142 , while @xmath273 is finite in the interior of @xmath142 . hence @xmath274 if we combine this with @xmath269 , then using we see that the leading coefficient becomes @xmath275 the leading term annihilated by the operator @xmath214 ( since @xmath276 is harmonic and @xmath116 at @xmath142 ) , so the error term @xmath262 is @xmath277 at @xmath142 an improvement of one order over what might be expected , and the main point of introducing the correction term @xmath269 . thus , we have @xmath278 note that both @xmath279 and @xmath280 are smooth on @xmath138 , the former function vanishing simply at @xmath281 and the latter vanishing simply at @xmath282 . thus implies that the kernel of @xmath262 is @xmath283 times a bounded function on @xmath138 . this implies that @xmath262 is hilbert - schmidt , uniformly for @xmath284 $ ] , hence compact for each @xmath284 $ ] . therefore @xmath285 has finite dimensional null space and cokernel of the same dimension on @xmath18 . we next show that we can modify our parametrix by the addition of a finite rank term so that the new error term is invertible for small @xmath91 . the correction term will be @xmath286 where the @xmath287 , @xmath288 are in @xmath289 and independent of @xmath91 . since @xmath290 maps into @xmath289 , the null space is contained in @xmath289 and hence is independent of the choice of @xmath291 . thus we choose @xmath288 to span the null space of @xmath285 , and we would like to choose @xmath287 so that @xmath292 span a space supplementary to the range of @xmath293 . this is possible since @xmath17 has trivial null space , and @xmath17 is self - adjoint , hence the range of @xmath17 on @xmath289 is dense in @xmath18 . choosing such @xmath287 , we define @xmath294 ( which is actually independent of @xmath91 ) by . we now define @xmath295 and define @xmath296 by setting @xmath297 @xmath296 enjoys all the properties of @xmath262 listed above . in addition , since @xmath298 is such that @xmath299 is invertible , it follows that actually @xmath300 is invertible for all sufficiently small @xmath91 ; we assume that @xmath301 is chosen so that @xmath300 is invertible for all @xmath284 $ ] . we now analyze the inverse of @xmath300 . let us write @xmath302 where this equation defines @xmath303 . the decay of the kernel @xmath296 at the boundary of @xmath138 implies that @xmath296 is hilbert - schmidt on @xmath18 . hence @xmath303 is also hilbert - schmidt . the regularity of @xmath262 on @xmath138 , and the fact that @xmath304 vanishes simply at @xmath305 , imply that @xmath306 ) \text { for all } n.\ ] ] using this and the formula @xmath307 shows that @xmath308 ) \text { for all } n. \label{sk}\ ] ] we are particularly interested in the kernel @xmath309 , which we shall call @xmath310 , since the addition of @xmath310 will correct the parametrix @xmath311 to the exact resolvent kernel . [ g5 ] let @xmath312 . then the kernel @xmath313 is in @xmath314 much more precise statements can be made about the kernels and , for example by using melrose s pushforward theorem @xcite , which shows that these kernels are actually conormal , with respect to the boundary and the diagonal , on @xmath84 . however , the @xmath315 statements will suffice for our purposes and are more straightforward to prove . let us break up @xmath311 into two parts @xmath316 , where @xmath147 is as in lemma [ conormal ] . thus @xmath317 is a smooth family of scattering pseudodifferential operators , while @xmath318 has no singularity at the diagonal . we first consider @xmath318 which is localized away from the diagonal . let @xmath319 denote the multiplication operator by @xmath320 on @xmath0 . then we have @xmath321 the kernel @xmath322 decays to order @xmath323 at @xmath141 and to order @xmath324 at @xmath89 and @xmath87 . if we multiply this kernel by @xmath325 , which corresponds to composing with @xmath326 on the right , then it also decays to order @xmath327 at @xmath140 and @xmath86 . this means that we can write @xmath328 ) , \quad x = \ang{z}^{-1}\ ] ] since @xmath108 is a product of boundary defining functions for @xmath281 , and @xmath329 vanishes to first order at @xmath89 and @xmath87 . in a similar way , using , we find that @xmath330));\ ] ] note that composing with @xmath331 on the left is harmless here because the kernel @xmath303 vanishes to infinite order on every boundary hypersurface where @xmath332 blows up . for this same reason the kernel is @xmath333 in the left variable , uniformly in the right variable and in @xmath91 . it follows that the composition @xmath334 ) . \label{linfty}\ ] ] now we integrate in @xmath91 . if we ignore the @xmath335 factor ( which is bounded ) then we find that @xmath336 because @xmath337 in exactly the same way we show that @xmath338 finally we consider the integral with @xmath311 replaced by @xmath317 . we may regard @xmath339 as a smooth family of scattering pseudodifferential operators , and @xmath303 as an element of @xmath340 since scattering pseudodifferential operators map @xmath289 to itself continuously , it follows that @xmath341 is also an element of the space . performing the @xmath91 integral we get an extra vanishing factor at @xmath342 , yielding @xmath343 , which proves the lemma for this piece . this completes the proof . recall that in section [ para ] we split @xmath344 , where @xmath124 was bounded from @xmath9 to @xmath9 for all @xmath345 , and @xmath346 hence , it remains to analyze @xmath125 . let us decompose @xmath347 as in the previous section and write @xmath348 correspondingly . the easiest kernel to deal with is @xmath349 ; this kernel is in @xmath350 , hence is bounded from @xmath9 to @xmath9 for @xmath351 . the kernel @xmath352 is bounded on @xmath9 for @xmath28 because it is a classical zero order pseudodifferential operator with proper support ; see chapter vi , section 5 of @xcite . the kernel @xmath353 we decompose further as as @xmath354 , where @xmath355and @xmath147 is as defined above lemma [ conormal ] . it is clear that @xmath356 is bounded on @xmath9 for @xmath28 , because t he riesz kernel on @xmath24 has this property . also , @xmath357 , like @xmath352 , it is a classical zero order pseudodifferential operator with proper support , hence bounded on all @xmath9 . we next consider @xmath358 . lemma [ g5 ] , with @xmath359 , shows that @xmath358 is in @xmath360 , where @xmath361 , for all @xmath362 , which implies that @xmath358 is bounded on @xmath9 for @xmath28 . thus we are left with @xmath363 . the kernel of @xmath363 such that @xmath364 moreover , the leading coefficient of @xmath363 at @xmath90 is a constant times @xmath365 . let us first consider the kernel of @xmath366 near @xmath90 and away from @xmath87 . this given by @xmath367 note that @xmath366 is supported away from @xmath89 , @xmath87 and @xmath86 since the support of @xmath368 is compact . it is a smooth function of @xmath30 and @xmath369 which is rapidly decreasing in @xmath370 . it vanishes to order @xmath371 at @xmath372 ; note that @xmath373 is a boundary defining function for @xmath90 in this region . moreover , it is given by @xmath374 changing variable of integration to @xmath370 and taking into account @xmath375 we see that the integral is @xmath376 at @xmath90 . if we do the analogous calculation for @xmath377 and add the results we find that the kernel of @xmath363 is given by @xmath378 at @xmath90 . this proves the last statement of the lemma . a similar computation can be done for @xmath377 at @xmath89 , but now the result vanishes to order @xmath11 at the left boundary , because the derivative @xmath379 , which is applied to the left variable of the kernel , increases the order of vanishing by @xmath13 at the left boundary . consider next the kernel @xmath377 near the triple intersection of @xmath90 , @xmath142 and @xmath140 . in this case , local boundary defining functions are @xmath380 , @xmath163 and @xmath381 . we claim that the kernel is actually a smooth function of @xmath382 , @xmath383 , @xmath108 , @xmath384 and @xmath385 in this region , which is a stronger statement , since @xmath386 . to see this , note that the kernel of @xmath377 is equal to @xmath387 times a @xmath62 function on @xmath138 . generally , if @xmath388 is any smooth function of @xmath207 and @xmath179 , @xmath389 , then @xmath390 is a smooth function of @xmath207 and @xmath391 . in other words , the function @xmath392 is smooth . ( this is easily checked directly by differentiating @xmath393 ; inverse powers of @xmath207 are harmless due to the @xmath394 factor . ) now let @xmath395 and @xmath396 , and treat the other coordinates as parameters , and the claim follows . the kernel @xmath377 vanishes to order @xmath397 at @xmath142 and @xmath398 at @xmath140 . we change variable of integration to @xmath370 as before , and the change of measure @xmath399 gives us additional vanishing at _ both _ @xmath142 and @xmath140 , since @xmath400 vanishes at both @xmath142 and @xmath140 . thus the result is a smooth function of @xmath401 which vanishes to order @xmath53 at @xmath402 and order @xmath403 at @xmath404 , which verifies the statement of the lemma near the corner @xmath405 . the other regions of @xmath84 are treated similarly . this lemma implies that , for @xmath4 , @xmath363 is an element of @xmath360 . moreover , for @xmath5 , this is not true since the function @xmath406 is not in @xmath407 then , and the coefficient of @xmath406 is @xmath365 which does not vanish identically . therefore @xmath363 can not be applied to any bounded function equal to @xmath408 near infinity , which lies in @xmath9 for @xmath5 . this completes the proof of theorem [ main ] . if @xmath0 has one euclidean end then the same argument shows that the riesz transform is bounded on @xmath9 for all @xmath28 . in this case , the parametrix @xmath264 can be taken to be ( compare with ) @xmath409 in this case the role of @xmath276 in the computation above is played by the constant function @xmath13 . the argument is the same as above , except that the gradient of @xmath13 vanishes so that we get @xmath410 instead of @xmath411 in ( as outlined in the introduction ) , leading to the boundedness for all @xmath20 strictly between @xmath13 and @xmath324 . as part of the analysis of the heat kernel we analyzed the structure of the resolvent @xmath126 for real @xmath91 , including an analysis of the asymptotics of its kernel when @xmath412 . this analysis remains valid for any cone @xmath413 for any @xmath46 . we can use this to obtain information about the heat kernel @xmath414 of @xmath415 on @xmath0 via the contour integral @xmath416 where @xmath417 is the contour @xmath418 . let us focus on the heat kernel in the following asymptotic regime : we fix a point @xmath419 , which we think of as being in the ` compact part ' of @xmath0 ( where the metric is not flat ) , and fix an end of @xmath0 and a point @xmath420 which we think of as a point at infinity for this end . consider the behaviour of the heat kernel @xmath421 where @xmath422 and @xmath151 , @xmath423 so that @xmath424 approaches a finite positive limit @xmath425 . assume that @xmath0 has euclidean ends , with the number of ends at least two . under the limiting process described above , @xmath426 approaches a limit , for any value of @xmath291 . indeed @xmath427 where @xmath428 is the harmonic function which tends to @xmath13 at the given end and tends to @xmath429 at all other ends . in particular , we have a lower bound on the derivatives of the heat kernel for large time : @xmath430 for @xmath431 this result is not surprising . the point of this proposition is that _ taking derivatives in the @xmath30 variable gives no additional decay in the heat kernel _ ( in this asymptotic regime ) . this contrasts with euclidean space where each additional derivative gives additional decay of @xmath432 . the @xmath91th @xmath30-derivative of the heat kernel is given by the contour integral with the resolvent replaced by the @xmath91th @xmath30-derivative of the resolvent . clearly , to prove the theorem we only have to consider the kernel of the resolvent in a neighbourhood of @xmath90 and @xmath142 . near the interior of @xmath142 , and away from @xmath90 the function @xmath433 is a smooth function , which goes to infinity at @xmath90 ; in fact , @xmath434 is a boundary defining function for @xmath90 . in this integral , the term @xmath435 therefore vanishes together with all its derivatives at @xmath90 , since @xmath436 by assumption , which means that we may ignore the expansion of the resolvent at @xmath90 . hence to find the asymptotics the heat kernel near @xmath90 in this regime we only need to consider the expansion of the resolvent at @xmath142 ( up to a correction that vanishes to infinite order as @xmath423 ) . using the @xmath315 bounds , we may write the @xmath91th derivative of the resolvent kernel in the form @xmath437 in the region of interest . let us first substitute @xmath438 for the resolvent into the integral . thus we want to compute the limit @xmath439 substituting @xmath440 and @xmath441 , and using @xmath442 , we get @xmath443 taking the limit is trivial , since is independent of @xmath444 . to perform the integral , consider the case of @xmath24 , with kernel @xmath445 with @xmath30 fixed to be the origin . this gives rise to an integral @xmath446 which is equal to @xmath447 multiplying through by @xmath448 gives @xmath449 hence , is equal to @xmath450 which is the right hand side of . if we now substitute @xmath451 for the resolvent in , which vanishes to an additional order as @xmath423 as compared to @xmath438 , then the integral also vanishes to an additional order , giving a zero contribution to the limit . this proves the proposition . it is also of interest to compute the leading behaviour of the heat kernel @xmath414 as @xmath151 and as @xmath80 both tend to infinity , but along different ends . suppose that @xmath452 , where @xmath453 is fixed and that @xmath454 , @xmath455 is fixed , and suppose further that @xmath456 where @xmath457 . under this asymptotic regime , the limit @xmath458exists and is finite . hence in this asymptotic regime the heat kernel has @xmath459 decay as @xmath151 . for @xmath63 this is faster than the usual @xmath460 decay . hence gaussian lower bounds do not hold for the heat kernel on @xmath0 . this was observed in @xcite , and can be heuristically explained in terms of bro wnian motion on @xmath0 . here we give an explicit quantitative description of the failure of this lower bound . we shall perform a similar computation as in the proof of the previous proposition . since @xmath461 under this limiting regime , and @xmath80 go to infinity along different ends , we end up at the ` anti - diagonal ' part of @xmath87 . hence we need to consider the resolvent kernel near the anti - diagonal part of @xmath87 and @xmath140 , where @xmath462 and @xmath463 . it is the @xmath269 term which is important here ; we need the leading behaviour of @xmath464 at the negative end . it is not hard to show that @xmath465at this end . indeed , the harmonic function @xmath466 is equal to @xmath467 as @xmath468 along this end , for some @xmath469 . the leading coefficient @xmath470 from lemma [ const ] must then be equal to a constant times @xmath471 , which follows readily from the fact that @xmath472 satisfies the equation @xmath473 . the specific structure of the parametrix @xmath311 , together with the estimate with @xmath431 , shows that in this region the resolvent kernel may be written as a sum@xmath474 in the region of interest . substituting @xmath438 for the resolvent into the integral , we obtain@xmath475let @xmath476 . substituting @xmath440 and @xmath477 , and using @xmath442 , we get @xmath478thus the limit exists and is finite , when @xmath438 is substituted for the resolvent . as in the previous proof , when @xmath451 is substituted for the resolvent the limit is zero , since @xmath451 decays to an additional order at infinity . this completes the proof . here @xmath479 is a complete riemannian manifold of dimension @xmath11 . we want here to discuss some consequence of the boundedness of the riesz transform on @xmath9 for some @xmath480 for the @xmath9 cohomology . on @xmath481 , the space of @xmath14 differential forms admits the hodge decomposition @xmath482 where @xmath483 ( see @xcite ) . let us recall now the definition of reduced @xmath9-cohomology : for @xmath484 , the first space of reduced @xmath9 cohomology of @xmath481 is @xmath485 where we take the closure in @xmath9 . the first space of reduced @xmath14 cohomology can be identified with @xmath486 . as noticed in @xcite , if we assume that for some @xmath487 the riesz transform @xmath488 is bounded on @xmath9 and on @xmath489 , then the hodge projector @xmath490 i.e. the orthogonal projector of @xmath19 onto the space of ` exact forms ' extends by continuity to a bounded operator @xmath491 we assume now that @xmath479 is a complete riemannian manifold , @xmath492 , satisfying the nash inequality @xmath493 for all @xmath494 and some @xmath495 , and that the volume growth of geodesic balls is uniformly bounded : @xmath496 it follows from @xcite that ( [ nash ] ) implies a matching lower bound : @xmath497 note that ( [ nash ] ) easily implies the faber - krahn inequality : @xmath498 for all @xmath499 with finite measure , where @xmath500 is the first eigenvalue for the laplacian on @xmath501 for the dirichlet boundary conditions ( in fact , ( [ nash ] ) and ( [ fk ] ) are equivalent , see @xcite ) . also , if @xmath502 , ( [ nash ] ) is equivalent to the sobolev inequality : @xmath503 for some @xmath504 ( see for instance @xcite ) . according to @xcite , we know that on @xmath481 the riesz transform is bounded on @xmath505 for @xmath5061,2]$ ] . hence if we assume that for some @xmath487 the riesz transform is also bounded on @xmath9 then , according to the above remark , the hodge projector @xmath507 extends by continuity to a bounded operator @xmath491 [ last ] under the hypotheses ( [ nash ] , [ volumgro ] ) , if the riesz transform is bounded in @xmath9 for some @xmath23 , then @xmath508 is the closure in @xmath65 of @xmath509 . according to @xcite , the nash inequality implies that the semigroup @xmath510 satisfies the bound @xmath511 a result of n. varopoulos ( @xcite ) then implies the following mapping property for @xmath5121,n[$ ] : @xmath513 in order to prove the lemma , we have to show that if @xmath514 then @xmath515 can be approximated in @xmath9 by a sequence of elements of @xmath509 . we seek a sequence @xmath516 of smooth functions with compact support such that @xmath517 since we assume that the riesz transform is bounded in @xmath9 , we know that its adjoint @xmath518 is bounded . hence we have @xmath519 . note that the condition @xmath520 is satisfied since we are assuming that @xmath521 , in which case @xmath522 . thus , by , @xmath523choose a point @xmath524 and choose @xmath525 @xmath526 since @xmath527 and @xmath528 obviously tends to @xmath515 in @xmath9 as @xmath529 , we need only to show that @xmath530 we know that @xmath531 is harmonic outside a big ball @xmath532 containing the support of @xmath533 . now the faber - krahn inequality ( [ fk ] ) implies a mean value inequality for harmonic functions ( see @xcite , lemma 6.9 ) which yields : @xmath534 provided that @xmath535 . in particular if @xmath536 we obtain @xmath537 hence we finally obtain , if say @xmath538 : @xmath539 which indeed goes to zero when @xmath529 . we have proved that @xmath540 the converse inclusion follows from the fact that @xmath541 and that @xmath183 , being a bounded projector , has a closed range . as a consequence of lemma [ last ] , if the assumptions ( [ nash],[volumgro ] ) are satisfied and if the riesz transform is bounded on @xmath9 for some @xmath23 , then @xmath542 can be identified with:@xmath543 moreover we also have under the hypotheses ( [ nash ] , [ volumgro ] ) , if the riesz transform is bounded in @xmath9 for some @xmath23 , then @xmath544 [ coclosed ] as a matter of fact , we have @xmath545 . the density of @xmath546 in @xmath65 and the boundedness of @xmath183 in @xmath9 imply that @xmath547 is dense in @xmath548 . but we have @xmath549 the latter space is closed , hence we have the inclusion @xmath550 now assume that @xmath551 is coclosed . we define a sequence of cutoff functions by @xmath552 then @xmath553 but now @xmath554 and @xmath555 , because @xmath556 ( here @xmath557 denotes the contraction of @xmath533 with the vector field @xmath558 ) . take @xmath5591,n [ $ ] with @xmath560 then we have @xmath561 but an easy computation leads to @xmath562 with , we have @xmath563 hence by continuity of @xmath15 we obtain @xmath564 . in particular we obtain a hodge - de rham interpretation of the @xmath9 cohomology : [ hdr ] under the hypotheses ( [ nash ] , [ volumgro ] ) , if the riesz transform is bounded in @xmath9 for some @xmath23 , then @xmath565 in general one can not compare @xmath9 cohomology for different values of @xmath20 . however let us now assume that the ricci curvature of @xmath0 is bounded from below : @xmath566 from the bochner formula @xmath567 we see that if @xmath533 is a harmonic @xmath13-form in @xmath14 , then it satisfies the subelliptic estimate @xmath568 indeed if for @xmath569 we define @xmath570 , then it is classical to show that the bochner formula and the kato inequality imply @xmath571 passing to the limit @xmath572 , we get the desired subelliptic estimate ( only in the distributional sense ) . hence with the nash inequality we can deduce that @xmath533 is in fact bounded and @xmath573 this can be done using a nash - moser iteration scheme @xcite , but we can also use our upper bound on the heat operator with the inequality ( [ subelliptic ] ) , to assert that for every @xmath574 @xmath575 ) we obtain @xmath576 with @xmath577 , we obtain the desired bound . hence there is a well defined map @xmath578 proposition [ hdr ] immediately implies [ inj ] assume that ( [ nash ] , [ volumgro ] , [ ricci ] ) hold , and that for some @xmath23 , the riesz transform is bounded on @xmath9 and on @xmath489 . then the natural map @xmath579 is injective . a corollary of this lemma is : [ nb]assume that @xmath502 , and that @xmath481 satisfies the assumptions ( [ nash ] , [ volumgro ] , [ ricci ] ) and that it has more than two ends , then for every @xmath27 the riesz transform is not bounded on @xmath9 . in @xmath502 , then @xmath481 satisfies the sobolev inequality ( [ sobo ] ) , and following @xcite , we know that if @xmath0 has more than two ends there exists a non - constant bounded harmonic function @xmath580 with finite dirichlet energy with @xmath228 compact and @xmath581 unbounded then @xmath582 . ] , hence @xmath583 is a harmonic @xmath14 @xmath584form . take @xmath516 as in . we have @xmath585 , but if @xmath27 , then if @xmath586 we have @xmath587 and integrating by parts we have : @xmath588 this quantity is bounded for @xmath25 and grows as @xmath589 if @xmath590 . hence we obtain that @xmath591 . hence @xmath583 is zero in reduced @xmath9 cohomology . since @xmath583 is non - zero in @xmath486 , and since under ( [ nash],[volumgro ] ) the riesz transform is bounded on @xmath9 for @xmath592 , lemma ( [ inj ] ) says that it can not be bounded on @xmath9 if @xmath27.this generalizes the unboundedness part of theorem [ main ] to the much larger class of manifolds satisfying , , . in this final section we discuss some questions suggested by the work above , and pose some open problems . for any complete riemannian manifold @xmath0 of infinite measure there are numbers @xmath593 such that the riesz transform is bounded on @xmath9 for all @xmath20 between @xmath594 and @xmath595 ( it may or may not be bounded at @xmath596 or @xmath597 ) . we may call these values the lower and upper thresholds for @xmath0 . there are a number of classes of manifolds on which the riesz transform is known to be bounded on @xmath9 for all @xmath20 ( in other words , @xmath598 and @xmath599 ) ; for example , manifolds with nonnegative ricci curvature @xcite , cartan - hadamard manifolds with a spectral gap @xcite , noncompact symmetric spaces @xcite and lie groups of polynomial growth @xcite ( see @xcite for more examples and references ) . on the other hand , coulhon and ledoux showed in @xcite that for any @xmath600 there is a manifold @xmath0 with bounded geometry such that @xmath601 . another example , with polynomial volume growth , was given in @xcite . since , as we mentioned earlier , it was shown by coulhon and duong that @xmath598 for a large class of complete manifolds , and that @xmath602 for certain simple surfaces , one could wonder whether @xmath190 is the upper threshold for a large class of manifolds . but h .- q . li @xcite proved that for @xmath11-dimensional cones with compact basis , @xmath603where @xmath604 is the smallest nonzero eigenvalue of the laplacian on the basis . note that @xmath605 here . manifolds with more than one euclidean end satisfy the doubling condition but not the scaled @xmath14 poincar inequality . our result sheds some light on the implications for the riesz transform of these conditions : it follows from @xcite and @xcite that doubling together with poincar ( equivalently , upper and lower gaussian estimates of the heat kernel ) are not sufficient for the riesz transform to be bounded for _ all _ @xmath23 . theorem [ main ] shows that these conditions are not necessary for the riesz transform to be bounded for _ some _ @xmath23 . the class of manifolds with euclidean ends is of course extremely special . one can attempt to enlarge the class of known examples synthetically , i.e. by creating further examples from known examples by performing various operations . our results may be seen as obstructions to the stability of the @xmath9 boundedness of the riesz transform under gluing for @xmath20 above the dimension . we only mention , without proof , one result along these lines . namely , if the riesz transform is bounded on @xmath9 on a complete riemannian manifold @xmath0 of infinite measure , and with ricci curvature bounded from below , then it is bounded on @xmath9 on @xmath607 for any compact @xmath608 . we also mention a conjecture on manifolds obtained by gluing several copies of a simply connected nilpotent lie group ( endowed with a left invariant metric ) . according to @xcite we know that the riesz transform is bounded for every @xmath20 on a simply connected nilpotent lie group . let @xmath609 be a simply connected nilpotent lie group of dimension @xmath502 ( endowed with a left invariant metric ) . according to @xcite we know that the riesz transform on @xmath609 is bounded for every @xmath20 . let @xmath610 be the homogeneous dimension of @xmath608 ; for instance we can set @xmath611 @xmath612 being a fixed point . let @xmath481 be a manifold obtained by gluing @xmath613 copies of @xmath609 . the manifold @xmath0 is diffeomorphic to the sphere with @xmath91 points removed . according to @xcite we know that on @xmath481 the riesz transform is bounded on @xmath9 for @xmath6141,2]$ ] . the argument of section [ coh ] can be applied ( changing the @xmath11 appearing in analytic inequalities by the homogeneous dimension @xmath610 ) . then ( [ nb ] ) implies that the riesz transform is not bounded on @xmath9 for @xmath615 . we can moreover compute the @xmath9 cohomology of @xmath0 . the arguments of proposition 3.3 of @xcite are given for the @xmath14 cohomology of such @xmath0 but they can be easily modified for @xmath9 cohomology . we find that for @xmath6141,\nu[$ ] , @xmath616 whereas for @xmath615 we find @xmath617 that is , the conclusion of lemma [ inj ] is satisfied if and only if @xmath618 . this gives another proof of the fact that the riesz transform is not bounded on @xmath9 for @xmath615 . finally , it would be interesting to get results for differential forms . one can consider either @xmath620 or @xmath621 , where @xmath622 is the laplacian on forms and @xmath39 really means @xmath623 where @xmath624 and @xmath625 for @xmath626 ( this projects off the @xmath14 null space of @xmath17 , i.e. the @xmath14-cohomology , which is trivial in the case of @xmath429-forms when @xmath0 has infinite measure ) . g. alexopoulos , an application of homogenization theory to harmonic analysis : harnack inequalities and riesz transforms on lie groups of polynomial growth , _ canad . j. math . _ 44 ( 1992 ) , 691727 . anker , sharp estimates for some functions of the laplacian on noncompact symmetric spaces , _ duke math . j. _ 65 ( 1992 ) , no . 2 , 257297 . p. auscher and t. coulhon , riesz transform on manifolds and poincar inequalities , preprint 2004 . p. auscher , t. coulhon , x.t . duong and s. hofmann , riesz transform on manifolds and heat kernel regularity , to appear in _ ann . _ d. bakry , etude des transformations de riesz dans les varits riemanniennes courbure de ricci minore , _ sminaire de probabilits _ , xxi , 137172 , lecture notes in math . , 1247 , springer , berlin , 1987 . i. benjamini , i. chavel and e. feldman , heat kernel lower bounds on manifolds using the old ideas of nash , unpublished manuscript , 1993 . d. bakry , t. coulhon , m. ledoux , l. saloff - coste , sobolev inequalities in disguise , _ indiana univ . j. _ 44 ( 1995 ) , no . 4 , 10331074 . p. brard , from vanishing theorems to estimating theorems : the bochner technique revisited , _ bull . a.m. s. _ 19 ( 1988 ) , 371406 . cao , y. shen and yingand s. zhu , the structure of stable minimal hypersurfaces in @xmath627 , _ math . _ 4 ( 1997 ) , no . 5 , carlen , s. kusuoka and d. stroock , upper bounds for symmetric markov transition functions , _ ann . h. poincar , probab . statist . _ 23 ( 1987 ) , 2 , suppl . , 245287 . g. carron , ingalits isoprimtriques de faber - krahn et consquences , in _ actes de la table ronde de gomtrie diffrentielle ( luminy , 1992 ) _ , smin . congr . , vol . 1 , soc . math . france , paris , 1996 , pp . g. carron , @xmath14-cohomologie et parabolicit , preprint 2004 math.dg/0407163 . t. coulhon and x.t . duong , riesz transforms for @xmath628 , _ trans . soc . _ 351 ( 1999 ) , 1151 - 1169 . t. coulhon and x.t . duong , riesz transform and related inequalities on non - compact riemannian manifolds , _ comm . in pure and appl . _ 56 ( 2003 ) , no . 12 , 1728 - 1751 . t. coulhon and m. ledoux , isoprimtrie , dcroissance du noyau de la chaleur et transformations de riesz : un contre - exemple , _ ark . _ 32 ( 1994 ) , no . 1 , 6377 . t. coulhon t. , li h.q . , estimations infrieures du noyau de la chaleur sur les varits coniques et transforme de riesz , _ archiv der mathematik _ 83 ( 2004 ) , 229242 . g. de rham , _ varits diffrentiables , formes , courants , formes harmoniques _ , hermann , 3rd edition , 1973 . a. grigoryan , heat kernel upper bounds on a complete non - compact manifold , _ rev . iberoamericana _ 10 ( 1994 ) , 2 , 395452 . a. grigor yan , estimates of heat kernels on riemannian manifolds , in _ spectral theory and geometry ( edinburgh , 1998 ) _ , 140225 , london math . lecture note ser . , 273 , cambridge univ . press , cambridge , 1999 . a. grigor yan and l. saloff - coste , heat kernel on connected sums of riemannian manifolds , _ math . _ 6 ( 1999 ) , no . 3 - 4 , 307321 . a. hassell and a. vasy , symbolic functional calculus and n - body resolvent estimates , _ anal . _ 173 ( 2000 ) , 257283 . a. hassell and a. vasy , the resolvent for laplace - type operators on asymptotically conic spaces , _ ann . fourier _ 51 ( 2001 ) , 1299 - 1346 . l. hrmander , _ the analysis of linear partial differential operators _ , vol . 3 , springer - verlag , berlin , 1985 . li , la transformation de riesz sur les varits coniques , _ j. funct . anal . _ 168 ( 1999 ) , 1 , 145238 . n. lohou , comparaison des champs de vecteurs et des puissances du laplacien sur une varit riemannienne courbure non positive , _ j. funct . anal . _ 61 ( 1985 ) , 2 , 164201 . r.b . melrose , spectral and scattering theory for the laplacian on asymptotically euclidian spaces , in _ spectral and scattering theory _ , m. ikawa , ed . , marcel dekker , 1994 . melrose , calculus of conormal distributions on manifolds with corners , _ int . math . res . not . _ 3 ( 1992 ) , 5161 . melrose , _ the atiyah - patodi - singer index theorem _ , a. k. peters , wellesley , ma , 1993 . stein , _ harmonic analysis _ , princeton university press , princeton , nj , 1993 . varopoulos , hardy - littlewood theory for semigroups , _ j. funct . _ 63 ( 1985 ) , no . 2 , 240260 .

for hole systems with an effective spin , we present an invariant decomposition of the spin density matrix that can be interpreted as a multipole expansion . the charge density corresponds to the monopole moment and the spin polarization due to a magnetic field corresponds to a dipole moment while heavy hole light hole splitting can be interpreted as a quadrupole moment . for quasi two - dimensional hole systems in the presence of an in - plane magnetic field the spin polarization is a higher - order effect that is typically much smaller than one even if the minority spin subband is completely depopulated . on the other hand , the field can induce a substantial octupole moment which is a unique feature of hole systems .

cond-mat0402360

a powerful approach to the quantum many - body problem is density functional theory . hohenberg and kohn @xcite showed that the ( nondegenerate ) ground state of a system of interacting particles is a unique function of the corresponding ground state density @xmath2 . the ground state of the system is thus fully characterized by its density @xmath2 ( ref . ) . for electrons with spin @xmath3 the density @xmath2 must be replaced by a @xmath4 spin density matrix @xmath5 . it is well - known that @xmath5 can be parameterized by the density @xmath6 and ( the magnitude of ) the spin polarization . @xcite the spin polarization at a magnetic field @xmath7 is thus an important quantity for our understanding of many - particle phenomena . recently , the interest in this subject has been renewed because experiments have indicated that the spin polarization of quasi two - dimensional ( 2d ) systems due to a magnetic field @xmath1 parallel to the 2d plane affects the apparent metallic behavior of these systems . @xcite furthermore , the spin polarization is an important parameter for possible applications in the field of spintronics . @xcite in quasi two - dimensional ( 2d ) systems the spin polarization has been studied by applying a magnetic field @xmath1 parallel to the 2d plane . @xcite if the zeeman splitting due to @xmath1 becomes sufficiently large , the minority spin subband is completely depopulated . in such a situation , electron systems are fully spin - polarized . this approach has been used by several groups to study the spin polarization in low - density 2d electron systems . @xcite recently , there has been considerable interest in the spin polarization of quasi 2d hole systems @xcite . hole systems in the uppermost valence band of many common semiconductors such as ge and gaas are different from electron systems due to the fact that the hole states have an effective spin @xmath0 ( ref . ) . usually , hole states in quasi 2d systems are discussed using a basis of angular momentum @xmath0 eigenfunctions with quantization axis perpendicular to the 2d plane . subband quantization yields so - called heavy hole ( hh ) states with @xmath8 component of angular momentum @xmath9 and light hole ( lh ) states with @xmath10 . often it is assumed that the spin degree of freedom of these hh and lh states behaves similar to the spin of @xmath11 electron states . however , recently it has been shown that the spin polarization of quasi 2d hh systems in the presence of an in - plane magnetic field behaves rather different from the more familiar case of @xmath12 electron systems . @xcite for example , the spin polarization of quasi 2d hh systems can change its sign at a finite value of @xmath1 . similar to the case of @xmath12 electron systems , the fundamental object for the characterization of @xmath0 hole systems is the spin density matrix @xmath5 from which we can derive , e.g. , the spin polarization . the dominant character of the occupied eigenstates of hole systems depends on the quantization axis of the underlying @xmath0 basis functions . obviously , all observable quantities such as the spin polarization may not depend on this choice . therefore , it is necessary to formulate the spin density matrix in a way such that observable quantities can be calculated independent of the particular choice for the basis functions that are used . it is the goal of this paper to present such a theory . using the theory of invariants @xcite we will derive an invariant decomposition of the spin density matrix of @xmath0 hole systems that can be interpreted as a multipole expansion . in general , the single - particle states in quasi 2d systems with effective spin @xmath13 ( the kohn - sham states ) can be characterized by multicomponent envelope functions @xcite @xmath14 here @xmath15 denotes the @xmath16th component of the envelope function that modulates the band - edge bloch function @xmath17 , @xmath16 labels the @xmath8 components of @xmath13 , @xmath18 is the subband index , @xmath19 is the spin index , and @xmath20 is the in - plane wave vector . the @xmath21-dimensional hermitian , positive definite spin density matrix @xmath5 has then the matrix elements ( temperature @xmath22 ) @xmath23{rl } \langle m | m ' \rangle = & { \displaystyle}\sum_{\alpha , \sigma } \int \ ! dk_\|^2 \ : \theta [ \pm ( e_{\mathrm{f}}- e_{\alpha\sigma } ( { { \ensuremath{\bm{k}}}}_\| ) ) ] \\ & { \displaystyle}\times \int \ ! dz \ : \xi^{m \,\ast}_{\alpha \sigma { { \ensuremath{\bm{k}}}}_\| } ( z ) \ : \xi^{m'}_{\alpha \sigma { { \ensuremath{\bm{k}}}}_\| } ( z ) \ ; . \end{array}\ ] ] here @xmath24 is the spin - split subband dispersion and @xmath25 is the fermi energy . the upper ( lower ) sign in eq . ( [ eq : spin - dens - mat ] ) applies to electron ( hole ) systems . we integrate over all orbital degrees of freedom that are irrelevant in the present discussion . the density is given by the trace of @xmath5 . obviously , the trace is independent of the particular choice for the set of basis function @xmath17 . for 3d _ homogeneous _ hole gases ( @xmath26 ) the matrix @xmath5 is proportional to the unit matrix , i.e. , we have equal contributions of hh and lh states because we have no preferential direction for the quantization axis of angular momentum . @xcite ( this was overlooked in ref . . ) for _ inhomogenous _ systems such as quasi-2d systems , the matrix @xmath5 has a more complicated structure , even at @xmath26 . in principle @xmath5 contains all information about the ground state of the system . however , as the hermitian @xmath27 matrix @xmath5 is characterized by 16 independent parameters , it appears impractical to characterize a @xmath0 system by its density matrix . thus , bobbert _ et al . _ @xcite suggested to parameterize the mixed hh - lh character of inhomogenous hole gases by the ratio between the lh and hh fermi wave vectors . @xcite proposed to use partial densities for hh and lh states . the latter approach was criticized by bobbert _ @xcite because the partial hole densities depend on the set of basis function @xmath17 . recently , krkkinen _ et al . _ @xcite suggested to treat mixed hh - lh systems as two - component systems where the hh and lh components are essentially independent of each other . until today , the interpretation of the @xmath28 spin density matrix @xmath5 of particles with @xmath0 has remained an open problem . we show here that the theory of invariants @xcite allows to decompose @xmath5 for systems with arbitrary point group @xmath29 into symmetry - adapted irreducible tensors which provide a rigorous yet physically transparent classification of the different terms in @xmath5 independent of the particular choice for the set of basis functions . we assume that the system is characterized by an @xmath30 multiband hamiltonian @xmath31 . the basis functions of @xmath31 transform according to an @xmath32-dimensional representation @xmath33 of @xmath29 . for simplicity , we assume that the representation @xmath33 is irreducible . the density matrix @xmath5 of this system can be written in the form @xmath34 here @xmath35 are @xmath30 matrices with all elements equal to zero , except for the element @xmath36 that equals one . the components @xmath37 and the matrices @xmath38 transform according to the product representation @xmath39 . in general , this representation is reducible , i.e. , we have @xmath40 equation ( [ eq : red_decomp ] ) implies that we can decompose the set of matrices @xmath38 ( the components @xmath37 ) into irreducible tensors @xmath41 ( @xmath42 ) that transform according to the representations @xmath43 of @xmath29 . the symmetry - adapted expansion of the density matrix @xmath5 is then given by @xcite @xmath44 where @xmath45 labels the elements of the irreducible representation @xmath43 . the terms @xmath46 can be interpreted as the projection of @xmath5 on the basis matrices @xmath47 . if the matrices @xmath48 are orthonormalized we thus have @xmath49 according to eq . ( [ eq : sym_dens_op ] ) , we get a decomposition of the density matrix @xmath5 in terms of invariants @xmath50 . often , these invariants provide already a clear interpretation of the different terms in @xmath5 . a second advantage of the invariant expansion ( [ eq : sym_dens_op ] ) is the following . using the clebsch - gordan coefficients for the irreducible representations @xmath51 of @xmath29 we can construct _ scalar _ quantities @xmath52 from the irreducible tensors @xmath53 which transform according to the unit representation @xmath54 ( ref . ) . these scalars are sums of products of the elements @xmath55 . by definition , they are independent of the particular choice for the set of basis functions that was used to evaluate the spin density matrix . therefore , the scalars @xmath52 represent a convenient set of independent variables for a parameterization and characterization of the spin density matrix of a system with point group @xmath29 . for a system with spherical symmetry we have @xmath57 . we denote the irreducible representations of @xmath56 by @xmath58 where @xmath59 is the angular momentum . here the invariant expansion ( [ eq : sym_dens_op ] ) is equivalent to a multipole expansion of the density matrix . it follows from the well - known relation for the addition of angular momenta , @xcite @xmath60 that for a system with angular momentum @xmath13 the invariant expansion ( [ eq : sym_dens_op ] ) contains multipole moments @xmath61 up to the order @xmath62 . the squared magnitude of the @xmath59th multipole moment is given by @xcite @xmath63 by definition , these scalars are independent of the quantization axis of angular momentum that is used for the basis functions @xmath64 in eq . ( [ eq : envelope_fun ] ) . @xmath65 } \begin{array}{c*{3}{@{\hspace{2em}}l } } \hline \hline \mathcal{d}_{\mathfrak{j } } & { \mathfrak{m } } & m_{{\mathfrak{j}}{\mathfrak{m } } } & \rho_{{\mathfrak{j}}{\mathfrak{m } } } \nl \hline \mathcal{d}_{0 } & 0 & \frac{1}{\sqrt{2 } } \ , \openone_{2\times 2 } & \frac{1}{\sqrt{2 } } \left({\left\langle{\left . { \frac{1}{2}}\right| } { \frac{1}{2}}\right\rangle } + { \left\langle{\left.{-\frac{1}{2}}\right| } { -\frac{1}{2}}\right\rangle}\right ) \nl \mathcal{d}_{1 } & 1 & -\frac{1}{2}\left(\sigma_x+i\sigma_y\right ) & -{\left\langle{\left.{-\frac{1}{2}}\right| } { \frac{1}{2}}\right\rangle } \\[1ex ] & 0 & \frac{1}{\sqrt{2 } } \sigma_z & \frac{1}{\sqrt{2 } } \left({\left\langle{\left . { \frac{1}{2}}\right| } { \frac{1}{2}}\right\rangle } - { \left\langle{\left.{-\frac{1}{2}}\right| } { -\frac{1}{2}}\right\rangle}\right ) \nl \hline \hline \end{array}$ ] @xmath65 } \begin{array}{c*{2}{@{\hspace{0.5em}}l } } \hline \hline \mathcal{d}_{\mathfrak{j } } & { \mathfrak{m } } & \mathcal{k}_{{\mathfrak{j}}{\mathfrak{m } } } \nl \hline \mathcal{d}_{0 } & 0 & k^2 = \{k_+ , k_-\ } + k_z^2 \nl \mathcal{d}_{1 } & 1 & - \frac{1}{\sqrt{2 } } b_+ \nl & 0 & b_z \nl \mathcal{d}_{2 } & 2 & - \frac{1}{2 } k_+^2 \nl & 1 & - \{k_+ , k_z \ } \nl & 0 & \frac{1}{\sqrt{6 } } ( 2 k_z^2 - \{k_+ , k_-\ } ) \nl \mathcal{d}_{3 } & 3 & - \frac{1}{\sqrt{8 } } k_+^2 b_+ \nl & 2 & \frac{1}{\sqrt{12 } } ( k_+^2 b_z + 2\{k_+,k_z\ } b_+ ) \nl & 1 & \frac{1}{\sqrt{120 } } ( k_+^2 b_- + 2 \ { k_+ , k_-\ } b_+ - 4 k_z^2 b_+ - 8 \{k_+ , k_z\ } b_z ) \nl & 0 & \frac{1}{\sqrt{10 } } [ ( 2 k_z^2 - \{k_+ , k_-\ } ) b_z - \{k_+,k_z\ } b_- - \{k_-,k_z\ } b_+ ] \nl \hline \hline \end{array}$ ] first we apply the above formalism to the well - known case of a density matrix for systems with angular momentum @xmath12 . using standard angular momentum algebra @xcite we obtain the invariant decomposition listed in table [ tab : spn12 ] , where @xmath66 , @xmath67 , and @xmath68 are the pauli spin matrices . we have omitted the terms with @xmath69 which can be obtained using @xmath70 and @xmath71 ( ref . ) . we note that the terms @xmath72 can be obtained either from eq . ( [ eq : dens_el ] ) or by using the clebsch - gordan coefficients of @xmath56 ( ref . ) . by summing over the invariants @xmath73 it is easy to check that these invariants provide a decomposition of the spin density matrix @xmath5 . using eq . ( [ eq : mult_mod ] ) we obtain the scalars [ eq : mom_e ] @xmath74 \rho_1 ^ 2 & = & \textstyle \frac{1}{2 } \left({\left\langle{\left . { \frac{1}{2}}\right| } { \frac{1}{2}}\right\rangle } - { \left\langle{\left.{-\frac{1}{2}}\right| } { -\frac{1}{2}}\right\rangle}\right)^2 + 2 \left| { \left\langle{\left . { \frac{1}{2}}\right| } { -\frac{1}{2}}\right\rangle } \right|^2 \ , . \label{eq : e_dip } \end{aligned}\ ] ] apart from a prefactor @xmath75 , the magnitude of the monopole moment , @xmath76 , is equal to the density . ( we remark that by definition @xmath77 is already a positive scalar that transforms according to @xmath78 so that we always have @xmath79 . ) apart from a prefactor @xmath80 , the ( squared ) magnitude of the dipole moment , @xmath81 , is equal to the ( squared ) magnitude of the polarization . due to the rotational invariance implied by @xmath57 , the density matrix is completely characterized by @xmath76 and @xmath81 . consistent with the invariant expansion of the density matrix one can perform an invariant expansion of the hamiltonian @xmath82 of @xmath12 electron systems . in lowest order of @xmath83 and @xmath84 we obtain the tensor operators listed in table [ tab : tensop ] . here @xmath85 , and @xmath86 denotes the symmetrized product of its arguments , e.g. , @xmath87 . we then obtain @xmath88 here @xmath89 is the external potential . the kinetic energy operator with effective mass @xmath90 transforms like a monopole and the zeeman term with effective @xmath91 factor @xmath92 transforms like a dipole . the symbol @xmath93 denotes the bohr magneton . we see here that for @xmath12 systems the @xmath59th invariant in @xmath94 characterizes the dynamics of the @xmath59th invariant of the spin density matrix . however , we want to emphasize that this simple correspondence between the different terms in the hamiltonian ( [ eq : ham_e ] ) and the multipole moments ( [ eq : mom_e ] ) is valid only because we can factorize the eigenfunctions of @xmath82 into an orbital part and a spin part . in general , no such factorization can be performed for @xmath0 hole states . this is the reason why we will find below that hole systems can have an octupole moment even though the corresponding hamiltonian does not contain any octupole term . we now perform the invariant decomposition of the spin density matrix for a system with @xmath0 . according to eq.([eq : add_ang_irep ] ) it contains multipole terms up to @xmath95 . using standard angular momentum algebra @xcite we obtain the invariant decomposition listed in table [ tab : spn32 ] . here @xmath96 , @xmath97 , and @xmath98 are the angular momentum matrices for @xmath0 and @xmath99 . once again , the ( squared ) magnitudes of the multipole moments can readily be calculated by means of eq . ( [ eq : mult_mod ] ) . we obtain [ eq : multpol_32 ] @xmath100 \rho_1 ^ 2 & = & { \textstyle}\frac{1}{20}\left(3 { \left\langle{\left . { \frac{3}{2}}\right| } { \frac{3}{2}}\right\rangle } + { \left\langle{\left . { \frac{1}{2}}\right| } { \frac{1}{2}}\right\rangle } - { \left\langle{\left.{-\frac{1}{2}}\right| } { -\frac{1}{2}}\right\rangle } - 3 { \left\langle{\left.{-\frac{3}{2}}\right| } { -\frac{3}{2}}\right\rangle}\right)^2 + \frac{1}{5 } \left| \sqrt{3 } { \left\langle{\left . { \frac{3}{2}}\right| } { \frac{1}{2}}\right\rangle } + 2 { \left\langle{\left . { \frac{1}{2}}\right| } { -\frac{1}{2}}\right\rangle } + \sqrt{3 } { \left\langle{\left.{-\frac{1}{2}}\right| } { -\frac{3}{2}}\right\rangle } \right|^2 \\[1ex ] \rho_2 ^ 2 & = & { \textstyle}\frac{1}{4}\left({\left\langle{\left . { \frac{3}{2}}\right| } { \frac{3}{2}}\right\rangle } - { \left\langle{\left . { \frac{1}{2}}\right| } { \frac{1}{2}}\right\rangle } - { \left\langle{\left.{-\frac{1}{2}}\right| } { -\frac{1}{2}}\right\rangle } + { \left\langle{\left.{-\frac{3}{2}}\right| } { -\frac{3}{2}}\right\rangle}\right)^2 + \left|{\left\langle{\left . { \frac{3}{2}}\right| } { \frac{1}{2}}\right\rangle } - { \left\langle{\left.{-\frac{1}{2}}\right| } { -\frac{3}{2}}\right\rangle } \right|^2 + \left|{\left\langle{\left . { \frac{3}{2}}\right| } { -\frac{1}{2}}\right\rangle } + { \left\langle{\left . { \frac{1}{2}}\right| } { -\frac{3}{2}}\right\rangle } \right|^2 \\[1ex ] \rho_3 ^ 2 & = & { \textstyle}\frac{1}{20}\left({\left\langle{\left . { \frac{3}{2}}\right| } { \frac{3}{2}}\right\rangle } - 3 { \left\langle{\left . { \frac{1}{2}}\right| } { \frac{1}{2}}\right\rangle } + 3 { \left\langle{\left.{-\frac{1}{2}}\right| } { -\frac{1}{2}}\right\rangle } - { \left\langle{\left.{-\frac{3}{2}}\right| } { -\frac{3}{2}}\right\rangle}\right)^2 + \frac{2}{5 } \left| { \left\langle{\left . { \frac{3}{2}}\right| } { \frac{1}{2}}\right\rangle } - \sqrt{3 } { \left\langle{\left . { \frac{1}{2}}\right| } { -\frac{1}{2}}\right\rangle } + { \left\langle{\left.{-\frac{1}{2}}\right| } { -\frac{3}{2}}\right\rangle } \right|^2 \nonumber \\ & & { \textstyle}+ \left|{\left\langle{\left . { \frac{3}{2}}\right| } { -\frac{1}{2}}\right\rangle } - { \left\langle{\left . { \frac{1}{2}}\right| } { -\frac{3}{2}}\right\rangle } \right|^2 + 2 \left|{\left\langle{\left . { \frac{3}{2}}\right| } { -\frac{3}{2}}\right\rangle } \right|^2 .\end{aligned}\ ] ] apart from a prefactor @xmath80 , the monopole moment @xmath76 is equal to the density , and apart from a prefactor @xmath101 , the squared dipole moment @xmath81 is equal to the squared magnitude of the polarization . @xcite the quadrupole moment @xmath102 and the octupole moment @xmath103 have no equivalent in @xmath12 systems . the magnitude @xmath104 of the quadrupole moment quantifies the effect of hh - lh splitting . if only the diagonal elements @xmath105 of @xmath5 are nonzero , @xmath104 is equal to the difference between the partial hh and lh densities used by enderlein _ @xcite ( apart from a prefactor @xmath80 ) . for @xmath0 systems , a complete characterization of the density matrix by means of scalar quantities requires to take into account additional terms beyond those obtained from eq.([eq : mult_mod ] ) . they characterize , e.g. , the interaction between different multipole moments . it follows from eq.([eq : add_ang_irep ] ) that these scalars are of higher order in @xmath106 . @xmath65 } \begin{array}{c*{3}{@{\hspace{2em}}l } } \hline \hline \mathcal{d}_{\mathfrak{j } } & { \mathfrak{m } } & m_{{\mathfrak{j}}{\mathfrak{m } } } & \rho_{{\mathfrak{j}}{\mathfrak{m } } } \nl \hline \mathcal{d}_{0 } & 0 & \frac{1}{2 } \openone_{4\times 4 } & \frac{1}{2}\left ( { \left\langle{\left . { \frac{3}{2}}\right| } { \frac{3}{2}}\right\rangle } + { \left\langle{\left . { \frac{1}{2}}\right| } { \frac{1}{2}}\right\rangle } + { \left\langle{\left.{-\frac{1}{2}}\right| } { -\frac{1}{2}}\right\rangle } + { \left\langle{\left.{-\frac{3}{2}}\right| } { -\frac{3}{2}}\right\rangle}\right ) \nl \mathcal{d}_{1 } & 1 & -\frac{1}{\sqrt{10 } } j_+ & -\sqrt{\frac{3}{10 } } \big({\left\langle{\left . { \frac{1}{2}}\right| } { \frac{3}{2}}\right\rangle } + \frac{2}{\sqrt{3 } } { \left\langle{\left.{-\frac{1}{2}}\right| } { \frac{1}{2}}\right\rangle } + { \left\langle{\left.{-\frac{3}{2}}\right| } { -\frac{1}{2}}\right\rangle } \big ) \nl & 0 & \frac{1}{\sqrt{5 } } j_z & \frac{1}{2\sqrt{5 } } \left ( 3 { \left\langle{\left . { \frac{3}{2}}\right| } { \frac{3}{2}}\right\rangle } + { \left\langle{\left . { \frac{1}{2}}\right| } { \frac{1}{2}}\right\rangle } - { \left\langle{\left.{-\frac{1}{2}}\right| } { -\frac{1}{2}}\right\rangle } - 3 { \left\langle{\left.{-\frac{3}{2}}\right| } { -\frac{3}{2}}\right\rangle}\right ) \nl \mathcal{d}_{2 } & 2 & \frac{1}{2\sqrt{6 } } j_+^2 & \frac{1}{\sqrt{2 } } \left({\left\langle{\left.{-\frac{1}{2}}\right| } { \frac{3}{2}}\right\rangle } + { \left\langle{\left.{-\frac{3}{2}}\right| } { \frac{1}{2}}\right\rangle } \right ) \nl & 1 & - \frac{1}{\sqrt{6 } } \{j_z , j_+ \ } & \frac{1}{\sqrt{2 } } \left(- { \left\langle{\left . { \frac{1}{2}}\right| } { \frac{3}{2}}\right\rangle } + { \left\langle{\left.{-\frac{3}{2}}\right| } { -\frac{1}{2}}\right\rangle } \right ) \nl & 0 & \frac{1}{6 } \left ( 2j_z^2 - j_x^2 - j_y^2 \right ) & \frac{1}{2 } \left ( { \left\langle{\left . { \frac{3}{2}}\right| } { \frac{3}{2}}\right\rangle } - { \left\langle{\left . { \frac{1}{2}}\right| } { \frac{1}{2}}\right\rangle } - { \left\langle{\left.{-\frac{1}{2}}\right| } { -\frac{1}{2}}\right\rangle } + { \left\langle{\left.{-\frac{3}{2}}\right| } { -\frac{3}{2}}\right\rangle } \right ) \nl \mathcal{d}_{3 } & 3 & - \frac{1}{6 } j_+^3 & - { \left\langle{\left.{-\frac{3}{2}}\right| } { \frac{3}{2}}\right\rangle } \nl & 2 & \frac{1}{\sqrt{6 } } \{j_z , j_+ , j_+ \ } & \frac{1}{\sqrt{2 } } \left({\left\langle{\left.{-\frac{1}{2}}\right| } { \frac{3}{2}}\right\rangle } - { \left\langle{\left.{-\frac{3}{2}}\right| } { \frac{1}{2}}\right\rangle } \right ) \nl & 1 & - \frac{1}{2\sqrt{15 } } \left ( 4\{j_z , j_z , j_+ \ } - \{j_x , j_x , j_+ \ } - \{j_y , j_y , j_+ \ } \right ) & - \frac{1}{\sqrt{5 } } \left({\left\langle{\left . { \frac{1}{2}}\right| } { \frac{3}{2}}\right\rangle } - \sqrt{3 } { \left\langle{\left.{-\frac{1}{2}}\right| } { \frac{1}{2}}\right\rangle } + { \left\langle{\left.{-\frac{3}{2}}\right| } { -\frac{1}{2}}\right\rangle } \right ) \nl & 0 & \frac{1}{3\sqrt{5 } } \left ( 2j_z^3 - 3 \{j_x , j_x , j_z \ } - 3 \{j_y , j_y , j_z \ } \right ) & \frac{1}{2\sqrt{5 } } \left ( { \left\langle{\left . { \frac{3}{2}}\right| } { \frac{3}{2}}\right\rangle } - 3 { \left\langle{\left . { \frac{1}{2}}\right| } { \frac{1}{2}}\right\rangle } + 3 { \left\langle{\left.{-\frac{1}{2}}\right| } { -\frac{1}{2}}\right\rangle } - { \left\langle{\left.{-\frac{3}{2}}\right| } { -\frac{3}{2}}\right\rangle } \right ) \nl \hline\hline \end{array}$ ] the dynamics of @xmath0 hole systems is characterized by the @xmath28 luttinger hamiltonian @xmath107 ( ref . ) . using tables [ tab : tensop ] and [ tab : spn32 ] we express @xmath107 in terms of spherical tensor operators @xcite @xmath108{rcl } \mathcal{h}_{4\times 4 } & = & { \displaystyle}- \frac{\hbar^2 \gamma_1}{m_0 } { \ensuremath{\bm{m}}}_0 \cdot { \ensuremath{\bm{\mathcal{k}}}}_0 - 2\sqrt{5 } \kappa { \mu_\mathrm{b}}{\ensuremath{\bm{m}}}_1 \cdot { \ensuremath{\bm{\mathcal{k}}}}_1 \\[2ex ] & & { \displaystyle}+ \sqrt{6}\frac{\hbar^2}{m_0 } \frac{2\gamma_2 + 3 \gamma_3}{5 } { \ensuremath{\bm{m}}}_2 \cdot { \ensuremath{\bm{\mathcal{k}}}}_2 + \zeta { \ensuremath{\bm{m}}}_3 \cdot { \ensuremath{\bm{\mathcal{k}}}}_3 \\[2ex ] & & { \displaystyle}+ v({{\ensuremath{\bm{r } } } } ) \openone_{4\times 4 } \ : , \end{array } \hspace*{-2em}\ ] ] where @xmath109 , @xmath110 , and @xmath111 are the luttinger parameters and @xmath112 is the isotropic effective @xmath91 factor . we neglect here the small terms with cubic symmetry . they are considered in the numerical calculations discussed below . once again , the magnetic field is acting in spin space as a dipole field ( the operator @xmath113 ) . obviously , the quadrupole and the octupole term in @xmath107 have no equivalent in @xmath12 systems . the quadrupole operator @xmath114 is responsible for the hh - lh splitting . @xcite we note that in bulk material , uniaxial or biaxial strain also gives rise to a splitting of hh and lh states because analogous to @xmath114 the strain tensor contains a term @xmath115 that transforms according to @xmath116 so that we get another invariant @xmath117 ( ref . ) . in eq.([eq : lutt_spher ] ) , the symbol @xmath118 denotes the prefactor of the octupole term . starting from the kane model we obtain in fourth order perturbation theory @xcite @xmath119 where @xmath120 is kane s momentum matrix element , @xmath121 is the fundamental gap , and @xmath122 is the spin - orbit split - off gap . if we insert typical values for @xmath120 , @xmath121 , and @xmath122 it can be seen that @xmath118 is rather small so that the octupole term in @xmath107 can safely be neglected . however , we show below that , nonetheless , the spin density matrix of hole systems can have a large octupole moment . as an example we consider next the spin density matrix of quasi 2d electron and hh systems in the presence of an in - plane magnetic field @xmath1 . for @xmath12 electron systems at @xmath123 we can choose the eigenstates independent of @xmath1 @xmath124 using eq . ( [ eq : e_dip ] ) we then obtain for the dipole moment normalized with respect to the total 2d density @xmath125 { \displaystyle}\frac{1}{2 } & b_\| \ge b_p \end{array } \right.\ ] ] where we have assumed that only the lowest orbital subband @xmath126 is occupied , @xmath32 denotes the total density in the 2d system , and @xmath127 is the magnetic field at which the system becomes fully spin - polarized . @xcite in eq . ( [ eq : envelope_qw_e ] ) the index @xmath8 at the spinor indicates that we have chosen the spin quantization axis in @xmath8 direction perpendicular to @xmath1 . of course , the dipole moment @xmath81 may not depend on this seemingly inconvenient choice . it is easy to check that we get the same result if the spin quantization axis is chosen parallel to @xmath1 . next we consider a 2d hh system in the presence of an in - plane magnetic field @xmath1 . we write the degenerate hh eigenstates at @xmath26 in the form @xcite @xmath128 here we have assumed that the luttinger hamiltonian ( [ eq : lutt_spher ] ) is expressed in a basis of @xmath0 angular momentum eigenfunctions in the order @xmath129 , @xmath80 , @xmath130 , and @xmath131 ; and the index @xmath8 at the spinor indicates that the quantization axis of these basis functions is perpendicular to the 2d plane . for simplicity we have neglected hh - lh mixing due to a nonzero in - plane wave vector @xmath132 and we also neglected the @xmath132 dependence of the envelope functions @xmath133 . these aspects are fully taken into account in the numerical calculations discussed below . a finite in - plane magnetic field @xmath1 gives rise to a mixing of hh and lh states . in first order degenerate perturbation theory we obtain the following expressions @xcite for the quasi 2d hh states in an in - plane field @xmath1 @xmath134 where [ eq : fak ] @xmath135 here @xmath136 ( @xmath137 ) is the energy of the @xmath18th hh ( lh ) subband at @xmath138 , and we have used the gauge @xmath139 . the expression for @xmath29 in ref . differs slightly from our eq.([eq : fak_g ] ) because we have used here the spherical approximation ( [ eq : lutt_spher ] ) of @xmath107 . as discussed in ref . we can often treat the hh - lh mixing characterized by @xmath140 and @xmath29 as a small parameter even if @xmath1 is so large that only the majority spin subband is occupied . this is a consequence of the frozen angular momentum of hh states in quasi 2d systems . using eq . ( [ eq : multpol_32 ] ) we then obtain for the multipole moments normalized with respect to the total 2d density [ eq : multpol_32_qw ] @xmath141 \tilde{\rho}_2 ^ 2 & = & \frac{1}{4 } + 3 \mathcal{k}^2 \left[(\delta n)^2 - 1 \right ] \\[1ex ] \tilde{\rho}_3 ^ 2 & = & \frac{(\delta n)^2}{2 } - \mathcal{k}^2 \left[3 ( \delta n)^2 - \frac{6}{5 } \right ] - \frac{18}{5 } \mathcal{g } \mathcal{k } \ , \delta n \nonumber \\ & & \label{eq : octupol_32_qw } - \frac{9}{5 } \mathcal{g}^2 \ , ( \delta n)^2 , \end{aligned}\ ] ] where @xmath142 denotes the normalized difference between the 2d densities @xmath143 in the majority and the minority spin subband ( with @xmath144 ) , and we have assumed that only the lowest hh subband @xmath126 is occupied ( which is usually the case for 2d hole systems ) . we note that prior to deriving the multipole moments ( [ eq : multpol_32_qw ] ) one must normalize the perturbed wave functions ( [ eq : envelope_qw_b ] ) . hh - lh mixing gives rise to a dipole moment @xmath145 proportional to the small parameters @xmath140 and @xmath29 ( ref . ) . in lowest ( i.e. , zeroth ) order of the hh - lh mixing , the quadrupole moment @xmath146 is equal to @xmath147 , i.e. , @xmath146 is essentially independent of @xmath1 , which is a consequence of hh - lh splitting . finally , we also get a substantial octupole moment even though we assumed @xmath148 in the luttinger hamiltonian ( [ eq : lutt_spher ] ) . if the minority spin subband is completely depopulated the normalized octupole moment equals @xmath80 in lowest ( i.e. , zeroth ) order of the hh - lh mixing . we note that the zeeman energy splitting of the hh states in the presence of an in - plane magnetic field @xmath1 is proportional to @xmath149 ( ref . ) . neglecting the nonparabolicity of the hh energy dispersion we thus have @xmath150 . it is interesting to trace back the physical origin of the multipole moments @xmath104 and @xmath151 . in spite of the fact that we do not have a simple correspondence between the multipoles in the luttinger hamiltonian [ eq . ( [ eq : lutt_spher ] ) ] and the multipole moments @xmath152 of the spin density matrix [ eq . ( [ eq : multpol_32_qw ] ) ] , we get nonzero multipole moments @xmath104 and @xmath151 only if the hamiltonian ( [ eq : lutt_spher ] ) contains a quadrupole term @xmath153 . if the quadrupole term in eq . ( [ eq : lutt_spher ] ) were zero the hh eigenstates ( [ eq : envelope_qw ] ) at @xmath26 were degenerate with the corresponding lh states . similar to the electron states ( [ eq : envelope_qw_e ] ) these four degenerate states could be factorized into a scalar orbital part times a spinor that depends only on @xmath154 , and we would have @xmath155 . to the best of our knowledge , the quadrupole term @xmath156 is a unique feature of the hamiltonian for the extended valence band states in semiconductors . @xcite it is instructive to check that the scalars @xmath157 in eq . ( [ eq : multpol_32_qw ] ) are indeed independent of the quantization axis @xmath158 of angular momentum used in eqs.([eq : envelope_qw ] ) and ( [ eq : envelope_qw_b ] ) . if @xmath158 is chosen parallel to @xmath159 eq.([eq : envelope_qw_b ] ) must be replaced by @xcite [ eq : envelope_qw_bb ] @xmath160 in this basis the dominant spinor components of the hh states have @xmath161 . nonetheless , we obtain the multipole moments ( [ eq : multpol_32_qw ] ) . to confirm the qualitative results in eq . ( [ eq : multpol_32_qw ] ) we show in fig . [ fig : mumo ] the self - consistently calculated multipole moments @xmath157 as a function of the in - plane magnetic field @xmath1 for a symmetric ( 100 ) gaas - al@xmath162ga@xmath163as quantum well with hole density @xmath164 @xmath165 and well width @xmath166 . the numerical calculations follow ref . . we evaluate eq . ( [ eq : spin - dens - mat ] ) by means of analytic quadratic brillouin zone integration . @xcite the solid lines in fig . [ fig : mumo ] have been obtained using the spherical approximation ( [ eq : lutt_spher ] ) of the luttinger hamiltonian @xmath107 . however , we have neglected the small octupole term in eq . ( [ eq : lutt_spher ] ) . for comparison we also show the results based on the @xmath167 kane hamiltonian @xmath168 which includes implicitly the octupole term . the dashed lines in fig . [ fig : mumo ] are based on the spherical approximation of @xmath168 while the dotted lines also take into account the cubic terms . @xcite ( a ) normalized spin subband density @xmath169 of the hh minority spin subband and ( b - d ) normalized multipole moments @xmath157 as a function of the in - plane magnetic field @xmath1 calculated self - consistently for a symmetric ( 100 ) gaas - al@xmath162ga@xmath163as quantum well with hole density @xmath170 @xmath165 and well width @xmath166 . the solid lines have been obtained by means of the spherical approximation ( [ eq : lutt_spher ] ) of the luttinger hamiltonian . the dashed ( dotted ) lines correspond to the spherical ( cubic ) approximation of the @xmath167 kane hamiltonian , respectively.,height=566 ] we can see in fig . [ fig : mumo ] that the complete depopulation of the hh minority spin subband due to @xmath1 does not imply full spin polarization of the system . @xcite in fig . [ fig : mumo ] ( solid lines ) , the minority spin subband is completely depopulated at @xmath171 t while @xmath172 . the latter value is much smaller than @xmath173 , the value of @xmath145 in a fully spin - polarized hh system . the zero of @xmath174 at @xmath175 t reflects a sign reversal of the spin polarization vector which is a unique feature of 2d hh systems . @xcite the derivative of @xmath174 is discontinuous at @xmath176 . at @xmath177 the quadrupole moment @xmath146 is slightly smaller than @xmath147 , the value of @xmath146 in a pure hh system . this is a consequence of the @xmath132-induced hh - lh mixing which was fully taken into account in fig . [ fig : mumo ] . for @xmath178 we observe only a small decrease of @xmath146 . this is due to the fact that the hh - lh splitting @xmath179 mev [ i.e. , the denominators in eq . ( [ eq : fak ] ) ] is the largest energy scale in the system so that the hh states have a frozen angular momentum perpendicular to the 2d plane . for comparison , we note that the zeeman energy splitting of the hh subband at @xmath180 is @xmath181 mev . the spin polarization due to @xmath1 is thus a higher - order effect in 2d hh systems . it is remarkable that the octupole moment @xmath182 at @xmath176 is close to @xmath80 , the largest possible value of @xmath182 in a 2d hh system . this value is essentially independent of whether we use @xmath107 without an octupole term or whether we use @xmath168 which includes implicitly the octupole term @xmath183 . for @xmath184 , the octupole moment @xmath185 remains essentially constant , consistent with eq . ( [ eq : octupol_32_qw ] ) . these findings suggest that an in - plane magnetic field @xmath1 provides an efficient tool to study 2d hh systems with a large octupole moment but with a small dipole moment ( i.e. , with a small spin polarization ) . we can obtain a 2d hh system with a large spin polarization but with a small octupole moment if the magnetic field is applied perpendicular to the 2d plane . figure [ fig : mumo ] shows that the cubic terms are only small corrections in @xmath168 . they can safely be neglected in a discussion of the main features in fig . [ fig : mumo ] . this is consistent with the fact that the cubic terms in @xmath168 ( and @xmath107 ) are proportional to @xmath186 which is typically a small parameter . @xcite on the other hand , it was important for the results shown in fig . [ fig : mumo ] that the numerical calculations took into account hh - lh mixing due to a nonzero in - plane wave vector @xmath132 . as the analytical expressions in eq.([eq : multpol_32_qw ] ) neglect this effect , we do not present a direct comparison between eq . ( [ eq : multpol_32_qw ] ) and the numerical results in fig . [ fig : mumo ] . we remark that it is straightforward to include the @xmath132-induced hh - lh mixing in eq.([eq : multpol_32_qw ] ) . however , we do not reproduce here the lengthy formulas . the numerical calculations shown in fig . [ fig : mumo ] have neglected many - particle effects beyond the hartree approximation . these effects will be important for a quantitative analysis of the spin multipole moments in the density regime discussed here . however , to the best of our knowledge an appropriate theory for particles with @xmath0 is currently not available . it represents a formidable task to develop such a theory . we note that the symmetry arguments developed here are not affected by many - particle effects . the multipole expansion of the spin density matrix can thus provide a useful starting point for a more quantitative theory . we expect that the essential features in fig . [ fig : mumo ] will be confirmed by such an investigation . our results can readily be generalized to spherically symmetric systems with arbitrary @xmath13 , both half integer and integer , @xcite as well as to systems characterized by the crystallographic point groups . @xcite as an example , we consider a @xmath12 electron system in a spatial environment with point group @xmath187 . here the @xmath188 spin density matrix is completely characterized by five independent scalars [ eq : dens_c2v ] @xmath189 \rho_1 & = & \textstyle \left| { \left\langle{\left . { \frac{1}{2}}\right| } { -\frac{1}{2}}\right\rangle } + { \left\langle{\left.{-\frac{1}{2}}\right| } { \frac{1}{2}}\right\rangle } \right|^2 \\[0.5ex ] \rho_2 & = & \textstyle \left| { \left\langle{\left . { \frac{1}{2}}\right| } { -\frac{1}{2}}\right\rangle } - { \left\langle{\left.{-\frac{1}{2}}\right| } { \frac{1}{2}}\right\rangle } \right|^2 \\[0.5ex ] \rho_3 & = & \textstyle \left| { \left\langle{\left . { \frac{1}{2}}\right| } { \frac{1}{2}}\right\rangle } - { \left\langle{\left.{-\frac{1}{2}}\right| } { -\frac{1}{2}}\right\rangle } \right|^2 \\[0.5ex ] \rho_4 & = & \textstyle \left ( { \left\langle{\left . { \frac{1}{2}}\right| } { \frac{1}{2}}\right\rangle } - { \left\langle{\left.{-\frac{1}{2}}\right| } { -\frac{1}{2}}\right\rangle } \right ) \left ( { \left\langle{\left . { \frac{1}{2}}\right| } { -\frac{1}{2}}\right\rangle } + { \left\langle{\left.{-\frac{1}{2}}\right| } { \frac{1}{2}}\right\rangle } \right ) \nonumber \\ & & \textstyle \times \left ( { \left\langle{\left . { \frac{1}{2}}\right| } { -\frac{1}{2}}\right\rangle } - { \left\langle{\left.{-\frac{1}{2}}\right| } { \frac{1}{2}}\right\rangle } \right ) . \end{aligned}\ ] ] each term in eq . ( [ eq : dens_c2v ] ) has a clear physical interpretation . once again , the scalar @xmath76 is the charge density . the scalars @xmath190 , @xmath104 , and @xmath151 describe the magnitude of the spin polarization in @xmath191 , @xmath192 , and @xmath8 direction , respectively . these quantities must be distinguished from each other in a system with point group @xmath193 . finally , @xmath194 describes the relative sign of the spin polarization in @xmath191 , @xmath192 , and @xmath8 direction . the five scalars @xmath152 thus form a decomposition of the four independent parameters of a hermitian @xmath4 matrix into their magnitudes and their relative signs . for the explicit formulas in eq . ( [ eq : dens_c2v ] ) we have assumed that the angular momentum quantization axis of the basis functions is parallel to the symmetry axis of the point group @xmath193 . we emphasize that the magnitude of the scalars @xmath152 in eq.([eq : dens_c2v ] ) does not depend on this choice . finally , we want to briefly address the question of how one can measure the multipole moments of the spin density matrix of hole systems . the polarization of @xmath12 electron systems can be probed by measuring the hyperfine interaction between the electrons and the atomic nuclei . electrons in the conduction band of common semiconductors like gaas originate in @xmath195-like atomic orbitals . therefore , the electrons have a large probability density at the atomic nuclei so that the hyperfine interaction is large . the @xmath0 hole states in the valence band , on the other hand , originate in @xmath196-like atomic orbitals which have a vanishing probability density at the atomic nuclei . the hyperfine interaction is therefore much weaker @xcite so that it will be difficult to use this technique to measure the polarization of @xmath0 hole states . indirect evidence for the unusual features of the spin density matrix of hole systems can be obtained by varying the confinement of quasi 2d hh systems perpendicular to the 2d plane . this will change the hh - lh splitting @xmath197 in the denominators of eq . ( [ eq : fak ] ) . here our results are in good qualitative agreement with the trends observed in recent experiments . @xcite a detailed comparison with experimental results will be the subject of a future publication .

brownian motion of free particles on curved surfaces is studied by means of the langevin equation written in riemann normal coordinates . in the diffusive regime we find the same physical behavior as the one described by the diffusion equation on curved manifolds [ j. stat . mech . ( 2010 ) p08006 ] . therefore , we use the latter in order to analytically investigate the whole diffusive dynamics in compact geometries , namely , the circle and the sphere . our findings are corroborated by means of brownian dynamics computer simulations based on a heuristic adaptation of the ermak - mccammon algorithm to the langevin equation along the curves , as well as on the standard algorithm , but for particles subjected to an external harmonic potential , deep and narrow , that possesses a mexican hat " shape , whose minima define the desired surface . the short - time diffusive dynamics is found to occur on the tangential plane . besides , at long times and compact geometries , the mean - square displacement moves towards a saturation value given only by the geometrical properties of the surface .

1211.5799

during last decades , the interest in diffusive processes has grown tremendously because of their universality in diverse physical areas ; ranging from condensed matter to elementary particle physics and gravitation @xcite . in particular , it has emerged an intense activity in the study of brownian motion in curved manifolds motivated by problems coming from biophysics @xcite . for instance , the lateral diffusion of proteins and lipids occurring inside cell membranes are interesting and complex since they determine the flux of nutrients between the cell and its exterior affecting , in consequence , the cell functionality @xcite . from the theoretical point of view , it is difficult to describe this phenomenon because the interactions with the remaining components of the membrane and the protein finite - size effects @xcite . besides , there are also curvature contributions @xcite and thermal fluctuations that produce shape undulations @xcite coupled to the lateral motion @xcite . and on top of that , protein diffusion is also affected by changing membrane thickness @xcite . the simplest approach to study this problem is to consider the brownian motion of a punctual particle on a frozen two - dimensional regular surface that represents the membrane @xcite . in this approximation , both thermal shape fluctuations and finite - size effects have not been taken into account explicitly but as an effective result reflected in the parameters of the model . as discussed below and although the results presented here are quite general for the brownian motion on a manifold , this work is primarily motivated by the aforementioned transport phenomena . although the understanding of brownian motion was established a century ago , it is noteworthy to mention that the study of brownian motion on manifolds started three decades ago . since the seminal work of n. g. van kampen @xcite , the fundamental equations of brownian motion on manifolds were established and the manifolds introduced , like in classical mechanics , as a result of the appropriate canonical transformations involved in the system with certain holonomic constraints . manifolds also appeared naturally in the dynamics of polymers in solution @xcite , when the polymer is modeled by means of the theory of brownian motion with constraints ( see , e.g. , @xcite , and more recently @xcite for a review ) . in this case , the number of constraints that take into account the bonds between monomers establishes the dimension of the manifold . however , in a real situation , the rigid constraints represent idealizations of stiff potentials that limit the motion in a certain spatial domain @xcite , whereas by including either thermal or statistical fluctuations the rigid constraints , in general , will no longer represent idealizations of elastic potentials @xcite . nonetheless , albeit the fluctuations are present , the rigid constraints may emulate real molecular bonds as illustrative toy models and , in some cases , realistic models can be also represented through a coarse - grain or large - scale description , like in the rigid - rod and wormlike chain models @xcite . in addition , brownian dynamics on curved manifolds becomes a natural framework to study diffusion on crystals with topological disorder , where the torsion of the manifold is crucial to quantify the degree of disorder @xcite . furthermore , smerlak has found that the eckart s heat flux in general relativity and the generalization of the tolman - ehrenfest relation to non - equilibrium stationary states , as well as gravitational corrections , can be best understood through the mean - square displacements of hypothetic particles in static isotropic curved space - times @xcite . although the diffusion equation is suited to study the brownian motion of free particles on curved surfaces , a more complete description is provided by the langevin equation . the latter is based on the newton s equation of motion but including a rapidly fluctuating force , gaussian distributed , representing the interaction among the particle and the solvent . it is well - known that in euclidean open spaces the mean - square displacement ( msd ) calculated from the langevin equation reproduces the standard einstein kinematical relation . in this kind of spaces , both langevin and diffusion equations describe the same dynamical behavior at the diffusive time regime , i.e. , @xmath0 , where @xmath1 is the friction coefficient of the solvent , @xmath2 is the particle mass and @xmath3 the momentum relaxation time @xcite . in a curved space , one might ask whether this property is preserved and , in general , what is the dependence of the dynamics on the geometry of the space . these points have been recently addressed by m. polettini @xcite whom posed a langevin equation , derived by a gauge principle and proved that its overdamped limit corresponds to the diffusion equation in curved manifolds . we here discuss the aforementioned points , but taking the damped ( @xmath4 ) and overdamped ( @xmath5 ) limits in the msd and look at its behaviour as a function on the geometrical properties of the space . in this work , we write down the langevin equation for manifolds following the same method introduced by e. j. hinch @xcite . the starting point is the newton s equation for free particles in a @xmath6-dimensional hypersurface @xmath7 . free means here that particles do not interact between each other and non external force is acting on them . nevertheless , they are restricted to move on @xmath7 . for the local momenta @xmath8 and local coordinates @xmath9 , the resulting langevin equations are , @xmath10 where @xmath11 is the riemannian metric tensor and @xmath12 the christoffel symbols . it turns out that these equations are the same found by kleinert and shabanov @xcite , who discussed its generalization to connections with torsion , as well as those derived in the work of m. polettini using a gauge principle ( invariance under local rotations ) @xcite . besides , the global version of these equations were obtained by e. j. hinch @xcite for the particular case of two monomers , one of them excessively massive , with one constraint . we also find that the msd , up to first order in curvature , calculated from ( [ locall-2 ] ) is given by @xmath13\nonumber\\ & -&\frac{2r_{g}}{3}\left(\tau_{b}d_{0}\right)^{2}\mathcal{j}\left(t/\tau_{b}\right)+\cdots , \label{result1}\end{aligned}\ ] ] where the terms in the square parenthesis are found to be the standard msd for the particle dynamics in the euclidean @xmath14 space and @xmath15 is a non - dimensional function that characterizes the particle dynamics coupled to the curvature ( see below at appendix b for its definition ) . @xmath16 is the ricci scalar curvature and @xmath17 is the geodesic distance of the general riemannian manifold . taking this equation , we are able to investigate the particle dynamics at different time scales : @xmath18 and for @xmath19 . @xmath20 is a characteristic time for the positions and momenta relaxation of the solvent molecules , at which the langevin description is not longer valid , and @xmath21 is the time scale when curvature effects become evident . it is shown that equation ( [ msd0 - 1 ] ) reproduces the same leading curvature effects in the diffusion regime as in ref . @xcite , which is based on the diffusion equation on curved manifolds . the geometrical properties become evident at times @xmath22 much more longer than @xmath21 . when this happens , the system reaches the thermodynamical equilibrium . in this regime , we reproduce the free - particle dynamics based on the diffusion equation on curved manifolds . the latter is explicitly compared with our computer simulations . using the well - known result that for compact supports ( @xmath23 ) the spectra of the laplace - beltrami operator , @xmath24 , is discrete . thus , it is easy to find that the expectation value for any observable @xmath25 in the overdamped limit is , @xmath26 where @xmath27 is the first non - zero eigenvalue of @xmath24 and @xmath28 is the volume element of the riemannian geometry @xcite . it is remarkable that the leading term obtained in this way allows us to determine the steady spacial density @xmath29 where @xmath30 and @xmath31 the volume of @xmath7 . this density ( see appendix d ) is consistent with the original calculation performed by kramers @xcite and recently discussed in @xcite . we explicitly analyze the dynamics of particles confined along a circle , as well as on a sphere . we test equations ( [ msd0 - 1 ] ) and ( [ obev1 ] ) by means of brownian dynamics computer simulations based on an heuristic adaptation of the ermak - mccammon algorithm @xcite to the langevin equation along curves , as well as on the standard algorithm . in the first case , which is here only applied to the circle , the particles are allowed to move in any direction with equal probability , but the geodesic distances they travel are gaussian randomly distributed . in the second case , the particles are subjected under the action of a spring - like force field in the @xmath32 ( @xmath33)-dimensional euclidean space , where the corresponding potential , with a mexican hat " shape , has its minima at the same points of the circle ( sphere ) . in the limit case of very stiff springs , we get the same results from both numerical routes , and analytical one , as we will see further below . we should mention that the inclusion of a spring - like potential to reproduce the holonomic constrain is a controversial issue because the agreement between theory and simulations establishes a clear example where the particles dynamics with lagrange constraints is equivalent with that using the stiff elastic potential even in systems with fluctuations . moreover , we should point out that this is not in contradiction with the work done by e. j. hinch @xcite and kampen and lodder @xcite . in particular , it is shown that the single canonical partition function using the stiff potential posed is the same for the single canonical partition function on the circle ( sphere ) in the limit of very stiff springs field , as far as the spring - like constant @xmath34 scales with the square of temperature . after the introduction , the manuscript is organized as follows . in section ii we present the langevin equation for curved manifolds , written in both global and local coordinates . in addition , we study the curvature effects on the msd at the following time regimes : @xmath18 and @xmath19 . in section iii we study the particle dynamics on the geometrical regime ( @xmath35 ) by means of the diffusion equation on curved manifolds . in section iv we explicitly compare the predictions for particles restricted to move along a circle and on a sphere with brownian dynamics computer simulations . finally , in section v we summarize some concluding remarks and perspectives of our work . we now specify the basis of langevin dynamics formalism following the method introduced by e. j. hinch @xcite . it is defined over an euclidean hypersurface @xmath36 , which is represented as the points @xmath37 such that @xmath38 . the langevin equation needs to include an holonomic constraint in order to bound a point particle on @xmath7 . let us denote the momentum @xmath39 , where @xmath40 is the tangent space at the point @xmath41 , i.e , the position @xmath42 of the particle . from a classical mechanics point of view , the addition of the term @xmath43 to the free - particle lagrangian allows us to impose an holonomic constraint on @xmath7 . indeed , the resulting equation of motion is @xmath44 and the required constraint is @xmath45 . we should remark that @xmath46 relaxes the constraint . then , for the langevin equation defined on @xmath7 , we simply include the previous constraint , a friction term and a stochastic force @xmath47 @xmath48 the second term of the right - hand side of equation ( [ leq1 ] ) represents the force caused by the holonomic constraint . the stochastic force is chosen such that it satisfies the standard fluctuation - dissipation relations @xmath49 where @xmath50 stands for the average in the _ ensemble _ of forces gaussian distributed over @xmath51 space ( see appendix a ) . remark that @xmath51 is a copy of @xmath52 , i.e. , the ensemble is given by all possible configurations of forces belonging to @xmath51 . the stochastic forces can be treated as vector fields in one dimension in the same spirit that zinn - justine introduced them in @xcite . the lagrange multiplier @xmath53 can be obtained using the constraint ( [ geq3 ] ) as follows . a time derivative on this constraint implies that @xmath54 where @xmath55 represents derivations in the space @xmath51 . since the momentum @xmath56 , then @xmath57 is normal to the tangent space . thus , the normal vector to the surface , i.e. , normal to @xmath58 , is given by @xmath59 . second derivative on equation ( [ constraint1 ] ) gives @xmath60 with @xmath61 . now , we get @xmath53 by equating ( [ constraint2 ] ) and the normal projection of ( [ leq1 ] ) . then , @xmath62 . therefore , the langevin equation involves a non - linear term proportional to a second power in momenta , @xmath63 and a projector , @xmath64 , that maps a vector @xmath65 into the tangent space . we point out that the @xmath66 matrix encodes the surface geometry . for instance , the constraint @xmath67 defines a plane in euclidean space , where @xmath68 is a constant vector and @xmath69 a real number . in this particular case , the @xmath66 matrix is zero and the normal vector of the surface is constant , @xmath70 , as it is required for a planar geometry . in the case of a sphere of radius @xmath71 , we have @xmath72 and the normal vector satisfies @xmath73 ; the matrix @xmath66 is given by @xmath74 . we should remark that the way in which the constraint affects the fluctuating force is through the projector @xmath75 . in other words , although there is a distribution of forces in @xmath51 for each point of the manifold , the langevin equation ( [ langp ] ) takes into account just those forces tangent to @xmath7 through the projector @xmath75 . it is also remarkable that for the constrained dynamics , for instance in a numerical routine , the fluctuating forces can be implemented in the same way as it is done for the three - dimensional euclidean spaces . also , one has to note that the quadratic term in the momentum is not a surprise since the left - hand side of equation ( [ langp ] ) corresponds to the ordinary kinetic term for a particle over a hypersurface . this means that the langevin equation reduces to the geodesic equation when both the friction and the stochastic force vanish together . this will be clarified further below when we write down the equation in local coordinates . we also have to mention that this equation is a particular case of a more general equation derived first by e. j. hinch @xcite within the context of polymers in solution for the case of two monomers , one of them excessively massive , with a single constraint . in addition , this global description is the natural starting point to introduce ambient interactions , where the extrinsic geometry may play a crucial role . it is also important to mention that constraints are merely the result of elastic forces excerted by connecting strings or rods , or other devices by which the free motion is hindered " @xcite and it would not be the exception for integral proteins or lipids in plasma membranes . thus , it is natural to ask whether the lagrange constraints are idealizations of elastic potentials when langevin - type of forces are present . to answer properly this question it would be necessary a careful analysis and it is out from the scope of the paper . however , following the analysis by n. g. van kampen and j. j. lodder @xcite one can conclude that a constraint system , with langevin - type of forces , could be the limiting case of an equivalent stiff system provided , minimally , that these langevin rapidly fluctuating forces , as well as all the remainder external forces , act upon the particle during a short - time @xmath76 with the requirement that @xmath77 , where @xmath78 is the stiffness parameter . for instance , one can choose @xmath20 i.e. , mean collision time of the solvent molecules , for @xmath76 . in general , as it is observed by e. j. hinch @xcite , it is necessary to introduce an extra _ pseudo - corrective force _ in order to convert the brownian motion of a constrained system into an equivalent very stiff system . in section iv , we test equation ( [ result1 ] ) , which is a consequence of the constrained langevin equation , in the cases of a sphere and a circle using the ermak - mccammon algorithm implemented by a particle immersed in stiff elastic potential . it will be proved that , in these particular cases , the constrained system is equivalent to that of very stiff potential . we now provide a description in local coordinates of the langevin equation ( [ langp ] ) . in local coordinates a hypersurface is parametrized by the mapping @xmath79 , where a particular point in @xmath7 is given by @xmath80 , being @xmath9 the local coordinates ( @xmath81 ) . in such coordinates , we have @xmath82 , @xmath83 , and @xmath84 , where @xmath85 is the local momentum and @xmath86 the tangent vectors ( note that @xmath87 ) . thus , the first derivative of the momentum is given by @xmath88 where @xmath89 . the partial derivative @xmath90 can be calculated using the weingarten - gauss equations @xmath91 , where @xmath92 are the components of the second fundamental form @xcite . by using these equations in the momentum time derivative we obtain @xmath93 the local coordinates version of the langevin equation can be straightforwardly obtained by substituting equation ( [ 1st ] ) into equation ( [ langp ] ) . hence , the tangent projection takes the form , @xmath94 while the normal projection is given by @xmath95 equations in ( [ locall2 ] ) are the local version of the langevin equation ( [ langp ] ) . they are the same derived by kleinert and shavanov who discussed the case of manifolds with torsion , see e.g. , @xcite . the same equations were also obtained by m. polettini from the local rotational invariance of wiener increments @xcite . this gauge invariance is also noted in the distribution of the forces ( [ distribution ] ) . as we mentioned above , the quadratic contribution in momentum is just the geodesic contribution . the normal projection ( [ normalr ] ) provides a geometrical identity that allows us to derive the extrinsic curvature in terms of the @xmath66 matrix . this identity is not casual ; it is actually the same found at the level set formulation of differential geometry @xcite . regarding the fluctuation - dissipation relations , the stochastic forces satisfy the following properties , @xmath96 where @xmath97 is the two - dimensional kronecker s delta . these relations are equivalent to their global version ( see appendix a ) . based on equation ( [ locall2 ] ) , it is clear that the particle dynamics does not depend on the extrinsic properties of the geometry . this means that the dynamics on a hypersurface can be studied in a riemannian geometry ; this is what we do from now on . we are mainly interested on the diffusion mechanisms in the weak curvature regime . let us recall that if @xmath98 is a local neighborhood of @xmath7 , the map @xmath99 is a local diffeomorphism @xcite , then @xmath100 . this implies that in a local neighborhood , we should have the same particle dynamics as found in planar spaces ( see , e.g. , ref . @xcite for the @xmath101 case ) . thus , it makes sense to study curvature effects around the euclidean solution . then , we first review the particle dynamics on the euclidean geometry @xmath102 , i.e. , when the curvature is zero , and , second , we expand the euclidean solution in order to study the leading curvature effects on the particle dynamics over the surface . in the euclidean geometry , both the global and local descriptions are the same ; the euclidean metric is simply @xmath104 and the chrystoffel symbols are zero . in this case , the langevin dynamics formalism reduces to the well - known standard equations @xcite @xmath105 and their solution can be written as @xcite , @xmath106 averaging equations ( [ pplanar ] ) over the ensemble of stochastic forces , one easily obtains @xmath107 we observe that the mean momentum decreases exponentially with time ( with the decaying time scale @xmath108 ) and the particle position is shifted by @xmath109 at long - times . we now consider for simplicity that @xmath110 and @xmath111 . other physical quantities of interest are the mean quadratic momentum , i.e. , @xmath112 , and the mean square displacement ( msd ) , @xmath113 . in order to calculate both , it is useful to find the temporal correlation function between two momenta , @xmath114 and @xmath115 , at times @xmath22 and @xmath116 , given by ( see appendix b for further details ) @xmath117.\end{aligned}\ ] ] using previous equation , it is straightforward to obtain the mean quadratic momentum : @xmath118 proceeding along the same lines , one can straightforwardly derive the msd : @xmath119 . \label{planesol}\end{aligned}\ ] ] in the diffusive regime , @xmath4 , the average kinetic energy reaches its equilibrium value . this allows us the evaluation of @xmath120 from the equipartition theorem . thus , @xmath121 and @xmath122 , where @xmath123 is the boltzmann constant and @xmath124 the absolute temperature . we also observe that in this time regime the msd reproduces the standard kinematical einstein relation @xmath125 , where @xmath126 is the free - particle diffusion coefficient @xcite . we should point out that the value of @xmath120 is independent of whether the space is curved or not , since it only depends on quantities intrinsic to the fluid , as solvent friction and particle dimension . higher order temporal correlation functions are also useful . in particular , we will see below that the four - point function @xmath127 is necessary in order to obtain the leading curvature corrections . this correlation function can be computed by using the wick s theorem @xcite , @xmath128 we turn now to the derivation of the leading weak curvature effects on the particles dynamics . as we already discussed , the langevin equation is quadratic in the momentum and that contribution is coupled to the particles positions through the chrystoffel symbols . the resulting equations are difficult to solve analytically , among other reasons because the left - hand side of equation ( [ locall1 ] ) involves a temporal derivative of the metric . using @xmath85 ( @xmath129 is independent of the metric ) , the local langevin equation allows us to obtain the following expressions , @xmath130 in order to explore curvature effects , we expand equation ( [ sing ] ) around the planar solution ( [ pplanar ] ) . to reach this goal , we use the riemann normal coordinates @xcite . in normal coordinates , we have @xmath131 where @xmath132 are the components of the riemann curvature tensor . using ( [ eqrt ] ) in ( [ sing ] ) one obtains @xmath133 we should notice that the langevin equation in euclidean geometries ( [ euclideang ] ) is recovered when the curvature vanishes . in order to find a solution around the euclidean case ( [ pplanar ] ) , we expand the momentum and position in the following way : @xmath134 and @xmath135 , where @xmath136 and @xmath137 are the solutions for zero curvature , given by ( [ pplanar ] ) . here , we have assumed that @xmath138 and @xmath139 when @xmath140 . if we consider only linear terms in curvature , we obtain the equation for @xmath141 , @xmath142 the second term of the right - hand side does not depend on @xmath143 ; it depends only on time . the integration of equation ( [ postime ] ) is similar to the one in the planar case . the initial condition for @xmath144 is @xmath145 , since @xmath146 satisfies @xmath147 . therefore , the momentum , up to linear terms , in an arbitrary riemannian geometry is given by @xmath148 and the position , up to linear terms as well , takes the form @xmath149 wick s theorem allows us to determine the temporal correlation functions of @xmath150 . therefore , we have found that the odd correlations vanish , as in the case of the mean values of the momentum and position : @xmath151 and @xmath152 . this means that there is not preferential points on the surface and the mean values are independent of the geometry . this result may change however for non - zero initial conditions . up to linear terms in the curvature , we obtain the following expectation value for @xmath153 @xmath154 where @xmath155 is the same as in equation ( [ planesol ] ) , @xmath156 and @xmath157 the quantity @xmath158 captures the dynamical contribution that appears in the weak curvature regime . in addition , the four - point correlation function @xmath159 is defined according to equation ( [ g4 ] ) ; it is built by the products of two - point correlation functions and each of them carries a kronecker s delta . hence , using the symmetries of the riemann tensor , the msd reduces to @xmath160 where @xmath16 is the ricci scalar curvature and @xmath161 equation ( [ jexp ] ) can be straightforwardly integrated ( see appendix b ) . equation ( [ result ] ) represents the msd ( geodesic mean square displacement ) in the weak curvature regime . as we can appreciate from equation ( [ jexp ] ) , the time scale @xmath108 defines two time regimes : the one with @xmath162 ( but very much larger than @xmath20 ) or the ballistic regime , and the one with @xmath4 called the diffusive regime . in the first case , the msd is given by @xmath163 the cubic term is the ordinary contribution to the ballistic regime when the initial condition is @xmath110 ( it becomes quadratic in @xmath22 for non - zero initial conditions @xcite ) . the next curvature contribution is of order @xmath164 ; typically negligible unless there is a region of very high curvature . in the diffusive regime , @xmath4 , the function @xmath165 reduces to @xmath166 . therefore , the msd becomes @xmath167 this result is the same found by one of us @xcite by means of the diffusion equation on curved manifolds . the msd shows a deviation from the planar result due to curvature effects . furthermore , equation ( [ msdcurv ] ) also shows the raise of two different diffusive regimes : the one with @xmath19 , and the overdamped regime , also called geometric regime , @xmath35 . here , @xmath168 stands for the time thereafter the curvature effects become dominant and it is the regime when the equilibrium is reached . this result is a confirmation that the langevin equation describes the same dynamics of the diffusion equation on curved manifolds in the diffusive regime . it is noteworthy to mention that this result has been recently obtained , using alternative methods , by m. polettini @xcite . it is important to mention that in the planar case , i.e. , @xmath169 , the particle can not feel any effect associated with the geometry ( @xmath21 is never reached , then it grows towards infinity ) . additionally , we should emphasize that in the particular case of @xmath170 the msd may exhibit deviations from the planar result that can not be associated to @xmath16 , since the gaussian curvature of lines is zero . in fact , as we will see further below , those effects are associated with the finite - size of the phase space . from now on , we use the fact that langevin equation and diffusion equation on curved manifolds describe the same dynamics in the diffusive and geometric regime . in the following section , we explicitly discuss some properties of the diffusive motion of the particles along a circle , @xmath171 , and on a sphere , @xmath172 . we now choose the diffusion equation in order to study the geometric regime ( @xmath173 ) in the manifolds @xmath171 and @xmath172 ( for a discussion on the diffusion on arbitrary hyperspheres see , for example , ref . @xcite ) . the diffusion equation on curved manifolds can be written as , @xmath174 where @xmath175 is the probability of finding the diffusing particles in the volume element @xmath176 , given that they began to move at @xmath177 . the probability density distribution @xmath178 is normalized with respect to the volume @xmath31 of the manifold and @xmath179 is the free - particle diffusion coefficient . the operator @xmath24 , called the laplace - beltrami operator , is defined by @xmath180 with @xmath181 and @xmath182 is a scalar function . the geometry is coupled to the brownian motion through the metric . it is clear that @xmath178 reaches a constant value when the system is under equilibrium conditions , i.e. , @xmath183 . the diffusion equation ( [ difeq ] ) is the same as the heat kernel equation and it has a lot of applications in the context of field theories on curved spaces @xcite . the expectation value of a scalar function @xmath25 defined on the manifold is given in the standard fashion , i.e. , @xmath184 and @xmath185 depends on the initial point @xmath177 . the characteristics of observables in manifolds are related with the particular structure of @xmath178 . besides , the probability density distribution @xmath178 can be determined by solving the eigenvalue problem @xmath186 , where @xmath187 is the eigenvalue corresponding to the eigenfunction @xmath188 . in addition , it is known that for compact manifolds , the spectra of @xmath24 is discrete and it can be written in a growing sequence @xmath189 , where @xmath190 @xcite . we also have a sequence of orthogonal eigenfunction @xmath191 in @xmath192 ( square - integrated functions of @xmath7 ) . in this sense , the probability density distribution can be formally written as @xcite @xmath193 with @xmath194 being the complex conjugate of @xmath188 . we note that degeneracy of eigenvalues is explicitly considered in the sum . now , let us consider an arbitrary observable @xmath25 . its dynamical behavior can be obtained using the formal expression for @xmath178 . the expectation value @xmath185 has a generic form ; its structure around the geometric regime is determined by the smallest eigenvalues . then , it can be written as follows : @xmath195 where @xmath196 ( it is also convenient to define @xmath197 ) . it is remarkable that the leading term obtained in this way allows us to determine the steady spacial density @xmath198 this is also consistent with the original calculation by kramers @xcite and recently discussed in @xcite . we can easily obtain some properties of any observable by looking at the particular form of equation ( [ obev ] ) . for example , at long times the expectation value @xmath185 becomes @xmath199 as a consequence of the finite size of the space . in physical terms , every observable that depends on the position will remain fixed , on average , and its distribution does not longer evolve with time . the quantity @xmath199 is the geometrical average of @xmath25 ; this is the reason we called this regime the geometric regime . although counterintuitive , the values of the observables do not depend on the temperature for @xmath200 ; it is only a function of the surface geometry . the value @xmath199 is also the mean - value in the equilibrium regime . this result is , indeed , the generalization to curved space of a classical ideal gas in the three - dimensional euclidean space @xmath101 . brownian motion on the circle represents , after the motion on the straight line , the simplest example where there is a clear manifestation of the geometrical effects on the particle dynamics , but it is also the most fundamental one , since it is fully described by a single physical variable . it is also relevant for the theoretical and experimental study of single - file diffusion in quasi - one - dimensional interacting systems ( see , e.g. , @xcite and references therein ) . the circle is the mappping @xmath201\to\mathbb{r}^{2}$ ] , where @xmath202 , with @xmath71 being the circle radius . the laplace - beltrami operator in this case takes the form @xmath203 . the eigenfunctions of this operator form the complete orthonormal set @xmath204 in @xmath205 and their corresponding eigenvalues are @xmath206 . in order to study brownian motion on @xmath171 , we choose the following initial and boundary conditions : @xmath207 and @xmath208 . after some simplifications , the explicit solution of the diffusion equation is @xmath209 in this case , the distribution is normalized with the perimeter of the circle , i.e. , @xmath210 , where @xmath211 and @xmath212 . the distribution is also symmetric under the interchange @xmath213 . the first moment , @xmath214 , and the second moment or msd , @xmath215 , of the distribution can be straightforwardly evaluated . the former is zero , since the distribution is an even function , whereas the msd has the form @xmath216 with @xmath217 being the arc - length . on the one hand , the msd given by equation ( [ msdcirculo ] ) reduces to @xmath218 for short times ( @xmath219 ) . on the other hand , at long times ( @xmath220 ) we have @xmath221 . in the geometric regime the dependence is only on the size of the circle . the numerical evaluation of equation ( [ msdcirculo ] ) is shown in figure 1 . as we mentioned previously , although the msd in ( [ msdcirculo ] ) deviates from the planar result , this difference is due to the finite size of the circle and not to curvature effects . we compare the predictions of equation ( [ msd ] ) with computer simulation results in figure 1 . the latter ones will be explained further below . [ fig1 ] ) , and the symbols to the brownian computer simulations results obtained by means of both the standard ermak and mccammon algorithm ( circles ) and its heuristic adaptation to curves ( triangles ) . the error bars of the simulation data are smaller than the size of the symbols . there is no appreciable difference between the results . the straight lines stand for the short and long - time limits as indicated.,title="fig:",width=302 ] we now study the brownian dynamics on the sphere putting special emphasis in the geometric regime . it is noteworthy to mention that several features of this special case have already been studied by several authors @xcite and it was originally used to study the rotational brownian dynamics of rods within the debye theory @xcite , where non - interacting rods can be cast into a diffusion equation on the unit sphere . here , this special case is emphasized in the geometric regime where we use the geodesic distance as the displacement of the particle as in ref . @xcite . in the sphere , the geodesic distance corresponds to a section of one riemann great circle . the geometry of a sphere is encoded into the metric given by @xmath222 where @xmath71 , @xmath223 and @xmath224 are the radius , polar and azimuthal coordinates of the sphere , respectively . the laplace - beltrami operator on the sphere has eigenvalues and eigenvectors given by @xmath225 and @xmath226 with @xmath227 and @xmath228 ; @xmath229 being the standard spherical harmonics . we choose @xmath177 to be on the north pole and take advantage of the rotational invariance . besides , the boundary condition ( [ difeq ] ) is explicitly taken into account . the solution of the diffusion equation is then @xmath230 , \label{eqprob}\end{aligned}\ ] ] where @xmath231 is the legendre polynomial of order @xmath232 . as in the previous case , we look for the information provided by @xmath214 and @xmath233 , but we have now that @xmath234 . by means of the operator method defined in @xcite , it is possible to show that the short - time behavior of the msd is given by equation ( [ msdcurv ] ) with the gaussian curvature of the sphere , @xmath235 . it is interesting to note that the terms in the msd that depend on the gaussian curvature are always negative . this means that curvature effects only contribute to reduce the particle diffusion with time . in the geometric regime , @xmath236 , we obtain from equation ( [ obev ] ) the following expressions , @xmath237 at the beginning the particles move around their initial position , i.e. , the north pole . after a long time , very much larger than @xmath238 , the expectation values @xmath239 and @xmath240 move towards the saturation values @xmath241 and @xmath242 , respectively . the particle has visited all the points on the surface and confinement dominates entirely the diffusive behavior ; the saturation values only depend on the size of the sphere . the behavior of equation ( [ limitcase ] ) is shown in figure 2 . [ fig2 ] ) , and the symbols to the brownian computer simulations results obtained by means of the standard ermak and mccammon algorithm ( circles ) . the error bars of the simulation data are smaller than the size of the symbols . there is no appreciable difference between the results . the straight lines stand for the short and long - time limits as indicated.,title="fig:",width=302 ] the expectation value of any observable @xmath243 on the sphere can be written as @xmath244 where @xmath245 is the projection of @xmath25 along the basis of legendre polynomial . we explicitly show the functional form of @xmath245 in appendix c , for both @xmath246 and @xmath247 . in 1978 , ermak and mccammon introduced a method for simulating the brownian dynamics of particles @xcite . this method , which has been adapted in euclidean coordinates , was derived from the langevin equation and became consistent with the fokker - planck equation . furthermore , such a method can be straightforwardly applied when either hydrodynamic interactions are considered explicitly or external forces act on the particles . this method has been successfully employed to study the structural and dynamic properties of a large variety of complex fluids , i.e. , colloids , polymers , etc . @xcite the algorithm of ermak and mccammon @xcite is given by @xmath248 where @xmath249 is the number of particles , @xmath250 is the inverse of the thermal energy . the hydrodynamic interactions ( hi ) are included through the diffusion tensor @xmath251 , @xmath252 is the total force exerted on the @xmath253-th particle and the index @xmath254 tells us that the variable must be calculated at the beginning in time at every step . the term @xmath255 represents a random displacement with a gaussian distribution function with mean value zero and a covariance matrix given by the elements @xmath256 ; these are the requirements needed to satisfy the fluctuation - dissipation theorem ( [ fluc - dissip ] ) . the indices @xmath257 and @xmath253 run over the particles , and the indices @xmath258 and @xmath259 over the cartesian coordinates . in our case , we do not consider hi and , therefore , @xmath260 , where @xmath179 is again the free - particle diffusion coefficient . with this assumption , the second term in the right - hand side of equation ( [ eermak ] ) disappears and allows us to simplify drastically the calculation of the third and fourth terms of the same side . as we mentioned previously , the algorithm of ermak and mccammon describes the temporal evolution of euclidean variables . however , it can be still used to describe the dynamics of particles on curved surfaces . this can be done by considering an external field that constrains the movement of the particles on the surface . we demand that the force coming from such a field does not contribute to the tangent displacements of the particles , i.e. , this force has to act normal to any point of the desired manifold ( i.e. @xmath171 or @xmath172 ) at any time to guarantee that it does not perform work on the system . then , the simplest vector force - field that satisfies such requirements can be written as @xmath261 where @xmath34 is a coupling constant , whose value is chosen in such a way that the particle displacements in the perpendicular direction to the surface is basically negligible , @xmath71 is a parameter of this force that we identify , here , with the radius of either the circle or the sphere and @xmath262 is a unit normal vector . this force can be thought as a spring - like force that attach the particle to a domain near the surface ; in this sense @xmath34 is a spring - like constant . in the two - dimensional case , this vector field can be explicitly visualized in figure ( 3 ) , where the circle ( solid line ) shows the separation of the plane in two regions defined by the sign of @xmath263 . [ fig41 ] separates the plane in two regions depending on the sign of the vector force - field @xmath264.,title="fig:",width=207 ] + it is also convenient to determine the potential energy associate to this force . this is given by @xmath265 this potential has a mexican hat " shape . in figure ( 4 ) , we plot the potential given by equation ( [ potential ] ) for the 2-dimensional case , where the points that minimize the potential correspond to the manifold ( in this case @xmath171 ) . [ fig31 ] ) has a mexican hat " shape , whose minima are the points of the manifold @xmath171 . the value of the coupling parameter , @xmath34 , controls the width of the well.,title="fig:",width=226 ] before we implement the external force ( [ forcer ] ) in the ermack - mccammon algorithm , let us provide two arguments that will help us to understand why it will reproduce the correct dynamics on either the circle or the sphere in the limit of very stiff potential , i.e. @xmath266 . we analyze this limiting process by choosing the situation within the context of the classical mechanics and statistical mechanics . thus , in the former case , we have a classical system consisting of a particle subjected to the vector force - field given by ( [ forcer ] ) . following n. g. van kampen and j. j. lodder @xcite the motion of the particle will not have rapid vibrations in the normal direction of the surface , as far as the force acting upon the particle , that initiates its movement , vary smoothly during a short - time @xmath76 and when the condition @xmath267 is entirely satisfied . under this assumption the result , as kampen and lodder pointed it out , is that the particle motion will be along the surface ( in this case either @xmath171 or @xmath172 ) . now , by using a statistical mechanics analysis , we perform the calculation of the single canonical partition function of the particle subjected to the external field and we compare it with the corresponding partition function on the sphere . on one hand , the latter partition function is given by @xmath268 where @xmath269 and @xmath270 is the de broglie wavelength ( see the appendix d for a derivation of this equation ) . on the other hand , the partition function for the particle subjected to the potential ( [ potential ] ) , after we integrate out the momenta , is given by @xmath271 where @xmath272 is some adimensional constant that does not change the physics . naively , it is expected that for large values of @xmath34 the only admissible value for @xmath273 should be any position @xmath274 with length @xmath275 leading to the effect of confinement on the sphere . however , this intuition is approximately correct because when temperature increase , the confinement effect would disappear . thus , the only way to maintain this confinement is scaling the value of the coupling constant @xmath34 with temperature in such a way that the particles are maintained on the sphere . indeed , this happens as we show in the following . using polar coordinates and performing several change of variables , last integral can be written as @xmath276 where @xmath277 . last integral can be perfomed exactly in terms of the error function ( see appendix d ) and for large value of @xmath34 it has an asymptotic value that goes to @xmath278 , therefore the partition function for @xmath266 is given by @xmath279 this means that the only way that this partition function converges to the one on the sphere ( @xmath280 ) is such that the coupling @xmath34 grows with temperature as @xmath281 the limit of large value of @xmath34 can be performed for a fixed temperature taking the adimensional constant @xmath282 . therefore , taking this scaling law for the coupling constant @xmath34 , we have that @xmath283 . thus , with these arguments in both classical mechanics and statistical mechanics we are confident that at least in these two extremal situations we can control the confinement effect of the particles on the surface by means of the stiffness parameter . hence , equation ( [ forcer ] ) is incorporated in the standard algorithm for brownian dynamics described in equation ( [ eermak ] ) to analyze the diffusion on the given surface . we should mention that the addition of force ( [ forcer ] ) into equation ( [ eermak ] ) has the same effect on the particle dynamics as the second term of the left - hand side in equation ( [ langp ] ) , i.e. , it only constrains the motion of the particles on the manifold . thus , this kind of trick allows us to study the diffusion on curved surfaces ( at least on @xmath171 and @xmath172 ) through the use of the standard ermak and mccammon algorithm . it is noteworthy to mention that according to e. j. hinch @xcite in order to convert the brownian motion of a rigid system to an equivalent very stiff system we have to add a _ pseudo - corrective _ force , @xmath284 , to the equation of motion . in the case of our interest , if we want to convert the brownian motion of a very stiff system to an equivalent rigid system one would have to apply a force opposite to @xmath284 . in the particular case of the sphere ( similar for the circle ) this force is given by @xmath285 ( see appendix d for details ) . the correction term would be @xmath286 which does not have any contribution in the particular cases of sphere and circle . similar result , is indeed , already found by grassia , hinch , and nitshe for the brownian motion on an ellipse in @xcite . in our brownian dynamics simulations , we have used @xmath287 free particles , a reduced time step @xmath288 and a reduced stiffness parameter @xmath289 . we also use @xmath290 time steps to reduce the statistical uncertainties . thus , the msd from the adaptation of the standard algorithm of ermak and mccammon is shown in figures 1 and 2 for @xmath171 and @xmath172 , respectively . equation ( [ eermak ] ) takes the simple form @xmath291 for free particles , with @xmath292 . in a @xmath6-dimensional euclidian open space this process is equivalent to allow the particles to move in any direction with equal probability , as long as the distances they travel are gaussian randomly distributed with variance @xmath293 . this is however the short - time behavior of the msd in @xmath6-dimensional manifolds ( [ msdcurv ] ) . hence , we heuristically extend the ermak and mccammon algorithm to curved manifolds by allowing the particles to move in any direction with equal probability , but the geodesic distances they travel are gaussian randomly distributed , i.e. , @xmath294 with @xmath295 , as long as @xmath296 . in the particular case of a circle , this idea leads to the following algorithm : a uniform random number is generated in the interval @xmath297 $ ] ; the particle in turn is allowed to move in the clock - wise direction if the result falls in @xmath298 $ ] , otherwise the particle moves in the opposite direction ; a gaussian randomly distributed number with variance @xmath299 is then generated in order to determine the arc - length the particle travels ; these steps are repeated for every particle , many times , in order to construct the dynamics of the system in its natural sequence . in our simulations , we let 1000 free particles to move in very short time steps , until they approximately cover a distance of 100 times the perimeter of the circle . the large number of particles allows to improve the numerical precision of our results . we expect , on the one hand , the short - time behavior @xmath300 , since this is included in the construction of the algorithm . on the other hand , for very long times ( @xmath200 ) the particles has to distribute uniformly along the perimeter of the circle . therefore , the geometric behavior of the msd must be given by the simple average of the geodesic square displacement @xmath301 which agrees with equation ( [ msdcirculo ] ) . these and the intermediate values of @xmath302 are shown in figure 1 . the extension of these ideas to the general case of curved surfaces will be presented elsewhere . in this work the diffusion of free particles on curved surfaces is studied . after writing the langevin equation and the fluctuation - dissipation theorem for curved surfaces , we solved the former in the riemann normal coordinates for weak curvatures , i.e. , up to linear terms in the riemann curvature tensor . from this solution , the dynamics of the particles can be clearly separated in three regimes ; the ballistic one , @xmath303 , and two diffusive regimes ; short times , @xmath304 , and long times , or geometric regime , @xmath305 . in the ballistic regime we find effects of the geometry up to order of @xmath164 typically negligible unless there is a region of high curvature . we therefore conclude that , typically , the local dynamics occurs in the plane tangent to the surface . nevertheless , in the long - time diffusive regime only the geometric effects take place . the free particle diffusion coefficient @xmath179 might be understood in terms of the short - time limit of the mean geodesic square displacement , @xmath306 , in a similar way as in the case of interacting particles . the geometry then appears as an external force acting on the diffusing particles , which can be recognized in the second term of the left side of equation ( [ langp ] ) . we should remark that in the short - time diffusive regime the langevin equation was found to have the same solution as the diffusion equation on curved surfaces @xcite , as it is expected and consistent with a work of m. polettini @xcite . we therefore used the latter in order to study the whole diffusive dynamics of free particles along a circle , @xmath307 , and over a sphere , @xmath308 . we do not expect curvature effects in @xmath307 since its gaussian curvature @xmath309 is zero . however , the msd displays a geometric diffusive regime due only to confinement effects , since the particles are unable to move beyond the region where the circle is placed . in @xmath308 the confinement and curvature effects act together to define the geometrical regime . the difference between curvature and confinement effects is subtle and somehow counterintuitive . this will be carefully reported somewhere else . we also reported some results from brownian dynamics computer simulations . we obtained them by implementing the standard ermak - mccammon algorithm , as well as its heuristic adaptation to curves . in the first case , we assumed that the particles are subjected to an external field that constrains the movement of the particle to the surface . the coupling constant @xmath34 can be thought as a spring - like constant that is adjusted to guarantee the particle dynamics very close to the surface . a particular test of this field was made on the ground of statistical mechanics by calculating the single canonical partition function of the particle in the field and compare it with the corresponding partition function on @xmath172 . it is found a curious effect , that may be experimentally tested , that the only way to maintain the confinement effect to the spherical surface is scaling the value of the coupling constant @xmath34 with temperature in a precise way . in the second case , which was only applied to the circle , we allowed the particles to move in every direction along the curve , every time displacing geodesic lengths given by random gaussian number with variance @xmath310 . the quantitative comparison of the theoretical results with the simulation data was shown in figures 1 and 2 . our approach can be extended in various directions . we could study the case of interacting particles where interaction may produce colored distributions for the stochastic forces in the langevin equation @xcite , so that wick s theorem , which is of central importance in our calculations , were not longer valid . nevertheless , it could be longer applied as an approximation , in the sense that the @xmath311-time correlation functions may be decomposed in terms of two - time correlation functions . in addition , both implementations of the ermak and mccammon algorithm may be further used for interacting particles , as well as for other physical circumstances . for instance , the rotational brownian motion of molecules can be studied by a diffusion equation on a manifold . for the case of the lateral diffusion of a protein or lipid we did not take into account the effects of the thermal fluctuations of the membrane and on top of that the finite size of these particles could involve local deformations on the membrane that can change the diffusion constant . furthermore , it could be interesting from the theoretical viewpoint to investigate if there are other realistic " circumstances where lagrange constraints represent idealization of a very stiff potential . the stochastic force is gaussian distributed for each point on the surface @xmath312 . in global coordinates this distribution is given by @xcite @xmath313\exp\left\{-\frac{1}{2\omega}\int^{t}_{0 } d\tau ~\delta^{ij}f_{i}\left(\tau\right)f_{j}\left(\tau\right)\right\ } , \label{distribution}\end{aligned}\ ] ] where @xmath314 $ ] is an appropiate functional measure . this is equivalent to a gaussian vector field theory in one dimension . the expectation values are defined by @xmath315 . in particular , the fluctuation - dissipation theorem ( [ fluc - dissip ] ) can be verified using ( [ distribution ] ) . the force distribution ( [ distribution ] ) also determines the fluctuaction - dissipation theorem in local coordinates ( [ locfluc - diss ] ) . to show this , let us separate the force in tangent and normal components . since , @xmath316 and @xmath317 are given for each point , thus @xmath318 is a biyective transformation between @xmath319 , with @xmath320 , and @xmath321 , with @xmath322 . thus the measure @xmath323 $ ] transforms to @xmath324\left[df_{n}\right]j$ ] , where @xmath325 is the jacobian . in addition , the argument of the boltzmann weight can be splitted in these coordinates . then , the measure @xmath326 can be written as @xmath327\left[df_{n}\right]j \exp \left\{-\frac{1}{2\omega}\int dt \left(g^{ab}{f}_{a}{f}_{b}+f^{2}_{n}\right)\right\}.\nonumber\\ \label{distribution2}\end{aligned}\ ] ] now , since the hypersurface is locally a plane we can always choose @xmath316 such that @xmath104 . therefore , the local fluctuation - dissipation relations ( [ locfluc - diss ] ) can be straightforwardly obtained from ( [ distribution ] ) . this technical detail allows us to establish that both global and local versions of the langevin equation on curved surfaces are equivalent . the correlation of two momenta for zero initial conditions can be computed from @xmath328 next , we use the fluctuation - dissipation theorem ( [ locfluc - diss ] ) . thus the integration over variable @xmath329 leads to the following result @xmath330 where @xmath331 is the heaviside step - function . the remaining integral over @xmath332 can be done for two cases @xmath333 and @xmath334 . if @xmath333 then @xmath335 for all @xmath336 $ ] , therefore @xmath337 . now , if @xmath334 then the integration for @xmath332 can be splitted in two parts @xmath338 in the first integral @xmath335 , since @xmath339 $ ] . then for this integral @xmath340 . for the second integral , we have @xmath341 , since @xmath342 $ ] . therefore @xmath343 . now , joining these results and performing the elementary integrals we reproduce equation ( [ pp2 ] ) . the determination of @xmath345 can be obtained from the calculation of @xmath346 we should remark that the integral @xmath347 appears in various places in the multiple integral ( [ jfunction ] ) . thus , the function ( [ jfunction ] ) can be written as follows @xmath348 the advantage to write @xmath349 in terms of @xmath350 is that @xmath351 . for the calculation of the function @xmath350 it is convenient to use the following equivalent expression for the green function @xmath352.\end{aligned}\ ] ] performing its integral we obtain @xmath353 now , we carry out the elementary integrations involved in @xmath354 . we then get the following expression @xmath355 + e^{-4\frac{t}{\tau_{b}}}\left[1+e^{\frac{t}{\tau_{b}}}\right.\right.\nonumber\\&\times&\left.\left.\left(-8+e^{\frac{t}{\tau_{b}}}\left(15 - 40e^{\frac{t}{\tau_{b}}}-6\frac{t}{\tau_{b } } + e^{2\frac{t}{\tau_{b}}}\left(32\right.\right.\right.\right.\right.\nonumber\\&+&\left.\left.\left.\left.\left.6\frac{t}{\tau_{b}}\left(\frac{t}{\tau_{b}}-4\right ) \right)\right)\right)\right]\right\}.\nonumber\\\end{aligned}\ ] ] the expectation values for @xmath243 can be calculated from @xmath356 where @xmath357 equation ( [ eggo ] ) depends explicitly on the chosen form of @xmath25 . in general , equation ( [ eqomega ] ) can not be written in a closed form and it must be studied numerically . in particular , we discuss here the mean values of the functions @xmath358 , and @xmath359 . in order to have a more manageable form for these expectation values we use the following identity @xmath360.\nonumber\\\end{aligned}\ ] ] now , in order to obtain ( [ g - function ] ) we perform the integration for even and odd vaues of @xmath232 . after performing the elementary integrations , we obtain the following results . for @xmath234 , @xmath361 is zero for even values of @xmath232 , and for odd values of @xmath232 it takes the form @xmath362 where the last sum does not take the values @xmath363 and @xmath364 . for @xmath365 , it is not difficult to show the identity @xmath366 for odd values of @xmath232 . however , for even values of @xmath232 we find @xmath367 where @xmath368 is a function defined as @xmath369 and @xmath370 is the binomial coefficient @xcite . _ partition function . _ the hamiltonian for a free particle of mass @xmath2 on a @xmath6-dimensional riemannian manifold @xmath7 is given by @xmath371 , where @xmath11 is the metric tensor of @xmath7 . the single partition function associated to this hamiltonian is given by @xmath372 where @xmath31 is the volume of @xmath7 , assumed it is compact . 0.5em _ the error function_. the exact value of the integral in equation ( [ integral ] ) is , @xmath373 @xmath374 using the asymptotic behaviour of the error function @xcite for large values of @xmath375 , we have the following asymptitic behavior @xmath376 0.5em _ pseudo - corrective force_. according to e. j. hinch @xcite in order to convert the brownian motion of a rigid system to that of an equivalent very stiff system , we should add the following pseudo - corrective force to the equations of motion , @xmath377 where @xmath378 , and @xmath379 is the constraint function of the system . for the corresponding conditions , the determinant is reduced to @xmath380 . for this case , the corrective force becomes @xmath381 where @xmath317 is the normal vector of the hypersurface and @xmath382 is the @xmath66-matrix defined above . for the spherical case , the constraint function is given by @xmath383 , where @xmath71 is the radius of the sphere . in this case the pseudo - potential force is @xmath384 . a similar result occurs for the circle . nina malchus and matthias weiss , biophysical journal , * 99 * , 1321 ( 2010 ) ; matthias weiss , hitoshi hashimoto , and tommy nilsson , biophysical journal * 84 * , 4043 ( 2003 ) ; valerii m. sukhorukov , j@xmath385rgen bereiter - hahn , plos * 4 * e4604 ( 2009 ) . ali naji , and franck l. h. brown , j. chem . phys . * 126 * , 235103 ( 2007 ) f. crdoba - valds , c. fleck and r. castaeda - priego , rev . . fis . * 53 * , 475 ( 2007 ) ; f. crdoba - valds , c. fleck , j. timmer and r. castaeda - priego , _ submitted_. ali naji , paul j. atzberger , and frank l. h. brown , phys . * 102 * , 138102 ( 2009 ) . ellen reister - gottfried , stefan m. leitenberger , and udo seifert , phys . e * 75 * , 011908 ( 2007 ) ; ellen reister - gottfried , stefan m. leitenberger , and udo seifert , phys . e * 81 * , 031903 ( 2010 ) . boris m. aizenbud and nahum d. gershon , biophys . j. * 38 * , 287 ( 1982 ) . d. anderson and h. wennerstr@xmath386 m , j. phys . chem . * 94 * , p. 8683 j. balakrishnan , phys . e * 61 * , 4648 ( 2000 ) r. holyst , d. plewczynski , a. aksimentiev , and k. burdzy , phys . e * 60 * , 302 ( 1999 ) . micheal christensen , journal of computational physics * 201 * , 421 - 438 ( 2004 ) . tomoyoshi yoshigaki , phys . e * 75 * , 041901 ( 2007 ) . marshall fixman , j. chem . phys . * 69 * , 1527 ( 1978 ) . j. m. rallison , j. fluid . mech . * 93 * , 251 ( 1979 ) . e. j. hinch , j. fluid . mech . * 271 * , 219 ( 1993 ) . hans christian @xmath387ttinger , phys . e * 50 * 2696 ( 1994 ) . hans christian @xmath387ttinger , _ stochastic processes in polymeric fluids_. ed . springer , ( 1996 ) . david c. morse , adv . phys . * 128 * , 65 ( 2004 ) . h. kleinert and s. v. shabanov , j. phys . a : math . gen . * 31 * , 7005 - 7009 ( 1998 ) . richard baush , rudi schmitz , and ukasz a. turski , z. phys . b. * 97 * , 171 ( 1995 ) ; richard baush , rudi schmitz , and ukasz a. turski * 73 * , 2382 ( 1994 ) . lingang chen and micheal w. deem , phys . e * 68 * , 021107 ( 2003 ) .

in many applications , such as development and testing of image processing algorithms , it is often necessary to simulate images containing realistic noise from solid - state photosensors . a high - level model of ccd and cmos photosensors based on a literature review is formulated in this paper . the model includes photo - response non - uniformity , photon shot noise , dark current fixed pattern noise , dark current shot noise , offset fixed pattern noise , source follower noise , sense node reset noise , and quantisation noise . the model also includes voltage - to - voltage , voltage - to - electrons , and analogue - to - digital converter non - linearities . the formulated model can be used to create synthetic images for testing and validation of image processing algorithms in the presence of realistic images noise . an example of the simulated cmos photosensor and a comparison with a custom - made cmos hardware sensor is presented . procedures for characterisation from both light and dark noises are described . experimental results that confirm the validity of the numerical model are provided . the paper addresses the issue of the lack of comprehensive high - level photosensor models that enable engineers to simulate realistic effects of noise on the images obtained from solid - state photosensors .

1412.4031

many modern measuring devices use cmos or ccd solid - state photosensors to convert light into a digital signal . due to imperfections of photosensors such a conversion is not ideal and leads to noise in the measured signal . therefore , one can either estimate and reduce the impact of noise from the sensor , or simulate and predict the impact of noise on the images quality . the problem of noise estimation has been addressed by the emva1288 standard @xcite . the high - level simulation of noise in photosensors , however , is still an area of active research . the main problem is that photosensors are affected by many different sources of noise , some of which can not be modelled adequately using only gaussian noise . [ review:2r2]different numerical models of cmos @xcite and ccd @xcite sensors are reported in the literature . the photosensor models differ by the generality and the scope of the noise sources being included . high - level simulators @xcite are used for the evaluation of how different noise sources influence image quality @xcite , while low - level simulator models specific aspects of the solid - sate physics of a sensor @xcite . the thoroughness of noise modelling differs as well ; for example , in some models of the photosensor , the dark current shot noise is described by its mean value @xcite , while other models use more sophisticated statistical description of noise @xcite . [ review2jei : r1q1 - 2 ] sophisticated software integrated suits such as the image systems evaluation toolkit @xcite ( iset ) were developed recently . [ [ review:1r2 ] ] models of cmos photosensors : + + + + + + + + + + + + + + + + + + + + + + + + + + + + both high - level @xcite and low - level @xcite numerical models of cmos photosensors are reported in the literature . low - level simulators like medici @xcite , which are able to model the 2-d distributions of potential and carrier concentrations , are used for simulation of solid - state physics in cmos photosensors @xcite . some of the papers deal with the numerical modelling of colour images that allow end users to see the effects of the noise on the images @xcite . other papers focus on the specific types of noise ; for example , the analysis of temporal noise in cmos active pixels sensors can be found in @xcite , an autoregressive model of fixed - pattern noise ( fpn ) was presented in @xcite along with other fpn models @xcite . high - level models ( e.g. , @xcite ) often assume that the adc , sense node and source follower are completely linear , which , as we will show in this paper , is not always the case . [ [ models - of - ccd - photosensors ] ] models of ccd photosensors : + + + + + + + + + + + + + + + + + + + + + + + + + + + the models of ccd photosensors @xcite and their noise parameters estimation @xcite are generally simpler than models of cmos sensors . the model by flory @xcite describes the noise of the sensor electronics as a combination of shot noise and amplifiers noise . the ccd camera noise model by cox @xcite includes photon and electronic shot noise , dark - current noise and readout noise ; however the model suggests that noise is stationary . healey and kondepudy @xcite model camera noise with offset fixed - pattern noise ( fpn ) , photon and dark - current shot noise , read noise and photo response non - uniformity ( prnu ) . [ review2jei : r2_q2]an adequate model was proposed by costantini and ssstrunk @xcite , where noises was categorised in four main types : photon shot noise , photo response non uniformity , dark current shot noise and read noise . [ [ review:18r2 ] ] the purpose of the article : + + + + + + + + + + + + + + + + + + + + + + + + + + + this article provides a literature review of the noise models that are used in the simulations of ccd and cmos photosensors and give a working numerical model of a photosensor . the aim of this article is to summarise the advances in numerical simulations of photosensors and provide a matlab implementation of a ccd / cmos sensor model , which has been demonstrated to adequately describe noise properties of a hardware cmos photosensor . the experimental results for a hardware 5 t cmos photosensor are provided in the article as validation for the developed numerical simulator . the article addresses the issue of the lack of high - level models of photosensors . it is shown that the dark noise may have more complicated structure than is typically assumed and therefore requires a more sophisticated statistical description . the formulated high - level model of a photosensor allows the simulation of realistic noise effects on the images and can be useful for testing of image processing algorithms . [ [ software - implementation ] ] software implementation : + + + + + + + + + + + + + + + + + + + + + + + + the matlab - based software implementation of the high - level ccd / cmos photosensor model described in this article is freely available on the website ` https://code.google.com/p/highlevelsensorsim/ ` and also on email request . the software simulator written as a series of functions that call the corresponding sub - functions sequentially to add the noise with specified parameters . two example test files are supplied that run : ( 1 ) a * simple * sensor model , which is completely linear and where all noise are gaussian , and ( 2 ) an * advanced * model , which has v / v and v / e non - linearities , lognormal noise models and other specific noises . both test files use the same code : adjusting the parameters in the test file , a user can turn the noise sources on or off and trigger non - linearities . the matlab code has a thorough documentation , both as comments inside the m - files and rendered for web view @xcite in a browser ( see _ /help _ directory and _ index.html _ file inside ) . introduction _ _ noise sources in solid - state photosensors : _ _ from photon to charge : _ _ _ photon shot noise and prnu _ _ dark signal , dark current fpn and dark current shot noise _ _ source follower noise ( johnson noise , flicker ( 1/f ) noise , and random telegraph noise ) _ from charge to voltage : _ _ _ sense node reset noise ( ktc noise ) _ _ offset fpn _ _ gain non - linearities ( v/@xmath0 gain and v / v gain non - linearity ) _ _ correlated double sampling _ from voltage to digital numbers _ ( including differential and integral linearity errors ) _ simulation methodology _ _ model : from photon to charge : _ _ _ simulating photon shot noise and prnu _ _ converting photons to electrons _ _ simulating dark signal , dark current fpn and dark current shot noise _ _ simulating source follower noise _ model : from charge to voltage : _ _ _ simulating sense node reset noise ( ktc noise ) and offset fpn _ _ converting electrons to voltage _ _ simulating v/@xmath0 and v / v gain non - linearity _ model : from voltage to digital numbers _ _ _ simulating adc non - linearity _ _ simulating quantisation noise _ experimental validation of the photosensor model _ _ radiometric function _ _ photon transfer curve _ _ signal - to - noise ratio _ _ photo response non - uniformity _ _ noise spectrogram _ _ dark signal performance for different integration times : _ _ _ short integration time _ _ long integration time _ conclusion _ _ many noise sources contribute to the resulting noise image that is produced by photosensors . noise sources can be broadly classified as either * fixed - pattern * ( time - invariant ) or * temporal * ( time - variant ) noise . fixed - pattern noise refers to any spatial pattern that does not change significantly from frame to frame . temporal noise , on the other hand , changes from one frame to the next . these noise sources will be described in detail in the following subsections . [ review2jei : r2_q3 ] the ability of a semiconductor to produce electrons from incident photons is referred to as _ quantum efficiency _ [ review:5r2][review:29r1 ] quantum efficiency can be supplied by the sensor manufacturer in the form of spectral responsivity as a function of wavelength , or measured empirically . [ review2jei : r2q9 - 1]for the sake of simplicty , we assume that the light is monochromatic , and therefore we need the spectral response of the sensor at one wavelength . the charge collected in each pixel of a sensor array is converted to voltage by a _ sense capacitor _ and a _ source - follower amplifier_. the sense node is shared for all pixels in the case of a ccd . in cmos sensors , the sense node is located inside each pixel . the process of photon capturing has an uncertainty that arises from random fluctuations when photons are collected by the photodiode . such uncertainty leads to _ photon shot noise _ and is described by the poisson process . in the case of a high level of light ( more than 1000 arrival events @xcite of photons ) , the poisson distribution may be approximated by a gaussian distribution . however , this is not justified for low - light conditions . the method of the approximation of the poisson distribution by gaussian is discussed in detail in @xcite . [ review:12r1]the photo response non - uniformity ( prnu ) is the spatial variation in pixel output under uniform illumination mainly due to variations in the surface area of the photodiodes . this occurs due to variations in substrate material during the fabrication of the sensor @xcite . the prnu is signal - dependent ( proportional to the input signal ) and is fixed - pattern ( time - invariant ) . [ review:30r1 ] the prnu factor is typically @xmath1 for a given sensor @xcite , but varies from one sensor to another . it is noteworthy that the prnu pattern can be used as a unique identification fingerprint for digital cameras @xcite . [ review:13r1 ] read noise is defined as any noise that is not a function of the signal @xcite ; it is a combination of the remaining noise generated between the photodiode and the output of the adc circuitry . the read noise consists of : _ dark current shot noise _ ( sec . [ subsec : dshot ] ) , _ dark current fixed pattern noise _ ( sec . [ subsec : dfpn ] ) , _ sense node reset noise _ [ sec : subsection : snresetnoise ] ) , _ source follower noise _ [ sec : sfnoise ] ) , _ adc quantisation noise _ ( sec . [ subsec : adcnoise ] ) . dark current is the thermally generated electrons @xcite discharging the pixel just as if a photon had hit the pixel . the dark current generally arises from surface defects in the @xmath2 interface @xcite , surface generation @xcite , thermal generation @xcite , and imperfections of the semiconductor manufacturing process @xcite . the average dark current @xmath3 [ e@xmath4/sec / pixel ] can be characterised by @xcite:[review:16r1 ] @xmath5 where @xmath6 is the pixel area [ @xmath7 , @xmath8 [ na/@xmath7 is the dark current [ review:7r2]figure - of - merit can be estimated from the measurements to ensure that the number that manufacturers publish is consistent with the sensor performance . using photon transfer methodology @xcite , one can estimate the dark current fpn factor @xmath9 and the dark current figure - of - merit @xmath8 . this can be done by plotting dark current shot noise and dark current fpn versus a dark signal . this is called `` dark transfer curve '' , out of which those parameters can be deduced . the procedure is described on page 170 - 171 in @xcite . ] ( varies with the sensor and usually reported by the manufacturer ) at 300k , @xmath10 [ ev ] is the band gap energy of the semiconductor which also varies with temperature , @xmath11 is the temperature in k , and @xmath12 is boltzman s constant . [ review:37r1]it was reported @xcite that for a ccd photosensor operated in multipinned phase ( mpp ) mode , the average dark current @xmath13 can be described as : @xmath14 where the first term is more important as the temperature increases , whereas the second term dominates at lower temperatures . temperature dependence of the dark current is explained in details in @xcite . the relationship between band gap energy and temperature can be described by varshni s empirical expression @xcite : @xmath15 where @xmath16 , @xmath17 and @xmath18 are material constants . for silicon @xmath19 $ ] , @xmath20 [ ev / k ] , and @xmath21 [ k ] . the dark signal varies from pixel to pixel , it is linear with respect to integration time , and doubles with every @xmath22 increase of temperature @xcite . when the exposure begins , a dark current is generated even if there is no light . the longer the exposure time @xmath23 , the more dark signal @xmath24 ( number of electrons per pixel ) will be generated : @xmath25 [ review:13 - 2r1]where @xmath3 is the average dark current from eq . [ eq : darkcurrentrate ] . in simulation we use eq . [ eq : darkcurrent ] to calculate the dark signal per pixel . but the dark signal @xmath24 is a subject of the _ dark current shot noise _ due to random generation of the electrons . [ review:37 - 1r1 ] an additional noise results from the dark signal due to the electrons being generated randomly by the photosensor . such noise is called dark current shot noise @xcite and is described by the poisson distribution as the random arrival of electrons @xcite as : @xmath26 where @xmath3 is the average dark current [ @xmath0/sec / pixel ] as described by eq . [ eq : darkcurrentrate ] . in a real sensor , the pixels differ slightly from one to another [ review2jei : r2q5 ] resulting in another source of noise called _ dark current fixed pattern noise_. [ review:33r1]pixels in a hardware photosensor can not be manufactured exactly the same from perfectly pure materials . there will always be variations in the photo detector area that are spatially uncorrelated @xcite , surface defects at the @xmath2 interface @xcite , and discrete randomly - distributed charge generation centres @xcite . these defects provide a mechanism for thermally - excited carriers to move between the valence and conduction bands @xcite . consequently , the average dark signal is not uniform but has a spatially - random and fixed - pattern noise ( fpn ) structure . the dark current fpn can be expressed @xcite as follows:[review:32r1 ] @xmath27 where @xmath23 is the integration time , @xmath3 is the average dark current described in eq.[eq : darkcurrentrate ] , and @xmath9 is the dark current fpn factor that is typically @xmath28 for ccd and cmos sensors . there are also so called `` outliers '' or `` dark spikes '' @xcite ; that is , some pixels generate a dark signal values much higher than the mean value of the dark signal . the mechanism of such `` dark spikes '' or `` outliers '' can be described by the poole - frenkel effect @xcite ( increase in emission rate from a defect in the presence of an electric field ) . [ review:4r1]in high - end ccd and cmos sensors the source follower noise has been decreased to a value of one electron rms @xcite . however , source follower noise for industry - grade sensors can be significant and therefore should be included in a photosensor model . the source follower noise consists of white noise , flicker noise ( 1/f noise ) , and random telegraph noise ( rtn ) . [ [ johnson - noise - white - noise ] ] johnson noise ( white noise ) + + + + + + + + + + + + + + + + + + + + + + + + + + + similar to the sense node , the source follower amplifier has a resistance that generates thermal noise . such noise is governed by the johnson white noise equation @xcite . the noise is commonly referred to as johnson noise , johnson - nyquist or simply as a white noise @xcite . [ [ flicker-1f - noise ] ] flicker ( 1/f ) noise + + + + + + + + + + + + + + + + + + + the flicker noise , also referred to as 1/f noise , is generally due to imperfect contacts between two materials at the junction @xcite , including metal - to - metal , metal - to - semiconductor and semiconductor - to - semiconductor . since mosfets are used inside each pixel , cmos image sensors exhibit higher 1/f noise than ccd sensors . more details and discussions about 1/f noise can be found in @xcite . [ [ random - telegraph - noise ] ] random telegraph noise + + + + + + + + + + + + + + + + + + + + + + as the ccd and cmos pixels are shrinking @xcite in dimension , the low - frequency noise is subsequently increasing @xcite . the origin of rtn is attributed to the random trapping and emission of mobile charge carriers resulting in discrete modulation of the channel current , which can be modelled @xcite as a random telegraph signal . however , the modelling and explanation of both flicker noise and rtn are still subjects of active research @xcite . it was shown @xcite that voltage fluctuations , which exhibit a @xmath29 power spectrum , can be described by the fisher - tippet - gumbel distribution . further details about rtn noise in photosensors can be found in @xcite . the conversion from charge to voltage is not perfect in the real photosensor : sense node reset noise , source follower noise , and offset fixed pattern noise add an uncertainty to the measured signal @xcite . these noise sources are described in this subsection . prior to the measurement of each pixel s charge packet , the sense node capacitor is reset to a reference voltage level @xcite . noise is generated at the sense node by an uncertainty in the reference voltage level due to thermal variations in the channel resistance of the mosfet reset transistor . the reference level of the sense capacitor is therefore different from pixel to pixel @xcite . it was reported @xcite that the reset noise can be a significant contributor to dark noise . the reset noise voltage is given by : @xmath30 where @xmath31 is boltzmann s constant , @xmath32 is the sense node capacitance [ f ] , and @xmath11 is the temperature [ k ] . because reset noise can be significant @xcite ( about 50 rms electrons ) , most high - performance photosensors incorporate a noise - reduction mechanism such as correlated double sampling ( cds ; see subsection [ subsec : cdsdescription ] and subsection [ subsec : cdssimulation ] ) . [ [ sense - node - reset - noise - for - ccd - and - cmos - sensors ] ] sense node reset noise for ccd and cmos sensors : + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + for ccd sensors , the sense node reset noise is removed by cds @xcite . in cmos photosensors , it is difficult to remove the reset noise given the specific cmos pixel architectures , even after application of cds . the cds suppresses the low frequency noise components , although it increases the thermal noise contributions @xcite . pixels in the same column of the photosensor share a column amplifier . [ review:36 - 1r1]differences in the gain and offset of these column amplifiers contribute to a column - wise offset fixed pattern noise . the offset fixed pattern noise ( offset fpn ) is caused by an offset in the integrating amplifier , size variations in the integrating capacitor , and as variation of bias / offset voltages @xcite . [ review:6r2 ] specifically , cmos offset fpn is due to threshold voltages differences in the pixel source follower amplifier @xcite . offset fpn arises from the fact that pixels in the same column of a cmos sensor share the same column amplifier . the difference in gain and the offsets of such an amplifier is a source of column - to - column offset fpn @xcite . this type of noise appears as `` stripes '' in the image and can result in significant image quality degradation . modelling of the offset fpn is discussed in subsection [ sec : subses : elgamalfpn ] . both ccd and cmos sensors may exhibit v / v ( voltage - voltage conversion ) [ review:8r2 ] and v/@xmath0 ( voltage - electrons conversion ) non - linearity . the v / v non - linearities is mainly @xcite due to the source follower amplifier . the v/@xmath0 non - linearity is due to the sense node capacitance @xmath32 . the v/@xmath0 non - linearity is a problem mostly for cmos sensors , whereas for ccd sensors the v/@xmath0 non - linearity is usually negligible . the v/@xmath0 non - linearity affects both fpn and shot noise . it can also be thought of as a sense node capacitor non - linearity . when a small signal is measured , the sense node capacitance @xmath32 may change negligibly . for a large signal , a change of @xmath32 affects the signal being measured . the v/@xmath0 non - linearity can be expressed as @xcite:[review:17r1 ] @xmath33,\ ] ] where @xmath34 is the total signal in electrons , the constant @xmath35 is @xmath36 , @xmath37 is the reference voltage to which the sense node is reset . the v / v non - linearity is caused by the non - linear response of source follower amplifier @xcite . in our simulations , we implemented the linear change of source follower gain @xmath38 similar to the model in @xcite . the v / v gain non - linearity is described as : @xmath39 [ review:17 - 1r1]where @xmath40 is the voltage that corresponds to the signal collected by the sense node , @xmath41 is the voltage that corresponds to the full - well signal , and @xmath42 is used to control the amount of non - linearity in the source follower gain @xmath38 . the parameter of non - linearity @xmath42 can be taken from the specifications of a photosensor . the new source follower gain @xmath43 from eq . [ eq : sfnewgain ] is then used for the conversion in eq . [ eq : conversiontosourcefollower ] in subsection [ sec : simulationsourcefollower ] to simulate the v / v non - linearity . [ review:3r1]fixed pattern noise performance of cmos sensors is usually lower than for ccd sensors @xcite . for this reason cmos sensors use noise reduction circuits such as correlated double sampling ( cds ) . the cds circuits are located per column and usually consist of two sample - and - hold ( s&h ) circuits @xcite . during the pixel read - out cycle , two samples are taken : the first when the pixel is in the reset state and the second when the charge has been transferred to the read - out node . during the reset stage , the photodiode capacitance is charged to a reset voltage . the reset voltage is read by the first sample - and - hold ( s&h ) in a correlated double sampling ( cds ) circuit @xcite . then the exposure begins : the photodiode capacitor is discharged during an exposure ( integration ) time at a rate proportional to the incident illumination . this voltage is then read by the second sample - and - hold of the cds . the cds circuit subtracts the signal pixel value from the reset value . the main purpose of cds is to eliminate fixed pattern noise caused by random variations in the threshold voltage of the reset and pixel amplifier transistors . many variants of cds were proposed and implemented @xcite . many sensors use correlated double sampling to eliminate column patters and pixels patterns . pixel noise reduced by the cds is known as `` crowbar '' that significantly reduces the effect of the reset column noise @xcite . [ review:34r1]an analogue - to - digital converter ( adc ) transforms a voltage signal into discrete codes . an @xmath44-bit adc has @xmath45 possible output codes with the difference between code being @xmath46 . the resolution of the adc indicates the number of discrete values that can be produced over the range of analogue values and can be expressed as : @xmath47 where @xmath48 is the maximum voltage that can be quantified , @xmath49 is minimum quantifiable voltage , and @xmath50 is the number of voltage intervals . therefore , the output of an adc can be represented as : @xmath51 the lower the reference voltage @xmath52 , the smaller the range of the voltages one can measure . in terms of the adc , the following non - linearity and noise should be considered for the simulations of the photosensors : integral linearity error , differential linearity error , quantisation error , and adc offset . [ [ differential - linearity - error ] ] differential linearity error + + + + + + + + + + + + + + + + + + + + + + + + + + + + + ( dle ) indicates the deviation from the ideal 1 lsb ( least significant bit ) step size of the analogue input signal corresponding to a code - to - code increment @xcite . assume that the voltage that corresponds to a step of 1 lsb is @xmath53 . in the ideal case , a change in the input voltage of @xmath53 causes a change in the digital code of 1 lsb . if an input voltage that is more than @xmath53 is required to change a digital code by 1 lsb , then the adc has _ dle error_[review:35r1 ] . in this case , the digital output remains constant when the input voltage changes from , for example , @xmath54 to @xmath55 , therefore corresponding the digital code can never appear at the output . that is , that code _ is missing_. [ [ integral - linearity - error ] ] integral linearity error + + + + + + + + + + + + + + + + + + + + + + + + ( ile ) is the maximum deviation of the input / output characteristic from a straight line passed through its end points @xcite . for each voltage in the adc input , there is a corresponding code at the adc output . if an adc transfer function is ideal , the steps are perfectly superimposed on a line . however , most real adc s exhibit deviation from the straight line , which can be expressed in percentage of the reference voltage or in lsbs . therefore , ile is a measure of the straightness of the transfer function and can be greater than the differential non - linearity . taking the ile into account is important because it can not be calibrated out . [ [ quantisation - errors ] ] quantisation errors + + + + + + + + + + + + + + + + + + + are caused by the rounding , since an adc has a finite precision . the probability distribution of quantisation noise is generally assumed to be uniform . hence we use the uniform distribution to model the rounding errors ( see subsection [ subsec : adcnoise ] ) . [ [ adc - offset - error ] ] adc offset error + + + + + + + + + + + + + + + + may occur due to the dc offset associated with the analogue inputs to the adc . the magnitude of the adc offset depends on the gain and input range selection . the photosensor model was implemented in matlab as a set of functions that generate noise according to the models discussed above . the functions sequentially transform the input to number of photons , to number of electrons , to voltages , and finally to a digital signal , as shown in fig . [ fig : camerascheme ] . one can simulate either ccd or cmos photosensors by turning the models of noise on and off . the simulation software contains routines for the simulation of the sensor and also scripts for the estimation of the noise characteristics . [ review:11-diagramr1 ] [ review:21 - 8r1 ] the numerical model of the photosensor simulates the temporal and fixed - pattern characteristics of noise sources accordingly . the matrices for the fixed - pattern ( time - invariant ) noise are calculated once and then saved to a file . those matrices will be loaded again to calculate the fixed pattern noise ; therefore , the noise will always remain the same reflecting the fixed pattern nature of the noise . on the other hand , the matrices for the temporal ( time - variant ) noise are recalculated each time the simulation begins . [ review:5r1 ] the input to the model of the photosensor is assumed to be a matrix @xmath56 that corresponds to the opical field . for the estimation purpose , the input can be a uniform light field ( i.e. , uniform image ) or [ review:5 - 1r1 ] gradient image ( i.e. , linear transition from no light to saturation level ) . [ review2jei : r1_q3 ] then the sensor s irradiance @xmath57 , which is in units of [ w / m@xmath58 , converted to the average number of photons @xmath59 collected by each pixel during the integration ( exposure ) time : @xmath60 where @xmath6 is the area of a pixel [ m@xmath58 , @xmath61 is integration ( exposure ) time , @xmath62 is the energy of a single photon at wavelength @xmath63 , @xmath64 is planck s constant and @xmath65 is the speed of light . therefore , each @xmath66-th element of the [ review:21 - 1r1 ] matrix @xmath59 contains the number of photons that have been collected by the @xmath66-th pixel of the `` software '' photosensor during the integration time @xmath61 . the photon shot noise is due to the random arrival of photons and can be described by a poisson process as discussed in subsection [ sec : photonshot ] . [ review:20r1]therefore , for each @xmath66-th element of the matrix @xmath59 that contains the number of collected photons , a photon shot noise is simulated as a poisson process @xmath67 with mean @xmath68 : @xmath69 in matlab , we use the ` poissrnd ` function that generates poisson random numbers with mean @xmath68 . [ review:23r1 ] that is , the number of collected photons in [ review:21r1 ] @xmath66-th pixel of the simulated photosensor in the matrix @xmath59 is used as the mean @xmath68 for the generation of poisson random numbers to simulate the photon shot noise . the input of the ` poissrnd ` function will be the matrix @xmath59 that contains the number of collected photons . the output will be the matrix @xmath70 , i.e. , the signal with added photon shot noise . [ review:21 - 7r1 ] the matrix @xmath71 is recalculated each time the simulations are started , which corresponds to the temporal nature of the photon shot noise . [ review:9r2 ] the matrix that contains the number of collected photons during the exposure time , including the photon shot noise @xmath59 , is converted to the matrix @xmath72 proportional to electrons for each @xmath66-th pixel as follows : [ review:19r1]@xmath73 where @xmath74 is the quantum efficiency [ e@xmath4/incident photons ] for the given wavelength . the quantum efficiency for a photosensor is a function of wavelength and is usually provided by the manufacturer . the photo response non - uniformity ( prnu ) is considered in our numerical model as a temporally - fixed light signal non - uniformity . according to our experimental results ( see subsection [ subsec : prnudiscussion ] ) , the prnu can be modelled using a gaussian distribution for each @xmath66-th pixel of the matrix @xmath75:[review:11r1][review:12 - 1r1 ] @xmath76 [ review:6r1]where @xmath77 is the prnu factor value . since the prnu does not change from one frame to the next , the matrix @xmath78 is calculated once and then saved to a file . [ review:21 - 4r1 ] in the simulations , the matrix @xmath78 can be loaded again to calculate the photo response non - uniformity . therefore , since the same matrix is used for all simulations of the photo response non - uniformity , it will always remain the same as it should be . the dark signal is calculated using eq . [ eq : darkcurrentrate ] for all pixels in the model . it is implemented using ` ones ` function in matlab as a matrix of the same size as the simulated photosensor . for each @xmath66-th pixel we have : @xmath79 where @xmath3 is the average dark current from eq . [ eq : darkcurrentrate ] , repeated here for convenience : @xmath80 in this article we conduct all measurements at the room temperature @xmath81 c. the matrix @xmath82 represents a constant value of dark signal generated during the integration time @xmath23 . however , since the electrons are generated randomly , the dark signal is a subject of a _ dark current shot noise_. [ review:22 - 1r1]similar to the photon shot noise , the _ dark current shot noise _ is due to the random arrival of the generated electrons and therefore described by a poisson process . the dark current shot noise can be simulated as a poisson process @xmath67 with mean @xmath68 as follows:[review:11 - 1r1 ] @xmath83 we use the ` poissrnd ` function that generates poisson random numbers with mean @xmath68 for each @xmath66-th element of the matrix @xmath82 from eq . [ eq : darkcurrentsignalsimulation ] that contains dark signal in electrons . the input of the ` poissrnd ` function will be the matrix @xmath82 that contains the number of electrons , and the output will be the matrix @xmath84 with added dark current shot noise . [ review:21 - 6r1 ] the matrix @xmath84 that corresponds to the dark current shot noise is recalculated each time the simulations are started , which corresponds to the temporal nature of the dark current shot noise . following the discussion in subsection [ subsec : dfpn ] , the dark current fixed pattern noise ( fpn ) is simulated using non - symmetric distributions to account for the `` outliers '' or `` hot pixels '' . it is usually assumed that the dark current fpn can be described by gaussian distribution . however , such an assumption provides a poor approximation of a complicated noise picture . this is shown in the experimental results in subsection [ subsec : darkfpnperf ] . [ review:31 - 6r1]studies show @xcite that a more adequate model of dark current fpn is to use non - symmetric probability distributions . the concept is to use two distributions to describe very `` leaky '' pixels that exhibit higher noise level than others . the first distribution is used for the main body of the dark current fpn , with a uniform distribution superimposed to model `` leaky '' pixels . for simulations at room - temperature ( @xmath85 c ) authors in @xcite use a _ logistic distribution _ , where a higher proportion of the population is distributed in the tails @xcite . for higher temperatures , inverse gaussian @xcite and log - normal @xcite distributions have been proposed . it was reported @xcite that the log - normal distribution works well for conventional 3 t aps cmos sensors with comparatively high dark current . in our simulations we use the log - normal distribution for the simulation of dark current fpn in the case of short integration times , and superimposing other distributions for long integration times . the actual simulation code implements various models , including log - normal , gaussian , and wald distribution to elumate the dark current fpn noise for short- and long - term integration times . the dark current fpn for each pixel of the matrix @xmath84 from eq . [ eq : dshotnoisesimulation ] is computed as follows[review:31 - 5r1 ] : @xmath86 where @xmath87 is from eq . [ eq : darkcurrentfpn ] , @xmath3 is the average dark current , and @xmath9 is the dark current fpn factor . since the dark current fpn does not change from one frame to the next , the matrix @xmath88 is calculated once and then can be re - used similar to the prnu simulations in subsection [ subsec : prnusimulation ] . the experimental results provided in subsection [ subsec : darkfpnperf ] confirm that non - symmetric models , and in particular the log - normal distribution , adequately describe the dark current fpn in cmos sensors , especially in the case of a long integration time ( longer than 30 - 60 seconds ) . our results are consistent with @xcite ; however , as we show in subsection [ subsec : verylongfpn ] , using only one distribution can not adequately describe the dark current fpn statistics . for long - exposure case , one needs to superimpose two ( or more , depending on the sensor ) probability distributions , as will be shown in subsection [ subsec : darkfpnperf ] . the components of the source follower noise were discussed in subsection [ sec : sfnoise ] . in the simulations , the source follower noise can be approximated as : @xmath89)},\ ] ] where @xmath90 is the source follower noise [ @xmath0 rms ] , @xmath91 is the readout clock frequency ( typically several mhz ) , @xmath92 is the cds sample - to - sampling time [ sec ] , @xmath93 is a sense node conversion gain from eq . [ eq : conversiontovoltage ] , @xmath38 is a source follower gain from eq . [ eq : conversiontosourcefollower ] , @xmath94 is the correlated double sampling ( cds ) dominant time constant usually @xcite @xmath95 [ sec ] . in order to approximate the source follower noise , we need to calculate the power spectrum of the noise @xmath96 and the cds transfer function @xmath97 . the power spectrum @xmath96 of the source follower noise can be approximated @xcite as : @xmath98 where @xmath99 is the thermal white noise in @xmath100 ( typically @xmath101 ) , @xmath102 is the flicker noise corner frequency in [ hz ] , and @xmath103 is the random telegraph noise ( rtn ) power spectrum that is given by @xcite : @xmath104 where @xmath105 is the random telegraph noise ( rtn ) characteristic time constant [ sec ] and @xmath106 [ a ] is the source follower current modulation induced by rtn . these constants are usually provided in the photosensor specifications and therefore can be used for the simulations . the @xmath97 is the cds transfer function and is given by the following expression @xcite : @xmath107 \cdot [ 2 - 2\cos(2\pi f t_s ) ] .\ ] ] in ccd photosensors , source follower noise is typically limited by the flicker noise . on the other hand , in cmos photosensors the source follower noise is typically limited by the rtn noise . [ review:4 - 1r1 ] using parameters provided in the specifications for the photosensors , one has to calculate eq . [ eq : sigma_sf ] using the vector @xmath97 from eq . [ eq : h_cds ] and @xmath96 from eq . [ eq : s_sf ] . the functions @xmath96 , @xmath103 , and @xmath97 can be calculated for each frequency value @xmath108 using vector notation to streamline the simulation code . we assume the source follower noise to be additive and gaussian - distributed : @xmath109,\ ] ] where the standard deviation of source follower noise @xmath110 is given by eq . [ eq : sigma_sf ] . the computed matrix @xmath111 represents the source follower noise ( in electrons ) . [ review:14r1]the simulation of charge to voltage conversion is performed as follows . the light signal @xmath112 contains photon shot noise ( already added to the matrix @xmath75 from eq . [ eq : photonstoelectrons ] ) and the prnu:[review:11 - 2r1 ] @xmath113 the dark signal matrix @xmath114 consists of dark signal , dark current shot noise , dark current fpn from eq . [ eq : darkfpnsimulation ] , and source follower noise @xmath111 is from eq.[eq : sourcefollowernoisesimulation ] as follows : @xmath115 then the matrices that correspond to the light signal @xmath112 and dark signal @xmath116 are summed together and rounded : @xmath117 [ review2jei : r1_q1]the number of electrons is then truncated to the full well ( the maximum number of electrons in the pixel ) and rounded . then the v/@xmath0 non - linearity is applied ( if desired ) to this sum using eq . [ eq : sensenodenonlin ] and converted to voltage by multiplication by the sense node conversion gain @xmath93 . the details are provided in the following subsections . the ktc noise is occurs in cmos sensors ( as we mentioned before in subsection [ sec : subsection : snresetnoise ] ) , while for ccd sensors the sense node reset noise is removed @xcite by correlated double sampling ( cds ) . random fluctuations of charge on the sense node during the reset stage result in a corresponding photodiode reset voltage fluctuation . the sense node reset noise is given by eq.[eq : ktcnoise ] , repeated here for convenience : @xmath118 the simulation of the sense node reset noise may be performed as an addition of non - symmetric probability distribution to the reference voltage @xmath37 . however , the form of distribution depends on the sensor s architecture and the reset technique @xcite . an inverse - gaussian distribution can be used for the simulation of ktc noise which corresponds to a hard reset technique in the cmos sensor , and the log - normal distribution can be used for soft - reset technique . the sense node reset noise can be simulated for each @xmath66-th pixel for the soft - reset case as : @xmath119 where @xmath120 is from eq.[eq : ktcnoise ] and then added to the matrix @xmath121 in volts that corresponds to the reference voltage . a model of the offset fpn in cmos sensors was represented @xcite as the sum of two components : a column and a pixel component . each component is modelled by a first order isotropic autoregressive process , where the processes are assumed uncorrelated . the model @xcite uses autoregressive processes to model the offset fpn since the parameters can be easily and efficiently estimated from the hardware sensor data @xcite.[review:36r1 ] [ review:31 - 7r1]the column offset fpn from the model @xcite can be described as follows . the noise is generated for each @xmath122-th column index of the matrix @xmath123 that corresponds to the voltage signal of the photosensor . the model assumes that the column offset fpn is a first order isotropic autoregressive process of the form : @xmath124 where the @xmath125 are zero mean , uncorrelated random variables with the variance @xmath126 , and @xmath127 $ ] is a parameter that characterises the dependency of @xmath128 on its two neighbours @xcite . the data in the matrix @xmath128 the repeated for every row to make the matrix of the appropriate size . more details are provided in @xcite . the simulation of the charge - to - voltage conversion uses the sense node gain @xmath93 [ v/@xmath0 ] as a parameter that is in the range @xmath130 \mu v / e^-$ ] ( sensor - dependent ) . conversion from charge to voltage is performed in the sense node as follows : @xmath131 where @xmath132 is the matrix of sense node voltages , @xmath37 is the reference voltage , @xmath133 is the matrix that corresponds to the total number of electrons collected by the `` software '' sensor ( see eq.[eq : totalsignalcollected ] ) during the integration time , and @xmath93 is the sense node gain in [ v/@xmath134 . in case when the sense node reset noise ( ktc noise ) is simulated , the matrix that contains the reset noise @xmath135 from eq . [ eq : snresetsimulation ] is added to the @xmath37 in eq . [ eq : conversiontovoltage ] : @xmath136 after that , the source follower gain @xmath38 [ v / v ] is applied : @xmath137 where @xmath38 is a source follower gain [ v / v ] . if the offset fpn is simulated , the matrix @xmath138 that corresponds to the offset fpn noise is added as follows : @xmath139 next , the v / v non - linearity from eq . [ eq : vvnonlin ] is applied if necessary and the resulting matrix of voltages is stored in @xmath140 . after that , the matrix @xmath140 that corresponds to voltages is converted to digital numbers . for the simulation model , the v/@xmath0 non - linearity can be expressed from eq.[eq : venonlinsignal ] as : @xmath141\ ] ] where @xmath142 is the charge of an electron , and @xmath17 is the coefficient of non - linearity . the parameter @xmath17 is usually provided in the photosensor specifications . the voltage matrix is multiplied by the new source follower gain @xmath43 : @xmath143 where @xmath132 is the sense node voltage signal , @xmath144 is the source follower voltage signal , and @xmath43 is a new source follower gain according to eq.[eq : sfnewgain ] repeated here for convenience : @xmath145 the next step involves the adc for quantisation to a digital signal . [ review:3 - 1r1]in this paper , we use the algorithm of correlated double sampling ( cds ) that is similar to the method described in @xcite . the algorithm is as follows . during the simulations , two images are acquired : a light image that contains both signal and noise , and a dark image with zero exposure time . the cds algorithm then subtracts these two images to eliminate fixed pattern noise . in the real hardware sensor , the cds techniques can be different and more sophisticated . this can contribute to mismatch between the data from the real sensor and from the numerical simulator . that is , the cds algorithms in the simulations can be less aggressive and therefore the residual noise on the image can be larger than on the actual sensor . after the electron matrix has been converted to voltages , the sense node reset noise and offset fpn noise are added , the v / v gain non - linearity is applied ( if desired ) , the adc non - linearity is applied ( if necessary ) . finally the result is multiplied by adc gain and rounded to produce the signal as a digital number : @xmath146 where @xmath147 is the total voltage signal accumulated during one frame acquisition , @xmath52 is the maximum voltage that can be quantified by an adc , and @xmath123 is the total voltage signal accumulated by the end of the exposure ( integration ) time and conversion . usually @xmath148 from eq . [ eq : sourcefollowervoltage ] after the optional v / v non - linearity is applied . in this case , the conversion from voltages to digital signal is linear . the non - linear case is considered below . in our model , we simulate the integral linearity error ( ile ) of the adc as a dependency of adc gain @xmath149 on the signal value . denote @xmath150 as an adc non - linearity ratio ( e.g. , @xmath151 ) . the linear adc gain can be calculated from eq . [ eq : kadc ] as @xmath152 and used as @xmath149 . the non - linearity coefficient @xmath153 is calculated as : @xmath154 where @xmath48 is the maximum voltage that can be quantified by an adc : @xmath155 where @xmath149 is the linear adc gain . the new non - linear adc conversion gain @xmath156 is then used for the simulations for eq . [ eq : adcconvertersimulation ] . it is assumed that the quantisation error is uniformly distributed between -0.5 and + 0.5 of the lsb and uncorrelated with the signal . denote @xmath157 the quantising step of the adc . for the ideal dc , the quantisation noise is : @xmath158 if @xmath159 then the quantisation noise is @xmath160 dn . the quantisation error has a uniform distribution . we do not assume any particular architecture of the adc in our high - level sensor model . using the formulated photosensor model , we were able to simulate realistically noised images similar to those obtained from a hardware cmos photosensor . a comparison with custom - made 5 t cmos photosensor with @xmath161 pixels of size @xmath162 is provided to validate the developed numerical model . [ review:10r2]the simulations were performed as follows . the parameters of the hardware cmos sensor were taken from the specifications ( see table [ tab : photosensorsim ] ) provided by the manufacturer and used in the simulations . a series of the measurements were conducted both for the hardware sensor and the simulated `` software '' photosensor . the methods and procedures of the measurement were the same for the hardware and the simulated photosensor . for example , in case of estimation of the radiometric function , the flat field for the hardware sensor was generated by an array of leds . the measurements were performed for monochromatic light with wavelength @xmath163 m ( [ review2jei : r2q9 - 2 ] we assume for the same of simplicty that the light is monochromatic , and therefore we need the spectral response of the sensor at one wavelength @xmath163 m ) . [ review:5 - 2r1 ] for the numerical model , the uniform image was used that corresponds to the light signal of the leds . [ review2jei : r2q7 ] the measurements were conducted at room temperature ( @xmath164c ) if not stated otherwise . [ text : nonlinearities ] .parameters of the simulated cmos sensor ( taken from the manufacturer s specifications ) . [ cols= " < , < " , ] as seen from the results in table [ tab : spectrogramcalc ] , the read noise can be indeed considered white , with the non - whiteness coefficient @xmath165 , which is consistent with the model @xcite . the non - flatness coefficient for the hardware sensor is slightly greater than that of the simulations ; the uncertainty of the measurements is larger as well . in this subsection we discuss the complexity of the dark current fpn models . it is shown that the assumptions usually made in the literature ( i.e. , that the dark signal is gaussian ) are not always consistent with the noise in the hardware photosensor . the experimental results of dark signal for long integration times ( see fig . [ fig : plotdarksignallongtimemultiple3in1_longtime_experiment ] ) illustrate the complexity of the dark signal : while for rather short integration time times ( less than 100 seconds ) the distribution can be approximated as gaussian , longer integration times have a lot more complicated structure . this is important is such applications as , e.g. , astronomy , where the integration times can be very long for dim objects . we measured the dark [ review2jei : r1q6 ] signal of the hardware cmos sensor for two different cases : 1 . _ short integration time _ ( less than 100 seconds ) : the only one probability distribution ( main distribution ) is used for description of the noise statistics ; 2 . _ long integration time _ ( longer than 100 seconds ) : a main distribution and a superimposed distribution are used ( for very long integration times the second superimposed distribution is used as well ) . we provide the analysis of dark signal statistics using superimposed distributions . the data provided in this subsection was measured at room temperature ( @xmath164c ) as before . [ [ method - of - measurements-3 ] ] method of measurements : + + + + + + + + + + + + + + + + + + + + + + + we took 32 dark frames for each integration time and averaged them in order to reduce the temporal noise and [ review2jei : r1q5 ] get the dark current fpn data . the normalised frequency of occurrences of noise values in the averaged dark image was estimated . for the estimation of statistical parameters of the dark current fpn we used a maximum likelihood estimator ( mle ) . all estimated statistical parameters reported are within a 95% confidence interval if not stated otherwise . for the description of the dark current fpn ( which is the main contributor of the dark noise , especially in short integration times ) , we use the following distributions : * log - normal distribution : @xmath166 * gamma distribution : @xmath167 $ ] as superimposed distributions , the uniform distribution and gaussian distribution were used , as discussed below . we measured the dark current fpn for the hardware cmos sensor and compared the results with our high - level sensor model . the 5 t cmos hardware sensor has in - built circuitry for noise cancelling that uses correlated double sampling ( cds ) technique . the details of the cds algorithm , however , have not been disclosed by the manufacturer . consequently , the noise on the images is reduced , so the comparison of the numerical models with the hardware data is approximate . the measurements presented in fig . [ fig : plotdarksignallongtimemultiple3in1_shorttimelinlin ] were taken for integration times of 0.001 second , which are typical for , e.g. , wavefront sensing in adaptive optics . the probability density function of the dark current fpn can be considered as gaussian for the short integration times , and the ml estimated values of @xmath168 and @xmath169 for the gaussian distribution were obtained from the experimental data for 0.001 second . while gaussian distribution provides and adequate description of the dark signal , it underestimates the `` outliers '' or `` dark spikes '' , and this is why we use the log - normal distribution ( see fig . [ fig : plotdarksignallongtimemultiple3in1_shorttimelinlin ] , data marked by `` @xmath170 ' ' ) . the mle parameters for the log - normal distribution were found as @xmath171 and @xmath172 . as integration time increases up to 100 seconds , the distribution of dark current fpn tends to be more asymmetric , however , the log - normal parameters remain similar to the very short integration time case ( @xmath173 and @xmath174 for the case of 100 second integration time ) . therefore , the log - normal distribution can still be considered as a reasonable model for the dark signal , which is consistent with the results reported in @xcite . it is noteworthy that the gamma distribution was found to be a better approximation for a 3 t cmos photosensor in @xcite , but for our 5 t cmos sensor the gamma distribution appears to underestimate the `` dark spikes '' , as can be noted from fig . [ fig : plotdarksignallongtimemultiple3in1_shorttimesemilog100_sec_img_2489pdfestimationexpvssim ] . [ review2jei : r1q6 - 1]also , one can notice that the dark current estimated from the measurements of the hardware sensor in this section are different from the model of dark current values in eq . [ eq : darkcurrentrate ] . this is expected , since our hardware cmos sensors has on - chip noise suppressing circuitry ( see subsection [ subsec : cdsdescription ] ) that removes dark noise levels . the non - gaussian nature of dark current fpn is more apparent as the integration time increases . one can see in fig . [ fig : plotdarksignallongtimemultiple3in1_shorttimelinlin150_sec_img_0012pdfestimationexpvssim ] that either the log - normal or gamma distributions alone do not describe the data adequately . therefore , in the case of long integration time , it is necessary to superimpose another probability distribution , as mentioned in @xcite , in order to create a distribution with a very long `` tail '' to simulate `` outliers '' . the overall distribution for the simulation of the dark signal was constructed as follows . first , the gamma distribution was generated with parameters @xmath175 and @xmath176 , which were estimated from the hardware cmos photosensor data . next , the product of two uniform distributions on the interval @xmath177 $ ] was superimposed ( the parameters were chosen to fit both the `` outliers '' and to widen the left part of the probability density ) . the results are shown in fig . [ fig : plotdarksignallongtimemultiple3in1_shorttimelinlin150_sec_img_0012pdfestimationexpvssim ] . one can see a better agreement of the superimposed distribution described above with the experimental estimation of the dark signal statistics . - ) and using a superimposed probability distribution ( -@xmath178- ) . ] the dark signal statistics is getting more complicated with very long integration time . the dark current fpn is no longer the main contributor to the noise statistics : one would expect the noise measurements to be dominated by thermally - generated electron noise . those are the two distributions ( see fig . [ fig : plotdarksignallongtimemultiple3in1_longtime ] ) that underlie the dark signal measurements . the dark signal in these cases may require modelling based on the experimental data for a particular type of a photosensor . in our example , we were able to compose a distribution to describe the experimental dark signal probability distribution using the superimposed distributions as follows : 1 . main distribution : gamma with parameters @xmath179 and @xmath180 ; 2 . superimposed : uniform distribution on interval @xmath181 $ ] ; 3 . superimposed : gaussian distribution with parameters @xmath182 and @xmath183 . the composed distribution describes the dark current fpn ( @xmath184 symbol ) with good accuracy , as seen in fig . [ fig : plotdarksignallongtimemultiple3in1_longtime ] . one can see from the provided data that the approximation of dark current fpn using only a gaussian distribution is poor . on the other hand , the superimposed distributions provide a better description of the complex structure of the dark current fpn . [ subsec : darknoisemeasurementdiscussion ] the modelling of the dark signal is complicated , since it is difficult to derive a model for the parameters of the superimposed probability distributions . the description of long - exposure dark current fpn is complicated because the statistical parameters are dependent on temperature and exposure ( integration ) time in a non - trivial way and likely to vary from one photosensor to another . an additional problem for the analysis of the dark signal is that the exact cds algorithm is usually not disclosed by the manufacturer , and even if disclosed , it is difficult to simulate precisely . furthermore , additional on - chip ( e.g. , transfer noise ) and off - chip noise ( clock - jitter noise , preamplifier noise , etc . ) can contribute @xcite to the resulting noise distribution . nonetheless , the description based on superimposed distributions gives acceptable results for the fixed sensor s parameters ( temperature , photosensor technology ) . the experimental data along with the models developed in this subsection describe the dark current fpn noise for different integration times . three cases were considered : * _ short integration time ( less than 100 seconds ) _ , where a lognormal distribution is adequate as a main distribution , * _ long integration time ( longer than 100 - 150 seconds ) _ , where another distribution must be superimposed to a main distribution ( or several , in case of integration times longer than 300 seconds ) . at short integration times , the noise is the variability of dark current across pixels , that is , dark current fpn ( or dark signal non - uniformity , dsnu ) . [ review2jei : r2_q15 ] at long integration times , one would expect the noise measurements to be dominated by thermally - generated electron noise . those are the two distributions ( see fig . [ fig : plotdarksignallongtimemultiple3in1_longtime_experiment ] and fig . [ fig : plotdarksignallongtimemultiple3in1_longtime ] ) that presumably underlie the measurements . as the integration time increases further , the probability distribution of the resulting dark signal gets more complicated and harder to model . the article provides a literature review of the noise models that are used for the simulations of noise in ccd and cmos photosensors . a high - level model of ccd and cmos photosensors based on a literature review is formulated in this paper . the experimental results for a hardware 5 t cmos photosensor are presented as a validation for the developed model of a photosensor . the formulated model of the cmos photosensor was compared with the data from a custom - made hardware photosensor for validation of the photosensor model . results of photon transfer curve ( ptc ) estimation show the importance of the non - linearity introduced in the model . using the models of v / v and the adc non - linearity allow to better match the properties of the hardware cmos sensor . it has been also demonstrated that the dark current fpn and in particular the dark `` spikes '' in the case of a long integration time has a more complicated distribution than previously discussed in the literature . the gamma distribution with a superimposed uniform and gaussian distributions are shown to be a reasonable approximation of the dark `` spikes '' for long integration time . the article summarises the efforts in numerical simulations of solid - state photosensors providing an extensive survey of literature on noise in photosensors . a high - level model of cmos / ccd sensors has been formulated in order to provide engineers with insight on the noise impact on images quality . using such a high - level photosensor model , one can simulate either ccd or cmos sensors by application of appropriate noise models . the formulated high - level simulation model can be used to synthesise realistically noised images for the development and testing of the image processing algorithms . the authors are gratefully acknowledge personal communications with _ dr . james r. janesick _ and his valuable comments . kolehmainen , j. aikio , m. karppinen , a.j . mattila , j.t . makinen , k. kataja , k. tukkiniemi , and p. karioja . . in _ proceedings of spie - optical modeling and performance predictions _ , volume 5178 , pages 204212 , 2004 . r. costantini and s. ssstrunk . . in _ proceedings of is&t / spie electronic imaging - sensors and camera systems for scientific , industrial , and digital photography applications _ , volume 5301 , pages 408419 , 2004 . mikhail v. konnik and james welsh . open - source framework for documentation of scientific software written on matlab - compatible programming languages . in _ proceedings of spie astronomical telescopes + instrumentation - software and cyberinfrastructure for astronomy ii _ , volume 8451 - 102 , 2012 . mo chen , jessica fridrich , miroslav goljan , and jan luk . source digital camcorder identification using sensor photo response non - uniformity . in _ proceedings of spie - security , steganography , and watermarking of multimedia contents ix _ , volume 6505 , page 65051 g , 2007 . yuan chen , s.m . guertin , m. petkov , d.n . nguyen , and f. novak . a chip and pixel qualification methodology on imaging sensors . in _ in proceedings of 42nd annual ieee reliability physics symposium _ , pages 435 439 , april 2004 . loukianova , h.o . folkerts , j.p.v . maas , d.w.e . verbugt , a.j . mierop , w. hoekstra , e. roks , and a.j.p . leakage current modeling of test structures for characterization of dark current in cmos image sensors . , 50(1):77 83 , 2003 . r. widenhorn , m.m . blouke , a. weber , a. rest , and e. bodegom . . in _ proceedings of spie - sensors and camera systems for scientific , industrial , and digital photography applications iii _ , volume 4669 , pages 193201 , 2002 . n. sakaguchi , n. nakamura , s. ohsawa , y. endo , y. matsunaga , k. ooi , and o. yoshida . dark current fixed pattern noise reduction for the 2/3-in two - million pixel hdtv stack - ccd imager . , 44(10):1658 1662 , 1997 . j. janesick , j. pinter , r. potter , t. elliott , j. andrews , j. tower , m. grygon , and d. keller . . in _ proceedings of spie - high energy , optical , and infrared detectors for astronomy iv _ , volume 7742 , pages 77420b77420b30 , 2010 . thomas j. fellers and michael w. davidson . ccd noise sources and signal - to - noise ratio . technical report , national high magnetic field laboratory , the florida state university , tallahassee , florida , usa , 2010 . h. rhodes , g. agranov , c. hong , u. boettiger , r. mauritzson , j. ladd , i. karasev , j. mckee , e. jenkins , w. quinlin , i. patrick , j. li , x. fan , r. panicacci , s. smith , c. mouli , and j. bruce . . , pages 718 , 2004 . b. fowler , m.d . godfrey , j. balicki , and j. canfield . . in _ proceedings of spie - sensors and camera systems for scientific , industrial , and digital photography applications _ , volume 3965 , pages 126135 , 2000 . g. deptuch , j.d . berst , g. claus , c. colledani , w. dulinski , u. goerlach , yu . gornoushkin , y. hu , d. husson , g. orazi , r. turchetta , j.l . riester , and m. winter . design and testing of monolithic active pixel sensors for charged particle tracking . in _ ieee nuclear science symposium conference record , volume 1 , pages 3/103 -3/110 _ , 2000 . c. xu , s. chao , and m. chan . a new correlated double sampling ( cds ) technique for low voltage design environment in advanced cmos technology . in _ proceedings of the 28th european solid - state circuits conference ( esscirc ) _ , pages 117120 , 2002 . gzen kkl , yusuf leblebici , and sandro carrara . a switched capacitor fully differential correlated double sampling circuit for cmos image sensors . in _ proceedings of 5th international symposium on medical information & communication technology ( ismict ) _ , pages 113116 , 2011 . r. turchetta , p.p . allport , g. casse , a. clark , j. crooks , a. evans , a. fant , a.r . faruqi , m.j . french , r. henderson , q. morrissey , d. prior , m. prydderch , j.j . velthuis , g. villani , n. waltham , b. willmore , and p. willmore . cmos monolithic active pixel sensors ( maps ) : new eyes for science . , 560(1):139 142 , 2006 .

we examine volume - limited samples of galaxies drawn from the sloan digital sky survey to look for relations among internal and collective physical parameters of galaxies as faint as . the internal physical properties of interest include morphology , luminosity , color , color gradient , concentration , size , velocity dispersion , equivalent width of line , and axis ratio . collective properties that we measure include the luminosity and velocity dispersion functions . we morphologically classify galaxies using the three dimensional parameter space of color , color gradient , and concentration index . all relations are inspected separately for early and late type galaxies . at fixed morphology and luminosity , we find that bright ( ) early type galaxies show very small dispersions in color , color gradient , concentration , size , and velocity dispersion . these dispersions increase at fainter magnitudes , where the fraction of blue star - forming early types increases . late types show wider dispersions in all physical parameters compared to early types at the same luminosity . concentration indices of early types are well correlated with velocity dispersion , but are insensitive to luminosity and color for bright galaxies , in particular . the slope of the faber - jackson relation ( ) continuously changes from to when luminosity changes from to . the size of early types is well - correlated with stellar velocity dispersion , , when km s . we find that passive spiral galaxies are well separated from star - forming late type galaxies at equivalent width of about 4 . an interesting finding is that many physical parameters of galaxies manifest different behaviors across the absolute magnitude of about . the morphology fraction as a function of luminosity depends less sensitively on large - scale structure than the luminosity function ( lf ) does , and thus seems to be more universal . the effects of internal extinction in late type galaxies on the completeness of volume - limited samples and on the lf and morphology fraction are found to be very important . an important improvement of our analyses over most previous works is that the extinction effects are effectively reduced by excluding the inclined late type galaxies with axis ratios of .

astro-ph0611607

with the large and homogeneous redshift surveys like the sloan digital sky survey ( sdss ; york et al.2000 ) and two degree field galaxy redshift survey ( 2dfgrs ; colless et al . 2001 ) , our view of the local universe became extended out to a few hundred mega parsecs . from such dense redshift surveys , the small - scale distribution of galaxies is also being revealed in more detail . it is now possible to make very accurate measurements of various relations among the physical properties of galaxies and the local environment . this is a difficult task , which should be completed before theoretical modeling and interpretations are attempted . the study is complicated because there are many physical parameters involved and their mutual relations are non - trivially interrelated . a further complication is the fact that definition of environment depends on the type of objects chosen to trace the large scale - structure . one could divide the physical properties of galaxies into internal and collective ones . examples of internal physical properties are color , luminosity , morphology , star formation rate , velocity dispersion , surface brightness , spectral type , size , and so on . collective properties can be the strength and statistical nature of spatial clustering , peculiar velocity field , luminosity function , velocity dispersion distribution function , halo mass function , axis ratio distribution function and so on ( see fig . 1 ) . on the other hand , the local environments can be spatial or temporal . dependence on the latter means the evolution in cosmic time . spatial environment can have various meanings . traditionally discrete types , such as cluster , group , field , or void , are used to distinguish among different environments of galaxies . recently , the smoothed number density of neighboring galaxies was also used to represent the environment . one could also use other parameters to define the environment of galaxies . density gradient , shear fields , and large - scale peculiar velocity fields can be important parameters affecting the properties of galaxies . the relations among all these physical observables and environment can give us information on galaxy formation and the background universe . many authors in the past have investigated the relations between environment and galaxy properties ( lewis et al . 2002 ; gmez et al . 2003 ; goto et al . 2003a ; balogh et al . 2004a , 2004b ; hogg et al . 2004 ; tanaka et al . 2004 ; kuehn & ryden 2005 ; desroches et al . 2006 ) . there has been much recent progress in studying the correlation between the distribution of galaxies and their basic properties , and the correlation among the properties . for example , the luminosity function of galaxies subdivided by various criteria has been measured . blanton et al . ( 2003a ) presented the luminosity functions of the sdss galaxies defined by galaxy light profile shape . madgwick et al . ( 2002 ) measured the luminosity function of 2dfgrs for different type of galaxies defined by their spectral properties . nakamura et al . ( 2003 ) studied the dependence of the luminosity function on galaxy morphology and the correlation between galaxy morphology and other photometric properties such as color and concentration index using a sample of sdss galaxies with visually identified morphological types . weinmann et al . ( 2006 ) performed an extensive study of the dependence of color , star formation , and morphology of galaxies on halo mass using galaxy groups selected in the sdss dr2 . shen et al . ( 2003 ) examined the correlation between the size distribution of galaxies and other properties such as their luminosity , stellar mass and morphological type . baldry et al . ( 2004 ) analyzed the bivariate distribution of the sdss galaxies in the color versus absolute magnitude space . the fundamental plane and the color - magnitude - velocity dispersion relation for early type galaxies with sdss have been investigated by bernardi et al . ( 2003a , 2005 ) very recently desroches et al . ( 2006 ) showed that the fundamental plane projections of elliptical galaxies depend on luminosity using dr4 of the sdss . in particular , they found that the radius - luminosity and faber - jackson relations are steeper at high luminosity , which we also found in our work using a larger sdss sample with morphologically cleaner early type samples . sheth et al . ( 2003 ) and mitchell et al . ( 2005 ) measured the velocity dispersion distribution function of early type galaxies carefully selected from the sdss samples using selection criteria similar to those of bernardi et al . ( 2003a ) . alam & ryden ( 2002 ) measured the axis ratio distributions for red and blue galaxies fitted by the de vaucouleurs and exponential profiles . this paper focuses on understanding of the relationship among many physical properties of galaxies using a set of volume - limited samples of the sdss galaxies , which are further divided into subsamples of constant morphology and absolute magnitude . this well - controlled experiment allows us to measure the relations among internal and collective properties as a function of morphology and luminosity . in our companion paper we have studied the dependence of various galaxy properties on the local density environment ( park et al . 2006 , hereafter paper ii ) . this paper is organized as follows . in section 2 , we describe our catalog and morphology classification and the physical parameters that we use in section 3 and 4 to investigate the relations among the galaxy properties . we summarize our results in section 5 . the sdss ( york et al . 2000 ; stoughton et al . 2002 ; adelman - mccarthy et al . 2006 ) is a survey to explore the large - scale distribution of galaxies and quasars by using a dedicated @xmath15 telescope at apache point observatory ( gunn et al . the photometric survey has imaged roughly @xmath16 sr of the northern galactic cap in five photometric bandpasses denoted by @xmath17 , @xmath18 , @xmath19 , @xmath20 , and @xmath21 centered at @xmath22 and @xmath23 , respectively , by an imaging camera with 54 ccds ( fukugita et al . 1996 ; gunn et al . the limiting magnitudes of photometry at a signal - to - noise ratio of @xmath24 are @xmath25 , and @xmath26 in the five bandpasses , respectively . the median width of the point - spread function ( psf ) is typically @xmath27 , and the photometric uncertainties are @xmath28 rms ( abazajian et al . 2004 ) . after image processing ( lupton et al . 2001 ; stoughton et al . 2002 ; pier et al . 2003 ) and calibration ( hogg et al . 2001 ; smith et al . 2002 ; ivezi et al . 2004 ; tucker et al . 2006 ) , targets are selected for spectroscopic follow - up observation . the spectroscopic survey is planned to continue through 2008 as the legacy survey and to yield about @xmath29 galaxy spectra . the spectra are obtained by two dual fiber - fed ccd spectrographs . the spectral resolution is @xmath30 , and the rms uncertainty in redshift is @xmath31 km s@xmath12 . because of the mechanical constraint of using fibers , no two fibers can be placed closer than @xmath32 on the same tile . mainly due to this fiber collision constraint , incompleteness of spectroscopy survey reaches about 6% ( blanton et al . 2003b ) in such a way that regions with high surface densities of galaxies become less prominent even after adaptive overlapping of multiple tiles . this angular variation of sampling density is accounted for in our analysis . the sdss spectroscopy yields three major samples : the main galaxy sample , the luminous red galaxy sample ( eisenstein et al . 2001 ) , and the quasar sample ( richards et al . 2002 ) . the main galaxy sample is a magnitude - limited sample with an apparent petrosian @xmath19-magnitude cut of @xmath33 , which is the limiting magnitude for spectroscopy ( strauss et al . it has a further cut in petrosian half - light surface brightness @xmath34 mag / arcsec@xmath35 . more details about the survey can be found on the sdss web site . in our study of galaxy properties , we use a subsample of sdss galaxies known as the new york university value - added galaxy catalog ( nyu - vagc ; blanton et al 2005 ) . this sample is a subset of the recent sdss data release 5 . one of the products of the nyu - vagc used here is a large - scale structure sample dr4plus ( lss - dr4plus ) . for local density estimation in the three - dimensional redshift space in our accompanying paper we use galaxies within the boundaries shown in figure 1 of paper ii , which improves the volume - to - surface area ratio of the survey . there are also three stripes in the southern galactic cap observed by sdss . density estimation is difficult within these narrow stripes , so for consistency with the samples examined in the accompanying paper , we do not use them . the remaining survey region covers @xmath36 deg@xmath35 . the primary sample of galaxies used here is a subset of the lss - dr4plus sample referred to as void0 , which is further selected to have apparent magnitudes in the range @xmath37 and redshifts in the range @xmath38 . these cuts yield a sample of 312,338 galaxies . the roughly 6% of targeted galaxies that do not have a measured redshift due to fiber collisions are assigned the redshift of their nearest neighbor . completeness of the sdss is poor for bright galaxies with @xmath39 because of both the spectroscopic selection criteria ( which excludes objects with large flux within the 3@xmath40 fiber aperture ; the cut at @xmath41 is an empirical approximation of the completeness limit caused by that cut ) and the difficulty of obtaining correct photometry for objects with large angular size . for these reasons , analysis of sdss galaxy samples have typically been limited to @xmath42 . this is unfortunate because it limits the range of luminosity that can be probed ; using the magnitude limits of the void0 sample the range of absolute magnitude is only about 3.1 at a given redshift . to extend the range of magnitude , we attempt to complete the redshift sample by supplementing the catalog with redshifts of bright galaxies , first from sdss itself , and then by matching to earlier redshift catalogs ( the missing galaxies are quite bright and easily observed by earlier surveys ) . we employ another subset of the lss - dr4plus sample referred as bvoid , which is made before redshift determination and includes galaxies with @xmath39 . galaxies in bvoid with sdss redshifts are added to the catalog . for those objects lacking redshifts , we searched earlier catalogs . out of @xmath43 bright objects with @xmath39 in the bvoid sample and within our survey region , we assign redshifts to 3945 galaxies cross matched with the updated zwicky catalog ( uzc ; falco et al . 1999 ) , 807 galaxies from the bvoid sample after redshift determination , 69 galaxies from the center for astrophysics ( cfa ) redshift catalog ] ( zcat 2000 version ) , and 4 galaxies from the _ iras _ point source catalog redshift survey ( saunders et al . we found 336 bright objects to be errors , such as outlying parts of a large galaxies that were deblended by the automated photometric pipeline , blank fields , stars , etc . out of the remaining 66 objects , 30 galaxies are included in our sample , using redshifts from the nasa extragalactic database ( ned ) . only 36 galaxies remain in the bright bvoid sample with no measured redshift . in addition to these bright bvoid galaxies , which have been selected based on their @xmath19-band magnitudes , we have included the redshifts of 340 uzc galaxies cross - matched with sdss objects in nyu - vagc , which were not included in the previous step . out of 6295 uzc galaxies within our survey region , 242 are still not matched with sdss objects in the nyu - vagc . ofthe missing uzc galaxies , 60% have redshifts smaller than our inner redshift cut of @xmath44 , leaving only 99 galaxies missing at larger redshift . in total , we add 5195 bright ( @xmath39 ) galaxies to the void0 samples , which yields a final sample of 317533 galaxies . volume - limited samples derived from the resulting catalog have nearly constant comoving number density of galaxies along the radial direction at redshift @xmath45 . thus , we treat our final samples as having no bright limit at redshift greater than @xmath44 . in this study we use only volume - limited samples of galaxies selected by absolute magnitude and redshift limits . figure 2 shows the definitions of our six subsamples in redshift - absolute magnitude space . the shallowest subsample is labeled cm , which stands for ` color - magnitude ' . we employ the cm sample to extend some of our analyses to fainter absolute magnitude . the deeper and thicker ( along the line of sight ) five samples , from d1 to d5 , are the main samples for study of galaxy properties . the definitions for all samples are summarized in table 1 . lccccc cm & @xmath46 & @xmath47 & @xmath48 & 11756 ( 3467)&3.00 + d1 & @xmath49 & @xmath50 & @xmath51 & 20288 ( 6256 ) & 3.41 + d2 & @xmath52 & @xmath53 & @xmath54 & 32550 ( 11341 ) & 3.78 + d3 & @xmath55 & @xmath56 & @xmath57 & 49571 ( 19270 ) & 4.18 + d4 & @xmath58 & @xmath59 & @xmath60 & 74688 ( 33039 ) & 4.58 + d5 & @xmath61 & @xmath62 & @xmath63 & 80479 ( 39333 ) & 5.56 + the comoving distance and redshift limits of each volume - limited sample defined by an absolute magnitude limit are obtained by using the formula @xmath64 where @xmath65 is the mean @xmath66-correction , @xmath67 is the mean luminosity evolution correction , and @xmath19 is the comoving distance corresponding to redshift @xmath21 . we adopt a flat @xmath68cdm cosmology with density parameters @xmath69 and @xmath70 to convert redshift to comoving distance . to determine sample boundaries we use a polynomial fit to the mean @xmath66-correction , @xmath71 we apply the mean luminosity evolution correction given by tegmark et al . ( 2004 ) , @xmath72 . the rest - frame absolute magnitudes of individual galaxies are computed in fixed bandpasses , shifted to @xmath73 , using galactic reddening corrections ( schlegel et al . 1998 ) and @xmath66-corrections as described by blanton et al . ( 2003c ) . this means that a galaxy at @xmath73 has a @xmath66-correction of @xmath74 , independent of its spectral energy distribution ( sed ) . we have applied the same mean luminosity evolution correction formula to individual galaxies . the physical parameters we consider in this study are @xmath75 color , absolute petrosian magnitude in the @xmath19-band @xmath76 , morphology ( see below ) , petrosian radius , axis ratio , concentration index , color gradient in @xmath77 color , velocity dispersion , and equivalent width of the h@xmath78 line . to compute colors , we use extinction and @xmath66-corrected model magnitudes . the superscript 0.1 means the rest - frame magnitude @xmath66-corrected to the redshift of 0.1 . all our magnitudes and colors follow this convention , and the superscript will subsequently be dropped . to measure some of the physical parameters of galaxies , we retrieve the @xmath18- and @xmath20-band atlas images and basic photometric parameters of all galaxies in our sample from the archive of photometric reductions conducted by the princeton / nyu group . to take into account flattening or inclination of galaxies , we use elliptical annuli in all parameter calculations , and the isophotal position angle and axis ratio in the @xmath20-band are used to define the elliptical annuli . the @xmath3 color gradient is defined to be the color difference between the region with @xmath79 and the annulus with @xmath80 , where @xmath81 is the petrosian radius . the ( inverse ) concentration index is defined by @xmath82 where @xmath83 and @xmath84 are the semi - major axis lengths of ellipses containing 50% and 90% of the petrosian flux in the @xmath20-band image , respectively . the @xmath3 color gradient , concentration index , and isophotal axis ratios are corrected for the effects of seeing . to do so , we first generate a large set of sersic model images with various scaling lengths , slopes , and axis ratios convolved with a range of psf sizes . for an observed galaxy image with a given size of the psf we look for a best - fit sersic model when the sersic model is convolved with the psf to find the true sersic index and the true axis ratio . the fit is made at radii between @xmath85 and @xmath86 to avoid the central region , whose profile is strongly affected by seeing . when the best - fit sersic models are found in the @xmath18 and @xmath20 bands , the @xmath3 color gradient and concentration index are calculated for the true sersic models and their differences from those of convolved images are used to correct the color gradient of the observed image for the seeing effects . we classify galaxy morphologies into early ( ellipticals and lenticulars ) and late ( spirals and irregulars ) types using the automated galaxy morphological classification method given by park & choi ( 2005 ) , who showed that incorporating the @xmath3 color gradient into the classification parameter space allows us to successfully separate the two populations degenerated in @xmath2 color and the concentration index helps us separate red disky spirals from early type galaxies ( see fig . 1 of park & choi 2005 ) . it should be noted that the `` morphological '' properties of galaxies can change when different bands are adopted . the difference in morphology in different bands is exactly the color distribution , which is reflected in our color gradient parameter . to define the classification boundaries in this multi - parameter space , they used a training set of 1982 sdss galaxies whose morphological types are visually identified . the distributions of the early and late type galaxies in sample d2 are shown in figures 6@xmath87 and 6@xmath88 , below . the reader is referred to park & choi ( 2005 ) for further details of the methods and various tests . we have visually examined color images of a set of faint galaxies retrieved by the sdss image list , and confirmed that the resulting morphological classification is highly successful with completeness and reliability reaching 90% . for the volume - limited samples d2 and cm we made an additional visual check to correct possible misclassifications of the automated scheme for blue early types ( those below the straight line in fig . 3@xmath87 ) and red late types ( those redder than @xmath89 in @xmath2 color ) . the morphological types of 1.9% of galaxies , which are often blended or merged objects , were changed by this procedure . this morphology classification allows us to make very accurate measurements in this study compared with previous work . we discuss this further in section 3.2 . we find that absolute magnitude and morphology are the most important parameters characterizing physical properties of galaxies , in the sense that other parameters show relatively little scatter once absolute magnitude and morphology are fixed . this is particularly true for early type galaxies . we begin this section by showing the dependence of various physical parameters of galaxies on their absolute magnitudes . in each case , we separately examine samples of early ( e / s0 ) and late ( s / irr ) morphological types . figure 3@xmath87 shows the color - absolute magnitude diagram of galaxies in the cm and d2 samples . galaxies in the cm sample are shown only for @xmath90 . the left panel shows the distribution of early types , and the right panel shows that of late types . the solid and dashed lines delineate the most probable @xmath2 color of galaxies in the d2 and cm samples , respectively . ( here and throughout , the `` most probable '' parameter value within 0.4 absolute magnitude bin is the mode of the distribution of that parameter . ) the most probable color of the red sequence galaxies has a break at @xmath91 and has slopes of about @xmath92 at the bright side and about @xmath93 at the faint side . we plot only those late type galaxies with axis ratios @xmath94 , to reduce the biases in absolute magnitude , color , color gradient , etc . , due to internal extinction . colors of late types have a much wider dispersion . their most probable color rapidly becomes bluer as galaxies become fainter , but is very insensitive to absolute magnitude fainter than @xmath95 , approaching @xmath96 at the very faint end ( @xmath97 , not shown ) . baldry et al . ( 2004 ) reported a slope of about @xmath98 for luminous red galaxies selected from the sdss , which is consistent with our measurement . one significant difference between baldry et al.s and our result is the color - magnitude relation of the late type ( blue distribution ) galaxies at the bright side . in their results the slope of the relation becomes less steep long before the color - magnitude relation for late types touches that of early types as one , toward brightest magnitude . on the other hand , our result shows no such bending until it approaches the relation for early types . the difference is probably caused by extinction of late type galaxies . figure 3@xmath87 shows that the color distribution of late type galaxies overlaps with early types at all magnitudes , and vice versa . the solid straight line in figure 3@xmath87 roughly divides the sequences of red and blue galaxies . however , about 24% of spiral / irregular galaxies brighter than @xmath99 are above this line when all inclinations are allowed . among all galaxies above this line , 32% are actually late types . therefore , an early type sample generated by a simple cut in the absolute magnitude versus color space must take into account this huge contamination . we also find a significant fraction of early types located outside the main red sequence at faint magnitudes . at absolute magnitudes between @xmath100 and @xmath101 , about 10% of early types are located below the straight line shown in figure 3@xmath87 . this outlying fraction is larger at fainter magnitudes . some of these blue early type galaxies may be actually spirals whose disks look too faint for both our automated and visual classification schemes to classify correctly , but the trend seems robust . again , it should be noted that even though the bimodality in galaxy color is mainly due to the bimodality in galaxy morphology , there is a significant overlap in color space between the two morphological classes . with few exceptions , very blue ( @xmath102 ) early type galaxies show strong emission lines in their spectra , indicating active star formation ( see fig . 6@xmath103 ) . the distributions of colors of early and late types at a given absolute magnitude are not gaussian . the colors of early types are skewed to blue color , and those of late types are skewed to red color . the blue sequence appears less prominent when the data is contaminated by internal extinction effects ( see fig . 12 , below ) . figure 3@xmath88 shows distributions of galaxies in color gradient , @xmath104 , versus absolute magnitude , @xmath76 . we find that the majority of e / s0 s ( _ left panel _ ) have a weak negative color gradient ; the envelope is slightly bluer than the core . their most probable color gradient is surprisingly constant at @xmath105 at all magnitudes explored , with a weak trend for more positive gradient toward faint and bright magnitudes . interestingly , the e / s0 s with @xmath106 have the minimum color gradient . it should be noted that the color gradient of the brightest early types approaches zero , meaning that the brightest early types are reddest and most spatially homogeneous in color . although the most probable color gradient has a nearly constant of absolute magnitude , the number of e / s0 s with blue cores increases strongly at absolute magnitudes fainter than @xmath107 . the color gradient of late type galaxies is more negative ( i.e. , bluer envelope ) than that of early types and is a slowly increasing function of absolute magnitude faint late types are more gas rich and have more uniform internal star formation activity , while bright late types have red central bulges and blue star forming disk structures . figure 3@xmath103 shows that the ( inverse ) concentration index @xmath108 of the early types ( _ left panel _ ) is nearly independent of absolute magnitude when @xmath109 , but rises rather steeply at fainter magnitudes . the most probable concentration index of the bright early types is close to that of a de vaucouleurs profile ( @xmath110 ) , even though it slightly increases at the brightest end . the brightest early types might be less concentrated because of more recent mergers , and the galaxies only slightly brighter than @xmath111 , with @xmath112 might be dynamically older ( more relaxed ) . for late type galaxies the most probable value is close to the value an exponential disk ( @xmath113 ) at the brightest end , albeit with a large dispersion . the petrosian radius in units of @xmath114 kpc as a function of absolute magnitude is shown in figure 4@xmath87 . the petrosian radius is calculated by using the elliptical annuli , and is typically larger than the value in the sdss photometric database which uses the circular annuli . early type galaxies ( _ left panel _ ) exhibit a tight correlation between physical size and luminosity , and the relation has a significant curvature . the slope is very steep ( @xmath115 at @xmath116 ) at the brightest end . the late types have larger @xmath81 , and their size relation with absolute magnitude is less curved . the curvature of the relation for both early and late types means that fainter galaxies have lower surface brightness . as a guide the radius - magnitude relation in the case of constant surface brightness ( @xmath117 where @xmath118 is luminosity and @xmath119 is surface brightness ) is drawn in both panels . the slope of this relation to the bright end of the late types tells that the brightest spiral galaxies , with @xmath120 , have surface brightness approximately independent of luminosity . this relation is also true for the early types in the regime @xmath121 fainter than for the late types . desroches et al . ( 2006 ) also found that the radius - luminosity relation of elliptical galaxies is steeper at high luminosity . shen et al . ( 2003 ) measured the size distributions of early and late type galaxies , divided by the concentration index or the sersic index , and found that they are well fitted by a log normal distribution . figure 4@xmath88 shows the variation with absolute magnitude of velocity dispersion , as measured by an automated spectroscopic pipeline called idlspec2d version 5 ( d. schlegel et al . 2007 , in preparation ) . galaxy spectra of the sdss are obtained by optical fibers with radius of @xmath122 . the finite size of this aperture smooths the central velocity dispersion profile . to correct the central velocity dispersion for this smoothing effects we adopt a simple aperture correction , @xmath123 ( jrgensen et al . 1995 ; bernardi et al . 2003b ) , where @xmath124 is the equivalent circular effective radius @xmath125 where @xmath126 is the seeing - corrected effective angular radius along major axis of the galaxy from model fits to a de vaucouleurs profile in the @xmath20 band . because the measured velocity dispersion is systematically affected by the finite resolution of the spectrographs at small values , the calculation of the most probable velocity dispersion is restricted to galaxies with @xmath127 km s@xmath12 . the errors in the measured velocity dispersions are about @xmath128 at 70 km s@xmath12 and @xmath129 at 50 km s@xmath12 . the dispersion in @xmath10 is typically @xmath130 , and the measurement error is typically 11 km s@xmath12 ( see section 4.2 ) . from these data we fit the slope , @xmath131 , defined by @xmath132 . at the bright end , near absolute magnitude @xmath133 , we find @xmath6 , while the slope at magnitudes between @xmath134 and @xmath9 drops to @xmath135 ( all five volume - limited samples are used for the measurements ) . for comparison , we plot two straight lines with slopes of @xmath136 ( _ upper left _ , fig . 4@xmath88 ) and 3 ( _ lower right _ , fig . 4@xmath88 ) , the former being the slope of the faber - jackson relation . recently , desroches et al . ( 2006 ) also found that the slope @xmath131 is steeper at high luminosity , which is consistent with our findings . the measured velocity dispersion of spiral galaxies brighter than @xmath137 is @xmath138% of that of early types ( or @xmath139 ) at a given absolute magnitude . because sdss fiber spectra sample the light from only the central @xmath140 radius region of galaxies , the measured spectral parameters are not fully representative of the physical properties of spiral galaxies , which tend to have large variation of star formation activity from the bulge to outer disk nevertheless , parameters based on fiber spectra , such as line widths and those from the principal component analysis , are often used to characterize the star formation activity , morphological types , etc . ( e.g. tanaka et al . 2004 ; balogh et al . thus , for comparison , we show such results for our samples . the right panel of figure 4@xmath103 shows the equivalent width of the @xmath141 line of late types as a function of @xmath76 . the equivalent width of the @xmath1 line is often used as a measure of recent star formation activity ( or nuclear activity ) . the majority of late types have large @xmath142 . the plot shows that fainter galaxies tend to have larger @xmath142 . this result should be considered carefully because the fiber spectra systematically miss the light from the outer disks of bright large galaxies , which might be active in star formation . the figure also shows that there is a class of late types that have weak @xmath141 line emission . the placement in this diagram of some of these objects may be due to the finite fiber size , but some of them are genuine passive spirals ( couch et al . 1998 ; poggianti et al . 1999 ; goto et al . 2003b ; yamauchi & goto 2004 ) . the left panel of figure 4@xmath103 shows @xmath142 of early type galaxies . we find that the number of early types with central star formation activity increases sharply at absolute magnitudes fainter than @xmath143 . we find similar trends for @xmath2 color and @xmath144 color gradient ( see also fig . 6 , below ) . early types with central star formation activities have been studied and some of them are @xmath145 galaxies ( goto et al . 2003c ; fukugita et al . 2004 ; quintero et al . 2004 ; goto 2005 ; park & choi 2005 ; yamauchi & goto 2005 ) . figure 5 shows the isophotal axis ratio of early ( _ left panel _ ) and late ( _ right panel _ ) types as a function of absolute magnitude . we choose the isophotal axis - ratio rather than axis ratios from de vaucouleurs or exponential profile fits that are computed by the automated sdss pipeline in our study because the isophotal position angles correspond most accurately to the true orientation of the major axis . the exponential profile fit is often bad particularly for barred spiral galaxies . curves in figure 5 are the median axis ratios of galaxies in the d2 ( _ solid line _ ) and cm ( _ dashed line _ ) samples . we also draw the median curves in red and blue colors by using _ only _ galaxies with @xmath146 . we clearly see that early type galaxies tend to be rounder at brighter magnitudes . figure 5 demonstrates that the absolute magnitudes of inclined late types with @xmath147 seriously suffer from dimming due to the internal absorption , while the effects are ignorable for those with @xmath94 . see also figure 12 , below and discussion in section 4.3 . figure 6@xmath87 shows the distribution of e / s0 ( _ left panel _ ) and s / irr ( _ right panel _ ) galaxies in the d2 sample in the @xmath2 color versus @xmath104 color gradient space . our morphological classification is based primarily on this parameter space . note that the sharp boundaries shown in figure 6a are due to a choice of our morphological criteria in the @xmath2 versus @xmath3 color gradient space , which is determined to correlate well with visually classified morphologies . scatter across the classification boundaries is caused by the concentration index constraint on early types ( see park & choi 2005 for details ) and corrections made by visual inspection . most early type galaxies fall within a strong concentration near @xmath148 and @xmath149 in this plane . because the color of the red sequence becomes bluer at fainter magnitudes ( see figure 3a ) , the center of this concentration moves to bluer @xmath2 when one examines fainter early types . there is also a trail of early types towards bluer colors and more positive color gradients . the classification boundary separating this blue early types from late types has been determined from a training set that showed such a trend ( park & choi 2005 ) . the majority of galaxies in this early type trail are fainter than @xmath111 ( _ blue points _ ) , often show emission lines ( see early types in fig . 6@xmath103 ) , and live mainly at intermediate and low density environments ( see early types in fig . 11@xmath87 of paper ii ) . these blue early types amount to 10% of early types , which affects the morphology fraction as a function of luminosity , the type - specific luminosity functions , and so on . in this manner , the color gradient and concentration constraints in our morphological classification reduce the misclassification of red late type as early type , and blue early type as late type . this is one of the major improvements we made in this study compared with previous works . late type galaxies brighter than @xmath111 are redder and have more negative color gradients than fainter late types . figure 6@xmath88 shows galaxies in the @xmath2 versus the ( inverse ) concentration index space . the brightest early types have @xmath150 and @xmath151 . there is a weak tendency for the redder ( @xmath152 ) bright early types to have smaller @xmath108 than the bluer ( @xmath153 ) fainter ones . late types are loosely concentrated at @xmath154 and @xmath155 . they have a broad overlap with the region occupied by early type galaxies . figure 6@xmath103 shows the relations between @xmath2 color and the equivalent width of @xmath1 line , @xmath156 . most red early type galaxies have weak emission , @xmath157 . because galaxies in the blue trail of early types usually have blue star - forming centers , @xmath156 in the light from their centers has a tight correlation with their color . the right panel of figure 6c shows that the @xmath1 emission line strength of the centers of late type galaxies ( plotted as @xmath158 $ ] ) is close to being linearly proportional to the overall @xmath2 color . this is particularly true for blue late types . this proportionality is as expected if @xmath1 emission is proportional to @xmath17-band luminosity . it is interesting to observe that the sequence formed by the blue early types is far from a straight line and has higher @xmath156 than the sequence of the blue late types . stellar velocity dispersion is a dynamical indicator of the mass of galaxies . figure 7@xmath87 shows that there is a good correlation between velocity dispersion and concentration index , and thus between mass and surface brightness profile . here it should be noted that the d2 sample starts to be incomplete at @xmath159 km s@xmath12 due to the absolute magnitude cut and the significant dispersion of @xmath10 at a given absolute magnitude ( see fig . 4@xmath88 ) . the relation becomes tighter for high-@xmath10 early type galaxies and approaches the de vaucouleurs profile value of @xmath160 at the highest end . the size of galaxies is well correlated with velocity dispersion when @xmath161 km s@xmath12 , but the dependence become weaker at lower @xmath10 ( figure 7b ) . note that the number of galaxies with @xmath162kpc is underestimated in the d2 sample . the error in @xmath10 is typically 11 km s@xmath12 ( see below ) , which somewhat broadens the size versus @xmath10 relation . these systematic effects should be taken into account in interpreting figure 7@xmath88 . figure 7@xmath103 indicates that there are progressively more centrally star - forming early type galaxies as the velocity dispersion decreases , consistent with the relation between @xmath156 and @xmath76 ( fig . 4@xmath103 ) . the upper panel of figure 8 shows the luminosity function ( lf ) of galaxies in the d3 sample . filled circles are the lf measured by using both early and late type galaxies . open circles are the lf of the e / s0 morphological types , and open squares are that of the s / irr types . symbols are calculated by binning galaxies in each magnitude bin . uncertainty limits are estimated from 16 subsets of each sample . curves are the best - fit schechter functions of the following form @xmath163 we use the minuit package of the cern program library to determine the parameters in the schechter function by the maximum likelihood method , which is applied to individual galaxies as described in sandage et al . the schechter function fits the data extremely well except for the highest luminosity bins . to reduce the bias in lf due to the internal absorption , we have included only those late types with axis ratios @xmath164 in our analyses . to estimate the number density ( or @xmath165 ) of all spiral galaxies , we need to know the fraction of late type galaxies with apparent axis ratios greater than @xmath166 . we adopt the fraction @xmath167 calculated from the cm sample ( see fig . 13@xmath88 , below ) , whose spiral membership is considered to be the least affected by the internal absorption . we use this fraction to scale up the late type galaxy lf estimated by using only those with @xmath164 . the results of the schechter function fits are summarized in table 2 . lcccc d1&@xmath49 & @xmath168 & @xmath169 & @xmath170 + d2&@xmath52 & @xmath171 & @xmath172 & @xmath173 + d3&@xmath55 & @xmath171 & @xmath174 & @xmath175 + d4&@xmath58 & @xmath176 & @xmath177 & @xmath178 + d5&@xmath61 & @xmath179 & @xmath180 & @xmath181 + d1&@xmath49 & @xmath182 & @xmath183 & @xmath184 + d2&@xmath52 & @xmath185 & @xmath186 & @xmath187 + d3&@xmath55 & @xmath188 & @xmath189 & @xmath187 + d4&@xmath58 & @xmath190 & @xmath191 & @xmath192 + d5&@xmath61 & @xmath193 & @xmath194 & @xmath195 + d1&@xmath49 & @xmath196 & @xmath197 & @xmath198 + d2&@xmath52 & @xmath199 & @xmath200 & @xmath201 + d3&@xmath55 & @xmath202 & @xmath203 & @xmath204 + d4&@xmath58 & @xmath205 & @xmath206 & @xmath207 + d5&@xmath61 & @xmath208 & @xmath209 & @xmath210 + since our volume - limited samples from the sdss do not extend to very faint magnitudes , there is a significant correlation between measured @xmath111 and @xmath78 . we choose the d3 sample to estimate a representative lf of the sdss galaxies as a compromise ; the uncertainty in @xmath107 is minimized for d4 , while the uncertainty in @xmath78 is minimized for d2 . however , the d3 sample covers the absolute magnitude space brighter than @xmath101 , only about 1.2 mag fainter than @xmath111 , and the correlation between the estimated parameters @xmath111 and @xmath78 makes their physical meanings less clear . the lf of the deeper d4 sample , which includes the sloan great wall ( gott et al . 2005 ) , shows the highest density of early type galaxies among all five volume - limited samples , and thus its lf seems less representative . the d5 sample does not extend faint enough to determine the @xmath78 parameter . the existence of the large scale structure affects both normalization and shape of lf ( e.g. croton et al . therefore , the trends in @xmath111 and @xmath78 are affected by the fact that nearby universe is a relatively underdense region and the shallower samples are more dominated by the relatively faint galaxies which prefers underdense regions ( see fig . 10 of paper ii for further details ) . when all spiral galaxies are used ( i.e. no internal absorption effect correction ) , the lf of all galaxies in the d3 sample is well fitted by @xmath211 and @xmath212 . for comparison , blanton et al . ( 2003a ) report @xmath213 , and @xmath214 for an apparent magnitude - limited sample called sdss lss sample 10 ( similar to sdss dr1 ) . if we adopt the faint end slope @xmath78 from the shallow sample d1 and @xmath111 from the deepest sample d5 , our estimates are @xmath215 and @xmath216 , which are very close to blanton et al.s . when the internal absorption effect in the sample is reduced , the lf of spirals changes , particularly in the @xmath78 parameter . the lf of early types in sample d3 has @xmath217 and @xmath218 while that of late types has a fainter characteristic magnitude of @xmath219 and a steeper faint - end slope of @xmath220 . if all spiral galaxies are used , the lf of late types has @xmath221 and @xmath222 . this demonstrates that the faint - end slope @xmath78 of late types would have been misleadingly measured steeper if the inclined late type galaxies had not been excluded . the blue galaxies classified as early types by our morphological classifier make an increasing contribution to the lf at faint magnitudes . we suspect that some of these objects are actually bulge - dominated spirals whose disks are lost to the background sky . however , the absolute contribution of blue early type galaxies to the lf is still small at the magnitudes studied , as seen in figure 3@xmath87 , and the type - specific lfs at magnitudes brighter than @xmath223 seems well determined by our samples . baldry et al . ( 2004 ) used @xmath2 color - selected early ( red ) and late ( blue ) type galaxy samples to measure their luminosity functions . the estimated @xmath111 parameter differed by 0.21 mag and the @xmath78-parameter was measured to be @xmath224 ( early ) and @xmath225 ( late ) . the difference in @xmath111 between early and late types is quite consistent with our result , but they obtained steeper @xmath78 s . the bottom panel of figure 8 shows the fraction of e / s0 galaxies in all five volume - limited samples as a function of @xmath19-band absolute magnitude . again , only late types with @xmath164 are used , and their number fraction of 0.505 is used to infer the total number of spiral galaxies . there is a surprisingly consistent monotonic relation between luminosity and early type galaxy fraction . the critical magnitude at which the early type fraction exceeds 50% is @xmath226 . even though the shape of lf itself shows significant fluctuations due to large scale structures when measured in different regions of the universe , the morphological fraction as a function of luminosity is relatively less sensitive to large scale structures and thus seems to be more universal . however , it is well - known that the early type fraction is a monotonically increasing function of local density ( dressler 1980 ; tanaka et al . 2005 ; goto et al . 2003a ; paper ii ) . weinmann et al . ( 2006 ) also measured the morphology fraction as a function of @xmath76 and found basically the same dependence , even though the sample they used is smaller . the velocity dispersion function ( vdf ) of early type galaxies can be used to derive gravitational lens statistics and to estimate cosmological parameters . to estimate the velocity dispersion function we need to have a sample of galaxy velocity dispersion data that is complete over a wide range . figure 4@xmath88 indicates that sample d2 , for example , starts to be incomplete at @xmath159 km s@xmath12 due to the absolute magnitude cut and the significant dispersion of @xmath10 at a given absolute magnitude . to lower the completeness limit down to @xmath227 km s@xmath12 we have used a series of our volume - limited samples an additional very faint volume - limited sample with @xmath228 , which is added because even the cm sample is not faint enough to give the @xmath10-distribution complete down to @xmath229 km s@xmath12 . the differences in volumes of these volume - limited samples are taken into account when we estimate the velocity dispersion function from a combination of seven samples . the circles in figure 9 show the vdf of early type galaxies measured in this way . the curve is the fit by a modified schechter function of the form @xmath230{\beta \over \sigma(\alpha/\beta)}{d\sigma \over \sigma},\ ] ] where @xmath231 is the characteristic velocity dispersion , @xmath78 is the low - velocity power - low index , and @xmath232 is the high - velocity exponential cutoff index ( mitchell et al . the vdf parameters obtained by the minuit package are @xmath233 when we made the fit with the above parameters , we assumed that the measurement error in @xmath10 was approximated by @xmath234 km s@xmath12 ( see below ) . the dashed curve in figure 9 is the modified schechter function best fit to the raw data points ( _ open circles _ ) which are broadened at high @xmath10 s due to the measurement error . the solid curve is the deconvolved vdf whose parameters are given above . it should be noted that there exists very strong correlations among the fitting parameters , and the results of the fitting depend sensitively on the range of the velocity dispersion used . our estimate of the vdf of early type galaxies is quite different from that of sheth et al . ( 2003 ) and mitchell et al . the dotted curve in figure 9 is the vdf best fit given by sheth et al . mitchell et al . ( 2005 ) presented the same function with the normalization @xmath165 lowered by 30% . the difference at small @xmath10 seems due to the fact that some of morphologically early type galaxies are discarded based on spectrum in sheth et al . ( cf . fig . 12 of mitchell et al . 2005 ) . it is not clear what makes the difference at large @xmath10 , but it could be due to the measurement error larger than that adopted here . in order to check whether or not our vdf is seriously affected by the measurement error in @xmath10 we have made the following test . we first adopt sheth et al . ( 2003 ) s vdf as the true one . we then look for galaxies having 4 or 5 repeat measurements of @xmath10 in southern stripe 82 ( plates 406 and 412 ) to estimate the error . the thick histogram in figure 10 shows the distribution of the standard deviations of @xmath10 of the galaxies with repeat measurements . the median size of error is 11 km s@xmath12 . one can note that the error estimated from the repeat measurements by the idlspec2d is larger than the formal error ( _ thin histogram _ ) given by the idlspec2d , but is comparable with that ( _ dotted histogram _ ) obtained from the repeat measurements by the official spectroscopic pipeline , spectro1d ( written by m. subbarao , m. bernardi and j. frieman ) . if the measurement error were larger than 20 km s@xmath12 , the dispersion of the relation between @xmath76 and @xmath10 shown in figure 4@xmath88 would be much larger . we then calculate the vdf affected by the error in @xmath10 . the dotted curve in figure 11 is sheth et al.s function , and the short - dashed curve is the vdf deformed due to the gaussian - distributed measurement error of @xmath234 km s@xmath12 , which reflects the fact that the error in @xmath10 estimated from the repeat measurements increases as @xmath10 increases . it demonstrates that our vdf can not be realized from sheth et al.s function by the convolution effects of the measurement error the scatter plot of galaxies in figure 5 indicates that the mean axis ratio of bright late type galaxies tend to be high . this is due to the internal absorption making inclined galaxies appear fainter . the bottom panel of figure 12 shows the late type galaxies of sample d3 in the @xmath76 versus @xmath2 color space as in the right panel of figure 3@xmath87 . galaxies with @xmath235 are shown as blue dots , and those with @xmath236 are shown as red dots . their most likely color curves are also drawn . we find that the group of highly inclined galaxies is significantly shifted with respect to the almost face - on ones toward fainter magnitudes and redder colors . we also find that the shapes of the two distributions are different , which can be explained if the galaxies with intermediate magnitude ( @xmath237 ) and color ( @xmath238 ) suffer from more dimming and reddening due to internal extinction . the color of very red or very blue galaxies seems to be less affected by the internal extinction . a detailed study of internal extinction is needed for the sdss galaxies . the early types are not much affected . if one does not take into account this inclination effect on magnitude , the measured lf of late type galaxies and the morphology fraction will be seriously in error . in analyses where these systematics are important , we have used late types with @xmath239 . figure 5 demonstrates that the inclination effects on magnitude is small for this set of late type galaxies . because of the inclination effects on magnitude , our volume - limited samples are not complete for late type galaxies near the absolute magnitude cuts . the lower panel of figure 13 shows distributions of isophotal axis ratio of late type galaxies in samples cm ( _ solid line with filled circles _ ) , d2 ( _ short - dashed line with open circles _ ) , and d4 ( _ long - dashed line with open squares _ ) . there appears to be no late type galaxy with nearly zero axis ratio because of the intrinsic thickness of disks of late type galaxies . the axis ratio of 0.2 due to the intrinsic thickness is consistent with previous works ( sheth et al . almost face - on galaxies are also lacking , because disks of late type galaxies do not appear exactly round even when they are face - on ( kuehn & ryden 2005 ) . the distributions are significantly different from one another because of the internal absorption effects , lacking highly inclined galaxies in the samples with bright cuts . we assume that the magnitude limit of cm sample ( @xmath240 ) is faint enough that its @xmath241 distribution is close to the intrinsic distribution . the fraction of galaxies with @xmath164 in the cm sample is 0.505 . the upper panel of figure 13 shows the distribution of projected axis ratio of early type galaxies . distributions obtained from three samples with different magnitude cuts match one another very well . this reflects the fact that internal absorption in early type galaxies is negligible . alam & ryden ( 2002 ) measured the axis ratio distributions for red and blue populations in the sdss edr sample . the distribution functions for red or blue galaxies well - fitted by a de vaucouleurs profile match well with our result for early types . sheth et al . ( 2003 ) also measured the distribution of axis ratio of early and late type galaxies in the sdss . their axis ratio distribution of early types agrees well with ours except at the @xmath242 bin . we have elaborately investigated the relations among various internal ( e.g. color , luminosity , morphology , star formation rate , velocity dispersion , size , radial color gradient , and axis ratio ) and collective properties ( e.g. luminosity function , velocity dispersion distribution function , and axis ratio distribution function ) . we will study other collective properties such as the correlation function , power spectrum , topology , and peculiar velocity field , in separate works . 1 . to extend the absolute magnitude range , we have added the redshifts of the bright galaxies with @xmath39 to the sdss - nyu - vagc . six volume - limited samples with faint limits from @xmath243 to @xmath244 , are used for more efficient use of the flux - limited sdss sample . we have divided the samples into subsamples of early ( e / s0 ) and late ( s / irr ) morphology types by using our accurate morphology classifier , working in the three - dimensional parameter space of @xmath2 color , @xmath3 color gradient , and concentration index . the morphology classifier is able to separate the blue early types from the blue late types , and also to distinguish the spirals from the red e / s0 types because it does not depend only on color . the late type galaxies with isophotal axis ratio of less than 0.6 have been excluded because they suffer severely from dimming and reddening due to internal extinction and then may cause biases in luminosity , luminosity function , color , color gradient , and so on . this correction makes the blue sequence of the late type galaxies bluer and brighter compared to that obtained from all late type galaxies . the blue sequence also become better - defined with smaller width . we have found that absolute magnitude and morphology are the most important parameters in characterizing physical properties of galaxies . in other words , other parameters show relatively small scatter once absolute magnitude and morphology are fixed for early type galaxies , in particular . one very interesting fact we have noticed in our work is that many physical parameters of galaxies manifest different behaviors across the absolute magnitude of about @xmath245 . for example , the red sequence of early types changes the slope at @xmath246 in the color - magnitude space . also , at a magnitude fainter than @xmath247 , the number of blue star - forming early types increases significantly , and the surface brightness profile as parameterized by @xmath108 becomes less centrally concentrated . the passive spirals with vanishing @xmath248 emission seem to have magnitudes brighter than @xmath249 . at fixed morphology and luminosity , we find that bright ( @xmath4 ) early type galaxies show very small dispersions in color , color gradient , concentration , size , and velocity dispersion . these dispersions increase at fainter magnitudes , where the fraction of blue star - forming early types increases . concentration indices of early types are well - correlated with velocity dispersion , but are insensitive to luminosity and color for bright galaxies , in particular . the slope of the faber - jackson relation ( @xmath5 ) continuously changes from @xmath6 to @xmath7 when luminosity changes from @xmath8 to @xmath9 . the size of early types is well - correlated with stellar velocity dispersion , @xmath10 , when @xmath11 km s@xmath12 . late type galaxies show wider dispersions in all physical parameters compared to early types at the same luminosity . we find that passive spiral galaxies are well - separated from star - forming late type galaxies at @xmath1 equivalent width of about 4 . we note that the estimated lf shows significant fluctuations due to large scale structures when measured in different regions of the universe . on the other hand , the morphological fraction as a function of luminosity is relatively less sensitive to large scale structures and thus seems to be more universal . in paper ii , it is shown that the morphology fraction is a monotonic function of local density at a fixed luminosity . since luminosity in turn depends on local density , this result indicates that the probability for a galaxy to be born as a particular morphological type is basically determined when its luminosity or mass is given . to this extent , morphology is determined by galaxy mass . the question here is what determines the morphology of a galaxy having a particular mass while keeping the type fraction corresponding to its mass scale . further investigations are required to answer this question . the authors thank chan - gyung park for making the velocity dispersion function fits and ravi sheth for helpful comments . cbp acknowledges the support of the korea science and engineering foundation ( kosef ) through the astrophysical research center for the structure and evolution of the cosmos ( arcsec ) . msv acknowledges support from nasa grant nag-12243 and nsf grant ast-0507463 . msv thanks the department of astrophysical sciences at princeton university for its hospitality during sabbatical leave . yyc , cbp , and msv thank the aspen center for physics , at which much of this paper was written . 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 web site is http://www.sdss.org/. the sdss is managed by the astrophysical research consortium for the participating institutions . the participating institutions are the american museum of natural history , astrophysical institute potsdam , university of basel , cambridge university , case western reserve university , university of chicago , drexel university , fermilab , the institute for advanced study , the japan participation group , johns hopkins university , the joint institute for nuclear astrophysics , the kavli institute for particle astrophysics and cosmology , the korean scientist group , the chinese academy of sciences ( lamost ) , los alamos national laboratory , the max - planck - institute for astronomy , the max - planck - institute for astrophysics , new mexico state university , ohio state university , university of pittsburgh , university of portsmouth , princeton university , the united states naval observatory , and the university of washington . , abazajian , k. , et al . 2004 , , 128 , 502 , adelman - mccarthy , j. k. , et al . 2006 , , 162 , 38 , alam , s. m. k. , & ryden , b. s. 2002 , , 570 , 610 , baldry , i. k. et al . , 2004 , , 600 681 , balogh , m. l. , et al . 2004a , , 615 , l101 , balogh , m. l. , et al . 2004b , , 348 , 1355 , bernardi , m. , sheth , r. k. , nichol , r. c. , schneider , d. p. , & brinkmann , j. 2005 , 129 , 61 , bernardi , m. , et al . 2003a , , 125 , 1866 , bernardi , m. , et al . 2003b , , 125 , 1817 , blanton , m. r. , et al . 2005 , , 129 , 2562 , blanton , m. r. , et al . 2003a , , 594 , 186 , blanton , m. r. , lin , h. , lupton , r. h. , maley , f. m. , young , n. , zehavi , i. , & loveday , j. 2003b , , 125 , 2276 , blanton , m. r. , et al . 2003c , , 125 , 2348 , couch w. j. , barger a. j. , smail i. , ellis r. s. , sharples r. m. , 1998 , , 497 , 188 , colless , m. , et al . 2001 , , 328 , 1039 , croton , d. j. , et al . 2005 , , 356 , 1155 , desroches , l .- b . , quataert , e. , ma , c .- p . , & west , a. a. 2006 , astro - ph/0608474 , dressler , a. , 1980 , , 236 , 351 , eisenstein , d. j. , et al . 2001 , , 122 , 2267 , falco , e. e. , et al . 1999 , , 111 , 438 , fukugita , m. , ichikawa , t. , gunn , j. e. , doi , m. , shimasaku , k. , & schneider , d. p. 1996 , , 111 , 1748 , fukugita , m. , nakamura , o. turner , e. l. , helmboldt , j. , nichol , r. c. 2004 , 601 , 127l , gmez , p. , et al . 2003 , , 584 , 210 , gott , j. r. , juric , m. , schlegel , d. , hoyle , f. , vogeley , m. s. , tegmark , m. , bahcall , n. , & brinkmann , j. 2005 , , 624 , 463 , goto 2005 357 , 937 , goto , t. , et al . 2003a , , 346 , 601 , goto , t. , et al . 2003b , , 55 , 757 , goto , t. , et al . 2003c , , 55 , 771 , gunn , j. e. , et al . 2006 , , 131 , 2332 , gunn , j. e. , et al . 1998 , , 116 , 3040 , hogg , d. w. , finkbeiner , d. p. , schlegel , d. j. , & gunn , j. e. 2001 , , 122 , 2129 , hogg , d. w. , et al . , 2004 , , 601 , l29 , ivezi , z. , et al . 2004 , an , 325 , 583 , jrgensen , i. , franx , m. , & kjrgaard , 1995 , , 276 , 1341 , kuehn , f. & ryden , b. s. 2005 , 634 , 1032 , lewis , i. , et al . 2002 , , 334 , 673 , lupton , r. h. , gunn , j. e. , ivezi , z. , knapp , g. r. , kent , s. , & yasuda , n. 2001 , in asp conf . 238 , astronomical data analysis software and systems x , ed . f. r. harnden , jr . , f. a. primini , & h. e. payne ( san francisco : asp ) , 269 , madgwick d. s. , et al . , 2002 , , 333 , 133 . mitchell , j. l. , keeton , c.r . , frieman , j.a . , & sheth , r.k , 2005 , , 622 , 81 , nakamura , o. , et al . 2003 , , 125 , 1682 , park , c. , et al . , 2006 , , submitted ( paper ii ) , park , c. , & choi , y .- y . , 2005 , , 635 , l29 , pier , j. r. , munn , j. a. , hindsley , r. b. , hennessy , g. s. , kent , s. m. , lupton , r. h. , & ivezic , r. 2003 , , 125 , 1559 , poggianti , b. m. , smail , i. , dressler , a. , couch , w. j. , barger , a. j. , burtcher , h. , ellis , r. s. , & oemler , a. , jr . 1999 , , 518 , 576 , quintero , a. d. , et al . , 2004 , , 602 , 190 , richards , g. t. , et al . , 2002 , , 123 , 2945 , sandage , a. , tammann , g. a. , & yahil , a. , 1979 , , 232 , 352 , saunders , w. et al . 2000 , , 317 , 55 , schlegel , d. j. , finkbeiner , d. p. , & davis , m. 1998 , , 500 , 525 , shen , s. , et al . 2003 , , 343 , 978 , sheth , r. k. , et al . 2003 , , 594 , 225 , smith , j. a. , et al . 2002 , , 123 , 2121 , stoughton , c. , et al . 2002 , , 123 , 485 , strauss , m. a. , et al . 2002 , , 124 , 1810 , tanaka , m. , et al . 2004 , , 128 , 2677 , tegmark , m. , et al . 2004 , , 606,702 , tucker , d. , et al . 2006 , an , in press , weinmann , s. m. , van der bosch , f. c. , yang , x. , & mo , h. j. 2006 , , 366 , 2 , yamauchi , c. , & goto , t. 2005 , , 359 , 1557 , yamauchi , c. , & goto , t. 2004 , , 352 , 815 , york , d. , et al . 2000 , , 120 , 1579

let be an elliptic surface defined over a number field , where is a smooth projective curve , and let be a section defined over with canonical height . in this article , we show that the function on is the height induced from an adelically metrized line bundle with non - negative curvature on . applying theorems of thuillier and yuan , we obtain the equidistribution of points where is torsion , and we give an explicit description of the limiting distribution on . finally , combined with results of masser and zannier , we show there is a positive lower bound on the height , after excluding finitely many points , for any non - special " section of a family of abelian varieties that split as a product of elliptic curves .

1701.07947

suppose @xmath14 is an elliptic surface defined over a number field @xmath1 , so @xmath2 is a smooth projective curve and all but finitely many fibers @xmath15 , @xmath16 , are smooth elliptic curves . we let @xmath17 denote the nron - tate canonical height of @xmath18 viewed as an elliptic curve over the function field @xmath19 ; we let @xmath20 denote the canonical height on the fibers for ( all but finitely many ) @xmath21 . suppose that @xmath22 is a section defined over @xmath1 for which @xmath23 , so , in particular , the points @xmath8 on the fiber are not torsion in @xmath15 for all @xmath24 . tate showed that the function @xmath25 is a weil height on @xmath6 , up to a bounded error @xcite . more precisely , there exists a divisor @xmath26 of degree equal to @xmath27 so that @xmath28 where @xmath29 is a weil height on @xmath6 associated to @xmath30 . in a series of three articles @xcite , silverman refined statement ( [ elliptic variation ] ) by analyzing the nron decomposition of the canonical height on the fibers @xmath31 where @xmath32 denotes the set of places of the number field @xmath1 , and @xmath33 are the integers appearing in the product formula @xmath34 for all @xmath35 . in this article , we explain how silverman s conclusions about the local functions @xmath36 are precisely the input needed to show that @xmath37 is a good " height function on the base curve @xmath2 , from the point of view of equidistribution . combining his work with methods from complex dynamics , as in @xcite , and the inequalities of zhang on successive minima @xcite , we prove : [ good height ] let @xmath1 be a number field and @xmath38 for a smooth projective curve @xmath2 defined over @xmath1 . fix any elliptic surface @xmath14 defined over @xmath1 and point @xmath39 satisfying @xmath40 . then @xmath41 for @xmath24 with smooth fibers , is the restriction of a height function on @xmath6 induced from an adelically metrized ample line bundle @xmath42 , with continuous metrics of non - negative curvature , satisfying @xmath43 theorem [ good height ] implies that our height function on @xmath2 satisfies the hypotheses of the equidistribution theorems of thuillier and yuan for points of small height on curves @xcite , and we deduce the following : [ equidistribution on b ] let @xmath1 be a number field and @xmath38 for a smooth projective curve @xmath2 defined over @xmath1 . fix any elliptic surface @xmath14 defined over @xmath1 and point @xmath39 satisfying @xmath40 . there is a collection of probability measures @xmath44 on the berkovich analytifications @xmath45 such that for any infinite , non - repeating sequence of @xmath46 such that @xmath47 as @xmath48 , the discrete measures @xmath49 converge weakly on @xmath45 to the measure @xmath50 at each place @xmath51 of @xmath1 . [ measures ] the measures @xmath50 of corollary [ equidistribution on b ] are not difficult to describe , at least at the archimedean places . at each archimedean place @xmath51 , there is a canonical positive @xmath52-current @xmath53 on the surface @xmath54 ( with continuous potentials away from the singular fibers ) which restricts to the haar measure on each smooth fiber @xmath55 . the measure @xmath50 on @xmath9 is just the pull - back of this current by the section @xmath12 . moreover , at every place , the measure @xmath50 is the laplacian of the local height function @xmath56 , away from its singularities . we give more details about ( and a dynamical perpective on ) the construction of the current @xmath53 in section [ dynamics ] . as a consequence of theorem [ good height ] , and combined with the work of masser and zannier @xcite , we obtain the so - called bogomolov extension of their theorems . suppose @xmath57 and @xmath58 are elliptic surfaces over a curve @xmath2 , defined over @xmath59 , and assume that at least one is non - isotrivial . we consider sections @xmath12 of the fiber product @xmath60 defined over @xmath59 . we say that a section @xmath61 is _ special _ if either * @xmath62 is torsion on @xmath63 for some @xmath64 , or * @xmath65 and @xmath66 are isogenous and there exists a pair of nonzero integers @xmath67 so that @xmath68 on a common cover @xmath18 of @xmath63 . if a family of abelian surfaces @xmath13 is isogenous to a fiber product ( after performing a base change @xmath69 if needed ) , we say that a section of @xmath70 is special if it is special on the fiber product . it is well known , assuming both @xmath65 and @xmath66 are non - isotrivial , that a special section will always pass through infinitely many torsion points in the fibers @xmath71 . that is , there are infinitely many @xmath72 for which @xmath73 if one of the curves , say @xmath66 , is isotrivial then a special section will pass through infinitely many torsion points unless the component @xmath74 is torsion and @xmath75 is constant and non - torsion on @xmath66 . is isotrivial if there exist a finite @xmath69 , an elliptic curve @xmath76 , and a birational @xmath77 . now let @xmath12 be a section of @xmath78 and denote by @xmath79 its extension to @xmath80 . then we say that @xmath12 is constant if @xmath81 is constant on @xmath80 . ] for proofs of these statements , see ( * ? ? ? * chapter 3 ) or , for dynamical proofs , see @xcite . the converse statement is also true , but it is much more difficult : masser and zannier proved that if @xmath82 for infinitely many @xmath83 , then the section @xmath12 must be special @xcite . we extend these results of masser - zannier from points of height 0 to points of small height : [ zhang language ] let @xmath2 be a quasiprojective smooth algebraic curve defined over @xmath59 . suppose @xmath84 is a non - isotrivial family of abelian varieties of dimension 2 defined over @xmath59 with a non - simple generic fiber . let @xmath42 be a line bundle on @xmath70 which restricts to an ample and symmetric line bundle on each fiber @xmath85 , and let @xmath86 be the induced nron - tate canonical height on @xmath85 , for each @xmath83 . for each non - special section @xmath87 defined over @xmath59 , there is a constant @xmath88 so that @xmath89 is finite . to put theorem [ zhang language ] in context , recall that ullmo and zhang proved the bogomolov conjecture @xcite , extending the manin - mumford conjecture ( a theorem of raynaud @xcite ) to points of small height on abelian varieties defined over @xmath59 . that is , for any abelian variety @xmath90 and choice of ample , symmetric line bundle @xmath42 on @xmath70 , and for any subvariety @xmath91 which is not itself an abelian subvariety or torsion translate of one , there is a positive constant @xmath92 so that @xmath93 is not zariski dense in @xmath94 . a key ingredient in their proofs is the equidistribution theorem of szpiro , ullmo , and zhang @xcite . in his 1998 icm lecture notes @xcite , zhang presented a conjecture about geometrically simple families of abelian varieties , which stated , in its most basic form : [ zhang ] [ zhang conj ] let @xmath2 be a quasiprojective smooth algebraic curve defined over @xmath59 . suppose @xmath84 is a non - isotrivial family of abelian varieties with fiber dimension @xmath95 , defined over @xmath59 with a simple generic fiber . let @xmath42 be a line bundle on @xmath70 which restricts to an ample and symmetric line bundle on each fiber @xmath85 , and let @xmath86 be the induced nron - tate canonical height on @xmath85 , for each @xmath83 . for each non - torsion section @xmath87 defined over @xmath59 , there is a constant @xmath88 so that @xmath89 is finite . when the dimension of the fibers @xmath85 is equal to 2 , the finiteness of @xmath96 for sections as in conjecture [ zhang conj ] was established recently by masser and zannier in @xcite . it is well known that the conclusion of conjecture [ zhang conj ] can fail to hold if @xmath70 is not simple and certainly fails if it is a family of elliptic curves , as mentioned above . however , the results of masser and zannier in their earlier work @xcite suggested a formulation of zhang s conjecture for the non - simple case in relative dimension 2 ; this is what we proved in our theorem [ zhang language ] . theorem [ good height ] , corollary [ equidistribution on b ] , and theorem [ zhang language ] were obtained in the special case of the legendre family @xmath97 over @xmath98 and the abelian variety @xmath99 , for sections @xmath12 with @xmath100-coordinates in @xmath101 in @xcite , using methods from complex dynamical systems , without appealing to silverman and tate s results on the height function . moreover , restricting further to sections @xmath12 with constant @xmath100-coordinate ( in @xmath102 ) , theorem [ zhang language ] was obtained without relying on the theorems of masser and zannier and gave an alternate proof of their result . this includes the special case treated by masser and zannier in their article @xcite . for sections with constant @xmath100-coordinate , the hypothesis on @xmath12 ( that @xmath40 ) is equivalent to asking that @xmath103 ( * ? ? ? * proposition 1.4 ) . the result of theorem [ zhang language ] extends to non - isotrivial families of abelian varieties @xmath13 in higher dimensions that split as a product @xmath104 of @xmath105 elliptic curves . indeed , if @xmath106 is a section , and if there exists an infinite sequence @xmath107 so that @xmath108 , then in fact @xmath109 for all @xmath64 , and we can reduce to theorem [ zhang language ] by considering pairs @xmath110 , @xmath111 . we conclude that every pair @xmath112 is special as a section of @xmath110 . * comments and acknowledgements . * this project was motivated , in part , by experiments to visualize silverman s results on the variation of canonical height @xcite in terms of the measures @xmath50 at archimedean places , and to examine their dependence on @xmath12 . in particular , the measure detects the failure of the local height function @xmath56 to be harmonic ; compare the comments on non - analyticity preceding theorem i.0.3 of @xcite . the images appearing in section [ experiment ] were first presented at the conference in honor of silverman s birthday , august 2015 . we thank charles favre , dragos ghioca , robert rumely , joseph silverman , and amaury thuillier for helpful suggestions and inspiring conversations . our research was supported by the national science foundation and the simons foundation . let @xmath113 be a product formula field of characteristic 0 , so there exists a family @xmath114 of non - trivial absolute values on @xmath113 and a collection of positive integers @xmath33 for @xmath115 so that @xmath116 for all @xmath117 . let @xmath118 be an elliptic curve with origin @xmath119 , expressed in weierstrass form as @xmath120 with discriminant @xmath121 . denote by @xmath122 the nron - tate canonical height function ; it can be defined by @xmath123p))}{n^2}\ ] ] where @xmath124 is the naive weil height on @xmath125 and @xmath126 is the degree 2 projection to the @xmath100-coordinate . for each @xmath127 , we let @xmath128 denote a minimal , algebraically closed field containing @xmath113 which is complete with respect to @xmath129 . for each @xmath51 , we fix an embedding of @xmath130 into @xmath128 . the canonical height has a decomposition into local heights , as @xmath131 for @xmath132 , with the local heights @xmath133 characterized by three properties ( * ? ? ? * chapter 6 , theorem 1.1 , page 455 ) : 1 . @xmath133 is continuous on @xmath134 and bounded on the complement of any @xmath51-adic neighborhood of @xmath119 ; 2 . the limit of @xmath135 exists as @xmath136 in @xmath137 ; and 3 . for all @xmath138 with @xmath139 p \not= o$ ] , @xmath140p ) = 4 \hat{\lambda}(p ) - \log|2y+a_1x+a_3|_v + \frac{1}{4 } \log|\delta|_v.\ ] ] now let @xmath1 be a number field and @xmath14 an elliptic surface defined over a number field @xmath1 with zero section @xmath141 . let @xmath142 be a non - zero section defined over @xmath1 , and assume that @xmath143 when viewing @xmath12 as a point on the elliptic curve @xmath18 defined over @xmath144 . for each @xmath16 such that the fiber @xmath15 is non - singular , we have point @xmath145 . we will investigate the function @xmath25 which is well defined at all but finitely many @xmath146 . furthermore , via the embedding of @xmath147 into @xmath128 for each place @xmath148 , we may view @xmath78 as defined over @xmath128 and consider the nron local heights @xmath36 on the non - singular fibers @xmath15 as functions of @xmath149 . let @xmath150 be the @xmath151-divisor on @xmath2 defined by @xmath152 here , @xmath153 is the local canonical height of the point @xmath12 on the elliptic curve @xmath18 over @xmath144 at the place @xmath154 , for each @xmath155 . the degree of @xmath150 is equal to @xmath27 . it follows from the definitions that @xmath156 is a subset of the finite set @xmath157 by enlarging @xmath1 , we may assume that the support of @xmath150 is contained in @xmath158 . that @xmath150 is a @xmath151-divisor is standard , following from the fact that the numbers @xmath153 can be viewed as arithmetic intersection numbers on a nron local model . see ( * ? ? ? * chapter iii , theorem 9.3 ) for a proof that @xmath159 ; see ( * ? ? ? * section 6 and p. 203 ) and ( * ? ? ? * chapter 11 theorem 5.1 ) for proofs that each local function @xmath133 also takes values in @xmath151 ; see ( * ? ? ? * theorem 1.1 ) for a dynamical proof . let @xmath160 be an analytic weil height on @xmath6 as defined in ( * ? ? ? * example 1(a ) ) . that is , we let @xmath161 be the genus of @xmath2 , and for each point @xmath162 , we choose an element @xmath163 of @xmath164 which has a pole of order @xmath165 at @xmath166 and no other poles . for each non - archimedean place @xmath51 of @xmath1 , set @xmath167 for all @xmath168 . for archimedean places @xmath51 , the local height is defined by @xmath169 we set @xmath170 for all @xmath7 . for fixed choices of @xmath163 , we will call the associated height @xmath160 our reference height " for the divisor @xmath150 . silverman proved : [ silverman triviality ] ( * ? ? ? * theorem [email protected] ) for any choice of reference height @xmath160 , there is a finite set @xmath171 of places so that @xmath172 for all @xmath173 and all @xmath174 . fix a point @xmath175 and a uniformizer @xmath176 for @xmath177 , and consider the function @xmath178 which is not _ a priori _ defined at @xmath177 . theorem [ silverman triviality ] implies that @xmath179 for all but finitely many places @xmath51 in a @xmath51-adic neighborhood of each @xmath177 . silverman also proved the following : [ silverman continuity ] ( * ? ? ? * theorem [email protected] ) fix @xmath180 and a uniformizer @xmath181 at @xmath177 . for all @xmath182 , there exists a neighborhood @xmath183 containing @xmath177 so that the function @xmath184 of extends to a continuous function on @xmath185 . recall that the nron - tate height @xmath17 and its local counterparts @xmath133 can be defined dynamically . letting @xmath18 be an elliptic curve defined over a number field @xmath1 , the multiplication - by-2 endomorphism @xmath186 on @xmath18 descends to a rational function of degree 4 on @xmath125 , via the standard quotient identifying a point @xmath12 with its additive inverse : @xmath187_\pi \ar[r]^\phi & e\ar[d]^\pi \\ { \mathbb{p}}^1 \ar[r]^{f_\phi } & { \mathbb{p}}^1 } \ ] ] an elementary , but key , observation is that a point is torsion on @xmath18 if and only if its quotient in @xmath125 is preperiodic for @xmath188 . the height @xmath17 on @xmath189 satisfies @xmath190 where @xmath124 is the standard logarithmic weil height on @xmath191 . now let @xmath78 be an elliptic surface defined over a number field @xmath1 , and let @xmath192 be a section , also defined over @xmath1 . in this section , we use this perspective to give a proof of subharmonicity of the local height functions @xmath193 and the extensions @xmath194 of . we will present this fact as an immediate consequence of now - standard complex - dynamical convergence arguments , at least when the fiber @xmath15 is smooth and the local height @xmath56 is finite . near singular fibers , we utilize the maximum principle and standard results on removable singularities for subharmonic functions . the same reasoning applies in both archimedean and non - archimedean settings . in [ subsection : measures ] we provide the background to justify the explicit description of the limiting distribution @xmath50 at the archimedean places @xmath51 of @xmath1 , as mentioned in remark [ measures ] . as in [ preliminaries ] , we let @xmath18 be an elliptic curve in weierstrass form , defined over a product - formula field @xmath113 of characteristic 0 . we define a rational function @xmath195 on @xmath125 by the formula @xmath196p)\ ] ] for all @xmath197 . here @xmath198 is the @xmath199coordinate for a point @xmath200 ; this function @xmath100 plays the role of @xmath201 in . in coordinates , we have @xmath202 and @xmath203 for @xmath204 . by a _ lift _ of @xmath205 , we mean any homogeneous polynomial map f on @xmath206 , defined over @xmath207 , so that @xmath208 , where @xmath209 is the tautological projection . a _ lift _ of a point @xmath210 is a choice of @xmath211 so that @xmath212 . the _ standard lift _ of @xmath205 will be the map @xmath213 defined by @xmath214 for each @xmath215 , the @xmath51-adic _ escape rate _ is defined by @xmath216 where @xmath217 any other lift of @xmath205 is of the form @xmath218 for some @xmath219 ; observe that @xmath220 note that @xmath221 for any choice of lift @xmath222 . furthermore , @xmath223 is continuous on @xmath224 , as proved in the archimedean case by @xcite . for non - archimedean absolute values @xmath51 , @xmath223 extends continuously to the product of berkovich affine lines @xmath225 ( * ? ? ? * chapter 10 ) . [ lambda and g ] for the standard lift @xmath222 of @xmath205 , and for each place @xmath51 of @xmath207 , the local canonical height function satisfies @xmath226 where @xmath227 . the proof is immediate from the properties of @xmath223 by checking the three characterizing conditions for @xmath133 . now let @xmath1 be a number field and @xmath14 an elliptic surface defined over a number field @xmath1 with zero section @xmath141 . let @xmath144 ; viewing @xmath18 as an elliptic curve defined over @xmath228 , we also fix a point @xmath39 . recall the function @xmath229 defined in . [ subharmonic ] for every @xmath180 and uniformizer @xmath181 in @xmath228 at @xmath177 , the function @xmath230 extends to a continuous and subharmonic function on a neighborhood of @xmath177 in the berkovich analytification @xmath231 . the continuity was already established in theorem [ silverman continuity ] , though it was not explicitly stated for the berkovich space . the argument below takes care of that . we begin with a lemma . [ harmonic ] fix @xmath232 and @xmath180 . let @xmath233 be a uniformizer at @xmath177 . for each place @xmath51 of @xmath1 , the function @xmath234 is harmonic in a neighborhood of @xmath177 in the berkovich analytification @xmath231 . this is silverman s ( * ? ? ? * lemma [email protected](c ) ) plus a removable singularities lemma for harmonic functions . see also ( * ? ? ? * proposition 7.19 ) for the extension of a harmonic function to a disk in the berkovich space @xmath231 . fix @xmath39 . let @xmath222 and @xmath94 be lifts of @xmath205 and @xmath198 to @xmath235 , respectively . iterating @xmath222 , we set @xmath236 and observe that @xmath237 from the definition of the escape rate . we let @xmath238 and @xmath239 denote the specializations of @xmath222 and @xmath94 at a point @xmath7 ; they are well defined for all but finitely many @xmath24 . observe that if @xmath222 is the standard lift for @xmath18 then so is @xmath238 for all @xmath24 . [ escape rate subharmonic ] fix @xmath39 , @xmath180 , and @xmath148 . for any choice of lifts @xmath222 of @xmath205 and @xmath94 of @xmath198 , the function @xmath240 extends to a continuous and subharmonic function in a neighborhood of @xmath177 in @xmath231 . first observe that the conclusion does not depend on the choices of @xmath222 and @xmath94 . indeed , @xmath241 for any @xmath242 . so by lemma [ harmonic ] the function @xmath243 is continuous and subharmonic for one choice if and only if it is continuous and subharmonic for all choices . let @xmath222 be the standard lift of @xmath205 . suppose that @xmath244 . since @xmath245 , we compute that @xmath246 now suppose that @xmath247 . fix @xmath248 and local uniformizer @xmath181 at @xmath177 . choose a lift @xmath222 of @xmath205 so that the coefficients of @xmath222 have no poles at @xmath177 , with @xmath249 . choose lift @xmath94 of @xmath198 so that @xmath239 is well defined for all @xmath24 near @xmath177 and @xmath250 . as above , we write @xmath251 and put @xmath252 so that @xmath253 for all @xmath254 and @xmath255 . set @xmath256 for each place @xmath51 , we set @xmath257 by construction , the limit of @xmath258 ( for @xmath24 near @xmath177 with @xmath259 ) is exactly the function @xmath260 for these choices . in fact , for @xmath24 in a small neighborhood of @xmath177 , but with @xmath261 , the function @xmath262 on @xmath125 is a well - defined rational function of degree 4 ; so the specialization of the homogeneous polynomial map @xmath238 satisfies @xmath263 . furthermore , as the coefficients of @xmath238 depend analytically on @xmath24 , the functions @xmath258 converge locally uniformly to the function @xmath260 away from @xmath264 . this can be seen with a standard telescoping sum argument , used often in complex dynamics , as in ( * ? ? ? * proposition 1.2 ) . in particular , @xmath260 is continuous on a punctured neighborhood of @xmath177 . at the archimedean places @xmath51 , and for each @xmath254 , the function @xmath258 is clearly continuous and subharmonic in a neighborhood of @xmath177 . at non - archimedean places @xmath51 , this definition extends to a berkovich disk around @xmath177 , setting @xmath265_t , \log[b_n(t)/t^{a_n}]_t\}\ ] ] where @xmath266_t$ ] is the seminorm on @xmath267 $ ] associated to the point @xmath24 . each of these functions @xmath258 is continuous and subharmonic for @xmath24 in a berkovich disk around @xmath177 . ( compare @xcite example 8.7 , proposition 8.26(d ) , and equation ( 10.9 ) . ) [ upper bound ] for all @xmath51 , and by shrinking the radius @xmath268 if necessary , the functions @xmath258 are uniformly bounded from above on the ( berkovich ) disk @xmath269 . as observed above , the functions @xmath258 converge locally uniformly away from @xmath264 to the continuous function @xmath270 . choose a small radius @xmath268 , and let @xmath271 which is finite by the convergence . because the functions are subharmonic , the maximum principle implies that @xmath272 throughout the disk of radius @xmath268 , for all @xmath254 . for the non - archimedean places , there is also a maximum principle on the berkovich disk , where the role of the circle of radius @xmath268 is played by the type ii point associated to the disk of radius @xmath268 ( see @xcite proposition 8.14 ) . we can now complete the proof of proposition [ escape rate subharmonic ] . as each @xmath258 is subharmonic , and the functions are uniformly bounded from above on the disk by lemma [ upper bound ] , we know that the ( upper - semicontinuous regularlization of the ) limsup of these functions is subharmonic . see @xcite proposition 8.26(f ) for a proof in the non - archimedean case . subharmonicity now follows from proposition [ lambda and g ] , lemma [ harmonic ] , and proposition [ escape rate subharmonic ] . the continuity at each archimedean place is the content of theorem [ silverman continuity ] . the continuity at each non - archimedean place is a combination of the continuity on the punctured berkovich disk ( as in the proof of proposition [ escape rate subharmonic ] ) and the continuity on type i ( classical ) points given in theorem [ silverman continuity ] . here we provide more details about the description of the measures appearing in the statement of corollary [ equidistribution on b ] , as discussed in remark [ measures ] . fix an archimedean place @xmath51 and any point @xmath180 . choosing a uniformizer @xmath181 at @xmath177 , recall the definition of @xmath194 from . we define @xmath273 on a neighborhood of @xmath177 in @xmath231 ; note that this is indepedent of the choice of @xmath181 . note that @xmath50 can be expressed as @xmath274 for @xmath24 outside of the finitely many points in the support of the divisor @xmath150 or where the fiber @xmath15 is singular . note , further , that @xmath50 assigns no mass to any individual point @xmath177 , because the potentials are bounded by theorem [ subharmonic ] . the details on the metric and the equidistribution theorem in section [ sec : metric ] will show that these are exactly the measures that arise as the distribution of the points of small height in corollary [ equidistribution on b ] . it is well known that the local height function on a smooth elliptic curve is a potential for the haar measure . that is , for fixed @xmath24 we have @xmath275 where @xmath276 is the normalized haar measure on @xmath15 and @xmath277 is a delta - mass supported at the origin of @xmath15 ; see , e.g. , ( * ? ? ? * theorem ii.5.1 ) . we present an alternative proof of this fact related to dynamics as part of proposition [ measure description ] , as a consequence of proposition [ lambda and g ] . [ measure description ] let @xmath14 be an elliptic surface and @xmath3 a section , both defined over a number field @xmath1 . let @xmath278 be the union of the finitely many singular fibers in @xmath18 . for each archimedean place @xmath51 of @xmath1 , there is a positive , closed ( 1,1 ) current @xmath53 on @xmath279 with locally continuous potentials so that @xmath280 is the haar measure on each smooth fiber , and @xmath281 is equal to the measure @xmath50 . as @xmath53 has continuous potentials , the restriction @xmath280 and the pullback @xmath281 are well defined . that is , we have @xmath282 where @xmath181 is a locally defined potential of @xmath53 , and @xmath283 locally on @xmath2 . the measure @xmath50 has no atoms , so it is determined by @xmath53 along the image of @xmath12 in @xmath279 . let us fix any small neighborhood @xmath185 in the base curve @xmath9 so that all fibers @xmath15 are smooth for @xmath284 . let @xmath262 be the map on @xmath125 defined in [ ss : dynamics ] ; by shrinking @xmath185 if necessary , we can find lifts @xmath238 of @xmath262 that are holomorphic in @xmath285 . from @xcite ( or the proof of ( * ? ? ? * proposition 1.2 ) ) , we know that the escape rate @xmath286 is continuous and plurisubharmonic as a function of @xmath287 . therefore @xmath288 projects to a closed and positive ( 1,1)-current @xmath289 on the complex surface @xmath290 , with locally continuous potentials . this current @xmath289 has the property that , restricted to each fiber @xmath125 , its total mass is 1 ; and the measure on the fiber is the measure of maximal entropy for the rational map @xmath262 @xcite . the restriction @xmath291 of the elliptic surface @xmath18 over @xmath185 maps with degree 2 to the complex surface @xmath292 by the projection @xmath201 of . the current @xmath289 can be pulled back to @xmath18 as @xmath293 where @xmath161 is a locally - defined continuous potential for @xmath289 . covering the base of @xmath279 by sets of the form @xmath185 , the local definitions glue to form the closed , positive @xmath52-current @xmath53 on @xmath279 . if @xmath3 is a section defined over the number field @xmath1 , then @xmath281 has potential given locally by @xmath294 for any lift @xmath239 of @xmath295 . proposition [ lambda and g ] yields that @xmath281 must coincide with @xmath50 . finally , to conclude that @xmath280 is equal to the normalized haar measure @xmath276 , we may use the well - known dynamical fact that for each fixed @xmath24 in the base , the measure @xmath276 projects by @xmath201 to @xmath125 to the unique measure of maximal entropy for the map @xmath262 ; see , e.g. , @xcite . in this section we give the proofs of theorem [ good height ] and corollary [ equidistribution on b ] . we first outline the proofs . let @xmath14 be an elliptic surface defined over a number field @xmath1 with zero section @xmath141 , and let @xmath142 be a section also defined over @xmath1 so that @xmath40 . recall from [ set up ] that we introduced a @xmath151-divisor @xmath296 on @xmath2 . by enlarging @xmath1 , we may assume that @xmath156 lies in @xmath158 . we will define an adelic metric on the ample line bundle @xmath42 associated to the divisor @xmath150 , inducing a height function @xmath297 such that @xmath298 and @xmath299 applying silverman s results on the variation of canonical height , theorems [ silverman triviality ] and [ silverman continuity ] , we will deduce that the metric is continuous and adelic . from theorem [ subharmonic ] , we will conclude that the metric is also semi - positive in the sense of zhang @xcite . we will use zhang s inequalities @xcite to deduce that @xmath300 consequently , we will be able to apply the equidistribution results of chambert - loir , thuillier , and yuan @xcite to complete our proofs . let @xmath301 be such that @xmath302 is an integral divisor . let @xmath303 be the associated line bundle on @xmath2 . note that @xmath304 so @xmath305 is ample ; by replacing @xmath306 with a multiple , we may assume that @xmath303 is very ample . fix a place @xmath51 of @xmath1 . let @xmath185 be an open subset of @xmath231 . each section @xmath307 is identified with a meromorphic function @xmath205 on @xmath185 satisfying @xmath308 we set @xmath309 taking the locally - defined uniformizer @xmath310 at @xmath24 in the definition of @xmath311 from . [ metric properties ] the metric @xmath312 on @xmath303 is continuous , semipositive , and adelic . the continuity and semipositivity follows from theorem [ subharmonic ] . ( in @xcite , semipositivity of a continuously metrized line bundle on a curve is defined terms of subharmonicity of potentials for the curvature form at each archimedean place , and as a uniform limit of smooth semipositive " metrics at each non - archimedean place . in @xcite , it is established that subharmonicity of potentials is a sufficient notion at all places , and he proves in ( * ? ? ? * theorem 4.3.3 ) that this notion of semipositivity coincides with that of zhang @xcite . see also ( * ? ? ? * lemma 3.11 , theorem 3.12 ) where this same argument is applied in a dynamical context . ) the adelic condition follows from theorem [ silverman triviality ] . a height function on @xmath6 is defined by setting @xmath313 where @xmath186 is any global section of @xmath303 which is nonvanishing along the galois orbit of @xmath24 , and @xmath314 is the metric of [ metric definition ] . recall that @xmath315 ; we may assume that our sections @xmath186 are defined over @xmath1 , and the product formula guarantees our height is independent of the choice of @xmath186 . our next goal is to prove the following two important facts about this height function @xmath316 . [ heightvsheight ] the height function @xmath316 satisfies @xmath317 for all @xmath21 such that the fiber @xmath15 is smooth . [ height is nonnegative ] the height function @xmath316 satisfies @xmath318 for all @xmath7 . first fix @xmath319 with smooth fiber @xmath15 . choose a section @xmath186 defined over @xmath1 that does not vanish along the galois orbit of @xmath24 , and let @xmath205 be the associated meromorphic function on @xmath2 . then @xmath205 takes finite and nonzero values along the galois orbit of @xmath24 . we have , @xmath320 where the second equality follows from the product formula . for @xmath321 such that @xmath322 is smooth , it is necessarily the case that @xmath323 , and therefore @xmath324 . to compute @xmath325 , observe that @xmath326 so its galois orbit is trivial ; fixing a uniformizer @xmath327 at @xmath177 , we have @xmath328 where @xmath329 is the function of associated to the uniformizer @xmath181 . we can compute @xmath325 using the dynamical interpretation of the local heights , described in section [ ss : dynamics ] . fix a weierstrass equation for @xmath18 in a neighborhood of @xmath177 and write @xmath330 . the assumption that @xmath331 is equivalent to @xmath332 . after possibly shrinking @xmath185 , write @xmath333 for the chosen uniformizer @xmath181 and a function @xmath334 that does not vanish in @xmath185 . we choose a lift @xmath94 of @xmath335 on @xmath185 defined as @xmath336 , where @xmath337 . notice that @xmath338 and @xmath339 are regular at @xmath177 . let @xmath222 be the standard lift in these coordinates , defined in ; it satisfies @xmath340 , and we have @xmath341 . since @xmath342 , proposition [ lambda and g ] implies that @xmath343 for all @xmath285 . therefore , @xmath344 the product formula now yields that @xmath345 , as claimed . to prove proposition [ height is nonnegative ] , we first reduce to the case that the elliptic surface @xmath78 has semi - stable reduction ; that is all of its fibers are either smooth or have multiplicative reduction . the next lemma describes how the height associated with the divisor @xmath150 behaves under base extensions of the elliptic surface @xmath78 . it is adapted from ( * ? ? ? * reduction lemma [email protected] ) . we include it here for completeness . [ reductionlemma ] let @xmath346 be a finite map of smooth projective curves , let @xmath347 be a minimal model for @xmath348 , and let @xmath349 be the extension of the section @xmath12 . for each @xmath180 and @xmath350 , there is a neighborhood @xmath185 of @xmath351 in @xmath352 and a regular non - vanishing function @xmath205 on @xmath185 such that @xmath353 on @xmath354 . in particular , @xmath355 let @xmath181 be a uniformizer at @xmath177 , @xmath356 a uniformizer at @xmath351 and @xmath357 . since local heights are invariant under base extension we have @xmath358 notice that for all @xmath359 in a punctured neighborhood of @xmath360 the fibers @xmath361 are smooth . hence the map @xmath362 gives an isomorphism between the fibers @xmath363 . under this isomorphism @xmath364 is mapped to @xmath365 . invoking now the uniqueness of the nron local heights , we have @xmath366 combining and we get that for @xmath359 in a punctured neighborhood of @xmath351 , @xmath367 the definition of @xmath254 yields that the function @xmath368 is regular and non - vanishing at @xmath351 . the first part of the lemma follows . finally , theorem [ silverman continuity ] allows us to conclude that @xmath369 at the point @xmath360 , as claimed . the following lemma will allow us to prove proposition [ height is nonnegative ] in the case that a fiber has multiplicative reduction . the proof is lengthy , but it is merely a collection of computations using the explicit formulas for the local height functions , as in ( * ? ? ? * theorem [email protected] , [email protected] ) . [ bad fibers ] let @xmath78 be an elliptic surface and let @xmath142 be a non - zero section defined over @xmath1 . then there exists a finite extension @xmath370 of the number field @xmath1 so that , for each @xmath175 such that @xmath322 has multiplicative reduction , there exists an @xmath371 so that @xmath372 at all places @xmath51 of @xmath370 . we let @xmath373 be a minimal weierstrass equation for @xmath18 over an affine subset @xmath374 defined over @xmath1 with @xmath375 . here @xmath376 are regular functions at @xmath177 . using this weierstrass equation we write @xmath377 where @xmath378 . since @xmath78 has multiplicative reduction over @xmath379 , we have @xmath380 let @xmath51 be a place of @xmath1 ( archimedean or non - archimedean ) . we denote by @xmath381 the @xmath382invariant of @xmath383 , given by @xmath384 notice that equation yields that @xmath381 has a pole at @xmath177 . hence , we can find a @xmath385adic open neighborhood @xmath185 of @xmath177 and an analytic map @xmath386 such that the following holds : if @xmath387 is the modular @xmath382invariant ( * ? ? ? * chapter slowromancap5@ ) , then we have @xmath388 the function @xmath389 is given as @xmath390.\end{aligned}\ ] ] in the following , we choose a uniformizer @xmath391 at @xmath177 , and we identify @xmath392 with its expression @xmath393 $ ] and write @xmath394 equation yields that @xmath395 and @xmath396 $ ] . following the proof of ( * ? ? ? * section 6 ) and after possibly shrinking @xmath185 we have isomorphisms @xmath397 for @xmath398 . under these isomorphisms , we have @xmath399 here @xmath400 are the modular invariants , given by their usual @xmath401series @xmath402 and @xmath403 is the weierstrass @xmath404function given by @xmath405 in view of ( * ? ? ? * lemma [email protected] ) , after possibly replacing @xmath12 by @xmath406 , we may assume that @xmath407 is an analytic map satisfying @xmath408 in the following we identify @xmath409 with its series in @xmath410 $ ] and write @xmath411 where @xmath412 and @xmath413 $ ] . we claim that @xmath414 $ ] . to see this , notice that from ( * ? ? ? * slowromancap3@ ) we have that for @xmath284 @xmath415 where @xmath416 in the equation above @xmath121 denotes the modular discriminant given by @xmath417 since the functions @xmath418 and @xmath121 are defined over @xmath1 , we have that @xmath419 is also defined over @xmath1 . since @xmath420 $ ] and @xmath421 $ ] we get that @xmath422 $ ] . therefore , there are non - zero constants @xmath423 , non - negative integers @xmath424 and functions @xmath425 $ ] such that for all @xmath284 @xmath426 next , we aim to express @xmath427 ( as in the statement of the lemma ) in terms of @xmath428 . using the isomorphisms in [ uni - iso ] , the uniqueness of the local canonical heights and the explicit formulas for the local canonical heights ( * ? ? ? * theorem [email protected] , [email protected] ) , we get @xmath429 where @xmath430 is the second bernoulli polynomial . since @xmath431 and using , we get @xmath432 in what follows , for @xmath433 $ ] we write @xmath434 in view of ( * ? ? ? * lemma [email protected] ) , we have @xmath435 using equations and , equation yields @xmath436 finally , notice that ( * ? ? ? * theorem [email protected] ) implies @xmath437 therefore @xmath438 where @xmath439 belongs to a finite extension of @xmath1 , denoted by @xmath370 . by ( * lemma [email protected] ) there is a finite map of smooth projective curves @xmath440 such that if @xmath347 is a minimal model for @xmath441 , then @xmath442 has semi - stable reduction over the singular fibers of @xmath78 . moreover , we may choose @xmath80 so that everything is defined over @xmath1 . thus , by lemma [ reductionlemma ] and using the product formula , we may assume that the singular fibers of our elliptic surface @xmath14 have multiplicative reduction . for all @xmath7 for which @xmath15 is smooth , we know from proposition [ heightvsheight ] that @xmath443 . the canonical height is always non - negative , so we may conclude that @xmath444 for all such @xmath24 . assume now that @xmath180 has a fiber with multiplicative reduction . enlarging the number field @xmath1 if necessary we may assume that @xmath379 and that its corresponding @xmath427 defined in the statement of lemma [ bad fibers ] is in @xmath445 . then , on using the product formula , lemma [ bad fibers ] implies that @xmath446 . this completes the proof . let @xmath447 be the line bundle on @xmath2 induced from the divisor @xmath150 . from theorem [ metric properties ] , we know that its @xmath306-th tensor power can be equipped with a continuous , adelic , semipositive metric , so that the corresponding height function is ( a multiple of ) the canonical height @xmath448 on the smooth fibers . thus , by pulling back the metric to @xmath447 , we obtain a continuous , semipositive , adelic metric on @xmath447 inducing the desired height function . it remains to show that this height @xmath316 satisfies @xmath449 . this is a consequence of propositions [ heightvsheight ] and [ height is nonnegative ] and zhang s inequalities on successive minima @xcite . recall that , since @xmath40 , we know that there are infinitely many @xmath21 for which @xmath450 ( for a complex - dynamical proof , see ( * ? ? ? * proposition 1.5 , proposition 2.3 ) . ) therefore , from proposition [ heightvsheight ] , we may deduce that the essential minimum of @xmath316 on @xmath2 is equal to 0 . on the other hand , from proposition [ height is nonnegative ] , we know that @xmath451 for all @xmath7 . therefore , from ( * ? ? ? * theorem 1.10 ) , we may conclude that @xmath449 . when combined with the equidistribution theorems of yuan and thullier @xcite , we immediately obtain the corollary from theorem [ good height ] . the measures @xmath50 are the curvature distributions associated to the metrics @xmath314 at each place @xmath51 . from the definition of the metric in [ metric definition ] , we see that they are given locally by @xmath452 in a @xmath51-adic neighborhood of any point @xmath175 , and for any choice of uniformizer @xmath181 at @xmath177 . we first show that , to prove the theorem , it suffices to prove the result for sections of the fiber product @xmath453 of elliptic surfaces @xmath78 and @xmath454 over the same base , and to assume that the line bundle @xmath42 is generated by the divisor @xmath455 . let @xmath2 be a quasiprojective smooth algebraic curve defined over @xmath59 . suppose @xmath84 is a non - isotrivial and non - simple family of abelian varieties of dimension 2 defined over @xmath59 . from the poincar reducibility theorem @xcite , we know that @xmath70 is isogenous to a fibered product of elliptic curves . that is , there is a branched cover @xmath69 and a pair of elliptic surfaces @xmath456 and @xmath347 that give rise to an isogeny @xmath457 over @xmath80 . now let @xmath42 be a line bundle on @xmath70 which restricts to an ample and symmetric line bundle on each fiber @xmath85 for @xmath458 . then the line bundle @xmath42 pulls back to a line bundle @xmath459 on @xmath460 , and it again restricts to an ample and symmetric line bundle on each fiber over @xmath461 . now suppose that we have a section @xmath87 . the section @xmath12 pulls back to a section @xmath462 , and this in turn pulls back to a ( possibly multi - valued ) section of @xmath460 . if multi - valued , we can perform a base change again , passing to a branched cover @xmath463 , so that the induced section @xmath464 is well defined . by definition , the assumption that @xmath12 is non - special on @xmath70 means that it is non - special as a section of @xmath465 . finally , we observe that the conclusion of theorem [ zhang language ] does not depend on the choice of line bundle . ( we thank joe silverman for his help with this argument . ) recall that , on any abelian variety defined over @xmath59 , the notion of a small sequence " of points is independent of the choice of ample and symmetric line bundle . that is , if we take two ample and symmetric divisors @xmath466 and @xmath467 , then we know that there exists an integer @xmath468 so that @xmath469 is ample ; similarly there exists @xmath470 so that @xmath471 is ample . it follows from properties of the weil height machine that the heights @xmath472 and @xmath473 will then satisfy @xmath474 for real constants @xmath475 . upon passing to the canonical height , we conclude that @xmath476 on the abelian variety . in particular , @xmath477 for some sequence in @xmath478 if and only if @xmath479 . now suppose we have a family of abelian varieties @xmath13 . two line bundles @xmath480 and @xmath481 associated to relatively ample and symmetric divisors induce a canonical heights @xmath482 and @xmath483 on each fiber @xmath85 . but recalling that amplitude persists on zariski open sets ( * ? ? ? * theorem 1.2.17 ) , there exist positive integers @xmath484 and @xmath485 so that the line bundles @xmath486 and @xmath487 are relatively ample on a zariski open subset of the base @xmath2 . passing to the canonical heights once again , we find that the relation holds uniformly over @xmath2 ( after possibly excluding finitely many points ) . therefore , for any section @xmath87 , there exists a positive constant @xmath488 of theorem [ zhang language ] for height @xmath489 if and only if it there exists such a constant @xmath490 for @xmath491 . now let @xmath57 and @xmath58 be a pair of elliptic surfaces over the same base curve @xmath2 , defined over @xmath59 . let @xmath12 be a section of the fiber product @xmath492 , so that @xmath12 can be expressed as @xmath61 with @xmath493 for @xmath494 . let @xmath42 be the line bundle on @xmath495 associated to the divisor @xmath496 . the canonical height @xmath497 on each fiber @xmath85 is easily seen to be the sum of canonical heights ( see , e.g. , @xcite for properties of the height functions ) , so that @xmath498 now assume that @xmath61 is not a special section . define @xmath499 for @xmath500 and for all @xmath72 where both @xmath501 and @xmath502 are smooth elliptic curves . it suffices to show that there is no infinite sequence @xmath503 for which @xmath504 as @xmath48 . we have assumed that @xmath12 is not special , so neither @xmath505 nor @xmath506 is identically 0 . if @xmath63 is isotrivial , and if @xmath62 is constant non - torsion on @xmath63 , then @xmath507 for all @xmath24 and we are done . thus , we may assume that the function - field heights @xmath508 are nonzero for @xmath500 . let @xmath509 and @xmath510 be the adelically metrized line bundles on the base curve @xmath2 associated to the height functions @xmath505 and @xmath506 , from theorem [ good height ] . these metrized line bundles @xmath511 are both equipped with continuous adelic metrics of non - negative curvature . suppose there exists a sequence @xmath512 such that holds . in theorem [ good height ] , we showed that @xmath513 . therefore , we may apply the observation of chambert - loir ( * ? ? ? * proposition 3.4.2 ) , which builds upon on zhang s inequalities @xcite , to conclude that there exist integers @xmath514 and @xmath515 so that @xmath516 and @xmath517 are isomorphic as line bundles on @xmath2 and their metrics are scalar multiples of one another . it follows that the height functions @xmath505 and @xmath506 are the same , up to scale , and in particular they have the same zero sets . in other words , @xmath518 is a torsion point on @xmath501 if and only if @xmath519 is a torsion point on @xmath502 ( for all but finitely many @xmath24 in @xmath2 ) . we have therefore reduced our problem to the main results of @xcite , and we conclude that @xmath61 must have been a special section . as this was ruled out by assumption , we deduce that there exists a constant @xmath520 so that the set @xmath521 is finite . this concludes the proof of theorem [ zhang language ] . in this final section , we provide a few illustrations of the distributions @xmath50 for an archimedean place @xmath51 , arising in corollary [ equidistribution on b ] . in proposition [ dense ] , we present a complex - dynamical proof that the archimedean measures @xmath50 will have support equal to all of @xmath2 . given @xmath78 and section @xmath12 , we plot the parameters @xmath24 where @xmath8 is a torsion point on the fiber @xmath15 of specified order . as proved in corollary [ equidistribution on b ] , the local height function at each place @xmath522 determines the distribution of the torsion parameters ; it is a potential for the measure @xmath50 ( away from the singularities ) . recall that if we have two sections @xmath12 and @xmath523 that are linearly related on @xmath18 , then the distributions of their torsion parameters in @xmath2 will be the same . figure [ example 1 ] , top , illustrates the example of silverman from ( * ? ? ? * theorem i.0.3 ) . here , we have @xmath524 with @xmath525 and @xmath526 in @xmath527-coordinates . plotted are the torsion parameters of orders @xmath528 for all @xmath529 ; that is , the points @xmath24 in the base @xmath2 where @xmath8 is torsion of order @xmath528 on the fiber @xmath15 . roughly , a smaller yellow dot corresponds to higher order of torsion . figure [ example 1 ] , bottom , is another section of the same family , where the @xmath100-coordinate of @xmath8 is constant and equal to @xmath530 . ( strictly speaking , this second @xmath12 is not a section of our given @xmath531 , because the @xmath532-coordinate will not lie in @xmath533 but in an extension ; however , the property of being torsion and the determination of its order is independent of which point in the fiber we choose . ) observe the distinctly different pattern of yellow dots in the first and second pictures , especially in the left half of the two pictures , illustrating the linear independence in @xmath534 of the two sections . figure [ example 2 ] illustrates the torsion parameters for two independent sections of the legendre family , @xmath535 over @xmath525 , studied in @xcite . the chosen sections are @xmath75 , with constant @xmath100-coordinate equal to @xmath536 , and @xmath537 , with constant @xmath100-coordinate equal to 5 . as in figure [ example 1 ] , we plot the torsion parameters of orders @xmath528 for all @xmath529 ; generally , a smaller yellow dot signifies higher order of torsion . it was proved in @xcite that the limiting distributions for sections with constant @xmath100-coordinate satisfy @xmath538 ( at an archimedean place ) if and only if @xmath539 . it was proved in @xcite and @xcite that there are no @xmath540 for which both @xmath541 and @xmath542 are torsion on @xmath15 . again , observe the difference in the geometry of the yellow dots for the two independent sections . figure [ example equidistribution ] illustrates our equidistribution result , corollary [ equidistribution on b ] , for the example of the legendre family with the section @xmath537 . plotted are the torsion parameters of orders @xmath528 with ( a ) @xmath543 , ( b ) @xmath529 , and ( c ) @xmath544 . observe how the yellow dots fill in the grid structure " in the base curve @xmath2 , exactly as do the torsion points for one elliptic curve . [ h ] and @xmath526 , shown in the region @xmath545 . the singular fibers occur at @xmath546 , and one sees the effects of numerical error in a small neighborhood of these parameters . at bottom , torsion parameters for section @xmath12 having @xmath100-coordinate @xmath547 for all @xmath24.,title="fig:",width=432 ] and @xmath526 , shown in the region @xmath545 . the singular fibers occur at @xmath546 , and one sees the effects of numerical error in a small neighborhood of these parameters . at bottom , torsion parameters for section @xmath12 having @xmath100-coordinate @xmath547 for all @xmath24.,title="fig:",width=432 ] [ h ] and @xmath97 . at left , the section @xmath75 with @xmath548 for all @xmath24 ; at right , the section @xmath537 with @xmath549 for all @xmath24 . both are shown in the region @xmath550 , title="fig:",width=288 ] and @xmath97 . at left , the section @xmath75 with @xmath548 for all @xmath24 ; at right , the section @xmath537 with @xmath549 for all @xmath24 . both are shown in the region @xmath550 , title="fig:",width=288 ] [ h ] and @xmath97 and @xmath537 as in figure [ example 2 ] . at top , torsion parameters of orders @xmath528 for @xmath551 ; bottom left , of orders @xmath528 for @xmath529 , and bottom right , of orders @xmath528 for @xmath544 . , title="fig:",width=288 ] + and @xmath97 and @xmath537 as in figure [ example 2 ] . at top , torsion parameters of orders @xmath528 for @xmath551 ; bottom left , of orders @xmath528 for @xmath529 , and bottom right , of orders @xmath528 for @xmath544 . , title="fig:",width=288 ] and @xmath97 and @xmath537 as in figure [ example 2 ] . at top , torsion parameters of orders @xmath528 for @xmath551 ; bottom left , of orders @xmath528 for @xmath529 , and bottom right , of orders @xmath528 for @xmath544 . , title="fig:",width=288 ] as mentioned above , the smaller yellow dots in the illustrations correspond , roughly , to higher orders of torsion . these images are produced with a standard escape - rate algorithm . we use the dynamical system @xmath262 on @xmath125 , induced from multiplication by 2 on the elliptic curve @xmath15 from section [ dynamics ] , line . the coordinates on @xmath125 are chosen so that @xmath552 is the image of the 0 of @xmath15 . we mark @xmath24 yellow if @xmath553 for some @xmath529 . in all of these examples , the yellow dots will fill in the picture as the order of torsion grows , and the support of the measures @xmath50 is equal to @xmath9 . in fact , this will always be the case , for any ( nontrivial ) section of a complex elliptic surface , as our final result , proposition [ dense ] , shows . let @xmath78 be an elliptic surface over a projective curve @xmath2 , defined over @xmath554 , and let @xmath192 a section for which @xmath40 ( over the function field @xmath555 ) . let @xmath556 be the measure on @xmath2 defined as in proposition [ measure description ] , as the pullback of the current @xmath557 that restricts to haar measure on each smooth fiber . in other words , @xmath556 is locally defined as the laplacian of the function @xmath270 introduced in proposition [ escape rate subharmonic ] , which is well defined when working over @xmath554 . [ dense ] let @xmath78 be an elliptic surface over a projective curve @xmath2 , defined over @xmath554 , and let @xmath192 be a section for which @xmath40 ( over the function field @xmath555 ) . then the set @xmath558 is dense in @xmath9 and @xmath559 we give a complex - dynamical proof , viewing proposition [ dense ] as a consequence of the main result of @xcite . ( we do not use the equidistribution result , corollary [ equidistribution on b ] . ) an analytic proof is also presented in ( * ? ? ? * notes to chapter 3 ) . let @xmath560 be a finitely - punctured riemann surface such that the fiber @xmath15 is smooth for all @xmath561 . let @xmath562 be the degree - two projection and @xmath563 be the rational map induced by multiplication - by-2 on @xmath15 , as defined in the introduction to section [ dynamics ] . it is well known that the holomorphic family @xmath564 is structurally stable ; see , e.g. , ( * ? ? ? * chapter 4 ) . thus , over any simply - connected subset @xmath185 of @xmath565 , there is a holomorphic motion of the periodic points of @xmath262 which extends uniquely to a holomorphic motion of all of @xmath125 , conjugating the dynamics . the key observation is that @xmath556 is precisely the bifurcation measure " of the pair @xmath566 on @xmath565 . see @xcite and @xcite for definitions . the support of @xmath556 is equal to the bifurcation locus of @xmath566 ; in particular , the parameters @xmath567 for which @xmath568 is preperiodic for @xmath262 are dense in @xmath569 . therefore , it suffices to show that @xmath570 . suppose to the contrary that there is an open disk @xmath571 for which @xmath572 . then the pair @xmath566 is stable on @xmath185 , and therefore @xmath568 can not be a repelling periodic point for any @xmath285 . from the uniqueness of the holomorphic motion , it follows that @xmath573 is part of the holomorphic motion on @xmath185 . by analytic continuity , then , we deduce that @xmath568 must follow the motion of a point over all of @xmath565 . this implies that the pair @xmath566 is stable throughout @xmath565 and the measure @xmath556 is 0 . but this is absurd by the assumption that @xmath40 ; see ( * ? ? ? * theorem 1.1 ) . antoine chambert - loir . . motivic integration and its interactions with model theory and non - archimedean geometry . volume ii _ , volume 384 of _ london math . lecture note ser . _ , pages 150 . cambridge univ . press , cambridge , 2011 .

this paper presents the results of a _ spitzer _ irac m photometric search for warm dust orbiting 17 nearby , metal - rich white dwarfs , 15 of which apparently have hydrogen dominated atmospheres ( type daz ) . g166 - 58 , g29 - 38 , and gd 362 manifest excess emission in their irac fluxes and the latter two are known to harbor dust grains warm enough to radiate detectable emission at near - infrared wavelengths as short as 2 m . their irac fluxes display differences compatible with a relatively larger amount of cooler dust at gd 362 . g166 - 58 is presently unique in that it appears to exhibit excess flux only at wavelengths longer than about 5 m . evidence is presented that this mid - infrared emission is most likely associated with the white dwarf , indicating that g166 - 58 bears circumstellar dust no warmer than k. the remaining 14 targets reveal no reliable mid - infrared excess , indicating the majority of daz stars do not have warm debris disks sufficiently opaque to be detected by irac .

0710.0907

the _ spitzer space telescope _ opened a new phase space to white dwarf researchers interested in the infrared properties of degenerate stars and their environments . the majority of white dwarfs are inaccessible from the ground beyond 2.4 @xmath1 m due to their intrinsic faintness combined with the ever - increasing sky brightness towards longer wavelengths @xcite . this limits any white dwarf science which aims to study matter radiating at @xmath3 k. prior to the launch of _ spitzer _ , only one previously published , directed mid - infrared study of white dwarfs exists ; an _ infrared space observatory _ search for dust emission around 11 nearby white dwarfs , 6 of which have metal - rich photospheres @xcite . owing to the superb sensitivity of _ spitzer _ @xcite , a cycle 1 irac program was undertaken to search for warm dust emission associated with cool , hydrogen atmosphere white dwarfs with photospheric metals , the daz stars . this paper presents a synopsis of the irac results , including the detection of @xmath4 @xmath1 m flux excess at g166 - 58 , @xmath0 @xmath1 m data on g29 - 38 and gd 362 , and also includes gemini @xmath5 @xmath1 m spectroscopy of g29 - 38 . the origin and abundances of photospheric metals in isolated white dwarfs has been an astrophysical curiosity dating back to the era when the first few white dwarfs were finally understood to be subluminous via the combination of spectra and parallax ; sirius b , 40 eri b , and van maanen 2 @xcite . in a half page journal entry , @xcite noted that his accidentally discovered faint star with large proper motion had a spectral type of `` about f0 '' @xcite . it was not until forty years ( ! ) later that it became clear that van maanen 2 was metal - poor with respect to the sun @xcite . over the course of the another decade and a half , it became gradually clear that white dwarfs in general both hydrogen atmosphere degenerates void of metallic features and helium atmosphere degenerates with photospheric metal lines had heavy element abundances a few to several orders of magnitude below solar @xcite . oxymoronically , white dwarfs with detectable , yet evidently sub - solar , heavy element abundances are now referred to as metal - rich . the two fundamental classes of metal - poor - yet - rich white dwarfs are essentially the same as their metal - free counterparts , which are designated by the main atmospheric constituent ; either hydrogen or helium . it is easiest to think of white dwarfs as type da ( hydrogen ) or non - da ( helium ) . for a wide range of temperatures , white dwarfs with hydrogen atmospheres will manifest balmer lines ( type da ) , excepting perhaps the very hottest cases where hydrogen may be ionized ( e.g. pg 1159 stars ) . white dwarfs with helium atmospheres display a variety of spectral behaviors which depend primarily on their effective temperatures ; helium ii lines ( type do ) for the hottest , helium i lines ( type db ) for a wide temperature range , and no lines ( type dc ) for the coolest . although there do exist cases of mixed atmospheres , either inferred or directly observed , typically one of the two light gases dominates the composition and photosphere ; a notable exception is the case of carbon opacity in a helium atmosphere ( type dq ) . there are both subtleties and complexities in the composition and spectral characteristics of white dwarfs which will not be discussed here , but for the full guided tour of the white dwarf spectral zoo , the reader is referred to both the definition of the current classification scheme @xcite and a broad spectral atlas @xcite . the important points to take away from a given designated spectral type is : 1 ) the first letter after d indicates the dominant gas in the atmosphere ; 2 ) when metals are detected in an ultraviolet or optical spectrum of a white dwarf , the letter z is added to the designation ( with dc becoming dz ) . any primordial heavy elements in white dwarfs can only be sustained in their photospheres for the brief ( @xmath6 yr ) period while the degenerate is still rather hot and contracting significantly , and then only to a certain degree . this is achieved through radiative levitation at @xmath7 k for hydrogen atmospheres and @xmath8 k for helium atmospheres @xcite . below these temperatures , the cooling degenerate stars develop significant convection zones , which enhance the gravitational settling of the elements , now unimpeded by radiative forces @xcite . diffusion timescales for the sinking of metals are always orders of magnitude shorter than the evolutionary ( cooling ) timescales of white dwarfs . therefore , external sources are responsible for the presence of the metals within cool white dwarf photospheres . although there were suspects and many spurious detections reported , all metal - rich white dwarfs were historically helium atmosphere degenerates until the confirmation of g74 - 7 , the first daz star @xcite . even so , a decade later all other known metal - bearing white dwarfs were still restricted to the helium - rich variety , despite reports to the contrary ; an examination of tables 1 and 2 in @xcite with hindsight reveals that only g74 - 7 remains classified as daz @xcite . this situation began to change a little less than a decade ago with a couple of individual discoveries , followed by many more , now totalling around fifty objects @xcite . the reason daz stars are latecomers to the metal scene is because helium atmospheres are quite transparent relative to hydrogen atmospheres ; all else being equal , a given calcium abundance will produce an equivalent width of order @xmath9 times stronger in a helium as opposed to hydrogen atmosphere , making it far easier to infer the presence of photospheric metals in helium - rich degenerates @xcite . convection zones can be many orders of magnitude larger in helium atmosphere white dwarfs , resulting in diffusion times for heavy elements up to @xmath10 yr @xcite . this relatively long - lived photospheric retention allows for the possibility that extant metals in such objects are the remnant of a long ago ( up to several diffusion timescales ) interstellar cloud encounter @xcite . however , it was shown that the galactic positions and space motions of dz and dbz stars are not correlated with local interstellar clouds @xcite . additionally with regards to cool helium atmosphere white dwarfs , there is the ever - present and still unexplained lack of detectable hydrogen ( or more specifically in the case of those bearing metal lines , the very high inferred metal to hydrogen abundance ratios ) , which should be expected in quantity , and readily visible in low opacity helium - dominanted photospheres , if accretion from the interstellar medium has occurred at any time during their cooling @xcite . the timecales for the diffusion of metals in hydrogen atmosphere white dwarfs diminish quite drastically relative to those in their helium atmosphere counterparts , owing to significantly smaller convection zones for @xmath11 k , and in warmer cases ( @xmath12 k ) where the convection zone is almost negligible , these diffusion times can be just a few days @xcite . unless such objects have just emerged from interstellar accretion episodes within these short periods ( unlikely ) , they must be currently accreting at rates sufficient to produce the observed abundances . @xcite find no correlation between enhanced interstellar medium densities and the current positions of accreting daz stars . in contrast , @xcite conclude that , with certain caveats , the warm , partially ionized medium can explain the observed accretion rates and abundances in daz stars . two decades ago it was proposed that accretion of circumstellar material might be the cause of the heavy metal abundances seen in some white dwarfs , so that the origin of contaminating elements seen is bimodal in nature ; either interstellar or circumstellar @xcite . the first such model invoked episodic cometary impacts from reservoirs which managed to survive the post - main sequence mass loss phases , particularly the asymptotic giant branch @xcite . this particular model fails to explain 1 ) the daz stars with the highest metal abundances , and 2 ) the observed distribution of abundances in general @xcite . a more promising model of circumstellar accretion invokes the tidal disruption of an asteroid , which goes on to form a ring of debris around the white dwarf , from which the photospheric heavy elements originate @xcite . given that a typical solar system asteroid is around @xmath13 times more massive than a typical comet , this model can explain both the relatively high metal abundance and the observed infrared excess seen at several metal - rich white dwarfs @xcite . while it may be the case that both mechanisms create contaminated white dwarf photospheres perhaps circumstellar in the case of high metal abundances and interstellar for those in the lower range there are a growing number of metal - rich white dwarfs which are either confirmed or suspected to harbor circumstellar dust @xcite . observations in the mid - infrared are most sensitive to both warm and cool orbiting dust at metal - bearing white dwarfs , providing a direct test of circumstellar accretion hypotheses . additionally , such a search can constrain the frequency of orbiting material more strongly than ground - based near - infrared observations , which are sensitive only to dust not far from its sublimation temperature . table [ tbl1 ] lists the 17 metal - rich white dwarfs observed with irac , taken from @xcite with the exception of gd 362 @xcite . all targets were chosen as daz stars , but recent evidence implies that two of the white dwarfs ( g77 - 50 and gd 362 ) are helium - rich ( @xcite ; d. koester 2007 , private communication ) . for the purposes of this paper and the statistics which follow , many targets are nominally referred to as daz , even though future observations may reveal helium - rich atmospheres in some . this fact is simply unavoidable as helium becomes spectroscopically undetectable in white dwarfs cooler than @xmath14 k and its presence can only be inferred by indirect methods . these stars were chosen among available _ spitzer _ cycle 1 observations for their high metal abundances . between 2004 november and 2005 august , observations were executed with the infrared array camera ( irac ; @xcite ) in all four bandpasses : 3.6 , 4.5 , 5.7 , and 7.9 @xmath1 m . a 20-point cycling dither pattern ( of medium step size ) was used for each target in each bandpass , with 30 s frame times at each position , yielding a total exposure time of 600 s at all wavelengths . the data were processed with the irac calibration pipeline ( versions @xmath15 ) to create a single , fully processed and reduced image ( @xmath16 pixels in all four channels ) upon which to perform measurements . aperture photometry was carried out with the standard iraf task apphot , and measured fluxes were corrected for aperture size , but not for color . generally , the flux and signal - to - noise were measured in a @xmath17 pixel aperture radius , dependent upon target brightness and neighboring sources , with a @xmath18 pixel sky annulus . the measured fluxes were converted to the standard irac aperture using corrections described in the most recent version of the irac data handbook @xcite . the results are listed in table [ tbl2 ] . listed in table [ tbl2 ] are the total errors in the measured and calibrated flux , together with irac pipeline versions with which the data were extracted from the archive . the photometric errors were estimated by taking the per pixel standard deviation in the extracted sky level and multiplying by the area of a 2 pixel radius photometric aperture , the smallest radius for which there exist derived aperture corrections in the irac data handbook @xcite . this approach is conservative and does not assume gaussian , random noise for the following reasons . it was found in general that the observations went sufficiently deep as to approach or reach the confusion limit , primarily , but not exclusively , at the two shorter wavelengths . additionally , there was sometimes low spatial frequency structure seen in the background of the two longer wavelength images , which appeared to be from either real diffuse sources such as cirrus , or from imperfections in the irac pipeline . these factors led to difficulty in the determination of the true sky level in the vicinity of the science target , the per pixel noise in the sky , and in the flux measurement itself . even in the smallest , 2 pixel radius , aperture used for photometry , there was occasional possible flux contamination from neighboring sources in this relatively large 18.1 square arcsecond area , compounded by the undersampled pixels . this possibility was evident from a few ( typical ) to several ( rare ) percent increase in photometric flux that was sometimes seen as the aperture radius was increased from 2 to 3 pixels , which should not occur for single point sources and accurate aperture corrections . all targets were unambiguously detected at all wavelengths . there are five other sources of photometric error in the irac camera which were considered @xcite . first , the absolute calibration uncertainty in the irac instrument is reported as 3% @xcite . second , color corrections have been ignored and are typically smaller than 1% . third , there is the pixel phase dependent correction , which is reported to be no more than 4% peak to peak or @xmath19% for a single image at 3.6 @xmath1 m @xcite . the images analyzed here were produced by a set of 20 pseudo - random dithers which essentially annihilates this source of error . fourth , there is the array location dependent correction for rayleigh - jeans type stellar sources . this effect is the largest source of photometric uncertainty in the irac instrument and may reach 10% peak to peak or @xmath20% , therefore a typical error of this type should be @xmath17% . however , experiments have shown that for well dithered data , as that analyzed here , the effect tends to average out ( as would be expected ) and is less than 1% @xcite . to assess the reliability of the absolute irac calibration , a couple of experiments were undertaken ( m. jura 2007 , private communication ) . first , the 3.6 to 4.5 @xmath1 m flux ratios of the target stars with @xmath21 k , and no evidence for the presence of warm dust , were examined and found to vary typically within 5% of the mean , however a couple of targets deviated by as much as 10% quite significant for such a small number of white dwarfs . second , a similar result was obtained by examining 4.5 to 7.9 @xmath1 m flux ratios of identically warm white dwarfs from @xcite . in this instance the relatively lower signal - to - noise at 7.9 @xmath1 m may have been a factor as evidenced by the result of @xcite who fit a subsample of the @xcite white dwarfs with current models to within 4% at 4.5 @xmath1 m , yet only to within 10% at 7.9 @xmath1 m . white dwarf models themselves can not account for the observed deviations in flux ratios at these warmer temperatures , which can only amount to around 1% for extreme values of surface gravity and temperature ( p. bergeron 2007 , private communication ) . furthermore , @xcite report observed irac fluxes for 33 main sequence stars which deviate from model predicted photospheric values , on average , by 8% at 3.6 @xmath1 m , 4% at 4.5 @xmath1 m , and 6% at 7.9 @xmath1 m . @xcite find similar deviations between measured irac fluxes and model predictions for 74 young main sequence stars , amounting to 5% on average at all wavelengths . based on these analyses and findings , it seems appropriate to assign a @xmath22 irac calibration uncertainty of 5% . for bright targets , the total error is dominated by the absolute calibration uncertainty , while for the faintest targets the total error is dominated by the uncertainty in the aperture photometry . for these reasons , the total errors represented in table [ tbl2 ] should be considered conservative . on three nights in 2006 january g29 - 38 was observed at gemini observatory at mauna kea with the near - infrared imager ( niri ; @xcite ) in spectroscopy mode using the @xmath23-grism , which covers @xmath24 @xmath1 m . spectra were taken at two positions along the @xmath25 slit with 1 s exposures and 60 coadds . overall , approximately 2 hours of usable science frames were gathered . calibration frames taken each night included spectral flats as well as observations of one of the two a0v telluric standards hip 110578 or hip 12640 , utilizing 0.2 s exposures and @xmath18 coadds nodded at two positions along the slit . both the science and telluric standard frames from each night were pairwise subtracted at the two nod positions in order to best remove the bright , variable sky at these wavelengths . the subtracted frames were flat fielded and then median combined , with bad pixels and cosmic rays fixed manually , creating two spectra of opposite polarity which were extracted using standard iraf tasks . the two extracted spectra from each night were wavelength calibrated using aborption lines from the night sky , and averaged . the science spectrum each night was divided by the telluric spectrum and multiplied by a blackbody of the appropriate temperature . the science spectra from all three nights were shifted and averaged to create the final spectrum . the flux was converted from @xmath26 to @xmath27 , normalized to one , then flux calibrated using irac 3.6 @xmath1 m photometry . the signal - to - noise was estimated by measuring the standard deviation along 20 sections of 50 pixels each spanning the entire wavelength range , yielding values between 4 and 10 ( average 7 ) over @xmath28 @xmath1 m , and between 8 and 15 ( average of 12 ) over @xmath29 @xmath1 m . these were calculated assuming continuum or pseudo - continuum over all regions . the irac fluxes for all targets , together with optical and near - infrared data , are plotted as spectral energy distributions in figures [ fig1][fig5 ] . in general , the short wavelength photometry was taken from the most accurate and reliable sources available , including but not limited to : @xcite and references therein ; 2mass @xcite ; denis @xcite ; and various other sources @xcite . because the aim of the survey was to identify significant mid - infrared photometric excess due to opaque dust , blackbody fits to the spectral energy distributions of the target white dwarfs will suffice to model their expected , essentially rayleigh - jeans , behavior at irac wavelengths . if the total uncertainty in the irac photometry decreases sufficiently , white dwarf models should prove more useful in this regard . of the 17 observed stars , three display excess radiation within their irac beams ( considered here to be the solid angle contained within a full width at half maximum airy disk , typically @xmath30 in diameter and @xmath31 square arcseconds in area at @xmath0 @xmath1 m ) , with a high degree of certainty . the case of g166 - 58 is discussed in some detail below and shows evidence for continuum emission from @xmath2 k dust . both g29 - 38 and gd 362 show warm ( @xmath32 k ) thermal continuum and strong silicate emission in their mid - infrared spectra and have therefore been confirmed to harbor orbiting rings of dust @xcite . for the remainder of the targets , the irac observations either rule out emission from warm , opaque debris or are of sufficiently low signal - to - noise as to preclude a definitive conclusion either way . unfortunately , there was a single target ( g21 - 16 ) for which accurate fluxes could not be extracted due to unavoidable source confusion within and around the irac beam . an unsuccessful attempt was made to photometrically isolate the white dwarf with the iraf task daophot ; based on the signifiant crowding , it is likely that the irac beam itself is contaminated at all wavelengths . from the present work , 3 of 17 metal - rich white dwarfs , or 18% display definite irac continuum excess consistent with orbiting dust . however , this fraction is only 2 in 15 daz stars , or 13% . if one counts all daz white dwarfs observed with irac in cycle 1 , the total which display irac flux excess is 3 of 25 targets or 12 % @xcite . the spectral energy distribution of gd 362 , together with its irac data , is displayed in figure [ fig5 ] . as expected on the basis of ground - based @xmath33 photometry @xcite , the @xmath0 @xmath1 m region is dominated by warm thermal emission from orbiting dust , whose flux can be reproduced out to @xmath34 @xmath1 m by a @xmath35 k blackbody . this approximation is physically nonviable as it corresponds to a ring ( or sphere ) with a single temperature and radius , yet does not deviate much from a ring model of substantial radial extent @xcite at these shorter mid - infrared wavelengths . the measured flux at 7.9 @xmath1 m has not been corrected for color and shows an indication that the broad and strong 10 @xmath1 m silicate emission feature @xcite contributes significantly into that relatively wide ( @xmath36 @xmath1 m ) bandpass . in figure [ fig6 ] , the @xmath37 @xmath1 m photometric data on gd 362 are plotted together with the more realistic model of @xcite . the model invokes a face - on geometrically thin opaque dust ring of finite radial extent , with an inner temperature of @xmath38 k and an outer temperature range @xmath39 k. for gd 362 , these temperatures correspond to a ring which extends from @xmath40 @xmath41 to @xmath42 @xmath41 @xcite . the data agree reasonably well with the higher temperature curve , excepting the 24 @xmath1 m flux ( disregarding the silicate emission - enhanced 7.9 @xmath1 m flux ) . there are two possibilities based on this model : 1 ) the outermost opaque grains have a temperature @xmath43 k or ; 2 ) the flux at 24 @xmath1 m is affected by another dust emission feature ( which would indicate forsterite , if present ) . photometric and spectroscopic observations of gd 362 utilizing all three instruments aboard _ spitzer _ , including a detailed model fit of the orbiting dust , its mid - infrared thermal continuum and emission features , dust mass estimates , temperature and particle size distribution , dimensions and types of the emitting regions are presented in @xcite . the spectral energy distribution of g29 - 38 , together with its irac data , is displayed in figure [ fig5 ] . while the overall similarity between g29 - 38 and gd 362 is apparent , there are some distinctions . although their inner dust temperatures are clearly similar , the thermal continuum flux of g29 - 38 over @xmath44 @xmath1 m appears to be falling , while for gd 362 it is rising ; likely an indication of varying amounts of opaque dust both slightly warmer and cooler than the @xmath35 k blackbody approximations for the excess at each of these stars . in order to fit the slope of the @xmath44 @xmath1 m photometry of g29 - 38 with a blackbody , a temperature near 1100 k is necessary , greatly overpredicting the near - infrared flux . yet as can be seen from the figure , the 3.6 @xmath1 m flux is somewhat underpredicted by 900 k. using the more plausible and physical model of @xcite , figure [ fig6 ] plots the thermal infrared excess of g29 - 38 from @xmath37 @xmath1 m from all available _ spitzer _ photometric data , together with the model curves . these disk models are exactly those applied to g29 - 38 in @xcite , now plotted with more accurate data with greater wavelength coverage . for the same inner and outer temperatures given above , the extent of the opaque ring is from @xmath40 @xmath41 to @xmath45 @xmath41 for g29 - 38 . in this case , unlike gd 362 , all the photometric data are fitted decently by the model where the outermost grains have a temperature near 600 k , with the possible exception of the 3.6 @xmath1 m flux . the apparently excessive flux from g29 - 38 at the shortest wavelength irac channel was first noticed when the data were initially retrieved from the _ spitzer _ archive using pipeline version 11.0 . it was present again one version later , and then triple checked in 2006 november with version 14.0 , which was used for all the irac data on g29 - 39 in this work . the deviation between the models and the measured flux at 3.6 @xmath1 m is @xmath46 mjy from the best fit ring model or the blackbody . taking photometric error into account , which is entirely due to instrument calibration uncertainty , the average deviation is 0.70 mjy ; approximately @xmath47 of the flux error , or 10% of the excess . @xmath1 m spectroscopy was undertaken to investigate possible sources of this extra emission , and to better assess if it is real . as can be seen from figure [ fig7 ] , the @xmath23-grism spectrum of g29 - 38 is essentially featureless , with an apparent , slight , upward slope towards 4 @xmath1 m ( a thermal continuum approximated by a 900 k blackbody would peak longward of 5 @xmath1 m in @xmath27 ) . previous to this investigation , the only spectral information in this wavelength regime came from data presented in @xcite , where a relatively low signal - to - noise spectrum is not inconsistent with emission over @xmath48 @xmath1 m . the spectral flux error in the figure [ fig7 ] data , translated to mjy via the irac 3.6 @xmath1 m flux , is typically 1.2 mjy over @xmath28 @xmath1 m ( outside the atmospheric transmission window , an area very sensitive to water vapor and prone to large variability ) , and 0.7 mjy over @xmath29 @xmath1 m ( atop the @xmath49-band ) . there exist a plethora of emission features in this region associated with polycyclic aromatic hydrocarbons and their multitudinous close relatives : specifically , features between 3.2 and 3.6 @xmath1 m seen in the interstellar medium @xcite ; circumstellar matter @xcite ; ultraviolet - excited ( planetary , proto - planetary , and reflection ) nebulae @xcite ; as well as in comets @xcite . the well known 3.3 @xmath1 m feature , when present , is always significantly weaker than its other family members at 6.2 , 7.7 , 8.6 , and 11.3 @xmath1 m @xcite . because these stronger features are absent from the irs @xmath50 @xmath1 m spectrum of g29 - 38 @xcite , it is expected no feature should be present at 3.3 @xmath1 m , although the irs data were published after the the @xmath23-grism observations were planned . there are additional features around 3.4 @xmath1 m ( primarily due to methanol , ethane , and other hydrocarbon species ) which typically dominate this region when observed towards comets @xcite , but which are much weaker than the 3.3 @xmath1 m feature in circumstellar environments @xcite . because of these observational facts , in addition to the relatively fragile nature of hydrocarbons ( some small species are more volatile than water ice ) in the vicinity of high density ultraviolet radiation fields @xcite , it is unlikely that a cometary feature would be seen in the vicinity of the @xmath51 @xmath41 circumstellar dust ring at g29 - 38 . the failure to detect any possible sources of excess emission in the @xmath5 @xmath1 m region at g29 - 38 leaves a few possibilities for the 2.8 @xmath52 disagreement between its measured irac flux and the applied models . the first is that the photometry in the 3.6 @xmath1 m bandpass is improperly calibrated , making it inaccurate . the second is that both the ring model of @xcite and the single temperature blackbody fail to predict the correct flux at this wavelength . a @xmath53 k blackbody ( figure [ fig6 ] ) is not inconsistent with the @xmath54 @xmath1 m photometry , but underpredicts the three longer wavelength _ data points , requiring that dust emission affects those bandpasses . the third possibility is the discrepancy arises from photometric variability . it is well - known that g29 - 38 is a pulsating white dwarf ; @xmath55-band light curves reveal periods of 615 , 268 , 243 , and 186 s with amplitudes of 0.12 , 0.03 , 0.03 , and 0.02 mag respectively , which are mirrored at @xmath56-band with corresponding strengths of 0.02 , 0.02 , 0.03 , and 0.03 mag respectively @xcite . although matching periods were searched for at @xmath23 , none were found , but with sufficiently large upper limits which do not exclude variations similar to those seen at @xmath56 @xcite . because the irac observations lasted 600 s , these photometric variability timescales can not explain the 3.6 @xmath1 m flux being possibly high . given that the irac channel 1 ( @xmath57 @xmath1 m ; @xcite ) data agree reasonably well with ground - based @xmath23-band ( @xmath58 @xmath1 m ; @xcite ) data with no color corrections to either , the discrepancy is likely to be model disagreement . the 4.5 and 7.9 @xmath1 m fluxes for g29 - 38 in table [ tbl2 ] agree well with the quoted ( uncorrected for color ) fluxes reported by @xcite . it is possible that the deviation arises from variability which has not yet been seen or reported , but no such claim is being made based on the present data . a fourth possibility might be a companion , but all substellar models of appropriate age range predict about twice as much excess at 4.5 @xmath1 m relative to 3.6 @xmath1 m , as well as a commensurate rise from 3 to 4 @xmath1 m which is not observed in the @xmath23-grism spectrum @xcite . figure [ fig4 ] displays the spectral energy distribution of g166 - 58 together with its irac flux measurements . the plotted data are optical @xmath59 from @xcite , the average of 2 @xmath60-band values cited in @xcite , and near - infrared @xmath61 which are the average of values given in @xcite and @xcite . the effective temperature of the white dwarf is taken from both @xcite and @xcite , where the determinations are both within 0.6% of 7400 k. g166 - 58 displays clear excess emission within its irac beam beginning at 5 @xmath1 m , making it unique among white dwarfs confirmed or suspected to harbor orbiting dust . in all other cases , the excess becomes unambiguous by 3 @xmath1 m , as in g29 - 38 and gd 362 @xcite . the implied temperature of the @xmath4 @xmath1 m excess can be reproduced by a blackbody of @xmath62 k , and is substantially cooler than the @xmath32 k temperatures inferred in all other dusty white dwarfs with @xmath17 @xmath1 m excess . yet circumstellar dust at 400 k is still considered warm relative to the overwhelming majority of main sequence stars with infrared excess attributed to a debris disk ; with a few notable exceptions , these are all kuiper belt analogs with @xmath63 k , and @xmath64 au @xcite . before proceeding further with any analysis and interpretation , owing both to the uniqueness of the irac data on g166 - 58 and the irac field in its vicinity , the nature and validity of the excess must be examined . there is another source @xmath65 from g166 - 58 seen in all four irac images , which are displayed in figure [ fig8 ] . this source is also present in the sdss photometric catalog ( release 5 ) where it is designated sdss j145806.96 + 293726.3 and classified as a galaxy @xcite . this object has optical and mid - infrared colors that are consistent with an extragalactic source . the irac images of g166 - 58 and the nearby galaxy overlap near @xmath66 from their image centers . both sources appear point - like at all four wavelengths . the percentage of flux from the nearby galaxy in a 2 pixel radius centered at g166 - 58 , measured relative to the total flux of the white dwarf over the same area , is relatively benign at @xmath17% , with the exception of 7.9 @xmath1 m where it is 12% . this was determined by folding the irac point spread function of g166 - 58 in two along the axis of symmetry ( up - down in the figure [ fig8 ] images ) , and subtracting one side from the other . exploiting this symmetry , it was straightforward to remove the contamination in the aperture photometry of g166 - 58 and vice versa for similar measurements of the nearby galaxy ( its flux is 0.18 , 0.17 , 0.13 , and 0.43 mjy at 3.6 , 4.5 , 5.7 , and 7.9 @xmath1 m respectively ) . the values in table [ tbl2 ] for g166 - 58 correspond to the signal after removal of this unwanted flux . two additional methods used to obtain the flux of g166 - 58 are discussed below , and a comparison of the fluxes obtained for the white dwarf by each method is shown in table [ tbl3 ] . first , the white dwarf and nearby galaxy were photometrically fitted and spatially deconvolved using daophot . second , photometry was obtained by radial profile analysis of the white dwarf . figure [ fig9 ] displays overlaying and identical linear contour plots for 7.9 @xmath1 m images of g166 - 58 both before and after the removal of the nearby galaxy . the parameters from a 2.0 pixel gaussian radial profile fit ( image centroid and full width at half maximum ) of the white dwarf remain essentially unchanged by inclusion of the galaxy . radial profile analysis of g166 - 58 at each irac channel yields a measurement for the white dwarf , via comparison with the radial profiles and fluxes of g29 - 38 and gd 362 . examination of g166 - 58 in a 7.9 @xmath1 m image where only the galaxy is fitted and subtracted reveals a clear point - like source and excess at the location of the white dwarf . in summary , table [ tbl3 ] illustrates excellent agreement among the three photometric methods used to determine the flux of g166 - 58 , with the exception of daophot at 3.6 @xmath1 m ( which may be due to undersampled data , which is most germaine at this shortest wavelength . ) further support that the two irac sources are separated photometrically is evidenced by their 5.7 to 7.9 @xmath1 m colors ; the nearby galaxy has a flux ratio of 0.30 , while the excess detected at g166 - 58 has a flux ratio of 0.46 . hence , with regards to the adjacent galaxy , there should be no doubt of its lack of influence in the white dwarf data presented . the known density of background galaxies , along with the presence of the relatively bright nearby source on the irac chips , prompted an evaluation of the probability that yet another , hidden source could be contaminating flux measurements within the beam of g166 - 58 itself , and therefore be responsible for the @xmath4 @xmath1 m excess . the white dwarf is located at galactic latitude @xmath67 , hence any contaminating source would almost certainly be extragalactic in nature . to assess the probability of such a line of sight coincidence , the following estimations were made . the 7.9 @xmath1 m excess emission at g166 - 58 is about 0.07 mjy , or 14.9 mag . _ spitzer _ irac 7.9 @xmath1 m source counts from @xcite indicate approximately 3,000 galaxies per magnitude per square degree at @xmath68 magnitude . taking the distribution to be flat over a 0.5 magnitude interval centered at 15.0 mag , this yields around 1500 galaxies per square degree of appropriate brightness to reproduce the 7.9 @xmath1 m excess seen at g166 - 58 . because the white dwarf and its excess display a point - like nature at this wavelength ( see figure [ fig9 ] ) , any unresolved background source would have to lie within a small fraction of the irac beam width of g166 - 58 in the plane of the sky , certainly within an area of 2 square arcseconds , conservatively speaking . therefore , the probability of finding a galaxy of 0.07 mjy brightness at 7.9 @xmath1 m within 2 square arcseconds of g166 - 58 is about 1 in 4300 . the odds that 1 in 17 target stars is contaminated by such a background galaxy should then be around 1 in 250 . therefore , the most likely explanation for the excess @xmath4 @xmath1 m emission at g166 - 58 is circumstellar dust associated with the white dwarf . if the @xmath4 @xmath1 m irac fluxes are the sum of the white dwarf and another source , one can immediately rule out a cold companion . a substellar object with an energy distribution similar to that implied by the 400 k blackbody fit , would have a radius of roughly @xmath69 m , more than twice the size of jupiter . furthermore , the combined spectral energy distribution appears nothing like what might be expected from such an orbiting cold degenerate ; it most notably lacks significant flux expected at 4.5 @xmath1 m @xcite . therefore , the excess emission must be due to warm circumstellar material . as a first stab at modeling the excess , one might invoke tidal dust rings similar to those which have been successful for both g29 - 38 and gd 362 @xcite . following the formalism of @xcite , a flat opaque ring ( or disk ) geometry , implies a dust grain temperature - radius relation given by @xmath70 where @xmath71 , @xmath72 , @xmath73 , and @xmath74 are the temperature of the emitting grains , their distance from the star , the stellar effective temperature , and the stellar radius , respectively . if one assumes that g166 - 58 is a single , carbon - oxygen core white dwarf with log @xmath75 and @xmath76 k @xcite , then its radius is @xmath77 @xmath41 @xcite . these stellar parameters yield a distance of @xmath78 @xmath41 to 400 k grains . if one assumes the ring extends to where the grain temperature is 200 k , then the ring would extend to @xmath79 @xmath41 . these inner and outer disk radii would be significantly larger than those implied for g29 - 38 , where the ring probably does not extend much further than @xmath80 @xmath41 ( around 30 stellar radii ) for outer grain temperatures of @xmath81 k , which seem to fit the data in figure [ fig6 ] quite well . this is also true of the implied size of the tidal ring about gd 362 , where models @xcite also yield an outer radius @xmath82 @xmath41 ( around 40 stellar radii ) . regarding the outer edge of the disk and an appropriate scale for the tidal breakup of a rocky body such as an asteroid or comet , @xcite provides a thorough review and revision of effective roche limits . the distance , @xmath83 , at which a small orbiting body will be disrupted by the gravitational field of a large body of radius @xmath74 , can be expressed , in simplified form , as @xmath84 where @xmath85 and @xmath86 are the densities of the large and small bodies , and @xmath87 is a constant which typically has a value in the range @xmath88 , but can be smaller . the coefficient @xmath87 depends on the model , which may include factors such as ; composition , heterogeneity , size , shape , rotation , orbital characterisitcs , shear and tensile strengths . the classical value of @xmath89 is for the case of a rotating , uniform , self - gravitating liquid in a circular orbit , whereas for a nonrotating , spherical satellite of solid rock or ice , the value becomes @xmath90 . this last approximation is likely to be valid for typical asteroids and comets with radii @xmath91 km , at least to within 50% @xcite . taking @xmath92 g @xmath93 , which lies within the range of densities for both asteroids and comets ( @xmath94 g @xmath93 , @xmath95 g @xmath93 ; @xcite ) , and calculating the average density of g166 - 58 with the parameters above , an estimate of the roche limit using @xmath90 is @xmath96 @xmath41 . hence , if an optically thick , flat ring orbits g166 - 58 at distances corresponding to 200 k dust grains , these would lie within a region consistent with tidal disruption of a minor planet . however , at grain temperatures of 100 k , the implied distance from g166 - 58 would be 2.4 @xmath41 , a region where asteroids or comets should remain intact , even if liquified @xcite . integrating the flux of the 400 k blackbody fit to the irac excess yields @xmath97 @xmath98 , while the stellar luminosity given by models for a hydrogen atmosphere white dwarf with @xmath99 k and log @xmath100 , yields @xmath101 @xmath98 @xcite . together , these determine @xmath102 , which is logarithmically about midway between @xmath103 found for the main sequence a3 star @xmath104 leporis , and @xmath105 for both g29 - 38 and gd 362 @xcite . therefore , instead of invoking an optically thick , flat ring , it may be more appropriate to suppose an optically thin shell of blackbody grains in radiative equilibrium , whose distance is given by @xcite @xmath106 in this case , 400 k grains would be located near 2.3 @xmath41 , about 6 times further out than predicted by the opaque disk model and well beyond the roche limit for large rocks . an advantage of this assumption is that it allows an estimation of the minimum dust mass contained in the disk . because white dwarfs have masses @xmath107 @xmath108 , yet greatly reduced luminosities @xmath109 @xmath98 , radiation pressure on dust grains can not compete with gravitational attraction . this can be seen by examining the ratio of these two forces , represented by the parameter @xmath110 where @xmath111 is the dust particle radius and @xmath112 is the radiation pressure coupling coefficient @xcite . assuming the case of geometric optics , where the effective grain cross section equals its geometric cross section and @xmath113 , yields the maximum possible value for @xmath114 near @xmath115 @xmath1 m . for @xmath92 g @xmath93 , @xmath116 @xmath98 , @xmath117 @xmath108 , @xmath118 and hence gravitational forces dominate over radiation pressure . this simply serves to show that sub - micron size dust and gas particles could certainly be present at white dwarfs without any danger of being lost to radiation pressure . the minimum dust mass of an optically thin disk is approximately @xmath119 although both smaller and larger particles are almost certainly present , the size of the thermally emitting dust is on the order of @xmath120 @xmath1 m . taking the density of silicate grains to be 2.5 g @xmath93 gives a rough lower limit to the mass of the the dust disk of @xmath121 g , a mass of a very large comet . while radiation pressure can not remove dust at white dwarfs , drag forces can . the timescale for poynting - robertson removal of particles is given by @xcite @xmath122 for white dwarfs in general , the ratio @xmath123 for optically thin dust will be @xmath9 times smaller than for main sequence stars . silicates of 1 @xmath1 m size and @xmath124 g @xmath93 orbiting g166 - 58 at 2.3 @xmath41 will be removed by the drag force in @xmath125 yr . the ratio @xmath126 increases to a maximum near @xmath115 @xmath1 m , implying removal timescales about 10 times more rapid . for smaller particles , the radiation coupling efficiency swiftly declines and this ratio decreases , levelling off near @xmath127 @xmath1 m at a value similar to that for 1 @xmath1 m grains @xcite . hence , in an optically thin disk , all particles up to 1 @xmath1 m in size should be removed by poynting - robertson drag within 500 yr . assuming a present balance between accretion and diffusion with the above stellar parameters for g166 - 58 , the rate at which it currently gains refractory circumstellar material is @xmath128 g @xmath129 ( koester & wilkens 2006 ; where a factor of 0.01 has been included to reflect the absence of accreted hydrogen and helium ) . if this accretion occurs over a single diffusion timescale of @xmath130 yr , the total accreted mass would @xmath131 g , equivalent to the mass of a small solar system asteroid . based on available data , it appears possible or perhaps likely that g166 - 58 is not a single white dwarf . although this target was not listed in table 2 of @xcite for white dwarfs known or suspected to be in binary systems , they determine a low surface gravity of log @xmath132 via a combination of optical through near - infrared photometry and parallax . at 7400 k , contemporary white dwarf models predict that such a low surface gravity implies a radius @xmath133 @xmath41 , which is about 25% too large for a normal carbon - oxygen core degenerate @xcite . an inferred oversized radius and corresponding overluminosity can be explained either by a single , low mass , helium core degenerate or via binarity involving two components of similar ( @xmath134 mag ) brightness a double degenerate . actually , the fact that g166 - 58 appears overluminous based on its spectral energy distribution and parallax has existed since the analysis of @xcite . there , using an essentially identical procedure , a low surface gravity of log @xmath135 was determined , carrying the same implication . the near - infrared @xmath61 measurements of both @xcite and @xcite agree to within their respective errors . because the implied overluminosity of g166 - 58 relies fairly heavily on its trigonometric parallax(es ) , a careful literature search was performed in order to assess all available astrometric data . the value employed by @xcite comes from the most recent version of the yale parallax catalog , and is given as @xmath136 @xcite . the source of the yale catalog parallax is the us naval observatory ( h. c. harris 2006 , private communication ) , first published in @xcite and then updated ( and possibly revised ) in @xcite . this parallax is given as @xmath137 and @xmath138 in their table 1 . @xcite quote a value of @xmath139 while citing @xcite , but this is merely the _ relative _ ( i.e. measured ) parallax quoted above . paying careful attention to detail , one can see the yale and us naval observatory values are not identical , but reflect slightly different astrometric corrections to obtain the absolute parallaxes , with the yale galactic model being more recent and likely more reliable . therefore , the only existing parallax measurement for g166 - 58 , upon which its overluminosity rests , is @xmath140 or @xmath141 pc . @xcite use this parallax for g166 - 58 , together with its optical and near - infrared spectral energy distribution to deduce @xmath142k , @xmath143 @xmath108 , and @xmath144 mag . however , fitting models to the slope and balmer line profiles of an optical spectrum , @xcite determine a nearly identical temperature but log @xmath75 , @xmath117 @xmath108 , @xmath145 , and @xmath146 pc instead . this disparity still can be explained by the presence of another white dwarf which contributes the extra flux at @xmath147 , but _ not _ by a single low mass white dwarf with an overly large radius , which is inconsistent with the balmer line profile fit . the apparent discrepancy might also be explained by supposing that g166 - 58 is located closer to 29 pc , which is near the 30.3 pc lower limit implied by the uncertainty in its parallax . @xcite remark that radial velocity measurements in 1998 june , 1999 april and july agree within their errors and hence there is some weak evidence against radial velocity variability in g166 - 58 . a da or dc white dwarf companion at @xmath148 mag would dilute the balmer lines and possibly redden the optical spectrum , depending on its effective temperature @xcite . it may be the case that such a companion should have already caused notice in the data analyzed by @xcite , but nothing was noted . figure [ fig4 ] appears to imply that a single temperature blackbody does not fit the optical and near - infrared data perfectly . in fact , there seems to be a slight near - infrared excess at @xmath61 , which would become more prominent if a higher temperature model were used to fit @xmath149 only . yet the irac @xmath5 @xmath1 m data lie very close to the plotted blackbody , so perhaps there is another explanation for any apparent mismatch between model and published flux , such as data which are not photometric or calibration errors . in any case , follow up observations of g166 - 58 would help to evaluate the possibility that it may be a double degenerate . specifically , a radial velocity study , a careful model analysis of its spectral energy distribution with accurate and precise photometry , or another trigonometric parallax measurement would all be useful . identification of a double degenerate suspect follows more or less as it does for main sequence stars ; the object lies superior to its expected position in a hertzsprung - russell ( or equivalent ) diagram . this requires that the distance and effective temperature of the star are known or constrained in some fashion . for white dwarfs in the field , the nominal sequence is located near and about a radius corresponding to @xmath150 @xmath108 , or log @xmath100 for cool to warm white dwarfs @xcite . a single target which lies above the log @xmath100 sequence can either be a single white dwarf responsible for the bulk of the luminosity via an overly large radius ( and a comparably low mass and surface gravity ) , or a near equal brightness binary . somewhat oxymoronically , single , low mass ( @xmath151 @xmath108 , helium core ) white dwarfs are understood to be the end products of close binary evolution and more likely than not still attached to their stellar cannibals @xcite . but a single white dwarf with low surface gravity will have appropriately thinned balmer lines due to a reduction in stark broadening , and this is not seen in g166 - 58 @xcite . therefore , if g166 - 58 is binary , it must be composed of two relatively normal mass white dwarfs of similar brightness . given the fact that several such double degenerate suspects have turned out to be bona fide @xcite , it is appropriate to consider dust models which conform to this distinct possibility in addition to those above for a single white dwarf ( see table [ tbl4 ] for a summary of possible parameters ) . the average stellar parameters for g166 - 58 from the two analyses which utilize its trigonometric parallax , both of which find a similar overluminosity , are log @xmath152 , and @xmath153 k @xcite . the resulting effective radiating surface for such a solution is @xmath154 @xmath41 @xcite . applying the opaque , flat ring model and equation ( 1 ) with these stellar parameters yields a distance of @xmath155 @xmath41 to 400 k grains . this is a valid model for a single , low mass white dwarf with the above large radius , but not for a binary . in short , it is not possible to fit two similar white dwarfs within such a tight disk while avoiding interactions ( mass transfer ) and at the same time maintaining relatively cool , and sufficiently distant 400 k grains . additionally , there would be gravitational interactions between the binary and such a disk which would probably preclude a flat geometry . returning to the optically thin case , equation ( 3 ) places 400 k grains at 2.7 @xmath41 . with some adjustment of parameters , this scenario allows ample space for a double degenerate to orbit without interaction @xcite , and sufficient distance from the binary in order to maintain 400 k dust . observations of g166 - 58 with the fine guidance sensors aboard the _ hubble space telescope _ show that it is spatially unresolved to approximately @xmath156 ( e. p. nelan 2006 , private communication ) . this precludes separations wider than 0.3 au or 60 @xmath41 , yet still permits binarity with dust located at one white dwarf component . in table [ tbl1 ] , there are three confirmed or suspected double degenerates : g77 - 50 , ec 1124@xmath157293 , g166 - 58 . these white dwarfs are relisted in table [ tbl5 ] together with their divergent spectroscopic and photometric parameters . in light of the potential binarity of g166 - 58 , a brief focus on the other similar systems is appropriate . _ g77 - 50_. this white dwarf has a recent parallax measurement from @xcite of @xmath158 over a 6.2 yr baseline , placing it firmly at @xmath159 pc . thus , its derived spectroscopic parameters in table [ tbl5 ] , which consist of a good @xmath73 determination from optical and near - infrared photometry , but only a crude log @xmath160 estimate from a rough fit to a weak h@xmath87 feature @xcite , are not consistent with the astrometric distance . as can be seen in the table , the relatively low spectroscopic log @xmath160 value makes g77 - 50 appear _ underluminous _ ; this is atypical as overluminosity is the signpost of binarity . this discrepancy is resolved by presuming the h@xmath87 profile examined by @xcite is due to two velocity - shifted lines ( indeed the line profile shown in figure 23 of @xcite appears asymmetric ) or perhaps weak magnetism which also causes balmer line broadening and mutation in cool white dwarfs where the lines are already weak @xcite . its binarity has been almost certainly confirmed via the detection of two ca lines with disparate velocities , observed at two epochs with both lines revealing individual velocity varation @xcite . therefore , g77 - 50 is a double degenerate in which both components are metal - rich , one or both of which have detectable hydrogen lines . for equally luminous components , models predict @xmath161 @xmath108 ( log @xmath162 ) , and @xmath163 mag for @xmath164 k hydrogen atmosphere white dwarfs . since both components are polluted with metals , only accretion from circumbinary or interstellar material is consistent with the observations and a circumstellar origin can be ruled out . @xcite lists this system as helium - rich based on the absence of h@xmath114 ( d. koester 2007 , private communication ) _ ec 1124@xmath157293_. the parallax reported in @xcite is an unpublished trigonometric measurement ( m. t. ruiz 2006 , private communication ) of @xmath165 which implies a very low mass white dwarf or a binary ( or both ) , if accurate . the difference in the absolute magnitudes of the spectroscopic ( @xmath166 mag ) and photometric ( @xmath167 mag ) parameter determinations implies an extra source with @xmath168 mag which is obviously much brighter than an equally luminous companion to the spectroscopically identified star . therefore , either the system contains a very low mass dz white dwarf with a proportionally large radius so that it dominates the binary spectral energy distribution , or the trigonometric parallax is inaccurate . given the spectroscopic analysis of @xcite , which yields a very normal 0.6 @xmath108 da white dwarf , it is difficult to currently reconcile this system as binary . radial velocity measurements by @xcite in 1998 december and 1999 april agree within the errors and give a gravitational redshift corrected velocity of @xmath169 km @xmath170 for the h@xmath114 line ( no errors are given , but are likely no greater than a few km @xmath129 ) , while similar measurements reported by @xcite yield @xmath171 km @xmath129 ( no epoch given ) . @xcite find narrow to damning evidence of binarity at some known or suspected double degenerates @xcite such as : broad h@xmath114 cores ( e.g. g141 - 2 , case 2 ) , presumably from two velocity - shifted yet unresolved cores ; two separate h@xmath114 cores ( e.g. g271 - 115 , g77 - 50 ) ; and a single , variable h@xmath114 core ( e.g. lhs 1549 ) . two of these have been confirmed as binary by other methods : g141 - 2 has been spatially resolved with the _ hubble _ fine guidance sensors ( e. p. nelan 2006 , private communication ) and radial velocity monitoring of lhs 1549 has determined its orbital period @xcite . the irac photometry of this target was problematic due to a nearby m dwarf common proper motion companion which is reported here for the first time . the companion , pg 0235@xmath172064b , is separated from the white dwarf primary by @xmath173 at @xmath174 in the irac 3.6 @xmath1 m image ( epoch 2005.6 ) , shown in figure [ fig10 ] . examining archival images from 2mass , and the dss reveals separations and position angles between the a and b components which remain essentially constant between 1950 and 2001 . blinking the 1950 and 1990 dss frames clearly shows the pair moving together over 40 yr . the usno - b1.0 catalog has @xmath175 @xmath176 at @xmath177 for the white dwarf @xcite , which would have caused the pair to separate by almost @xmath178 over the last 55 yr if the secondary were background . hence the pair is bound . pg 0235@xmath172064b has reliable 2mass photometry consistent with an early m dwarf which has very likely contaminated some previously published photometry and spectroscopy of the white dwarf , even as far as the blue optical region , causing it to appear too cool ( red ) @xcite . figure [ fig1 ] shows @xmath179 k ( da3.4 ) provides a better fit to the white dwarf data @xcite than previously published , lower effective temperatures corresponding to da4.4 and da8 @xcite . the shorter wavelength photometric data in figure [ fig1 ] were selected as to be minimally , or not at all contaminated by the cool companion : @xmath55 from @xcite ; @xmath180 from @xcite ; and @xmath181 from @xcite . assuming log @xmath100 , very close to that determined by @xcite from a spectrum likely to be contaminated by the m dwarf , the white dwarf would lie at @xmath182 pc . using the _ hubble _ gsc2.2 @xcite blue and red magnitudes , one can estimate @xmath183 mag and @xmath184 for the red dwarf . this corresponds to a spectral type near m3 and agrees reasonably well with the expected absolute magnitude of @xmath185 mag at the estimated white dwarf distance @xcite . for all of the irac images , the task daophot was used in an attempt to remove the light of the m dwarf , but it was found that this generally oversubtracted its flux in the region of the white dwarf , and combined with the @xmath186 mag difference in brightness between components , proved unreliable . instead , the symmetry of the point spread function was exploited to self - subtract the flux on the opposite side of the m dwarf at the location of the white dwarf , introducing an additional error component equal to the square root of the percent flux removed . although a survey of only 17 stars does not allow robust statistics , it is clear that the majority of daz white dwarfs do not harbor warm dust of sufficient emitting surface area to be detected with irac . if most or all of these stars do host circumstellar material , the fractional luminosities must be relatively low compared to currently known dusty white dwarfs . this could result from a modest amount of dust ( as in the zodiacal cloud ) , large particle sizes , or cooler material further from the star . another possibility is that any warm dust produced within the roche limit of a white dwarf is swiftly destroyed through mutual collisions , not unlike ice in a blender , as it orbits with keplerian velocities near @xmath187 . in optically thin disks , particles with orbital period @xmath188 will collide on a timescale given by @xmath189 . the tidal rings at g29 - 38 and gd 362 have been modeled to extend from approximately @xmath190 @xmath41 , where a typical orbital period is only @xmath191 hr and the resulting collision timescale for @xmath192 is @xmath193 dy . if a sizeable fraction of dust produced in a tidal disruption event is initially optically thin , then both collisions and poynting - robertson drag will compete to quickly annihilate this material . the ratio of these two timescales for dust particles orbiting a distance @xmath72 from a star of mass @xmath194 can be written as where @xmath194 and @xmath23 are in solar units , @xmath72 is in au , @xmath86 is in g @xmath93 , and @xmath111 is in microns . this fraction reaches a minimum for 0.1 @xmath1 m grains with 1 g @xmath93 at the inner disk edge , and yields @xmath196 for all possible white dwarf disk parameters . table [ tbl6 ] gives minimum values for @xmath197 at the inner edges of the disks at g29 - 38 , gd 362 , and g166 - 58 . therefore collisions will erode dust grains faster than they can be removed by angular momentum loss . this is also true in the case of optically thick disks where : 1 ) the bulk of material is shielded from starlight , and hence the poynting - robertson effect is diminished , and 2 ) the collision timescale is less than half the orbital period @xcite . hence , for a wide range of disk densities , it is plausible that mutual collisions within an evolving dust ring at a typical white dwarf will result in the relatively rapid self - annihilation of the micron size grains required to radiate efficiently at @xmath198 @xmath1 m . following the tidal disruption of an asteroid , if one models the dust produced as a collisional cascade , the expected particle size distribution behaves classically as @xmath199 @xcite . for dust at main sequence stars , this distribution is reshaped on short timescales as sub - micron size grains are removed by radiation pressure which , as shown above , does not apply in the case of white dwarfs . in the absence of radiation pressure , the average particle in will have a size @xmath200 @xmath201 . for practical purposes , at white dwarfs one can assume @xmath202 @xmath1 m , where particles are already inefficient absorbers and emitters of infrared radiation , and anything smaller approaches the size of gas molecules and atoms . with such a distribution of extant dust , 99.7% of the particles will have sizes @xmath203 @xmath1 m , leaving a paltry fraction of larger particles which could effectively support infrared emission from the disk . clearly , such small particles and gas might be present at most or all white dwarfs which show signs of circumstellar accretion such as the daz stars . on the other hand , the persistence of warm dust disks at several white dwarfs @xcite must be explained despite the fact that in some or most cases where it is produced , it may also be efficiently destroyed . one possibility is that the disk density ( which could contain gas ) becomes sufficiently high as to damp out collisions in the disk and also make it optically thick , thus somewhat protecting it from self - erosion and drag forces simultaneously . the evolution of such a dense , fluid - like ring is then dominated by viscous forces ( differential rotation and random motions ) which cause it to spread , losing energy in the process @xcite . the maximum lifetime of such a ring occurs at minimum viscosity where @xmath205 is the radial extent of the ring and @xmath52 is the surface mass density @xcite . if the mass of a large solar system asteroid , @xmath206 g , were spread into a tidal ring of negligible height ( @xmath207 m ) , a radial extent @xmath190 @xmath41 , consisting of micron size particles orbiting a typical white dwarf , the resulting volume mass density ( 0.55 g @xmath93 ) would be sufficiently high that the mean free path of particles is on the same order as their size . this could effectively damp out collisions , thus minimizing viscosity , and with a resulting surface mass density of @xmath208 g @xmath209 , permit a potential disk lifetime in the absence of competing forces longer than the gyr white dwarf cooling timescales . however , for sustained accretion rates as low as @xmath210 g @xmath129 , a @xmath206 g disk would become fully consumed within several myr . large rocks and colder material orbiting at @xmath211 @xmath41 will be unaffected by any of the aforementioned processes , and such a reservoir of material is strictly necessary to supply some fraction of daz white dwarfs with photospheric metals , regardless of circumstellar dust production ( collisional versus tidal ) and evolution ( persistence versus destruction ) . the overall number of white dwarfs with remnant planetesimal belts may be rather high based on a growing number of detections . if one takes 12% ( 4.1 ) as the fraction of daz stars with circumstellar dust as observed by _ spitzer _ to date , 20% as the fraction of daz stars among cool da white dwarfs @xcite , and 80% as the number of cool da stars among all white dwarfs in the field @xcite , then a lower limit to the number of white dwarfs with asteroid - type belts is at least 2% . this fraction could be as high as 20% if the majority of metal - rich white dwarfs harbor circumstellar matter , which raises important questions about the implied frequency of planetesimal belts around main - sequence stars and the current detection rate ( see the appendix of @xcite ) . owing to their low luminosities , white dwarfs which may have been polluted by heavy elements in winds or transferred material from substellar companions @xcite are easily identified with irac observations @xcite down to t dwarf temperatures . there is no evidence of such companions in the data presented here , ruling out all but the coldest brown dwarfs , active planets and moons as close orbiting , companion - like polluters ( farihi et al . 2008 , in preparation ) . j. farihi thanks m. jura for helpful discussions on circumstellar dust , s. fisher for his expertise on mid - infrared detectors and photometry , t. geballe for encyclopedic assistance with @xmath5 @xmath1 m spectra , s. wachter and s. carey for sharing their familiarity with _ spitzer _ instruments and data , and p. bergeron for kindly providing access to current white dwarf models . this work is 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 . support for this work was provided by nasa through an award issued by jpl / caltech . spectroscopic observations for this work were taken as part of the gemini director s discretionary time gn-2005b - dd-1 . gemini observatory is operated by the association of universities for research in astronomy , inc . , under a cooperative agreement with the nsf on behalf of the gemini partnership : the national science foundation ( united states ) , the particle physics and astronomy research council ( united kingdom ) , the national research council ( canada ) , conicyt ( chile ) , the australian research council ( australia ) , cnpq ( brazil ) , and conicet ( argentina ) . 0032@xmath157175 & g266 - 135 & 9240 & 14.94 & -10.20 & 1,2 + 0235@xmath172064 & pg & 15000 & 15.5 & -9.03 & 1,3 + 0322@xmath157019 & g77 - 50 & 5220 & 16.12 & -11.36 & 1,4 + 0846@xmath172346 & gd 96 & 7370 & 15.71 & -9.41 & 1,5 + 1102@xmath157183 & ec & 8060 & 15.99 & -10.43 & 1,5 + 1124@xmath157293 & ec & 9680 & 15.02 & -8.53 & 1,4,6 + 1204@xmath157136 & ec & 11500 & 15.67 & -7.72 & 1,7 + 1208@xmath172576 & g197 - 47 & 5880 & 15.78 & -10.96 & 1,8 + 1344@xmath172106 & g63 - 54 & 7110 & 15.12 & -11.13 & 1,8 + 1407@xmath172425 & pg & 10010 & 15.03 & -9.87 & 1,9 + 1455@xmath172298 & g166 - 58 & 7390 & 15.60 & -9.31 & 1,8,9 + 1632@xmath172177 & pg & 10100 & 13.05 & -10.75 & 1,9 + 1633@xmath172433 & g180 - 63 & 6690 & 14.84 & -8.63 & 1,8,9 + 1729@xmath172371 & gd 362 & 10500 & 16.23 & -5.1 & 7,10 + 1826@xmath157045 & g21 - 16 & 9480 & 14.58 & -8.83 & 1,8 + 1858@xmath172393 & g205 - 52 & 9470 & 15.63 & -7.84 & 1,5 + 2326@xmath172049 & g29 - 38 & 11600 & 13.04 & -6.93 & 1,5 + 0032@xmath157175 & @xmath216 & @xmath217 & @xmath218 & @xmath219 & 11.0 + 0235@xmath172064 & @xmath220 & @xmath221 & @xmath222 & @xmath223 & 12.4 + 0322@xmath157019 & @xmath224 & @xmath225 & @xmath226 & @xmath227 & 12.4 + 0846@xmath172346 & @xmath228 & @xmath229 & @xmath230 & @xmath231 & 10.5 + 1102@xmath157183 & @xmath232 & @xmath233 & @xmath234 & @xmath235 & 11.0 + 1124@xmath157293 & @xmath236 & @xmath237 & @xmath238 & @xmath239 & 12.4 + 1204@xmath157136 & @xmath240 & @xmath241 & @xmath242 & @xmath243 & 12.4 + 1208@xmath172576 & @xmath244 & @xmath245 & @xmath246 & @xmath247 & 11.0 + 1344@xmath172106 & @xmath248 & @xmath249 & @xmath250 & @xmath251 & 12.4 + 1407@xmath172425 & @xmath252 & @xmath253 & @xmath254 & @xmath255 & 12.4 + 1455@xmath172298 & @xmath256 & @xmath257 & @xmath258 & @xmath259 & 12.4 + 1632@xmath172177 & @xmath260 & @xmath261 & @xmath262 & @xmath263 & 11.4 + 1633@xmath172433 & @xmath264 & @xmath265 & @xmath266 & @xmath267 & 11.4 + 1729@xmath172371 & @xmath268 & @xmath269 & @xmath270 & @xmath271 & 12.4 + 1826@xmath157045 & @xmath272 & @xmath273 & @xmath274 & @xmath275 & 12.4 + 1858@xmath172393 & @xmath276 & @xmath277 & @xmath278 & @xmath279 & 10.5 + 2326@xmath172049 & @xmath280 & @xmath281 & @xmath282 & @xmath282 & 14.0 +

we construct and evolve three one - parameter families and one two - parameter family of steady - state models of stellar disks embedded in live dark matter ( dm ) halos , in order study the dynamical and secular phases of bar evolution . these models are tested against those published in the literature in order to extend them and to include the gaseous component in the follow up paper . specifically , we are interested in the angular momentum , , redistribution in the disk - halo system during these two evolutionary phases without distinguishing between the resonant and non - resonant effects . we confirm the previous results and quantify for the first time the dual role that the dm halos play in the bar evolution : more centrally concentrated halos dilute the dynamical processes of the initial bar growth , such as the spontaneous bar instability and the vertical buckling instability , and slow down the transfer , while facilitating it in the secular phase . the rate of transfer in the disk and the halo is followed up in order to identify sites and times of peak activity in emission and absorption . within the corotation radius , , the disk remains nearly constant in time , as long as stays within the disk a sign that the lost angular momentum to the outer disk and the halo is being compensated by an influx of fresh due to the outward motion of . we demonstrate that this is feasible as long as the bar slowdown dominates the loss of inside . next , we find that in some models the bar pattern speed stalls for prolonged time periods , i.e. , the bar exhibits a constant rate of tumbling when is located outside the disk . this phenomenon appears concurrent with the near absence of transfer between the disk and the halo , and is associated with the halo _ emitting _ at the corotation resonance and absorbing it at the inner lindblad resonance . furthermore , we confirm that stellar bars generally display the corotation - to - bar size ratios in the range of , but only between the times of the first buckling and leaving the disk . hence , the corotation - to - disk size ratio emerges as an important dynamic discriminator between various stages of barred disk evolution . finally , we analyze a number of correlations between the basic parameters of a barred disk and a halo , some already reported in the literature and some new .

0904.0646

bars are expected to have a profound effect on disk galaxy evolution because they constitute a major departure from an axial symmetry and hence facilitate angular momentum and mass redistribution . despite prolonged and focused investigation of bar formation and evolution many theoretical and observational issues remain unresolved . among these are the origin of bars in the universe spontaneous or tidally induced , evolution of bar fractions with redshifts , and theoretical predictions of bar pattern speeds . the intricacies of the bar growth or decay with time are obscure how do bars capture orbits ? do all bars extend to their corotation radii ? lastly , various aspects of nested bar systems are only now beginning to be analyzed . because the underlying dynamics of bars is strongly nonlinear , their numerical modeling has spearheaded the efforts to understand their orbital structure , ability to channel the angular momentum to the outer disk and dark matter ( dm ) halo , efficiency in triggering the radial gas inflows , and more . while we have gained some insight to all of these processes , much remains to be done and one expects additional effects to surface . in this work we aim at deeper understanding of angular momentum redistribution in the disk - halo system facilitated by the stellar bars in the presence of the gas component . here , we analyze a set of equilibrium _ collisionless _ models which differ by a single parameter from each other . in the associated paper , we add the gas component to the system without affecting the mass distribution there ( villa - vargas et al . , paper ii , in preparation ) . our collisionless modeling is in a way complementary to that of athanassoula ( 2003 ) . we reproduce some of the correlations between the bar , disk and halo properties discussed in the above work and quantify additional correlations found . all these will be compared to models with gas . importance of a galactic spheroidal component in bar evolution has been gradually understood , from its supposedly stabilizing effect on the disk ( ostriker & peebles 1973 ; efstathiou et al . 1982 ) , to serving as disk angular momentum sink ( e.g. , sellwood 1980 ; weinberg 1985 ; debattista & sellwood 2000 ; athanassoula 2002 ) , and appears to be full of controversies ( e.g. , christodoulou , shlosman & tohline 1995a , b ; athanassoula 2008 ) . study of the dominant role of resonance interactions between the bar and surrounding orbits was pioneered by lynden - bell & kalnajs ( 1972 ) and applied to bar and halo orbits by tremaine & weinberg ( 1984 ) , athanassoula ( 2002 ) , martinez - valpuesta , shlosman & heller ( 2006 ) , ceverino & klypin ( 2007 ) , weinberg & katz ( 2007a , b ) and dubinski , berentzen & shlosman ( 2009 ) . toomre ( 1981 ) argued that the bar growth can be damped by the introduction of an inner lindblad resonance ( ilr ) , by cutting off the swing amplification mechanism for the @xmath3 modes . hotter stellar disks are expected to be more stable as well ( e.g. , athanassoula & sellwood 1986 ) . counter - intuitively , in the long run , a disk embedded in a more centrally concentrated halo develops a stronger and longer bar ( athanassoula & misiriotis 2002 ) . a comprehensive analysis by athanassoula ( 2003 ) has shown that the efficiency of resonances in angular momentum transfer depends on the dm and stellar dispersion velocities and the dm densities in the vicinity of the major resonances . as the resonances sweep across the phase - space , about @xmath4 of dm particles are locked in the lower order resonances at any given time ( dubinski et al . additional aspects of the bar evolution are known to be influenced by the spheroidal components , such as the vertical buckling instability in the bar ( berentzen et al . 2007 ) . in general , @xmath0 transfer in the disk - halo systems can depend on three parameters , namely , the particle population near low level resonances and the velocity dispersions in the disk and halo ( e.g. , athanassoula 2003 ) . in principle , this means dependence on the halo central mass concentration and on radial velocity dispersions in the disk and the halo . we , therefore , focus on how parameters of the dm halo distribution , such as its mass , concentration , extent , and rms velocities affect the disk - halo evolution and the angular momentum transfer mediated by the stellar bar . specifically , we are interested in how the presence of gas in the disk influence the angular momentum redistribution in the system . for the purpose of clarity , we separate our discussion of collisionless models ( this paper i ) from models with gas ( paper ii ) . collisionless models are used in order to verify the basic details of bar evolution in order to serve as benchmarks for models with gas . they also are used to calibrate our results to those of athanassoula ( 2003 ) . our paper is structured as follows : 2 describes the numerics , initial conditions and model parameters . 3 provides results on basic evolution of stellar bars in our sets of models and analyzes the redistribution of angular momentum in these systems , while 4 is focused on correlations between various parameters in the bar - disk - halo models . we use the @xmath5-body part of the ftm-4.4 hybrid code ( e.g. , heller & shlosman 1994 ; heller , shlosman & athanassoula 2007b ) to evolve the stellar disks and dm halos . the gravitational forces are calculated using the falcon routine ( dehnen 2002 ) which scales as @xmath6 . the units of mass and distance are taken as @xmath7 and 10 kpc respectively . this makes the unit of time as @xmath8 when @xmath9 , and the velocity unit @xmath10 . the gravitational softening is @xmath11 for stars and dm particles . the models consist of a stellar disk with @xmath12 and of dm halo with @xmath13 particles . the initial conditions were created following the prescriptions and density profiles from hernquist ( 1990 ) . the mass volume density distribution in the disk is given in cylindrical coordinates by @xmath14 where @xmath15 is the disk mass , @xmath16 is a radial scale length and @xmath17 is a vertical scaleheight . the density of the spherical halo is given by @xmath18 where @xmath19 is the mass of the halo , @xmath20 is a gaussian cutoff radius and @xmath21 is the core radius . @xmath22 is the normalization constant defined by @xmath23 \ } ^{-1}\ ] ] with @xmath24 . the particle velocities , dispersion velocities and asymmetric drift corrections have been calculated using moments of the collisionless boltzmann equation . since models thus constructed are not in exact virial equilibrium , the halo component was relaxed for @xmath25 in the frozen disk potential . our goal is to investigate the effect of the dm mass distribution on the evolution of stellar bars in a live disk halo system . we choose a specific model , defined here as the standard model ( sd ) , where the dm density profiles are specified by three parameters , @xmath19 , @xmath21 and @xmath20 , corresponding to the dm halo mass , dm core radius and the dm gaussian cutoff radius . based on the sd model , three sequences of models have been created , each sequence resulting from varying one of the parameters only . in the fourth sequence , we simultaneously modify two parameters at the time , @xmath21 and @xmath20 , in attempt to target the outer halo only , while trying to keep the inner halo unchanged . in the sd model , the halo - to - disk mass ratio in the inner @xmath26 is kept to unity . table [ sd_param ] lists the values of the sd parameters . .parameters of the standard model [ cols= " < , > , < , > " , ] [ sd_param ] the initial mass volume density profile in the disk is kept unchanged in all models . the spatial distribution of the stellar particles is thus identical in all realizations of the initial conditions . on the other hand , the velocities of the stellar particles have been adjusted in each models to provide for the changing rotational support against the combined gravitational potentials of the varying disk - halo systems . the four model sequences are as following : @xmath19 sequence : : : the mass of the halo , @xmath19 , has been reduced to @xmath27 and @xmath28 ( models m70 and m40 respectively ) of the sd model . @xmath21 sequence : : : the dm core radius @xmath21 was increased by a factor of 2 and 4 with respect to the sd model ( models c30 and c57 respectively ) . @xmath20 sequence : : : the dm gaussian cuttoff radius @xmath20 was decreased to @xmath29 and @xmath30 of that in the sd model ( models t22 and t16 respectively ) . @xmath31 sequence : : : in these hybrid models , mt1 and mt2 , both @xmath20 and @xmath19 have been varied , to keep the inner @xmath32 dm density profile close to that of the sd model . l r r r + model & @xmath33 & @xmath21 & @xmath20 + sd & 3.15 & 0.1425 & 2.85 + + m70 & 2.20 & 0.1425 & 2.85 + m40 & 1.26 & 0.1425 & 2.85 + + c30 & 3.15 & 0.30 & 2.85 + c57 & 3.15 & 0.57 & 2.85 + + t22 & 3.15 & 0.1425 & 2.20 + t16 & 3.15 & 0.1425 & 1.60 + + mt1 & 2.40 & 0.1425 & 2.20 + mt2 & 1.71 & 0.1425 & 1.60 + [ seqpar ] after halo relaxation in the frozen disk potential . each panel displays the stellar disk rotation curve ( red , long dashed ) , halo ( blue , dashed ) and the total curve ( black , solid ) . ] the dm halo parameters in each model are listed in table [ seqpar ] . the number of particles and the numerical truncation radii are the same as in the sd model . the halo mass @xmath19 is conserved in sequences @xmath21 and @xmath34 . the dm density profiles of all models after relaxation in the frozen disk potential are shown in figure [ rhomods ] . the resulting dm density profiles exhibit monotonic decrease along the @xmath35 sequence , show a progressively larger core in the @xmath21 sequence , move the outer halo mass inward to the inner halo , across @xmath36 , in the @xmath34 sequence , and increase the outer halo mass while leaving the core unchanged in the hybrid models sequence . the initial circular velocity curves , showing the contributions from the disk , the halo and the total , are displayed in figure [ vcir3 ] . models m40 , c30 and c57 host maximal disks which dominate the potential of the inner part . models sd , m70 , mt1 and mt2 have equal contributions of disk and halo at @xmath37 , while t22 and t16 are halo dominated . with the development of the stellar bar , the disk typically becomes even more dominating , although the inner halo is also dragged in an adiabatic contraction ( e.g. , dubinski et al . the model sequences described in table 2 differ only by the parameters of dm distribution as given by eq . 2 . these include the total mass of the halo , its core size and the gaussian truncation radii . as the bar properties and angular momentum evolution are both heavily dependent on the mass distribution in the system , the choice of the free parameters allows us to fine - tune the changes . because the dm appears to be on the receptive side of the angular momentum transfer , we target the halo as a whole , and its inner and outer parts separately . discussion and comparison with published models is made in 4 . to gauge the strength of the bar we use the fourier amplitude @xmath38 of the @xmath3 mode normalized by the @xmath39 mode . it is obtained by various methods here we show the results of integration over all the disk , and integration over restricted cylindrical volumes of a particular interest . we define @xmath38 in the integration limit over the fixed radial range @xmath40 . second , we define @xmath41 when integrate over @xmath42 range , where @xmath43 is the bar size defined in 3.2 . intuitively , the second definition reflects the bar properties more fully because @xmath41 is not diluted by the disk properties . both @xmath38 and @xmath41 show some similarities , e.g. , the same peaks , raises and drops , and differences , e.g. , the relative strength of the peaks ( fig . [ rb_a2 ] , first and second columns ) . most models show common stages in the evolution of @xmath38 and @xmath41 : ( 1 ) an initial exponential steep rise , ( 2 ) a subsequent peak or plateau followed by a sudden drop , ensued by ( 3 ) a more gradual and sustained rise , ending by ( 4 ) a saturation of the bar strength . the duration of each stage varies from model to model , but stages `` 1 '' and `` 2 '' are much shorter than stage `` 3 '' . we refer to stages `` 1 '' and `` 2 '' as the dynamical evolution of the bar , and stages `` 3 '' and `` 4 '' as the secular evolution of the bar . models m40 and t16 are the exceptions . the former model has the least massive halo and exhibits a prolonged secular growth of the bar after the first peak . the bar amplitude does not reach saturation even in the hubble time . the latter model has the most massive halo core and the bar strength never shows a clear first peak seen in all other models . however a careful analysis of @xmath38 and especially @xmath41 reveals that the exponential growth is terminated at @xmath44 and the subsequent evolution lucks the vertical buckling instability ( see 3.4 ) , but otherwise follows the path of other models . we take a closer look at these trends in 4 . we note that our decision to create one - parameter sequences naturally leads to a wide spectrum of bar properties and can include some of the extreme behavior . some clear trends of the bar strength behavior can be observed along each of the sequences defined in 2.2 . first , in all models , except the two exclusions mentioned above , @xmath38 and @xmath41 have saturated by @xmath45 . because @xmath41is tailored for the bar , the saturation of this parameter is more obvious ( more about this in 3.3 ) . second , making the _ inner _ halo less massive ( i.e. , less centrally concentrated ) , either by reducing the total halo mass or by increasing the size of the central core ( sequences @xmath35 and @xmath21 , respectively ) , results in a shorter rise time of the bar instability and hence brings up the bar earlier . third , increase in the mass of the inner halo and decrease in the outer one , has the most dramatic effect on the bar strength , substantially increasing the timescale of the bar instability . this hints at dm mass concentration being important rather than the total halo mass . varying the mass of the outer halo alone has a much smaller effect on the bar . finally , in all one - parameter sequences , the first @xmath38 or @xmath41 peak forms a progressively more extended plateau for less concentrated halos , respectively . again t16 is an exclusion . evolution in the sd model , showing the high frequency oscillations during @xmath46 . no smoothing has been applied to the data . ] furthermore , @xmath38 and @xmath41 show high frequency variability in some models ( e.g. , figure [ a2sd ] ) , being strongest in models sd , c30 and t22 . this variability can also be seen in @xmath43 and ps . evolution , the bar length and its pattern speed . this variability is limited to the time of the bar secular growth and dies out with saturation of the bar strength , @xmath38 and @xmath41 . these oscillations coincide in time with the presence of four short arms located close to the bar end , and two spirals in the outer disk beyond the bar radius . both sets of spirals have pattern speeds larger than that of the bar . the presence of more than one strong pattern of waves or modes in the disk can be the result of a non - linear mode coupling ( e.g. , tagger et al . 1987 ; sygnet et al . 1988 ) that gives rise to beat waves of a frequency @xmath47 and the azimuthal wave number @xmath48 , where @xmath49 and @xmath50 are the ( azimuthal ) wave numbers of the interacting waves and @xmath51 and @xmath52 are their frequencies , @xmath53 , where @xmath54 s are their pattern speeds . this type of coupling has been observed and quantified in numerical simulations ( e.g. , masset & tagger 1997 ; martinez - valpuesta 2006 ) . debattista & sellwood ( 2000 ) has also observed the high frequency variability , but no quantitative analysis of the mode coupling was attempted . we test whether the bar and the outer two spirals , both @xmath3 modes , couple to give rise to the observed beat mode @xmath55 arms . we verify that the frequency @xmath56 of the @xmath55 mode is actually the sum of the frequencies of the interacting modes . table [ m2m4spectra ] shows the frequencies @xmath57ps . , @xmath58 and @xmath56 obtained with the mode coupling analysis , with an error of @xmath59 . here @xmath60 is the pattern speed of the outer @xmath3 spirals . the measured frequency of the beat mode ( column 5 ) matches the frequency predicted by the non - linear coupling ( column 6 ) . we can explain the oscillations in @xmath38 as resulting from superposition of the major axis of the bar with the outer spirals , which are the second strongest mode ( after the bar itself ) present in the disk . the frequency @xmath61 of the @xmath38 oscillations must thus be equal to the frequency at which the spirals align with the bar @xmath62 . as can be seen from columns 7 and 8 , these two frequencies coincide within the error limits . l * 7r + model & time & @xmath63 & @xmath64 & @xmath56 & @xmath65 & @xmath61 & @xmath66 + sd & 106 & 1.80 & 4.86 & 6.66 & 6.66 & 0.49 & 0.49 + c30 & 113 & 1.32 & 4.24 & 5.58 & 5.56 & 0.45 & 0.46 + t22 & 149 & 1.80 & 5.72 & 7.58 & 7.52 & 0.61 & 0.62 + [ m2m4spectra ] figure [ rb_a2 ] shows the evolution of ps . in our models , which is well defined by the time @xmath38 @xmath67 . in all models , the initial value of ps is very similar , with the scatter of @xmath68 . although the details of the evolution of ps . differ from model to model , most importantly , often shows an ( anti)-correlation with @xmath38 and @xmath41 , as emphasized already by athanassoula ( 2003 ) . this is not always observed in cosmological simulations of disk evolution ( heller , shlosman & athanassosula 2007a ; romano - diaz et al . we find , however , that the anti - correlation between the bar strength and its pattern speed are limited to times prior to saturation of @xmath38 and @xmath41 . in 3.3 we show that this corresponds to times when the bar corotation radius lies within the disk . in three models , m70 , t22 and mt1 , the @xmath38ps . correlation is maintained even after the corotation leaves the disk in these models the total @xmath0 of the disk is conserved during this time period of flat @xmath38 and ps . . a close examination of ps . evolution in fig . [ rb_a2 ] reveals prolonged periods of ps . @xmath69 these are found either in the dynamical phase , prior to the first buckling but when the bar is already sufficiently strong ( sd , m70 , c57 ) , or during the late stage of secular evolution , especially pronounced in m70 , t22 and mt1 , and to a lesser degree in c57 . these latter cases are of most interest to us . a similar behavior was discussed before by valenzuela & klypin ( 2003 ) , who related it to an abnormally low dynamical friction of the bar against the background . we return to this issue in 3.6 . the semi - major axis of the bar , @xmath43 , is defined here as the radius where the bar equatorial ellipticity drops by @xmath70 off its peak . martinez - valpuesta et al . ( 2006 ) has tested this method by contrasting it with the most reliable way of determining the bar size using the last stable orbit supporting the bar ( martinez - valpuesta et al . it seems robust when applied after the first maximum of the bar strength . the ellipticity of the bar at different radii is obtained by fitting ellipses to the isodensity contours in the face - on disk . figure [ rcd ] displays the evolution of @xmath43 for all the models . similarly to the evolution of @xmath38 and @xmath41 , the bar length has initial period of a fast growth that reaches maximum , sometimes followed by a drop and subsequent sustained but slower rise , in most models . some models show a degree of saturation , toward the end . the first peak in @xmath43 always coincides in time with that of @xmath38 , except in the model t16 which lacks it . the peak in @xmath43 , however , is not as pronounced as that in @xmath38 and there are additional caveats . as in @xmath38 , the initial growth varies with a model in such a way that , in models with the less centrally concentrated halos , the time span of the initial growth is shorter . furthermore , in some cases , additional drop in the bar length can be seen after some time of the secular growth , in the time range of @xmath71 . this drop appears to be related to the formation and disappearance of the _ ansae _ at each tip of the bar ( martinez - valpuesta et al . the ansae eventually ` detach ' from the bar and becomes misaligned with its major axis . the transition to this misalignment results in a fast decrease of the bar length , taking place at @xmath72 for model sd , @xmath73 for m70 , 22 for c30 , 190 for c57 , 205 for t22 , and 165 for mt1 and mt2 . in general , evolution in the bar length does not necessarily go in tandem with the changes in @xmath38 and is much less monotonic . for example , the drop in @xmath43 around @xmath72 in the sd model has no counterpart in the @xmath38 or @xmath41 evolution . of course , a stronger @xmath38 does not mean a longer bar , as seen in the last @xmath74 of the evolution of the sequence @xmath19 models . overall , the @xmath35 and @xmath34 sequences show a substantial dispersion in the bar sizes at the end of the simulations , while two other sequences end up with very similar bars . in all sequences , the bars differ substantially during the intermediate times , especially after the first buckling time . the bar corotation radius @xmath1 has been computed using linear approximation ( the second column in fig . [ rcd ] ) . it grows with time as a consequence of the bar slowdown . @xmath1 keeps growing throughout the simulation , the exceptions are models m70 , t22 and mt1 , in which a plateau is reached . we use this characteristic radius in order to construct and follow the evolution of two important ratios , namely , @xmath1/@xmath43 and @xmath1/@xmath75 . the position of @xmath1 has a profound effect on the angular momentum transfer from the bar region to the outer disk and the dm halo ( athanassoula 2002 ; martinez - valpuesta et al . the dm particles at the corotation resonance are responsible for much of the angular momentum absorption by the halo . hence , the availability of halo or outer disk resonant particles is of the prime importance to the evolution of the bar . specifically , it is important to know whether @xmath1 lies within the stellar disk in our models at all times . we define @xmath75 as the disk radius which encloses 98% of the disk mass , and the evolution of the ratio @xmath1/@xmath75 is shown in figure [ rcd ] . at the time of the bar formation , @xmath1/@xmath75 lies between 0.5 and 0.8 in all models . this ratio raises with time enough to move the corotation out of the disk with the exception of models m40 and t16 , where bars grow at the slowest pace during secular evolution . the growth of the bar seems to be sensitive to the moment at which @xmath1 reaches the disk edge even if the growth continues , it proceeds at a slower pace . this change happens with a slight but measurable delay of @xmath76 . moreover , as noted before , the detachment of the ansae from the bar happens just before @xmath1/@xmath77 , which manifests itself by a sharp drop in the bar length . we pay a special attention to the bar evolution when @xmath1/@xmath77 . [ rcd ] displays an interesting behavior of this ratio in the above regime it flattens with time before resuming its growth . analyzing this behavior , we find that as the @xmath1 moves outside the disk , some stellar particles become trapped at @xmath1 . @xmath3 spiral arms appear connecting the outer edge of the disk to these trapped particles . this is characteristic of all models where the above conditions exist . at @xmath78 , depending on the model , the spirals dissolve , and their particles form an amorphous cloud outside the disk . we return to this issue in 3.6 . the bar must lie of course inside its @xmath1 , because orbits outside @xmath1 do not support the bar . the morphology of the offset ( gas ) shocks in the numerical bars has been argued to constrain the ratio @xmath1/@xmath43 , restricting it to @xmath1@xmath79 ( athanassoula 1992 ; see also early work by van albada & sanders 1982 ) . limiting our discussion of this ratio only to times when @xmath1/@xmath80 , we observe it dropping to the range @xmath81 after the first buckling , and staying generally below @xmath82 until the @xmath1 moves out of the disk . model t16 ( no buckling ! ) does not follow this trend and rather shows a monotonic decay to @xmath83 at the end of the run . on the other hand , m40 the only other model with @xmath1 always in the disk , displays a remarkably constant @xmath1@xmath79 almost over the entire run ( figure [ rcd ] ) . naturally , after @xmath1 moves out of the disk , this ratio is bumped to above 2 . in paper ii , this ratio will be tested against the shape of the offset shocks in the gas . of the @xmath84 mode in the @xmath85-plane corotating with the bar major axis . no filtering was applied to this data . ] the @xmath5-body bars undergo phases of vertical buckling instability where transient asymmetries develop in the form of vertical oscillations of the equatorial plane and asymmetric thickening of the disk ( e.g. , combes & sanders 1981 ; combes et al . 1990 ; pfenniger & friedli 1991 ; raha et al . 1991 ) , and results in the formation of the so - called peanut / boxy - shaped bulges ( kormendy & kennicutt 2004 and refs . therein ) . these phenomena can be recurrent and affect different parts of the bar ( martinez - valpuesta et al . 2006 ; berentzen et al . we measure the buckling asymmetry with the fourier coefficient @xmath86 of the @xmath84 component in the @xmath85-plane , again normalized by the m=0 mode . in all our models , except t16 , we detect at least a single phase of the vertical buckling . the buckling always happens simultaneously with the first @xmath38 drop ( e.g. , martinez - valpuesta & shlosman 2004 ) , which we confirm here . in all the model sequences , models with less centrally concentrated halo exhibit earlier and stronger buckling this happens because @xmath38reaches its first maximum earlier in these models . we do not find any clear correlations between the strength and time of the second buckling and the initial properties of the models . in all cases of repeated bucklings , @xmath1 lies within the disk . we note that t16 does not exhibit the drop in @xmath38 or @xmath41 , but does show a break in their slopes around @xmath44 . moreover , while all our models develop peanut / boxy bulges abruptly after these drops , t16 develops this bulge gradually over the secular stage of its evolution . this model appears as a nice example of a secular buildup of peanut / boxy bulges as a result of the diffusive action of the vertical inner lindblad resonance ( friedli & pfenniger 1990 ; martinez - valpuesta & shlosman 2004 ) . berentzen et al . ( 2007 ) have shown that mass concentrations resulting from gas accretion to the center damp the vertical buckling and alter the bulge shape , making it more elliptical , if the gas fraction in the disk is high . t16 is a purely collisionless model with the most centrally concentrated halo . it is possible that a qualitatively different behavior is the corollary of this concentration . clearly , the action of the vertical resonance leads to a _ secular _ and not dynamical buildup of a peanut - shaped bulge in this case . we also note that in all models the nonlinear inner lindblad resonance ( ilr ) first appears only after the first buckling , although the linear analysis claims their existence from the start ( pfenniger & friedli 1991 ; martinez - valpuesta et al . the appearance of this resonance is the consequence of the increase in the central mass concentration , by a factor of @xmath87 , as a result of the bar buckling , which drags inward the dm as well ( dubinski et al . 2009 ) . . ] the bar increases its mass by capturing particles from the disk within its @xmath1 . in figure [ mbmd ] we plot the fraction of the stellar mass contained inside the bar , @xmath88 . to estimate the bar , we sum the particle masses within a rectangular box aligned with the bar major axis , with dimensions given by the major and minor axes of the bar and @xmath89 . by the end of the run , the bar has captured between @xmath90 and @xmath91 of the disk mass ( the exception is model t16 whose bar growth has been severely delayed by the halo mass concentration ) . the initial stage of bar formation is associated with an intense capture of disk particles . eventually this rate declines , and even though the bar mass keeps growing it does so at a much more modest pace . the growth of @xmath88 clearly correlates with the evolution of the bar size in fig . [ rcd ] ( left frame ) an increase / decrease in one of them corresponds with an increase / decrease in the other . typically , the growth of the bar size and of @xmath88 saturates when the @xmath1/@xmath75 is driven above unity . however , surprisingly , there are exclusions of this behavior , e.g. , c57 . we do not find correlation between the bar mass and @xmath38 . it may happen that the bar keeps capturing mass while @xmath38 is saturated , as in model sd at @xmath92 and later on . in summary , the ratio @xmath88 saturates around 0.8 in in all models , except t16 . components of the halo angular momenta , @xmath93 ( solid , black ) and @xmath94 ( dashed , green ) , enclosed by the cylindrical volumes @xmath95 @xmath1 and @xmath96 @xmath1 , and @xmath97 . ] the backbone of the bar is composed from material moving along eccentric orbits of relatively low angular momentum , compared to the circular orbits of the same energy . moreover growing bars consume more circular orbits , i.e. , those with larger @xmath98 ratio , redistributing their angular momentum to the outer disk and halo . the capacity of different regions of the disk and halo to emit and absorb angular momentum is thus of a prime importance to the bar evolution , and largely determine its ability to grow by trapping additional disk orbits . rather than estimate the effect of the resonances on this process ( athanassoula 2002 ; martinez - valpuesta et al . 2006 ) , we focus on the total @xmath0 transfer within the disk and to the halo , i.e. , accounting for the resonant and non - resonant @xmath0 redistribution . in figure [ jzd ] we plot the evolution of the @xmath99 component of the total angular momentum of the disk , @xmath100 , as well as in the cylindrical volumes centered on the symmetry axis of the disk and separated by @xmath1 , i.e. , @xmath101@xmath1 ( @xmath102 ) and @xmath1 @xmath103 ( @xmath104 ) , and similarly for the halo ( fig . [ jzh ] ) . since the total angular momentum in each model is conserved , the evolution of @xmath105 in the halo is an inverted mirror image of that in the disk . the initial angular momentum , @xmath100(@xmath37 ) , of the disk varies from model to model due to the differences in the halo mass distribution . it changes very little before the vertical buckling in the bar , while @xmath102 and @xmath104 anticorrelate . this means that the halo gains a negligible amount of @xmath0 , which flows nearly exclusively to the outer disk during this phase ( see also fig . [ jzh ] ) . after the buckling , @xmath100 and @xmath104 drop monotonically till the end of the run . this is not true about @xmath102 which stays approximately constant until the time when @xmath1 leaves the disk ( fig . [ jzd ] ; see also martinez - valpuesta et al . how this is possible ? while it is not surprising that the loss of @xmath0 within the corotation leads to an outward motion of @xmath1 , the apparent ` conspiracy ' in @xmath102@xmath69 over such an extended time period is somewhat puzzling . it hints at some regulation mechanism which fine tunes the loss of angular momentum to the outer disk and to the halo by gravitational torques with its influx across the corotation . we attempt to answer this question , using obvious simplifications : the explanation of this phenomenon lies in the intricacies of @xmath102 balance and the resulting behavior of ps . . the angular momentum within @xmath1 is @xmath106 where @xmath107 is the average tangential velocity of particles inside @xmath1 , which also defines the coefficient @xmath22 . we require @xmath102@xmath69 between the times of buckling , as variation of @xmath102 is relatively small during the buckling ( e.g. , fig . [ jzh ] ) ] and when @xmath1@xmath108 , and take the time derivative of @xmath102 using the leibniz formula , @xmath109 dr + \nonumber\\ + v_{\rm c } r_{\rm cr}^2 \sigma(r_{\rm cr},t ) \frac{dr_{\rm cr}}{dt } = 0,\end{aligned}\ ] ] where @xmath110 is the disk surface density which can be obtained from eq . [ eq : rho_disk ] by integrating over @xmath99 . we assume @xmath111 this is justified because contribution of @xmath0 within the velocity turnover radius is small and at larger radii , @xmath112 is dominated by the isothermal halo and is nearly independent of radius ( fig . [ vcir3 ] ) . the first term in eq . [ eq : derivjd ] describes the emission of @xmath0 by the _ disk , within an instantaneous @xmath1 , due to the torque @xmath113 imposed by the dm and the outer disk . for brevity , we assume that the stellar bar extends to @xmath1 and this torque acts upon it . in other words , all the material within @xmath1 resides in the bar . the second term in eq . [ eq : derivjd ] represents the influx of @xmath102 due to the advance of @xmath1 and the resulting addition of a new mass within this radius on nearly circular orbits . next , we relate the motion of @xmath1 to the slowdown of the bar , @xmath114 . the torque on the bar gives the rate of change of its angular momentum ( and , therefore , change in the angular momentum of the inner disk ) : @xmath115 where @xmath116 is the moment of inertia of the bar and @xmath117 is the angular momentum of the internal circulation within the bar . ( the second term does not appear in eq . 10 of athanassoula ( 2003 ) . ) hence , @xmath118 . assuming that the dominant loss of the angular momentum by the bar is due to the slowdown of its tumbling @xmath119 ( not a trivial assumption and definitely not a general one ! ) , we can re - write eq . [ eq : derivjd ] , replacing its first term by the action of @xmath113 , as @xmath120 where we have taken @xmath121 . if the second term in eq . [ eq : torque2 ] loosely represents the moment of inertia of the bar , the net change in @xmath102 is indeed negligible , explaining its near constancy in fig . [ jzd ] ( mid column ) up to the time when @xmath1 leaves the disk . in summary , we show that the constancy of the angular momentum within @xmath1 can be indeed explained if a number of straightforward assumptions is made to allow for an analytical estimates . these assumptions are as following : ( 1 ) the bar extends to near @xmath1 during this time and the disk mass within this radius lies mainly in the bar ; ( 2 ) the rate of change of the angular momentum in the bar due to the change in its moment of inertia and internal circulation is smaller than that resulting from the bar slowdown ; ( 3 ) the circular velocity in the disk around the @xmath1 region and beyond is independent of radius ; and ( 4 ) the average tangential velocity of the material in the bar is a fixed fraction of the disk circular velocity . while this behavior shows up in all our models , as long as @xmath1 lies within the disk , a more general set of mass distributions should be tested before making a final conclusion . as a function of radius and time for the sd model ( see also the text ) . the top row corresponds to the halo and the bottom row to the disk . red / blue colors represent the absorption / emission of @xmath0 using a linear scale in color . the main resonances are indicated by the solid lines : the outer / inner lindblad resonances ( olr / ilr ) , the outer / inner ultraharmonic resonances ( ouhr / iuhr ) and the corotation ( cr ) . the cylindrical shells have @xmath122 and @xmath97 . time smearing is @xmath123 . ] while the halo absorbs all @xmath0 emitted by the disk , it is instructive to look into which parts of the halo are especially active in this process . figure [ jzh ] displays the evolution of the absorbed @xmath93 and @xmath94 . we observe that the inner halo becomes more receptive to the process with time , except in m40 and t16 . this is clearly related to the outward motion of @xmath1 . we note also , that the rate of @xmath0 absorption by the inner halo increases substantially after @xmath1/@xmath77 , and the halo serves as the only sink of the angular momentum . asymptotically , one can divide the models in two classes : when the rate of @xmath0 absorption by the inner halo goes to zero or even becomes negative ( e.g. , m70 , t22 , mt1 ) , or when it declines only slightly ( e.g. , sd , m40 , c30 ) . an alternative way to detail the exchange of the angular momentum in the disk - halo system is to divide the disk and halo into a number of concentric cylindrical shells . we have constructed a 2-d map of the angular momentum in each shell as a function of radius and time . it is more revealing , however , to plot the time derivative of @xmath100 and @xmath105 at each radius . this analysis was performed separately for the halo and the disk . in figures [ jz_diag_sd ] to [ jz_diag_mt ] , we display color - coded diagrams of the rate of change of angular momentum @xmath124 with radius and time for each run , and the same for the halo . red color corresponds to emission and blue to absorption of the angular momentum . all the diagrams corresponding to disks have been calibrated identically . dm halos have also been calibrated uniformly among themselves . as a function of radius and time for models m70 ( left column ) and m40 ( right column ) . the top row corresponds to the halo and the bottom row to the disk . color code and lines as in fig . [ jz_diag_sd ] . ] as a function of radius and time for models c30 ( left column ) and c57 ( right column ) . the top row corresponds to the halo and the bottom row to the disk . color code and lines as in fig . [ jz_diag_sd ] . ] as a function of radius and time for models t22 ( left column ) and t16 ( rigt column ) . the top row corresponds to the halo and the bottom row to the disk . color code and lines as in fig . [ jz_diag_sd ] . ] as a function of radius and time for models mt1 ( left column ) and mt2 ( right column ) . the top row corresponds to the halo and the bottom row to the disk . color code and lines as in fig . [ jz_diag_sd ] . ] the most striking features in figures [ jz_diag_sd ] to [ jz_diag_mt ] is the continuity of color , i.e. , emission and absorption bands , both in the disk and the halo , and their outward shift . the bands appear nonuniform in time and space . the halos are almost exclusively absorbers of the angular momentum , while the disks show regions of absorption and emission . from the halo evolution in the sd model ( fig . [ jz_diag_m ] ) , we note that @xmath105 is absorbed preferentially at three distinct bands which shift outwards with time . other and weaker absorbing regions are observed between them and at larger radii . the absorption / emission bands in the disk are alternating , unlike in the halo . to relate the emission / absorption bands to dynamical characteristics of the disk - halo system , we specify the positions of linear resonances on the color map . these are used for guidance purpose only , as the positions of linear resonances maybe misleading , especially for the strong bars , and especially for the inner lindblad resonance(s ) . we find a close correspondence between the behavior of @xmath125(r ) bands and the location of the linear resonances ( superposed solid lines ) , namely , the corotation , inner / outer lindblad resonances ( ilr / olr ) , and the inner / outer ultra - harmonic resonances ( uhr ) . while it is clear that the redistribution of @xmath0 in the disk and the halo is a strongly nonlinear process , some general conclusions can be made . first , as expected , the emission and absorption between the inner disk and the inner halo components strongly correlate , with a certain interference of the outer disk ( figs . [ jz_diag_m][jz_diag_mt ] ) . here we refer to inner / outer with respect to the @xmath1 . the peak emission of @xmath0 by the disk happens at the band associated with the ilr . a weaker emission band lies at the inner uhr ( see also fig . 10 of martinez - valpuesta et al . the strongest absorption in the disk lies at the corotation and extends to the outer uhr . weaker absorption is associated with the olr in the disk . the halo absorption bands are centered on the ilr , inner uhr / corotation and the outer uhr / olr . however , it is nonnegligible in other regions as well . we note that producing horizontal slices in the halo reveals that the lower halos , i.e. , @xmath126 , absorb mostly at the ilr while the upper halo absorb mostly at @xmath1 . this is a refinement to previous results which shown that the main absorption is at the cr resonance ( athanassoula 2002 ; martinez - valpuesta et al . 2006 ) . second , we observe temporal correlations between the emission and absorption peaks in the disk and the halo . over the length of the run , both the disk and the halo experience from two to three ` times of activity , ' except t160 . the first such activity time is clearly associated with the first peak in @xmath38 and the subsequent buckling . the other activity time lies toward the saturation of @xmath38 , late in the evolution of the bar . some models , like sd , show an additional activity in between . this seems to be associated with @xmath1 crossing @xmath75 the halo absorption peaks at the corotation during this time . the resonances move outwards faster during intense @xmath0 absorption . the outward migration of the resonances is of course related to the bar slowdown , resulting in the resonances sweeping across the disk and halo material . third , @xmath125(r ) and @xmath127(r ) in figs . [ jz_diag_m][jz_diag_mt ] appear to correlate with @xmath102 , @xmath104 , @xmath93 and @xmath94 in figs . [ jzd ] and [ jzh ] . it also correlates with @xmath38 and especially with @xmath41 . lastly , the maxima of disk emission correspond to the maxima in the halo absorption . next , we return to the atypical behavior of some stellar bars in their secular phase , which exhibit ps . @xmath128 , discussed in 3.1 . in all the models where we observe this phenomenon , the ratio @xmath1/@xmath129 at the time of detection . this behavior , as expected , is associated with the rate of @xmath0 transfer from the disk to the halo : in all these models , @xmath130 , both in the halo absorption or disk emission during @xmath131 ( figs . [ jz_diag_m ] to [ jz_diag_mt ] ) . the outward motion of the disk resonances stalls as well ( figs . [ rcd ] and [ jz_diag_m ] [ jz_diag_mt ] ) . moreover , there is an indication that this is concurrent with the reversal of @xmath0 flow : while the disk still shows a weak emission at its ilr , the halo shows a weak _ emission _ at @xmath1 and a strong absorption at the ilr . the latter remains the most active resonance in the halo . the emission of @xmath0 by the halo at @xmath1 is rather unusual and has not been observed before in any models in the literature , to the best of our knowledge . lastly , the disk inspection reveals the presence of two trailing spiral arms extending to @xmath1 , and when @xmath1/@xmath129 , connecting the outer edge with @xmath1 ( see 3.3 ) . some stellar particles appear to be trapped at @xmath1 in its outward motion up to @xmath132 , depending on the model . these particles appear to concentrate at specific azimuthal locations of the cr `` circle '' with respect to the bar ( in the bar frame of reference ) . the particles in the spirals are not trapped . these spirals are prominent in figs . [ jz_diag_m][jz_diag_mt ] , and are responsible for emission at @xmath1 and absorption of @xmath0 in the iuhr - cr band , as @xmath1 moves outside the disk . while the number of these particles is not large , their specific angular momenta is the largest in the disk . we have constructed and evolved three one - parameter and one two - parameter sequences of dm halos hosting a standard asymmetric stellar disk . all models developed bars and we have followed their evolution over a hubble time . the properties of the bars , such as their strength , pattern speed , size , corotation radius , ratios of corotation to bar lengths , and more , have been related to those of the disks and host halos . based on our results , we now test a number of correlations found by athanassoula ( 2003 ) . we also obtain additional ones all of these will be used in order to compare with the gas models ( paper ii ) . for this purpose , we deal with three entities : the inner , bar - forming disk ( defined within @xmath1 ) , the disk as a whole , and the dm halo . we adopt the halo - to - disk mass ratio , @xmath133 within the disk velocity turnover radius , @xmath134 ( e.g. , sackett 1997 ) at @xmath37 , as an independent variable , where @xmath16 is the disk radial scalelength defined in section [ s_inicon ] . in the following , we also use the radial and vertical dispersion velocities in the disk and radial dispersions in the halo , and test how the efficiency of @xmath0 redistribution within the system depends on these parameters . , by the inner stellar disk during its dynamical ( upper ) and secular ( lower ) phases of the bar evolution as a function of the @xmath133 ratio . the rate @xmath135 has been calculated within @xmath136 ( upper ) and within @xmath75 ( lower ) . the mass ratio has been calculated within @xmath136 . the averaging was performed over the time period from @xmath37 to the end of an exponential growth of the stellar bars ( upper ) and from the time of the first minimum of @xmath38 , i.e. , the of the buckling , to its saturation , @xmath137 ( about 10 gyr ) ( lower ) . the green circles represent the two hybrid models . the attempted linear fit in the lower frame does not include the lower right corner point ( t16 ) . ] to quantify the role of the halo in the bar instability , we calculated the exponential timescale of @xmath38 growth , @xmath138 , against @xmath133 within @xmath136 , which represents the halo concentration , normalized by the disk mass within the same radius . a linear correlation was obtained , namely , that an increase in the dm mass concentration leads to a slower bar growth before the first buckling . however , the bar evolution in the secular phase can not be described by an exponential growth . looking for a more universal measure of the bar growth at all times , we decided in favor of angular momentum change in the inner and full disks . figure [ djmhmd_in ] displays the average rate of angular momentum change , @xmath139 , in the dynamical and secular phases of the bar evolution , within @xmath136 and @xmath75 respectively . the former choice of a fixed radius ( @xmath136 ) results from total @xmath100 being nearly unchanged before the buckling @xmath0 is exchanged predominantly between the inner and outer disks across @xmath1 at this stage . we have refrained from using @xmath1 as an inner / outer disk separator here because its outward motion brings in fresh , high @xmath98 material which ` contaminates ' the @xmath0 transfer due to the resonant and non - resonant interactions between the inner and outer disks . the latter choice of @xmath75 in the secular phase results from @xmath102 being nearly constant @xmath0 is mostly exchanged between the disk and the halo . figure [ djmhmd_in ] shows that clear correlations of @xmath139 with @xmath133 persist over the hubble time more concentrated halos provide an increasingly efficient damping of the bar instability by reducing the rate of @xmath0 transfer to the outer disk ( upper frame ) . they do facilitate such a transfer in the secular phase ( lower frame ) . significantly , these correlations extend down at least to @xmath142 the disk - dominated models . we note that t16 exhibits a qualitatively and quantitatively different behavior and was not included in the attempted linear fit to the secular evolution ( lower frame ) . we have discussed this model in 3.4 . the explanation for such a dual role of a dm halo in the bar evolution lies in determining what serves as a sink of @xmath102 from the inner disk . as shown in 3.6 , the outer disk beyond @xmath1 absorbs nearly all of @xmath0 during the bar instability . a more concentrated halo during this phase will dilute the disk gravity and , therefore , will act against the resonant orbit coupling ( i.e. , resonant torques ) between the inner and outer disks . on the other hand , during the secular phase of the bar instability , after the first buckling , the halo serves as the sink of @xmath0 from the disk . in this case , if increase of the dm mass density leads to a concurrent increase in the dm phase space density near the resonances , it facilitates the bar - halo resonance coupling , unless counterbalanced by a ` hotter ' halo . the @xmath135 rates appear similar during the dynamical and secular phases . but the duration of each phase can differ considerably . therefore , one expects that the amount of @xmath0 acquired by the halo will be much more substantial during the secular phase compared to the angular momentum lost by the inner disk ( within a fixed radius ! ) during the dynamical phase . it is known that the dm halo affects the bar evolution . in particular , it was shown that the dm halo concentration anti - correlates with the bar growth time . athanassoula ( 2003 ) has also demonstrated the ( anti)-correlation between @xmath143 and @xmath38 . our figure [ djmhmd_in ] quantifies the dual role played by the dm halo during the bar evolution and ties this explicitly to the rate of the angular momentum transfer within the disk halo dynamical system . at @xmath137 ( @xmath144 gyr ) , as a function of the halo - to - disk mass ratio within @xmath136 at @xmath37 . two models do not reach the saturation and are shown as lower limits : m40 ( left ) and t16 ( right ) . the green circles represent the two hybrid models . ] amplitude on the ratio of vertical - to - radial velocity dispersions , @xmath145 . t16 is excluded as it does not buckle . the green circles represent the two hybrid models . these velocity dispersions have been measured in a ring defined by @xmath146 and @xmath147 . ] , the average growth rate of the bar , between @xmath148 and @xmath137 ( 10 gyr ) , vs @xmath149 , the ratio of the radial dispersion in the disk to the circular velocity at @xmath136 . the latter ratio is calculated at @xmath148 . t16 is omitted as it does not buckle . the green circles represent the two hybrid models . ] stronger bars are expected to be more efficient in redistribution of the angular momentum in the disk - halo systems . in our _ collisionless _ models , bars strengthen monotonically only prior to the first peak and after the first buckling . despite the overall non - monotonic behavior of the bar , its final strength correlates with the amount of @xmath150@xmath100 lost by the disk and acquired by the halo ( fig . [ a2vsdj ] ) , in agreement with athanassoula ( 2003 ) . for the same reason , we expect and obtain correlation between @xmath38 at @xmath137 ( @xmath144 gyr ) , and @xmath133 ratio within @xmath136 , i.e. , the halo mass concentration ( fig . [ a2_hmassin ] ) , thus confirming athanassoula ( 2003 ) claim . by this time , nearly all modeled bar strengths reach saturation ( e.g. , fig . [ rb_a2 ] ) , except in two models , m40 and t16 , shown as lower limits . as shown in 3.3 , @xmath38 and @xmath41 saturate shortly after @xmath1/@xmath77 . these models represent two extreme trends in the halo mass concentration . t16 has the most concentrated halo , does not buckle and its secular evolution differs from other bars . m40 represents the least concentrated halo with a fractionally most massive disk ( figs . [ rhomods ] and [ vcir3 ] ) . it has the shortest exponential growth for the bar , @xmath151 . therefore , it is puzzling why its bar strength does not saturate . while m40 exhibits buckling and subsequent drop in @xmath38 and @xmath41 , its secular growth is the slowest of all models and comparable to that of t16 . its bar is the longest of all models as the ansae do not disappear . furthermore , its @xmath1 lies always within the disk , while @xmath1/@xmath43@xmath152 and remains flat for a hubble time . all these properties appear unique in our models . while the behavior of t16 is partially understandable , that of m40 requires more explanation . it is difficult to explain within the context of the halo mass concentration only , and seems to require additional parameter . we invoke the disk radial dispersion velocity , @xmath153 , which in a way also governs the bar vertical buckling instability through its ratio to the vertical velocity dispersion , @xmath154 . we test this idea by first plotting the amplitude of buckling , @xmath86 , vs this ratio calculated just before the first buckling ( fig . [ a1_sigzsigr ] ) . indeed , m40 has the smallest @xmath155 among our models , and , therefore , is expected and , in fact , displays the largest buckling amplitude @xmath86 ( fig . [ rb ] ) . a side effect of this buckling is the overall heating of the disk . consequently , the disk will be less susceptible to the bar growth in the secular phase . [ a2deriv_sig_vc ] shows the growthrate of the bar strength averaged over the secular phase , @xmath156 , i.e. , between the buckling time , @xmath148 , and the saturation time , @xmath140 ( @xmath144 gyr ) , vs @xmath157 measured at @xmath148 ( i.e. , at the minimum of @xmath38 ) . a clear correlation exists between these parameters ` hotter ' disk impairs the bar growth in the secular phase of its evolution . note that m40 has the hottest disk in this stage , which would explain its anomalously slow bar growth . the same correlation against @xmath153 looks weaker and shows much larger dispersion of individual models . as mentioned before , in each separate sequence , the @xmath86 peak is stronger for models progressively less dominated by the halo . it is interesting that an anticorrelation between the initial central density of the halo and the strength of the first @xmath86 peak holds for our models altogether , regardless of which halo parameter is varied ( figure [ a1_hdens ] ) . moreover , the anticorrelation exists between the strength of the first @xmath86 peak and the ratio of the vertical - to - radial dispersion velocities , shown in figure [ a1_sigzsigr ] . to summarize , we confirm the previous results and quantify for the first time the dual role that the dm halos play in stellar bar evolution ( fig . 15 ) : more centrally concentrated halos slowdown dynamical processes in the disk , such as spontaneous bar instability and vertical buckling instability , as well as angular momentum redistribution in the system . they reverse this trend and facilitate the angular momentum transfer during the secular stage of bar evolution , following the buckling . we follow the angular transfer in the disk - halo systems by varying one basic parameter at the time in order to identify the sites and times of intense angular momentum flows . while we confirm the earlier works which have identified the ilr and corotation resonances as being primarily responsible for @xmath0 emission by the disk and absorption by the halo , we also find few caveats . first , the total angular momentum in the disk is largely conserved before the first buckling , thus @xmath0 flows across the corotation to the outer disk , although the amount is relatively low compared to the subsequent exchange . second , during the secular stage of the bar evolution , the angular momentum within the corotation resonance is largely conserved : the loss of @xmath0 by this region due to the gravitational torques on the bar ( resonant and non - resonant ) is compensated by the influx of new material , rich in @xmath0 , across the corotation , due to its outward motion . we also elaborate under what conditions this ` conspiracy ' law operates . in view of this , the ratio of @xmath1/@xmath75 emerges as important dynamic discriminator between various paths in barred disk evolution its value can be determined if bar pattern speeds are known . third , we find that in some models the bar pattern speed stalls becoming nearly constant for prolonged time periods . all stellar bars which show this behavior have their corotation lying outside the disk . the disk - halo angular momentum exchange nearly vanishes and the halos display a weak _ emission _ of @xmath0 at the corotation and _ absorption _ at the ilr . while otherwise the bar pattern speeds are generally strongly decaying over the secular timescale , the addition of the gas component can modify this trend and soften the bar braking . we postpone our conclusions on this issue to paper ii . we also find that @xmath1/@xmath43 ratio stays within 11.4 range , only occasionally spiking above it to @xmath82 . this behavior is typical after the first buckling and as long as @xmath1/@xmath158 . lastly , we confirm some known correlations ( figs . 16 and 17 ) between the basic parameters of the disk - halo system , e.g. , between the final bar strength and @xmath133 ratio . model m40 with the least concentrated halo does not follow this trend and we test the possibility that an additional parameter plays the role in this , i.e. , the dispersion velocities ratio @xmath159 . because m40 experiences the strongest buckling , the disk is heated up , which impairs the subsequent bar growth and can explain the behavior of m40 in the previous correlation . figs . 18 , 19 and 20 display new correlations between various parameters . the bar average growthrate over its secular evolution time decreases with increasing dispersion velocities in the disk , @xmath160 . we also show that in a closely associated relation the angular momentum lost by the disk , @xmath161 , over the evolution correlates with the final bar strength . finally , the amplitude of the first buckling depends on the central density in the dm halo . these correlations will be followed up when the gas component is present . we expect substantial modifications in the relations between various basic parameters in the disk - halo system . we are grateful to ingo berentzen for comments on the manuscript , and to ron buta for answering our questions . we also thank ingo berentzen and inma martinez - valpuesta for help with numerical issues . this research has been partially supported by nasa / ltsa / atp , nsf and stsci grants to i.s . stsci is operated by the aura , inc . , under nasa contract nas 5 - 26555 . i.s . is grateful to the jila fellows for support .

we present low - resolution mmt hectospec spectroscopy of 594 candidate monoceros stream member stars . based on strong color - magnitude diagram overdensities , we targeted three fields within the stream s footprint , with and . by comparing the measured iron abundances with those expected from smooth galactic components alone , we measure , for the first time , the spectroscopic metallicity distribution function for monoceros . we find the stream to be chemically distinct from both the thick disk and halo , with}=-1 $ ] , and do not detect a trend in the stream s metallicity with galactic longitude . passing from= to= , the median monoceros metallicity trends upward by 0.1 dex , though uncertainties in modeling sample contamination by the disk and halo make this a marginal detection . in each field , we find monoceros to have an intrinsic [ fe / h ] dispersion of- dex . from the caiik line , we measure [ ca / fe ] for a subsample of metal poor program stars with}<-0.5 $ ] . in two of three fields , we find calcium deficiencies qualitatively similar to previously reported [ ti / fe ] underabundances in monoceros and the sagittarius tidal stream . further , using 90 spectra of thick disk stars in the monoceros pointings with , we detect a 0.22 dex north / south metallicity asymmetry coincident with known stellar density asymmetry at kpc and kpc . our median monoceros } = -1.0 $ ] and its relatively low dispersion naturally fit the expectation for an appropriately luminous dwarf galaxy progenitor .

1205.0807

the monoceros stream ( * ? ? ? * ; * ? ? ? * hereafter y03 ) comprises a ring - like stellar overdensity in the plane of the galactic disk . between @xmath13 , the kinematically cold ( y03 ) stream covers @xmath14 at galactocentric radius r@xmath1517 - 19 kpc with radial thickness @xmath16 kpc @xcite . the true nature of the structure is still uncertain , with the leading scenarios being : a ) remnant of an accretion event @xcite , with the canis major dwarf galaxy @xcite as a possible progenitor , b ) disturbance due to a high - eccentricity flyby encounter @xcite , and c ) disk flare or warp ( e.g. , * ? ? ? * ) discerning between these scenarios , especially the first two , may have important theoretical consequences as mergers with orbits and mass ratios implied by @xcite are deemed unlikely in current @xmath17cdm models ( e.g. , * ? ? ? * ) . in characterizing any stellar population , metallicity is considered a fundamental parameter . unfortunately , the literature regarding the chemical composition of the monoceros stream is sparse and seemingly inconsistent . for example , y03 estimate @xmath2 } \sim -1.6 $ ] at @xmath18 , @xmath19 using the caiik line and colors of turnoff stars . on the other end , @xcite ( henceforth c03 ) measure a mean of @xmath20 using m giants , sampled mainly from observations of the northern component of the stream . most recently , using photometric metallicities , @xcite ( hereafter i08 ) find the msto stars in the northern anticenter portion of the stream to have a median metallicity of @xmath2 } = -0.95 $ ] , with intrinsic rms scatter of only 0.15 dex . hypotheses invoked to reconcile these values range from multiple stellar populations ( e.g. , * ? ? ? * ) , to systematic errors in calibrations ( c03 ) , and to questioning whether the northern and southern part of the ring share the same origin @xcite . also , except for the i08 photometric study , the full metallicity distribution function ( mdf ) of stars in the monoceros stream has actually never been measured from spectroscopy . to better characterize the chemical composition of monoceros , we have obtained spectra of @xmath15@xmath21 candidate member stars . this large sample size allows us to probe the monoceros mdf with good statistics . also , by pointing in multiple directions along the stream , we can search for metallicity gradients . specifically , we set out to characterize the [ fe / h ] mdf of monoceros in each of three fields with @xmath22 and @xmath23 ( see table [ obs ] for field details and naming conventions ) . additionally , we aim to gauge the @xmath24-enhancement of monoceros from a [ ca / fe ] measurement . this allows us to address the origin of monoceros by comparing its [ ca / fe ] to the differing @xmath24-abundance patterns found in the milky way versus dwarf spheroidal ( dsph ) galaxies . rrrrrrrrrrr + monoceros & 187.245@xmath25 & 24.401@xmath25 & 115.741@xmath25 & 32.662@xmath25 & 2@xmath267200 & 199 & 49 & 19.55 & 3.871 & 0.12 + thick disk & 187.245@xmath25 & 24.401@xmath25 & 115.741@xmath25 & 32.662@xmath25 & 2@xmath267200 & 30 & 123 & 17.62 & 1.603 & 0.12 + + monoceros & 202.884@xmath25 & @xmath2724.183@xmath25 & 76.908@xmath25 & @xmath272.587@xmath25 & 2@xmath267200 & 146 & 47 & 19.83 & 4.911 & 0.30 + thick disk & 202.884@xmath25 & @xmath2724.183@xmath25 & 76.908@xmath25 & @xmath272.587@xmath25 & 2@xmath267200 & 60 & 137 & 17.44 & 1.788 & 0.30 + + monoceros & 177.879@xmath25 & 18.055@xmath25 & 104.834@xmath25 & 38.852@xmath25 & 2@xmath267200,5400 & 249 & 84 & 19.08 & 2.688 & 0.29 + + calibration & 58.999@xmath25 & 40.917@xmath25 & 250.416@xmath25 & 36.454@xmath25 & 3600 & 189 & 39 & 18.34 & 5.028 & 0.05 + recognized as a stellar overdensity between @xmath28 for @xmath13 , monoceros appears most strikingly in the number counts of msto f / g dwarfs at galactocentric distances r@xmath1515 - 20 kpc . near the galactic anticenter , these stars correspond to apparent magnitude @xmath29-@xmath30 and color @xmath31-@xmath32 , with @xmath33 referring to sdss photometric passbands . by examining ( @xmath34 , @xmath35 ) color - magnitude diagrams ( cmds ) of objects classified by the sdss ( dr7 ) as stars within the joint monoceros / sdss footprint , we identified many degree - scale pointings in which the monoceros msto was obviously visible above the galactic background ( for background estimation details see @xmath36[sec : decomp ] ) . in the best cases , the relevant cmd region of such pointings exhibited a factor of @xmath15@xmath37 overdensity , with total projected density @xmath15@xmath38@xmath27@xmath39 stars / deg@xmath40 . this field size and target density matches that of mmt / hectospec , which , with its large 6.5 m mirror size , can obtain at @xmath41 s / n sufficient to measure [ fe / h ] with integration times of @xmath154 hours . hoping to gauge both @xmath42 and north / south gradients in the stream s metallicity , six such fields were initially targeted for mmt / hectospec spectroscopy , covering @xmath43 in the galactic north and sampling four galactic latitudes with @xmath44 . in several of these fields , a relatively small subsample of much brighter thick disk targets were chosen from a similar range of msto @xmath35 colors and with @xmath45 . the observations were carried out in mmt queue - scheduled multi - object observing mode . due to a combination of factors including weather , only three fields were observed , spanning a limited range of ( @xmath42 , @xmath3 ) , though still including both galactic north and south . table [ obs ] summarizes the targets / observations , and figure [ cmd ] shows the cmd of each field , indicating targeted regions . hectospec can observe up to 300 objects at once . in general , we typically devoted @xmath1550 fibers to measuring the sky , with the remainder targeting monoceros candidates subject to the constraints of the instrument ( e.g. , the availability of the targets in the desired magnitude range , and limits on how closely spaced the fibers could get ) . any remaining fibers were set to target the thick disk stars , to investigate the thick disk asymmetry present towards the galactic anti - center ( see @xmath36[thickdisk ] ) . we used the 270 gpm grating yielding a resolving power of @xmath46,000 near @xmath47 . the low resolution warrants full wavelength coverage of @xmath48-@xmath49 . the data were reduced with customized iraf procedures as part of the sao telescope data center service @xcite . the s / n ratio of the final spectra is typically @xmath1550 - 80 per pixel at @xmath50 for the monoceros stars and @xmath15135 for the brighter thick disk stars . we used the iraf / rvsao routine ` xcsao ` to measure the radial velocity ( @xmath51 ) of each star from cross - correlation with template stellar spectra on a per - exposure basis , correcting for heliocentric velocity before stacking the exposures of each object . typical errors on @xmath51 ranged from 10 - 20 km s@xmath52 , varying based on s / n . abundance extraction requires estimates of several stellar parameters of each object , namely : the surface gravity log@xmath34 , temperature @xmath53 , microturbulence @xmath54 , and metallicity . as our targets were selected from sdss imaging , each star has @xmath33 photometry available . using i08 equation ( 3 ) , which gives @xmath53 as a function of @xmath35 , we determined effective temperatures for all program stars . the statistical error per star due to photometric errors is @xmath15@xmath55k for thick disk targets and @xmath15@xmath56k per star for the fainter monoceros candidates . @xmath53 generally ranged from 5800 - 6300k within each sample of monoceros / thick disk stars . our targets were selected to be msto f / g dwarfs . based on the stellar parameters of 46 dwarfs from @xcite with @xmath57 } \le 0.0 $ ] and @xmath58 k@xmath596300 k , we adopt @xmath60 km s@xmath52 and log@xmath61 for all monoceros candidates and thick disk stars . in @xmath36[sec : err ] we discuss the impact of these assumptions . of course , we wish to derive rather than assume metallicities . to extract abundances , we first determine [ fe / h ] from iron absorption features assuming [ x / fe]=0 for all x , choosing features where this assumption has negligible effect ( see figure [ fe_plot ] and @xmath36[sec : err ] ) . the [ fe / h ] value of each object is then used as the metallicity input for subsequent ca abundance measurements . to infer iron abundances from our spectra , we focus on the iron line complex near @xmath62 . the absorption equivalent width ( ew ) in the region with @xmath63 is dominated by fe , mainly fei , but with some feii . the primary fe lines of interest and major contaminating lines due to other elements are labeled in figure [ fe_plot ] . all abundances , including those drawn from the literature for calibration , were standardized with respect to the current solar values of @xcite . given our relatively low resolution and moderate s / n , continuum normalization is a sensitive matter . we base our continuum fitting procedure on that used by the segue stellar parameter pipeline ( sspp , * ? ? ? * ) to measure [ fe / h ] , since our spectra have similar s / n and resolution to those for which their pipeline was designed . in particular , we iteratively fit the region of the spectrum blueward of @xmath64 with a ninth order polynomial , masking the strong balmer lines . the sspp additionally computes abundances with local continuum estimates based on sidebands . however , our resolution is a factor of @xmath152 lower than that of sspp spectra , making it unfeasible to isolate sideband regions which attain the continuum and are sufficiently wide to be robust . we take the calculated @xmath53 and assumed log@xmath34 , @xmath54 for each star and compare the observed spectrum to a library of synthetic spectra generated based on these fixed stellar parameters , but with varying [ fe / h ] . in particular , synthetic , continuum normalized spectra were generated using moog ( the latest version , 2010 ; * ? ? ? * ) , with one - dimensional no - overshoot model atmospheres ( assuming local thermodynamic equilibrium ) from @xcite . this version of moog accounts for the fact that rayleigh scattering becomes an important source of continuum opacity at short wavelengths blueward of 4500 @xcite . for each [ fe / h ] value with @xmath65 } < 0.5 $ ] in gradations of 0.01 dex , we compute the @xmath66 between the synthetic spectrum and the observed , normalized spectrum over the fe absorption region . we assign each star the [ fe / h ] corresponding to the minimum @xmath66 value thus obtained . the @xmath66 is computed with the variance taken from poisson statistics to be @xmath67 , where @xmath68 is the counts per pixel in adu , @xmath69 adu is the readnoise and @xmath70 is the gain . we compute the @xmath66 within the range @xmath71 , a region determined to optimize the fe absorption relative to absorption by other elements and random noise . an example best - fit result for typical s / n and iron absorption strength is plotted in figure [ fe_plot ] . the line list employed in our syntheses is provided in table [ linelist ] . to test our [ fe / h ] fitting procedure , we apply the above methodology to 189 mmt / hectospec spectra of globular cluster stars in m13 with comparable s / n and sdss photometry . we again estimate @xmath53 from @xmath35 and adopt @xmath72 km s@xmath52 , log@xmath34 = 4.25 based on @xcite table 2 . m13 is known to have @xmath2}=-1.55 $ ] @xcite , and we recover a median [ fe / h ] of @xmath73 . we thus apply a @xmath74 dex correction factor to all of our [ fe / h ] measurements . based on visual inspection of the faintest and apparently most metal poor objects in our monoceros candidate sample , it became clear that several of these lacked sufficient iron ew to measure [ fe / h ] in the presence of statistical noise and minor continuum placement errors . to remove these outliers , we computed @xmath75 , the @xmath76 error in total fe fitting region ew due to poisson noise and continuum placement errors . the continuum error was estimated by fitting a polynomial two orders lower than the default continuum , and generally amounted to @xmath77@xmath78 continuum level difference . we then rejected all stars for which the s / n and ew were sufficiently low that @xmath75 translated to @xmath790.5 dex [ fe / h ] uncertainty . for monoceros targets in fields north , south , north18 the percentage of stars rejected was 7.5% , 6.8% and 0.4% respectively . all rejected stars had estimated @xmath2 } < -2.05 $ ] . no thick disk [ fe / h ] estimates were rejected , as these stars generally have higher s / n and metallicity than the monoceros candidates . in performing maximum likelihood fits to our [ fe / h ] distributions ( see @xmath36[sec : decomp ] ) we also identified , among monoceros candidates , three [ fe / h ] outliers at @xmath2 } > 0 $ ] . these objects also displayed unreasonably low @xmath80 } < -1 $ ] , again suggesting a dramatic overestimate of [ fe / h ] ; these three objects were also excluded from the sample . there are several uncertainties on our derived metallicities . first there is statistical scatter from poisson noise , which , for typical s / n is a function of [ fe / h ] . we gauge this error with monte carlo generation of spectra to which we have added random poisson noise . for typical @xmath2 } = -1 $ ] , we find an rms scatter of @xmath81 dex . at typical halo @xmath2 } = -1.6 $ ] we find a scatter of 0.25 dex , while for typical thick disk @xmath2}=-0.7 $ ] we find an rms scatter of only 0.05 dex . with sample sizes of 100@xmath27250 stars per field , this statistical error is thus not a primary concern . another source of uncertainty stems from our assumptions about stellar parameters . we find the effects of reasonably large systematic over / underestimates in each of log@xmath34 ( @xmath82 dex ) , @xmath53 ( @xmath83 k ) , and @xmath54 ( @xmath82 dex ) , to be @xmath84 dex , @xmath85 dex , @xmath85 dex in [ fe / h ] respectively . summing these effects in quadrature yields an uncertainty of @xmath86 dex on our derived metallicities . a further possible source of bias in [ fe / h ] is contamination of the iron abundance fitting region by lines of other elements , most notably ca ( see figure [ fe_plot ] ) . in determining [ fe / h ] , we generated synthetic spectra assuming that [ x / fe]=0 for all x. while [ x / fe ] is not known for monoceros , the most relevant parameter of [ ca / fe ] for the galactic disk / halo tends to lie between 0.0 and 0.4 . we address the resulting [ fe / h ] bias by using moog to simulate populations of stars with [ ca / fe ] following the known galactic trend with [ fe / h ] ( see @xmath36[ca_cont ] ) , adding appropriate noise to the synthetic spectra before feeding them back into our [ fe / h ] fitter . the resulting bias is very small , @xmath870.02 dex . a final important source of uncertainty is the continuum normalization procedure . with our low resolution , the blending of lines means the continuum is rarely achieved , and could potentially be systematically over / underestimated throughout the sample , thereby raising / lowering all metallicities . to test the systematic effect due to differing continuum normalizations , we again re - fit abundances with new continua parameterized by a polynomial two orders lower than the default continuum . we find in general a resulting systematic offset of 0.07 dex in [ fe / h ] . we assign a final systematic uncertainty of 0.13 dex to our [ fe / h ] distributions , adding the errors from continuum fitting and stellar parameters in quadrature . this value is somewhat conservative , as much of this uncertainty ( particularly the continuum normalization ) has been calibrated out with the m13 data . we find , through simulations similar to those previously described , that the observed [ fe / h ] scatter is well described by a combination of both poisson noise and statistical scatter of the underlying stellar parameters , again at the level of @xmath88k in @xmath53 , @xmath89 dex in both @xmath54 and log@xmath34 . thus , summing these sources of scatter in quadrature , we find for typical s / n and @xmath2 } = -1 $ ] a single - star statistical error of 0.16 dex . to characterize the monoceros mdf , we must first decompose the distribution of reliable [ fe / h ] estimates in each field into galactic components ( halo , disk ) and monoceros members . to estimate the expected background in both halo and disk stars , we simulate cmds for each pointing with both the besanon model @xcite and _ galfast _ , using the best fit galactic parameters of @xcite . all simulations included realistic photometry errors and were extinction free , for comparison to observed sdss cmds dereddened according to @xcite , henceforth sfd . we created a matched filter in target density for the stars with reliable [ fe / h ] in each field , then integrated against the observed , besanon and _ galfast _ star counts to determine the expected number of contaminating stars from the disk and halo . for each field , overall cmd normalization constants close to unity were applied by requiring agreement between the total number of stars redder than @xmath90 and falling within the targeted range of @xmath91 values . this analysis suggests @xmath1565% of monoceros candidates with well - measured [ fe / h ] are true stream members . in no case do we expect any thin disk stars , as the monoceros targets have @xmath92 kpc . figure [ decomp ] shows a decomposition of the observed distributions of reliable [ fe / h ] for each field . in principle , there are nine free parameters : the normalization of monoceros , thick disk , and halo components , their median metallicities and their dispersions . the following subsections describe our assumptions for the thick disk and halo distributions , and procedure for fitting the monoceros component . in each field , the thick disk normalization was fixed by the cmd analysis to be the average of the predictions from the _ galfast _ and besanon simulations . in fields north , south we fix the median metallicity to that of the thick disk targets in these fields , plus a small @xmath93 dex correction from @xcite equation ( a2 ) to account for the vertical disk metallicity gradient ( our thick disk targets are at lower @xmath94 than the more distant thick disk stars which contaminate the monoceros sample ) . for field north18 , in which we did not target any thick disk stars , the median [ fe / h ] used for field north was adopted . a @xmath95 dex intrinsic dispersion calculated from the kinematically selected thick disk stars in @xcite was adopted for the thick disk in all fields . in fields north , north18 , the halo normalization was fixed to the average of the _ galfast _ and besanon predictions . a cut on @xmath96 km s@xmath52 , yielding a subsample which should be dominated by large velocity dispersion halo members , suggests that the absolute number of halo stars in each field is roughly constant . however , in field south , _ galfast _ gives an unreasonably large halo contribution relative to this expectation , so in that case we simply use the besanon value . for field north18 , the halo median / dispersion were fit by examining the top quartile of stars in s / n ( 62 stars , s / n@xmath97 ) , which displayed a relatively narrow peak at @xmath2 } = -1.57 $ ] . for field north , we used sdss proper motions from @xcite to isolate 7 halo stars with the cut @xmath98 km s@xmath52 and @xmath99 km s@xmath52 . this yielded a median metallicity of @xmath271.80 . for field south , the kinematic data were of lower quality , and we adopt @xmath2 } = -1.6 $ ] from @xcite . for fields north , south the dispersion was calculated by adding the errors due to poisson noise and stellar parameter scatter in quadrature with a 0.3 dex intrinsic halo [ fe / h ] dispersion ( i08 ) . the normalization of the monoceros distribution is fixed by requiring that the components sum to the total number of targets . this leaves two free parameters characterizing the monoceros mdf : its median and dispersion . all components were assumed to be intrinsically gaussian . to account for detection inefficiency at very low [ fe / h ] , we modulated the distributions by a linear ramp defined to be zero at ( and below ) the minimum detected [ fe / h ] and unity at ( and above ) the maximum rejected [ fe / h ] . for each field , we fit the monoceros [ fe / h ] median @xmath100 and dispersion @xmath101 by maximizing the likelihood : @xmath102 \it _ i;\mu_{mon},\sigma_{mon})\}\ ] ] where @xmath103 is the total number of reliable [ fe / h ] measurements in each field and @xmath104)$ ] is the normalized sum of the halo , thick disk and trial monoceros modulated gaussian distributions . table [ dist ] lists the fitting assumptions and results , which agree with expectations based on visual inspection . we have also measured [ ca / h ] using the caiik line for program stars sufficiently metal poor to avoid saturation effects ( @xmath2 } < -0.5 $ ] ) . our [ ca / h ] determination follows a procedure very similar to that described for [ fe / h ] in @xmath36[fe_fit ] . for each star , we adopt the same stellar parameters as in the [ fe / h ] fits , but now input the measured [ fe / h ] as the kurucz atmosphere metallicity . to determine each star s [ ca / h ] value , we minimize the same @xmath66 statistic , this time for @xmath105 . the line list for caiik syntheses is provided in table [ linelist ] . for the ca fitting region , we used a local continuum normalization procedure , iteratively fitting a second order polynomial to the region @xmath106 . we found that this continuum essentially matched the values / slopes of the continuum on the two `` sideband '' regions blueward / redward of the caii h / k absorption . this fits the continuum shape well , but underestimates the normalization by neglecting absorption which causes the continuum to be rarely if ever realized . for this reason , these continuum fits caused [ ca / h ] to be systematically underestimated , with the effect becoming more pronounced with increasing [ fe / h ] . we recalibrate the continuum by requiring that our thick disk sample exhibit a trend of [ ca / fe ] vs. [ fe / h ] matching that from the literature . the expected trend was computed using 150 milky way field stars from @xcite . we found good agreement in slope could be achieved by mulitplying the initial continua by a factor of ( @xmath107 ) , with @xmath108 increasing linearly from zero at low metallicity , @xmath2}_{\rm m13 } = -1.55 $ ] , to 13% at @xmath2 } = -0.50 $ ] . agreement in offset was achieved with subsequent addition of 0.03 dex to all [ ca / fe ] . applying this procedure to our 189 m13 spectra yielded @xmath80 } = 0.25 $ ] , in agreement with the estimate of @xcite , who find @xmath80 } = 0.19\pm0.07 $ ] from high - resolution spectroscopy of 25 cluster members . we apply the local continuum fit and subsequent recalibration to monoceros candidates with @xmath109 } < -0.5 $ ] , where we had sufficient disk stars to verify the recalibration procedure . figure [ ca_comparison ] overplots these monoceros results with the literature trend and recalibrated disk , where the lines are constructed by taking the median of [ ca / fe ] in bins 0.2 dex wide in [ fe / h ] . also plotted are linear fits to recent [ ti / fe ] abundances of 70 sagittarius and 21 monoceros stream members @xcite . the @xmath24-enhancements [ ti / fe ] and [ ca / fe ] are expected to display similar trends , as both reflect the relative heavy element contributions of type ii ( @xmath24-elements@xmath110fe ) vs. type ia supernovae ( fe only ) , and hence probe the star - formation history . dsph galaxies typically show deficiencies in @xmath24-elements relative to the milky way , owing to a slower star formation rate in which fewer type ii explosions occur before the onset of type ia explosions at @xmath1111 gyr . cccccccccccc north & 183 & 40.8 & 23.2 & 119.2 & @xmath112 & 0.35 & @xmath271.80 & 0.42 & @xmath113 & @xmath114 & @xmath115 + south & 131 & 33.3 & 13.0 & 84.7 & @xmath116 & 0.35 & @xmath271.60 & 0.42 & @xmath117 & @xmath118 & @xmath119 + north18 & 247 & 28.7 & 56.4 & 161.8 & @xmath112 & 0.35 & @xmath271.57 & 0.25 & @xmath120 & @xmath121 & @xmath122 as the monoceros members comprise @xmath152/3 of targets in each field , it can be seen by visual inspection that monoceros has a median @xmath2 } \sim -1 $ ] , distinct from both the thick disk and halo . indeed , our maximum likelihood fits yield @xmath2 } \sim -1.0 $ ] for each subsample , showing no trend with @xmath42 ( see table [ dist ] ) . i08 analyzed @xmath15@xmath123 msto stars in a similar spatial region to that of our targets ( @xmath124/kpc @xmath125 , @xmath126/kpc @xmath127 , @xmath128 , see their figure 18 ) ; our inferred monoceros medians @xmath129 } < -0.98 $ ] are consistent with the i08 photometric estimate of @xmath2 } = -0.95 $ ] to within the systematic uncertainty of @xmath150.13 dex on our spectroscopic abundances . it should be noted that we have used the same @xmath35 temperature calibration as i08 . a more stringent test of sdss photometric metallicities would be provided by a study deriving @xmath53 independently . see @xmath36[sec : phot ] for a detailed comparison of photometric vs. spectroscopic metallicities . i08 also inferred a small intrinsic monoceros [ fe / h ] dispersion of only @xmath130 dex . for each field , we simulated the distributions of [ fe / h ] arising from a monometallic population at @xmath2 } = \mu_{mon}$ ] , given our s / n distribution , @xmath53 distribution , continuum placement errors and expected stellar parameter scatter . we found rms measurement - induced scatter of 0.21 dex , 0.23 dex , and 0.16 dex for fields north , south , north18 , implying intrinsic scatters of 0.21 dex , 0.22 dex , 0.10 dex respectively . applying the same methodology to m13 stars of comparable s / n ( @xmath7929 ) , we find a measurement error of 0.36 dex , and observed scatter of 0.42 dex , yielding an intrinsic m13 scatter of 0.20 dex . globular clusters generally have rms iron abundance dispersion @xmath131 dex @xcite . however , with only a single exposure , the m13 sample is subject to additional sources of error ( e.g. , cosmic rays ) which we have not modeled , perhaps reconciling this difference . if we are indeed slightly underestimating the observation - induced scatter , a very conservative interpretation would regard the intrinsic monoceros dispersions we obtained as upper limits . for fields north , north18 , a single gaussian component is sufficient to describe the monoceros mdf . for field south , there appears to be an excess at @xmath2 } \sim -1.2 $ ] . this could arise from a multimodality in the monoceros mdf or , more likely , indicates contamination by the metal weak thick disk ( mwtd , see @xmath36[mwtd ] ) . inspection of the cmd in field south also suggests multiple main sequence turnoffs aside from that of the disk . given these uncertainties in population modeling , the inferred gradient in monoceros median metallicity with @xmath3 ( @xmath132 dex from @xmath133 to @xmath134 ) appears not to be significant . in two of three fields , the monoceros [ ca / fe ] dips below the galactic trend as [ fe / h ] increases from @xmath135 to @xmath136 ( see figure 3 ) . in fields north and south , this [ ca / fe ] trend is qualitatively similar to the recently measured [ ti / fe ] trends of monoceros and sagittarius stream members @xcite . in field north18 , the observed monoceros trend appears to exactly match that of the milky way . however , the field north18 decomposition in figure [ decomp ] suggests the measured [ fe / h ] distribution cuts off more sharply than expected at high metallicity . if indeed [ fe / h ] has been systematically underestimated for the highest metallicity stars in this field , [ ca / fe ] would be artificially inflated . it should be noted that our monoceros sample plotted is a mixture of monoceros and disk stars , driving down the apparent contrast between the monoceros and milky way trends . in addition to the monoceros candidates , in fields north and south we measured [ fe / h ] for 30 and 60 thick disk stars respectively . these fields exhibit a strong stellar number density asymmetry , with the southern field showing an excess of more than 50% compared to its northern counterpart . the asymmetry is easily detected in the sdss data , is also detectable in pan - starrs1 @xmath137 survey data , and can not be accounted for by photometric errors or reddening uncertainties . juri et al . ( in preparation ) speculate it may be a signature of yet another , more nearby , stream , or otherwise a global disturbance ( a `` warp '' or a `` bend '' ) in the thick disk , of uncertain origin . figure [ decomp ] shows the measured distributions ; the median and dispersion of [ fe / h ] in field south are @xmath270.65 , 0.29 dex respectively , while for field north , the corresponding values are @xmath270.87 , 0.22 dex . the statistical scatter in our [ fe / h ] measurments @xmath150.05 dex per star has negligible effect on these values . the median @xmath94 for the north , south thick disk targets is respectively 1.60 kpc and 1.79 kpc , using the photometric parallax relation of i08 equations ( a6 ) and ( a7 ) . according to the gradient in disk metallicity from @xcite this corresponds to an expectation , given perfect disk symmetry , that south would have [ fe / h ] 0.01 dex _ lower _ than north . the asymmetry is not simply reflecting different scale heights . although the adopted systematic error on our metallicities is nominally 0.13 dex , this represents an offset with respect to the true abundances ; our internal consistency should be much better , especially for these two thick disk samples which have very similar colors and s / n , as well as relatively strong fe features . reconciling the median metallicities of north / south thick disk sample would require that stellar parameters such as log@xmath34 and @xmath54 differ greatly between the two populations , which itself would require invoking disk asymmetry about the midplane . a kolmogorov - smirnov test between the two measured [ fe / h ] distributions yields a consistency probablility of @xmath77@xmath138 , even excluding the added 0.01 dex contrast from the @xmath94 gradient in [ fe / h ] . the dispersion in [ fe / h ] for field south is 0.22 dex , versus 0.29 dex for field north and 0.35 dex for the thick disk ( averaged over many directions ) according to @xcite . a disrupted dsph origin for the overdensity would naturally explain the lower [ fe / h ] dispersion in field south , but also incorrectly predict an [ fe / h ] lower than that of the northern disk for the southern stellar overdensity . thus , we have spectroscopically identified an asymmetry in the thick disk towards the galactic anticenter , at @xmath10 kpc and @xmath11 kpc , which we expect to be confirmed and characterized further via stellar counts analysis . with @xmath33 photometry available , we can compute [ fe / h]@xmath139 from equation ( a1 ) of @xcite for the entire sample . [ fe / h]@xmath139 is reddening - sensitive . @xcite have recalibrated sfd , finding the need for a substantial , @xmath1545% , global change in @xmath140 , but only a @xmath37% change in @xmath141 . thus , our @xmath53 are negligibly affected at a @xmath8710k level . but for fixed @xmath35 , and near @xmath2}_{phot } = -1 $ ] , [ fe / h]@xmath139 has a strong @xmath142 gradient , @xmath790.04 dex per 0.01 mag @xmath142 . the @xcite recalibration implies @xmath1430.06 in portions of our sample , a major effect . however , the [ fe / h]@xmath139 estimator of @xcite is tied to sfd reddenings , and we therefore calculate [ fe / h]@xmath139 using @xmath144 ; it may be necessary to revise the sdss [ fe / h]@xmath139 relation using updated @xmath140 values if these photometric metallicities are to be independent of dust column . furthermore , single - pointing samples of [ fe / h]@xmath139 and [ fe / h]@xmath145 are sensitive to local reddening in differing ways ( [ fe / h]@xmath145 is a function of @xmath35 reddening through @xmath53 ) . near [ fe / h ] = @xmath146 , using conventional reddening laws , overestimating @xmath147 leads to an underestimate of [ fe / h]@xmath139 , but to an overestimate of [ fe / h]@xmath145 for fixed absorption ew . any local ( degree scale ) error in @xmath147 drives our photometric and spectroscopic metallicities apart . nevertheless , we find a highly significant correlation between [ fe / h]@xmath139 and [ fe / h]@xmath145 . the correlation is strongest for the highest s / n spectra , corresponding to the brightest objects with least noisy @xmath148 photometry and best measured [ fe / h]@xmath145 . m13 , closely approximating a monometallic population , shows only a marginal correlation , in accordance with the expectation that the observed spread in [ fe / h ] values arises predominantly from random scatter in the photometric / spectroscopic measurements . figure [ comparison ] overplots the distributions of photometric / spectroscopic metallicities for each of our subsamples . in general , the median photometric and spectroscopic metallicities agree at the level of @xmath15@xmath130 dex , which can be reconciled by the estimated systematic uncertainties of 0.13 dex on [ fe / h]@xmath145 and 0.1 dex on [ fe / h]@xmath139 ( i08 ) . except for the m13 sample , the [ fe / h]@xmath145 distributions are narrower than their [ fe / h]@xmath139 counterparts , by a median of @xmath149 dex in dispersion . of course , much of the total dispersion is intrinsic . the superiority of m13 [ fe / h]@xmath139 owes to the targets being relatively bright ( @xmath150 ) , yielding good photometry , yet receiving only a single , short hectospec exposure . as a byproduct of our abundance extraction procedure , we have heliocentric radial velocities in hand for our entire sample . typical statistical errors are 10 - 20 km s@xmath52 , depending on s / n . we gauged our systematic error using m13 as a calibration sample , since this globular cluster has a known @xmath51 = @xmath27244.2 km s@xmath52 @xcite . we find for our 189 m13 spectra a median of @xmath51 = @xmath27241.1 km s@xmath52 , and therefore have applied a correction of @xmath151 km s@xmath52 to all radial velocity measurements . @xcite found that monoceros is kinematically cold , consistent with the expectation for a disrupted dsph galaxy , measuring radial velocity dispersions @xmath152 = 22 - 30 km s@xmath52 . for stars with @xmath153 [ fe / h ] @xmath154 in fields north , south , north18 , we find @xmath15535.7 , 31.3 , 26.3 km s@xmath52 after correcting for median measurement uncertainties of 17 , 24 , 11 km s@xmath52 respectively . for comparison , the _ galfast _ ( besanon ) predictions for @xmath152 of thick disk stars in these fields is 57.7 ( 38.4 ) , 56.6 ( 36.3 ) , 54.3 ( 35.8 ) km s@xmath52 . all dispersions were calculated as half of the difference between the 84th and 16th percentile @xmath156 values . in all cases , the metallicity - selected monoceros sample is kinematically cold relative to the thick disk model , and appreciably so if we to are adopt the _ galfast _ prediction . our range of @xmath152 is slightly higher than that found by @xcite , but this may be due to contamination of the sample by disk and halo stars , both of which would tend to increase @xmath152 . we can also compare the median @xmath51 for the metallicity - selected monoceros samples to the predictions of @xcite . figure [ rv_longitude ] overplots measured radial velocities with the three prograde orbit models of @xcite . figure [ rv_longitude ] also plots @xmath42 vs. @xmath3 for our monoceros targets and the @xcite orbits , showing that field south is on a different `` wrap '' of the stream than are north , north18.there is indeed qualitative agreement between all of the models and our measured median @xmath51 for north , south , north18 of 7 @xmath157 2.9 , 49 @xmath157 4.5 , @xmath2720 @xmath157 2.4 km s@xmath52 , trending upwards with increasing @xmath42 . quantitative agreement is only achieved at the @xmath1510 km s@xmath52 level , with no single model particularly favored , as shown by the histograms of the metallicity - selected monoceros @xmath51 measurements in each field ( figure [ rv_hist ] ) . the median monoceros @xmath51 values generally differ substantially from the _ galfast _ ( besanon ) thick disk @xmath51 predictions of 30 ( 22 ) , 54 ( 59 ) , 5 ( 6 ) km s@xmath52 for fields north , south , north18 respectively . in summary , the measured radial velocities of our metallicity - selected monoceros subsamples are kinematically cold relative to the thick disk , have different median values than expected for thick disk stars , and generally agree with the model predictions of @xcite . since , as discussed in @xmath36[thickdisk ] , we find that the thick disk south sample is chemically distinct from its northern counterpart , it is of interest to compare the measured thick disk radial velocities with _ galfast_/besanon predictions . for both the north and south thick disk samples , we find rather poor agreement in terms of median @xmath51 , with measured values of @xmath278 , 31 km s@xmath52 versus _ galfast _ ( besanon ) predicted values of 21 ( 18 ) , 52 ( 49 ) km s@xmath52 . the measured dispersions @xmath152 for north , south are 42 , 25 km s@xmath52 , versus _ galfast _ ( besanon ) predicted values of 46 ( 47 ) , 46 ( 49 ) km s@xmath52 . the measured velocity dispersion in field north agrees with the thick disk predictions , whereas the south targets are much more kinematically cold than expected . thus , the thick disk kinematics also display a north / south asymmetry , with a low velocity dispersion for the southern overdensity suggestive of a coherent stream . we have presented the first ever spectroscopic monoceros mdf based directly on fe absorption line measurements , finding @xmath2 } = -1.0\pm0.1 $ ] . our results confirm the photometric metallicity analyses of i08 ( @xmath2 } = -0.95 $ ] ) and @xcite , who also used the @xcite calibration to derive results consistent with @xmath2 } \sim -1.0 $ ] towards ( @xmath42,@xmath3)=(232@xmath25,26@xmath25 ) . however , these earlier results are vulnerable to systematic problems ( e.g. , @xmath158 band extinction uncertainties ) which are largely circumvented by our spectroscopic analysis . while c03 and y03 inferred [ fe / h ] for monoceros spectroscopically , neither measured fe absorption directly . instead , both relied upon indices calibrated to mg and ca features . our [ ca / fe ] results suggest that such calibrations of [ fe / h ] to standard @xmath24-element trends may not be justified . further , in the case of y03 , whose technique measured caiik ew , our analysis shows that ca and fe absorption in this region are highly degenerate , implying large uncertainties for the y03 procedure . there does not appear to be a simple means of bringing about agreement between our result @xmath2 } = -1 $ ] and that of y03 , given the similar location of their targets at ( @xmath42,@xmath3)=(@xmath159,@xmath160 ) , @xmath161 kpc and our field south at ( @xmath42,@xmath3)=(@xmath162,@xmath163 ) , @xmath164 kpc . to whatever extent the present results disagree with those of c03 and y03 , our results should take precedence , as we have measured [ fe / h ] directly , analyzing spectra with depth - of - exposure @xmath157@xmath26 that of y03 and s / n comparable to that of c03 . our sample size also eclipses that of c03 by a factor @xmath7910@xmath26 and that of y03 by a factor @xmath151.5@xmath26 . assuming the monoceros progenitor contained stellar populations of varying age , we can still reconcile our result with that of the c03 m - giant sample , @xmath2 } = -0.4\pm0.3 $ ] . m - giants preferentially trace younger , higher metallicity populations than msto stars . @xcite have noted that this bias reconciles their msto sgr stream @xmath2 } = -1.15 $ ] with literature m - giant values of @xmath2 } \sim -0.6 $ ] . the monoceros case appears directly analogous given our conclusion that monoceros has msto @xmath2 } = -1.0 $ ] . this reasoning can also bring about agreement between our @xmath2 } = -1 $ ] and the median @xmath2 } = -0.71 $ ] of the 21 kinematically - selected m - giants studied by @xcite . since we have measured [ fe / h ] directly from the fe absorption lines of msto stars ( a population with relatively little metallicity bias ) and with s / n and sample size comparable to or better than previous spectroscopic efforts , we recommend adoption of our value @xmath2 } = -1.0\pm0.1 $ ] as the metallicity of the monoceros stream . @xcite have speculated that the mwtd and monoceros may share a common origin . this notion is based in part on the coincidence between the preliminary @xcite estimate of monoceros metallicity , @xmath2 } = -1.37 $ ] , and their inferred @xmath165}\rangle_{mwtd } = -1.3 $ ] . our measurements argue against this mwtd - monoceros connection , as we find a @xmath150.3 dex offset in [ fe / h ] between these populations . all of our systematics are characteristically @xmath150.05 - 0.10 dex , and crudely speaking we rule out a median @xmath2 } = -1.3 $ ] at the 2 - 3@xmath166 level . decreasing our median metallicity to @xmath2}_{spec } = -1.3 $ ] would require , for example , an unreasonably large temperature scale miscalibration of @xmath15330k . much controversy and speculation has surrounded the origin of the monoceros structure since its discovery . to characterize the stream s nature , y03 used star counts to estimate the total stellar mass of monoceros at @xmath15@xmath167-@xmath168 , a range consistent with the content of a relatively luminous dwarf galaxy . @xcite specifically simulated the case of a dsph progenitor , fitting observations well with an initital progenitor of total mass 3 - 9@xmath169 . still , detractors ( e.g. , * ? ? ? * ) argue that there is no need for an extragalactic origin of monoceros . how can our observations further constrain the nature of the monoceros progenitor ? the trends of @xmath170[fe / h]@xmath171 and its dispersion , @xmath172}$ ] , with dwarf galaxy luminosity have been studied by @xcite , and are reproduced in figure [ dwarf ] . dwarf galaxy @xmath170[fe / h]@xmath171 increases monotonically with luminosity ; by fitting a linear trend in @xmath170[fe / h]@xmath171 vs. @xmath173 for @xmath174}\rangle < -0.9 $ ] , we can translate our mdf measurement @xmath165}\rangle \approx -1.0 \pm 0.1 $ ] into a constraint on the monoceros progenitor , @xmath175@xmath176@xmath173@xmath176@xmath177 , assuming the progenitor was a dwarf galaxy ( see the dashed line and blue rectangle in figure [ dwarf ] ) . using a typical stellar mass - to - light ratio for dwarf galaxies @xmath178@xmath1793 @xcite , @xmath165}\rangle=-1 $ ] implies @xmath180@xmath181 of stellar mass in the monoceros progenitor . provided the fraction of the monoceros progenitor which has been stripped is similar to that of sgr ( @xmath792/3 , * ? ? ? * ) , we conclude that our estimate of the stellar mass in the monoceros stream based on chemistry and the assumption of a dwarf galaxy origin is consistent with the independent y03 estimate from observed star counts . having constrained the progenitor luminosity , we can also place monoceros on the figure [ dwarf ] plot of @xmath173 vs. @xmath172}$ ] , finding very good qualititave agreement between monoceros ( blue rectangle ) and other bright , low metallicity dispersion dwarf galaxies . instead of deriving @xmath173 from @xmath170[fe / h]@xmath171 , we can check whether a dsph progenitor with best - fitting mass determined by @xcite conforms the @xcite mdf trends . taking the monoceros stream to have the same total mass - to - light ratio as the sgr core ( @xmath173=@xmath182 , @xmath183=@xmath184@xmath181 , * ; * ? ? ? * ) , the @xcite mass range translates to @xmath185@xmath77@xmath173@xmath77@xmath186 . we have combined this constraint with our measured mdf to plot the @xcite monoceros progenitor ( red rectangles ) in figure [ dwarf ] . in both cases , @xmath173 vs. @xmath170[fe / h]@xmath171 and @xmath173 vs. @xmath172}$ ] , monoceros shows qualitative agreement with the dwarf galaxy trend . thus , we conclude that our abundance study supports a dwarf galaxy origin of monoceros , in that : ( 1 ) the mdf median and dispersion both conform to the expectations for an appropriately luminous dwarf galaxy progenitor , and ( 2 ) we have detected [ ca / fe ] deficiencies typical of dwarf galaxy stellar populations in two of three fields observed . in the future , similar observations / analysis could be applied to monoceros fields that more optimally search for @xmath42 and @xmath3 abundance gradients in the stream . for example , fixing @xmath42 and observing two @xmath94 values symmetrically above / below the galactic midplane would better isolate monoceros abundance trends with @xmath3 . detecting or ruling out a gradient toward the midplane in this way would significantly constrain models which claim the stream to be a disk warp " or flare " ( e.g. , * ? ? ? observing an extended range of @xmath42 values at fixed @xmath3 would better constrain a metallicity gradient along the stream . for example , sgr exhibits such a gradient , but only at the level @xmath152.4@xmath187 dex per degree along the debris trail @xcite . to probe a gradient at this level with spectroscopy comparable to that reported here , observations would need to span @xmath15100@xmath25 in @xmath42 , a factor of four increase relative to our current analysis , @xmath188 . we warmly thank nelson caldwell for his advice on mmt observations , hectospec details and queue scheduling , as well as evan kirby and john norris for fruitful discussions . we furthermore thank the queue observers and the sao telescope data center for reducing the data . this research made extensive use of the vienna atomic line database ( vald ) . a. m. m. is supported by a national defense science & engineering graduate fellowship . a. f. is supported by a clay fellowship administered by the smithsonian astrophysical observatory . m. j. acknowledges support by nasa through hubble fellowship grant # hf-51255.01-a awarded by the space telescope science institute , which is operated by the association of universities for research in astronomy , inc . , for nasa , under contract nas 5 - 26555 . asplund , m. , grevesse , n. , sauval , a. j. & scott , p. 2009 , , 47 , 481 bensby , t. , zenn , a. r. , oey , m. s. , & feltzing , s. 2007 , , 663 , l13 bond , n. a. , ivezi , . , sesar , b. , et al . 2010 , , 716 , 1 briley , m. m. , & cohen , j. g. 2001 , , 122 , 242 carlin , j. l. , majewski , s. r. , casetti - dinescu , d. i. , et al . 2011 , arxiv:1111.0014 carollo , d. , beers , t. c. , chiba , m. , et al . 2010 , , 712 , 692 carretta , e. , bragaglia , a. , gratton , r. , d2019orazi , v. , & lucatello , s. 2009 , , 508 , 695 chilingarian , i. v. , mieske , s. , hilker , m. , & infante , l. 2011 , , 412 , 1627 chou , m .- y . , cunha , k. , majewski , s. r. , et al . 2010 , , 708 , 1290 chou , m .- y . , majewski , s. r. , cunha , k. , et al . 2010b , , 720 , l5 cohen , j. g. , & melndez , j. 2005 , , 129 , 303 conn , b. c. , lane , r. r. , lewis , g. f. , et al . 2007 , , 376 , 939 crane , j. d. , majewski , s. r. , rocha - pinto , h. j. , et al . 2003 , , 594 , l119 ( c03 ) fulbright , j. p. 2000 , , 120 , 1841 gratton , r. g. , carretta , e. , claudi , r. , lucatello , s. , & barbieri , m. 2003 , , 404 , 187 harris , w. e. 1996 , , , 112 , 1487 ivezi , . , sesar , b. , juri , m. , et al . 2008 , , 684 , 287 ( i08 ) juri , m. , ivezi , . , brooks , a. , et al . 2008 , , 673 , 864 keller , s. c. , yong , d. , & da costa , g. s. 2010 , , 720 , 940 kurucz , r. l. 1993 , vizier online data catalog , 6039 law , d. r. , & majewski , s. r. 2010 , , 714 , 229 lee , y. s. , beers , t. c. , sivarani , t. , et al . 2008 , , 136 , 2022 majewski , s. r. , skrutskie , m. f. , weinberg , m. d. , & ostheimer , j. c. 2003 , , 599 , 1082 marigo , p. , girardi , l. , bressan , a. , et al . 2008 , , 482 , 883 martin , n. f. , ibata , r. a. , bellazzini , m. , et al . 2004 , , 348 , 12 mink , d. j. , wyatt , w. f. , caldwell , n. , et al . 2007 , in astronomical society of the pacific conference series , vol . 376 , astronomical data analysis software and systems xvi , ed . r. a. shaw , f. hill , & d. j. bell , 249 momany , y. , zaggia , s. , gilmore , g. , et al . 2006 , , 451 , 515 munn , j. a. , monet , d. g. , levine , s. e. , et al . 2004 , , 127 , 3034 newberg , h. j. , yanny , b. , rockosi , c. , et al . 2002 , , 569 , 245 norris , j. e. , wyse , r. f. g. , gilmore , g. , et al . 2010 , , 723 , 1632 pearrubia , j. , martnez - delgado , d. , rix , h. w. , et al . 2005 , , 626 , 128 robin , a. c. , reyl , c. , derrire , s. , & picaud , s. 2003 , , 409 , 523 schlafly , e. f. , & finkbeiner , d. p. 2011 , , 737 , 103 schlegel , d. j. , finkbeiner , d. p. , & davis , m. 1998 , , 500 , 525 sesar , b. , juri , m. , & ivezi , . 2011 , , 731 , 4 sneden , c. a. 1973 , phd thesis , the university of texas at austin sobeck , j. s. , kraft , r. p. , sneden , c. , et al . 2011 , , 141 , 175 wilhelm , r. , beers , t. c. , allende prieto , c. , newberg , h. j. , & yanny , b. 2005 , in astronomical society of the pacific conference series , vol . 336 , cosmic abundances as records of stellar evolution and nucleosynthesis , ed . t. g. barnes iii & f. n. bash , 371 yanny , b. , newberg , h. j. , grebel , e. k. , et al . 2003 , , 588 , 824 ( y03 ) younger , j. d. , besla , g. , cox , t. j. , et al . 2008 , , 676 , l21 cccccccccccc 5268.6 & tiii & 2.59 & @xmath189 & 3922.1 & fei & 3.29 & @xmath190 & 3933.3 & sci & 2.3 & @xmath191 + 5268.9 & tii & 3.32 & @xmath192 & 3922.7 & fei & 2.99 & @xmath193 & 3933.4 & sci & 0.02 & @xmath194 + 5269.5 & fei & 0.86 & @xmath195 & 3922.9 & fei & 0.05 & @xmath196 & 3933.6 & fei & 3.07 & @xmath197 + 5269.9 & tii & 1.87 & @xmath192 & 3923.0 & sci & 1.99 & @xmath198 & 3933.6 & fei & 3.27 & @xmath199 + 5270.3 & cai & 2.52 & @xmath200 & 3923.0 & fei & 3.25 & @xmath201 & 3933.7 & caii & 0.0 & @xmath81 + 5270.4 & fei & 1.6 & @xmath202 & 3923.5 & scii & 0.32 & @xmath203 & 3933.9 & tii & 2.29 & @xmath204 + 5271.6 & tii & 2.77 & @xmath205 & 3924.5 & tii & 0.02 & @xmath206 & 3934.2 & tii & 0.05 & @xmath207 + 5272.0 & cri & 3.44 & @xmath208 & 3925.2 & fei & 3.29 & @xmath209 & 3935.3 & fei & 2.85 & @xmath210 + 5273.2 & fei & 3.29 & @xmath211 & 3925.6 & fei & 2.83 & @xmath120 & 3935.8 & fei & 2.83 & @xmath212 + 5273.4 & fei & 2.48 & @xmath213 & 3925.8 & sci & 1.99 & @xmath201 & 3935.9 & fei & 3.27 & @xmath214 + 5273.4 & cri & 3.45 & @xmath215 & 3925.9 & fei & 2.86 & @xmath206 & 3936.3 & tii & 2.48 & @xmath216 + 5274.4 & tii & 3.09 & @xmath217 & 3926.0 & fei & 3.24 & @xmath218 & 3936.5 & sci & 2.32 & @xmath219 + 5274.6 & tii & 2.42 & @xmath220 & 3926.3 & tii & 2.58 & @xmath221 & 3936.8 & fei & 3.25 & @xmath222 + 5275.2 & cri & 3.37 & @xmath223 & 3927.7 & sci & 1.99 & @xmath224 & 3937.0 & tii & 2.29 & @xmath225 + 5275.3 & cri & 2.88 & @xmath226 & 3927.8 & sci & 1.99 & @xmath227 & 3937.3 & fei & 2.69 & @xmath228 + 5275.8 & cri & 2.88 & @xmath229 & 3927.9 & fei & 0.11 & @xmath230 & 3937.6 & tii & 2.3 & @xmath231 + 5276.0 & feii & 3.19 & @xmath232 & 3927.9 & fei & 2.83 & @xmath201 & 3937.7 & fei & 3.02 & @xmath233 + 5276.1 & cri & 2.88 & @xmath234 & 3928.1 & fei & 3.21 & @xmath218 & 3937.9 & tii & 2.17 & @xmath235 + 5277.0 & tii & 3.33 & @xmath236 & 3928.8 & tii & 2.31 & @xmath237 & 3938.0 & tii & 2.27 & @xmath238 + 5278.2 & sci & 3.05 & @xmath192 & 3929.0 & tii & 2.04 & @xmath239 & 3938.6 & sci & 2.3 & @xmath219 + 5278.2 & tii & 2.34 & @xmath240 & 3929.0 & tii & 2.3 & @xmath241 & 3938.6 & sci & 2.34 & @xmath242 + 5278.3 & sci & 3.05 & @xmath243 & 3929.1 & fei & 2.76 & @xmath244 & 3939.7 & tii & 2.31 & @xmath245 + 5279.7 & fei & 3.3 & @xmath246 & 3929.2 & fei & 3.25 & @xmath202 & 3940.0 & sci & 2.61 & @xmath247 + 5280.3 & cri & 3.36 & @xmath248 & 3929.9 & tii & 0.0 & @xmath249 & 3940.1 & tii & 2.5 & @xmath250 + 5281.8 & fei & 3.03 & @xmath251 & 3930.2 & sci & 1.99 & @xmath209 & 3940.9 & fei & 0.96 & @xmath252 + 5282.4 & tii & 1.05 & @xmath253 & 3930.2 & tii & 1.5 & @xmath254 & 3941.3 & fei & 3.27 & @xmath255 + 5283.4 & tii & 1.87 & @xmath256 & 3930.3 & fei & 0.09 & @xmath257 & 3941.4 & sci & 2.0 & @xmath239 + 5283.6 & fei & 3.24 & @xmath258 & 3930.5 & fei & 3.21 & @xmath259 & 3942.4 & fei & 2.99 & @xmath260 + 5284.1 & feii & 2.89 & @xmath261 & 3930.9 & fei & 2.45 & @xmath262 & 3942.4 & fei & 2.85 & @xmath263 + 5284.4 & tii & 1.04 & @xmath264 & 3931.1 & tii & 2.29 & @xmath265 & 3942.8 & fei & 3.27 & @xmath266 + 5285.0 & sci & 2.5 & @xmath267 & 3931.1 & fei & 3.27 & @xmath268 & 3943.3 & fei & 2.2 & @xmath269 + 5285.6 & cri & 3.36 & @xmath270 & 3931.3 & fei & 3.24 & @xmath271 & 3943.7 & sci & 2.0 & @xmath272 + 5285.8 & sci & 2.5 & @xmath273 & 3931.6 & tii & 2.32 & @xmath274 & 3944.3 & sci & 2.61 & @xmath275 + & & & & 3932.0 & sci & 1.99 & @xmath202 & 3944.4 & tii & 2.32 & @xmath276 + & & & & 3932.0 & tiii & 1.13 & @xmath196 & 3944.7 & fei & 2.85 & @xmath277 + & & & & 3932.6 & sci & 1.85 & @xmath192 & 3944.9 & fei & 2.99 & @xmath278 + & & & & 3932.6 & fei & 2.73 & @xmath197 & & & & +

we present wide - field multiband photometry of globular cluster ( gc ) systems in ngc 1399 , ngc 1404 , and ngc 1387 located at the central region of the fornax galaxy cluster . observation was carried out through , , , and bands , which marks one of the widest and deepest-band studies on extragalactic gc systems . the present-band photometry enables us to significantly reduce the contamination by a factor of two for faint sources (.5 ) . the main results based on some 2000 gc candidates around ngc 1399 , ngc 1404 , and ngc 1387 are as follows : ( 1 ) the gc system in each galaxy exhibits bimodal color distributions in all colors examined , but the shape of color histograms varies systematically depending on colors ; ( 2 ) ngc 1399 shows that the mean colors of both blue and red gcs become bluer with increasing galactocentric radius ; ( 3 ) ngc 1399 shows overabundance of gcs in the directions of ngc 1404 and ngc 1387 , indicating their ongoing interactions ; and ( 4 ) ngc 1399 also exhibits a offset between the center of the inner gc distribution and the galaxy s optical center , suggesting that ngc 1399 is not yet dynamically relaxed and may be undergoing merger events .

1301.2323

the fornax galaxy cluster is the second nearest cluster of galaxies after virgo , with a distance modulus of @xmath6 @xcite corresponding to a distance of 20.0 mpc . it consists of two major components of the main fornax cluster centered on ngc 1399 and a subcluster dominated by ngc 1316 @xcite . there have been many photometric and spectroscopic studies of globular cluster ( gc ) systems in the central region of the fornax cluster , and a fair number of them concentrated on the gc system associated with ngc 1399 . @xcite and @xcite presented the first ccd photometry for ngc 1399 gcs . both studies presented an unusually high gc specific frequency of @xmath7 ( i.e. , number of gcs normalized to the galaxy luminosity ; see @xcite ) . although subsequent studies ( @xcite ) derived a lower value ( @xmath8 ) , it is still higher than those of normal elliptical galaxies . @xcite suggested that ngc 1404 gcs may have been tidally stripped by ngc 1399 , resulting in a high @xmath9 value of ngc 1399 and a lower @xmath9 value of ngc 1404 . @xcite also proposed that the over abundance of ngc 1399 gcs can be explained by tidal stripping of gcs from neighboring galaxies and by the accretion of gcs in the gravitational potential of the fornax cluster . @xcite supported this idea by showing that gcs in ngc 1399 and ngc 1404 are statistically indistinguishable in terms of their color distributions and luminosity functions in their @xmath10 photometry . @xcite demonstrated the tidal stripping and accretion of gcs between ngc 1399 and ngc 1404 through their numerical simulations . @xcite presented the asymmetries in the azimuthal distribution of ngc 1399 gcs and suggested that they may be explained by the tidal stripping of gcs from ngc 1387 . since @xcite first suggested a possible bimodal @xmath11 color distribution in the gc system of ngc 1399 , many studies have confirmed the bimodal or multimodal gc color distribution in this galaxy through various passbands ( e.g. , @xcite ) . several studies have investigated the properties of blue and red gcs separately . @xcite and @xcite showed that red gcs appear more centrally concentrated than blue gcs . @xcite suggested that the distribution of blue gcs is similar to that inferred for dark matter . @xcite presented that the surface density profile of blue gcs is shallower than that of red gcs within 8@xmath12 but they are not distinguishable beyond the radius . they also showed that the specific frequency of blue gcs is a factor of three larger than that of the red gcs within 7@xmath12 . @xcite suggested that the specific frequencies for red and blue gcs are 3.3 and 14.3 , respectively . @xcite showed that the peak of luminosity function for blue gcs is 0.36 mag brighter in @xmath1 than that for red gcs . @xcite found that the optical infrared color distribution of ngc 1399 gcs is unimodal , whereas the optical colors of the gcs exhibit bimodality . @xcite investigated the connection between gcs and the stellar population of ngc 1399 using their @xmath11 photometry . they suggested that the metal - poor subpopulation is homogeneous and exhibits an extended spatial distribution contributing @xmath520% of the total stellar mass , while the metal - rich subpopulation is heterogeneous and dominates the inner region of the galaxy . ccccc ngc 1387 & 03:36:56.84 & 35:30:23.85 & 12.3 & sb0 + ngc 1389 & 03:37:11.67 & 35:44:39.73 & 12.8 & sb0 + ngc 1399 & 03:38:29.14 & 35:27:02.30 & 10.6 & e0 + ngc 1404 & 03:38:52.08 & 35:35:37.67 & 10.9 & e2 + [ tbl - target ] recent studies based on the advanced camera for surveys ( acs ) fornax cluster survey ( fcs ; @xcite ) have looked into various properties of gc systems in the fornax galaxy cluster . @xcite detected color magnitude relation for gcs ( i.e. , blue tilt ) and found that the slope is shallower than that found for virgo cluster survey gc samples . @xcite measured the half - light radii of gcs in acs fcs galaxies , and suggested the mean half - light radius as a distance indicator . @xcite found that the dispersion of the gc luminosity functions varies systematically with the brightness of parent galaxies . @xcite investigated the radial color gradients of gc systems in early - type galaxies in the fornax galaxy cluster , and found that both red and blue gc subpopulations exhibit significant color gradients . they also found that the slope of the gc color gradients depends on mass of host galaxies . there have been a few studies for the gc systems of ngc 1404 and ngc 1387 . for ngc 1404 , @xcite presented @xmath2 , @xmath13 photometry of the gc system , and derived a low specific frequency of @xmath14 . they suggested that ngc 1404 may have its origin outside the fornax cluster based on the low specific frequency similar to that of isolated galaxies together with its high radial velocity . @xcite presented broad @xmath15 distributions for gcs in the inner and outer regions of ngc 1404 and tentatively suggested a presence of two gc subpopulations . the bimodal distribution in @xmath15 was later confirmed by @xcite and @xcite . for ngc 1387 , @xcite found that the @xmath11 distribution of the gcs presents a distinct separation between red and blue gcs , and the radial distribution of red gcs is more centrally concentrated than blue gcs . ccccc 2006 nov 24 & @xmath0 & 1800 & 6 & 0.81 - 1.08 + 2006 nov 24 & @xmath1 & 900 & 3 & 0.911.20 + 2006 nov 24 & @xmath2 & 240 & 3 & 0.941.10 + 2006 nov 24 & @xmath3 & 300 & 6 & 0.630.80 + 2006 nov 28 & @xmath0 & 1800 & 9 & 1.141.43 + 2006 nov 28 & @xmath1 & 900 & 3 & 1.151.35 + 2006 nov 28 & @xmath2 & 360 & 3 & 0.861.07 + [ tbl - obs ] several studies have been done on kinematics and dynamics for gcs around ngc 1399 . @xcite suggested that , based on the low - dispersion spectra of 47 gcs , the large velocity dispersion of the gcs indicates that they are associated with the gravitational potential of the fornax cluster rather than with that of ngc 1399 . @xcite supported this suggestion by finding a strong radial increase of the velocity dispersion for 74 gcs . they found no compelling evidence for rotation and could not detect differences between the kinematics of the blue and red gcs . @xcite obtained medium - resolution spectra of 468 gcs , and they did not find any signature of rotation . they suggested that the large velocity dispersion of the gcs could arise from the contamination by foreground objects . @xcite investigated kinematical and dynamical properties of the gc system of ngc 1399 using the spectra of some 700 gcs . they found that blue and red populations are kinematically distinct , and the properties of red population resemble those of field stellar populations . several spectroscopic studies investigated chemical abundance and ages of ngc 1399 gcs . @xcite obtained spectra of 18 gcs in ngc 1399 and found that their abundances span the range observed in the milky way and m31 . they pointed out that @xmath16 correlates well with metallicity but the slope of the relation becomes flatter toward redder colors if the values of milky way gcs are considered together . @xcite investigated the ages of the gcs and found that the majority of their sample ( 10 gcs ) have ages of around 11 gyr . in this paper we provide a new @xmath0 , @xmath1 , @xmath2 , and @xmath3 photometric catalog of gc candidates in the central region of the fornax galaxy cluster . this is one of the widest and deepest @xmath0-band studies on extragalactic gc systems . section 2 describes the observations and data reduction . section 3 gives the gc catalog for ngc 1399 , ngc 1404 , and ngc 1387 , discussing the selection method and the foreground and background contamination level . section 4 presents the properties of gc systems around the galaxies , including spatial distributions , color magnitude diagrams ( cmds ) , color distributions , and the radial variations of color bimodality . section 5 summarizes our results . the observations were carried out on the nights of 2006 november 24 and 28 ( ut ) with the mosaic ii ccd imager mounted on the prime focus of the 4 m blanco telescope at cerro tololo inter - american observatory ( ctio ) . the observing conditions were clear and photometric with an average seeing of around 1@xmath17 . the mosaic ii consists of eight 2048@xmath184096 pixel ccds with a pixel scale of 027 providing a total field of view of 36@xmath1936@xmath12 . our target field was chosen such that the brightest elliptical galaxy in the fornax cluster , ngc 1399 , and its three neighboring galaxies , ngc 1404 , ngc 1387 , and ngc 1389 , were all imaged in a single telescope pointing . figure [ fig1_n1399b ] shows the median combined image of our target field as imaged in the @xmath1-filter , and table [ tbl - target ] lists the basic properties of the four members of the fornax cluster . we observed the target field with @xmath20 , @xmath1 , @xmath2 , and @xmath3 filters . we note that the standard johnson @xmath0 filter was not available at the time of our observations , so we used the sloan digital sky survey ( sdss ) @xmath20 filter as an alternative and later calibrated the photometry to the johnson @xmath0 system . the resulting effect of this calibration procedure is discussed below . the total exposure times in @xmath20 , @xmath1 , @xmath2 , and @xmath3 were 27000 , 5400 , 1800 , and 1800 s , respectively . we split our observations in each band into two sets of three dithered exposures ( five sets for @xmath20 ) to fill in the gaps between the individual ccd chips in the mosaic , and to minimize the impact on our photometry from ccd blemishes ( e.g. , hot pixels and bad columns ) . a summary of observations is provided in table [ tbl - obs ] . all images were pre - processed using the mscred package @xcite in iraf . we used the ccdproc task to correct for cross - talk between the ccd chips , apply and trim the overscan , correct for the bias level , and apply the flat - field correction . we used mscpixarea to correct for the variations in pixel scales across the frame produced by geometric distortion in the mosaic ii imager . each mosaic image was then split into eight individual images using the mscsplit task . to produce a clean high signal - to - noise image in each filter , we combined the images using the montage2 routine in the allframe package @xcite . we then applied a ring median filter @xcite having an inner radius of 2.5 times the mean full width at half - maximum ( fwhm ) and a width of 5 pixel on the combined images and subtracted off these images from the original images to remove the rapidly varying background light of the galaxies while preserving the point - sources . objects were detected from the background - subtracted images using the daophot ii / allstar routines through the following steps . first , we derived a list of objects and created new images with these objects subtracted . from these images , we detected fainter objects that were missed in the previous step and created another set of object - subtracted images . this procedure was iterated five times , and all detected objects were merged into a master catalog . the master catalog was then input to the allframe program @xcite along with all of the images and their point - spread function ( psf ) models for final psf photometry . aperture corrections were computed through the daogrow routine @xcite using bright and isolated point - sources , and we applied them to our photometry above . finally , an error - weighted mean instrumental magnitude for each object was obtained using the daomatch / daomaster routines @xcite . cccccc 2006 nov 24 & @xmath21 & 2.1194 @xmath22 0.0127 & 0.4495 @xmath22 0.0072 & 0.1458 @xmath22 0.0038 & nan + & @xmath23 & 2.0699 @xmath22 0.0380 & 0.4442 @xmath22 0.0144 & 0.0819 @xmath22 0.0581 & 0.1390 @xmath22 0.0259 + & @xmath1 & 0.3711 @xmath22 0.0045 & 0.2236 @xmath22 0.0037 & 0.0803 @xmath22 0.0015 & nan + & @xmath2 & 0.6402 @xmath22 0.0035 & 0.1386 @xmath22 0.0028 & 0.0297 @xmath22 0.0011 & nan + & @xmath3 & 0.0578 @xmath22 0.0039 & 0.0576 @xmath22 0.0030 & 0.0099 @xmath22 0.0011 & nan + + 2006 nov 28 & @xmath21 & 2.1641 @xmath22 0.0128 & 0.4495 @xmath22 0.0072 & 0.1458 @xmath22 0.0038 & nan + & @xmath23 & 2.1196 @xmath22 0.0383 & 0.4442 @xmath22 0.0144 & 0.0819 @xmath22 0.0581 & 0.1390 @xmath22 0.0259 + & @xmath1 & 0.3243 @xmath22 0.0048 & 0.2236 @xmath22 0.0037 & 0.0803 @xmath22 0.0015 & nan + & @xmath2 & 0.6061 @xmath22 0.0037 & 0.1386 @xmath22 0.0028 & 0.0297 @xmath22 0.0011 & nan + & @xmath3 & 0.0350 @xmath22 0.0041 & 0.0576 @xmath22 0.0030 & 0.0099 @xmath22 0.0011 & nan + [ tbl - std ] cccc chip 1 & @xmath0 & 0.0347 @xmath22 0.0020 & 0.0334 @xmath22 0.0055 + & @xmath1 & 0.0236 @xmath22 0.0007 & 0.0147 @xmath22 0.0022 + & @xmath2 & 0.0165 @xmath22 0.0004 & 0.0029 @xmath22 0.0014 + & @xmath3 & 0.0037 @xmath22 0.0005 & 0.0010 @xmath22 0.0017 + + chip 2 & @xmath0 & 0.0323 @xmath22 0.0019 & 0.0609 @xmath22 0.0055 + & @xmath1 & 0.0162 @xmath22 0.0007 & 0.0094 @xmath22 0.0024 + & @xmath2 & 0.0034 @xmath22 0.0005 & 0.0046 @xmath22 0.0015 + & @xmath3 & 0.0023 @xmath22 0.0005 & 0.0037 @xmath22 0.0017 + + chip 3 & @xmath0 & 0.0087 @xmath22 0.0019 & 0.0675 @xmath22 0.0057 + & @xmath1 & 0.0197 @xmath22 0.0007 & 0.0125 @xmath22 0.0021 + & @xmath2 & 0.0020 @xmath22 0.0004 & 0.0038 @xmath22 0.0014 + & @xmath3 & 0.0085 @xmath22 0.0005 & 0.0013 @xmath22 0.0016 + + chip 4 & @xmath0 & 0.0208 @xmath22 0.0020 & 0.0156 @xmath22 0.0062 + & @xmath1 & 0.0195 @xmath22 0.0007 & 0.0093 @xmath22 0.0021 + & @xmath2 & 0.0060 @xmath22 0.0005 & 0.0016 @xmath22 0.0015 + & @xmath3 & 0.0055 @xmath22 0.0005 & 0.0026 @xmath22 0.0016 + + chip 5 & @xmath0 & 0.0409 @xmath22 0.0020 & 0.0013 @xmath22 0.0057 + & @xmath1 & 0.0024 @xmath22 0.0007 & 0.0018 @xmath22 0.0022 + & @xmath2 & 0.0149 @xmath22 0.0005 & 0.0022 @xmath22 0.0016 + & @xmath3 & 0.0019 @xmath22 0.0007 & 0.0006 @xmath22 0.0014 + + chip 7 & @xmath0 & 0.0183 @xmath22 0.0020 & 0.0016 @xmath22 0.0060 + & @xmath1 & 0.0052 @xmath22 0.0007 & 0.0047 @xmath22 0.0021 + & @xmath2 & 0.0023 @xmath22 0.0004 & 0.0002 @xmath22 0.0014 + & @xmath3 & 0.0031 @xmath22 0.0005 & 0.0033 @xmath22 0.0017 + + chip 8 & @xmath0 & 0.0136 @xmath22 0.0020 & 0.0069 @xmath22 0.0060 + & @xmath1 & 0.0083 @xmath22 0.0007 & 0.0037 @xmath22 0.0021 + & @xmath2 & 0.0046 @xmath22 0.0005 & 0.0005 @xmath22 0.0015 + & @xmath3 & 0.0122 @xmath22 0.0005 & 0.0035 @xmath22 0.0017 + [ tbl - c2c ] the photometric calibration was achieved using standard stars in the fields of @xcite and @xcite . during the course of our observing runs , we imaged eight standard fields placing the stars of interest on chip 6 . each image contains 1525 standard stars , which covers an appropriately large range in color . we also observed the standard fields in many different airmasses . in the calibration of @xmath0 band , however , we used all the standard stars that are observed in subsequent observing runs during 2006 , 2007 , and 2008 , because the photometric uncertainty in @xmath0 band is quite large and the number of standard stars is too small to derive reliable solutions . to calibrate the instrumental magnitudes to the johnson - cousins standard magnitudes , we used the transformation equation of the form @xmath24 where @xmath25 is the calibrated magnitude , @xmath26 is the instrumental magnitude that is normalized to 1 s , c1 is the photometric zero point offset , c2 is the extinction coefficient for airmass corrections , and both c3 and c4 are the color coefficients . the colors used in the calibration are @xmath27 , @xmath28 , @xmath28 , and @xmath16 for @xmath0 , @xmath1 , @xmath2 , and @xmath3 , respectively . the photometric coefficients are determined by an iterative process of error - weighted fitting of the transformation equation to the standard magnitudes . the quadratic color term is included only in the equation of @xmath0 band , because a significant deviation from a linear fit occurs in the transformation of @xmath0 band . figure [ fig2-std ] shows the @xmath0 magnitude difference between the standard and instrumental magnitudes corrected for both of the photometric zero point and the atmospheric extinction , as a function of @xmath27 . the dashed lines represent two best - fit functions for blue and red standard stars , respectively . however , if we divide the observed data into two groups by @xmath27 and use the two functions in calibration separately , the discontinuity between two functions may lead to an artificial clump in color distributions of the data . to avoid this problem , we connected the two functions ( red lines ) at the internal division points of them in the range @xmath29 ( two vertical lines ) . finally , the instrumental magnitudes were converted to the standard johnson - cousins @xmath30 system using the set of photometric coefficients listed in table [ tbl - std ] calibrating the sdss @xmath20 photometry to the johnson @xmath0 system may cause systematic shifts in the colors based on this band since the system throughputs are different between the two filters . to estimate these shifts , we simulated our observations using the system throughputs of the two filters , synthetic spectra of gcs from the yeps model @xcite , and stars from the kurucz atlas9 model . we calculated the magnitude through each filter for each model gc , and converted it to the standardized johnson @xmath0 magnitude using the synthetic spectrum of a star having the same color of the gc . we then calculated the differences between these standardized magnitudes from the respective filters . the differences range from 0.002 mag to 0.022 mag , depending on the colors of the model gcs . we conclude that these are small enough to have negligible impact on our photometric analyses throughout the paper . as the individual ccd chips of the mosaic may have different color terms , we attempted to correct for this `` chip - to - chip variation '' as follows . on the night of 2008 may 5 , we took multiple @xmath31 observations of the landolt standard fields sa98 and sa107 with the mosaic ii , such that the same set of standard stars was positioned on each chip . using these data , we derived the color terms that correct for the chip - to - chip variation . although these color terms were very small , we applied them to our data in hopes of making the multiple measurements of the same source more consistent . the equation used in correction is of the form @xmath32 where @xmath33 is the magnitude converted to that on the reference chip 6 , @xmath34 is the magnitude measured on each chip , c1 is the zero point offset , and c2 is the color coefficient . the colors used in the calibration are @xmath27 , @xmath28 , @xmath28 , and @xmath16 for @xmath0 , @xmath1 , @xmath2 , and @xmath3 , respectively . table [ tbl - c2c ] lists the zero point offsets and color coefficients for each chip . we corrected for the foreground galactic extinction using the reddening maps provided by @xcite . to account for the atmospheric dispersion corrector ( adc ) of the 4 m blanco telescope that significantly modifies the system throughputs , we computed the effective wavelengths of our four bands from the total system throughputs including the effect of the adc , and we derived @xmath35 values based on the extinction law provided by @xcite . the mean extinction values in @xmath0 , @xmath1 , @xmath2 , and @xmath3 are 0.06 , 0.05 , 0.04 , and 0.02 mag , respectively . our images were calibrated for astrometry using the usno - b 1.0 catalog stars @xcite and the msctpeak task . the typical rms of our astrometric solutions is at the level of 02 . the r.a./decl . coordinates of our subsequent source catalogs use the positions derived from these images . we performed artificial star tests to estimate the completeness of our photometry . we used the addstar routine of daophot ii to generate 4000 artificial stars in each chip with magnitudes and color ranges consistent with those of the point - sources detected and measured above . the luminosity function of our artificial stars was chosen to be flat to preserve statistical significance in all magnitude ranges . artificial stars were added to the psf - subtracted images using the empirical psfs derived during our photometry . we used the same photometry procedure as above to detect and measure artificial stars in these images . this process was repeated seven times with the positions of artificial stars altering randomly each time , resulting in a total of 28,000 artificial stars per chip . in each 0.3 mag bin , we calculated the detection efficiency , i.e. , the ratio of the number of recovered artificial stars to the number of added ones . central regions of galaxies and the gaps between ccd chips were masked for the calculation . to obtain a limiting magnitude at which the completeness drops to 50% , the detection efficiencies were fitted to the analytic function by @xcite , @xmath36 where @xmath37 is the limiting magnitude , and @xmath38 is the parameter that controls the steepness of the completeness function near the limiting magnitude . figure [ fig3_comtest ] shows the measured completeness ( dots ) and the fitted functions ( solid lines ) in @xmath0 , @xmath1 , @xmath2 , and @xmath3 bands depicted by purple , blue , green , and red colors , respectively . the limiting magnitudes of the four bands are 24.4 , 24.4 , 23.7 , and 22.8 mag in @xmath0 , @xmath1 , @xmath2 , and @xmath3 bands , respectively . we note that the limiting magnitudes slightly vary in the @xmath390.2 mag level among individual ccds due to variations in detector characteristics . we also investigated the radial variations of completeness around ngc 1399 , ngc 1404 , and ngc 1387 . in the central region of a galaxy , the detection effectiveness at a certain magnitude decreases as the surface brightness of the galaxy increases , and it finally becomes zero at a saturated region . to quantify this effect in our observations , we performed separate artificial star tests near the galaxy centers . we added 8000 artificial stars on the individual ccd chips that contain the galaxies ( ngc 1399 , ngc 1404 , and ngc 1387 ) , this time with a centrally concentrated radial distribution centered on each galaxy . this experiment was repeated five times with the positions of artificial stars altering each time . the completeness was then calculated as a function of magnitude in concentric circles with radial intervals of 02 . figure [ fig4_com_radial ] shows the @xmath2-band completeness functions at various radial bins from the center of each galaxy . the thick black line represents the completeness function for the overall region , as derived above . as expected from the strong background lights near the centers of our target galaxies , we find that the completeness is very low near the center . the completeness increases as one moves out from the center , and then flattens out beyond 13 for ngc 1399 , 07 for ngc 1404 , and 05 for ngc 1387 . at the distance of the fornax cluster ( @xmath520.0 mpc ; * ? ? ? * ) , 1 pc is equal to 001 , so a typical gc appears as a point - source in our mosaic images . we therefore can not select gcs based on their apparent sizes but must rely on photometric colors . details on the selection process of gc candidates are given below . we start by using the allframe parameters chi and sharp to select point - sources from our photometric catalog . figure [ fig5_chi ] shows the sharp , chi , and the allframe photometric error as a function of @xmath40 magnitude with solid lines showing the selection criteria . while the figure only shows the @xmath0-band selection , we applied the same selection criteria to all bands simultaneously , and the resulting selection includes 5500 point - source objects . we then employ color cuts using our multi band photometry as demonstrated in figure [ fig6_twocol ] to select gc candidates . the blue selection boxes shown in the three color color diagrams were determined on the basis of the findings below . first , we matched our point - source objects with the spectroscopic catalog of @xcite and found that the ngc 1399 gcs , as identified by their measured line - of - sight velocities , form a tight sequence ( red dots in figure [ fig6_twocol ] ) in the color color diagrams , whereas foreground stars are located along an extended sequence ( cyan dots in figure [ fig6_twocol ] ) . second , we downloaded @xmath41/acs images of ngc 1399 from the archive and identified genuine gcs based on their sizes . the sizes were measured using ishape @xcite and following the same procedure outlined in @xcite . we then cross - identified the gcs with our point - source objects , and found that most of them lie inside the tight sequence ( orange dots in figure [ fig6_twocol ] ) . to select gc candidates , we further applied a magnitude cut of @xmath42 ( corresponding to @xmath43 ) to the color color - selected sample . this magnitude limit was determined based on the brightest gc in the spectroscopic catalog of @xcite , and we note that it is consistent with the limit used by @xcite to separate gcs from their ultra compact dwarf galaxies sample . the color color and magnitude cuts result in a final sample of 2037 gc candidates . figure [ fig6_twocol ] demonstrates the advantage of using @xmath0 band for separating star - forming background galaxies from the extragalactic gcs . the background galaxies have @xmath44 colors similar to the bluer gcs , but due to their star - forming nature , galaxies exhibit significantly bluer colors in @xmath45 and @xmath46 than the gcs . this allows us to effectively reduce contamination by background galaxies . the same effect was found by @xcite using the @xmath11 washington color index . nonetheless , our selection box still contains several foreground stars , and it is important to quantify the contamination level of our color color - selected objects . we provide the estimation of the contamination level below . to estimate the number of contaminating sources ( foreground stars and background galaxies ) in our gc candidate list , it would be best to use a control field observed near the science field with the same observational conditions . however , this approach is impractical for us because the @xmath0-band observations require very long exposure times . as an alternative , we used two independent methods to estimate the contamination level of our gc candidate list . ccccccccccc 1 & 54.4670781 & 35.8412594 & 23.863 & 0.057 & 23.466 & 0.018 & 22.611 & 0.025 & 21.614 & 0.018 + 2 & 54.4048663 & 35.8368808 & 23.099 & 0.028 & 22.941 & 0.011 & 22.069 & 0.015 & 21.119 & 0.010 + 3 & 54.1532595 & 35.8313662 & 23.048 & 0.038 & 22.743 & 0.012 & 21.970 & 0.015 & 20.920 & 0.020 + 4 & 54.3887923 & 35.8307234 & 24.085 & 0.070 & 23.983 & 0.029 & 23.384 & 0.049 & 22.462 & 0.030 + 5 & 54.3871824 & 35.8300091 & 23.772 & 0.054 & 23.865 & 0.025 & 23.147 & 0.038 & 22.327 & 0.028 + 6 & 54.3545131 & 35.8284302 & 23.147 & 0.030 & 22.633 & 0.008 & 21.661 & 0.011 & 20.490 & 0.006 + 7 & 54.5147090 & 35.8246663 & 22.822 & 0.023 & 22.949 & 0.014 & 22.247 & 0.015 & 21.430 & 0.013 + 8 & 54.3057038 & 35.8223111 & 24.007 & 0.061 & 24.027 & 0.028 & 23.363 & 0.045 & 22.429 & 0.028 + 9 & 54.2008513 & 35.8137756 & 24.064 & 0.070 & 24.111 & 0.030 & 23.383 & 0.045 & 22.486 & 0.030 + 10 & 54.4062928 & 35.8090401 & 21.291 & 0.007 & 21.338 & 0.004 & 20.674 & 0.005 & 19.854 & 0.003 + & & & & & & & & & & + & & & & & & & & & & + & & & & & & & & & & [ tbl - cat ] on average , chip 1 is located farthest away from the target galaxies ngc 1399 , ngc 1387 , and ngc 1404 , so it has the least number of gcs expected among the eight ccd chips . in figure [ fig7_ctoc ] , we compare the @xmath46 versus @xmath47 color color diagrams of the 5500 point - source objects for each chip . in chip 1 , there are 41 objects that are selected as gc candidates using the color color selection box as in figure [ fig6_twocol ] . if we assume that all 41 objects are contaminants , this implies that our catalog of gc candidates has a contamination level of 16% . we note that 16% is in fact an upper limit because chip 1 , which is at an average distance of 24(@xmath5140 kpc ) from the center of ngc 1399 , is likely to contain several gcs given that the ngc 1399 gcs are known to extend out to a projected distance of at least 45 ( @xmath5260 kpc ; @xcite ) . we further estimated the number of foreground stars using the besanon model of the milky way population @xcite . we generated a synthetic catalog of stars in the direction of the sky consistent with our target field and then selected stars using the same color color cut as used in figure [ fig6_twocol ] . the estimated number of foreground stars in our gc candidate list turned out to be 180 and the corresponding contamination level is 9% . as expected , this is lower than the upper limit set by using chip 1 as the control field above . based on the two estimates , the contamination level of our gc candidate list is likely to be in the 9%16% range . although the contamination level itself may not be that high , it is important to test whether the contaminating sources will affect our gc color distribution analyses . figure [ fig8_colbe ] shows the comparison of @xmath46 color distributions for all point - sources ( shaded histogram ) , our gc candidate list ( blue ) , and synthetic stars from the besanon model ( red ) between chips 7 and 1 . based on the similarities found in the red and blue color distributions in the upper panel of figure [ fig8_colbe ] , chip 1 seems to be a good control field . from the same figure , we find that the foreground mw stars are expected to have a @xmath46 color distributions that is peaked toward the bluer side similar to the @xmath46 color distribution of gc candidates . however , as demonstrated in the lower panel of figure [ fig8_colbe ] , the number of these foreground stars , when compared to that of the gc candidates , is too small to have any impact on our color distribution analyses presented in sections 4.3 and 4.4 . table [ tbl - cat ] presents the photometric catalog of total 2,037 gc candidates . the columns give the following information . + @xmath48 . identification number + @xmath49 . right ascension and declination ( j2000 ) + @xmath50 . @xmath40 magnitude and its rms uncertainty + @xmath51 . @xmath52 magnitude and its rms uncertainty + @xmath53 . @xmath4 magnitude and its rms uncertainty + @xmath54 . @xmath55 magnitude and its rms uncertainty . + we note that all magnitudes in the catalog are corrected for galactic extinction . l c c c c c c c c c + + @xmath27 & 0.016 @xmath22 0.006 & 0.102 @xmath22 0.003 & 0.372 @xmath22 0.021 & 0.188 @xmath22 0.009 & 0.34 & 1.84 & 0.010 & 0.090 & 0.860 + @xmath56 & 0.670 @xmath22 0.007 & 0.113 @xmath22 0.005 & 1.147 @xmath22 0.022 & 0.269 @xmath22 0.008 & 0.42 & 2.38 & 0.010 & 0.100 & 0.140 + @xmath57 & 1.557 @xmath22 0.008 & 0.134 @xmath22 0.006 & 2.188 @xmath22 0.025 & 0.357 @xmath22 0.008 & 0.45 & 2.66 & 0.010 & 0.080 & 0.010 + @xmath28 & 0.673 @xmath22 0.004 & 0.058 @xmath22 0.002 & 0.849 @xmath22 0.007 & 0.060 @xmath22 0.003 & 0.34 & 1.03 & 0.010 & 0.070 & 0.010 + @xmath15 & 1.571 @xmath22 0.005 & 0.087 @xmath22 0.003 & 1.925 @xmath22 0.014 & 0.135 @xmath22 0.007 & 0.37 & 1.55 & 0.010 & 0.050 & 0.010 + @xmath16 & 0.910 @xmath22 0.005 & 0.063 @xmath22 0.003 & 1.108 @xmath22 0.013 & 0.068 @xmath22 0.007 & 0.30 & 1.08 & 0.010 & 0.060 & 0.010 + + @xmath27 & 0.004 @xmath22 0.012 & 0.079 @xmath22 0.008 & 0.490 @xmath22 0.024 & 0.120 @xmath22 0.018 & 0.39 & 1.52 & 0.001 & 0.017 & 0.001 + @xmath56 & 0.682 @xmath22 0.015 & 0.101 @xmath22 0.011 & 1.393 @xmath22 0.026 & 0.136 @xmath22 0.017 & 0.39 & 1.35 & 0.001 & 0.001 & 0.001 + @xmath57 & 1.631 @xmath22 0.018 & 0.124 @xmath22 0.015 & 2.583 @xmath22 0.029 & 0.161 @xmath22 0.019 & 0.39 & 1.30 & 0.001 & 0.001 & 0.001 + @xmath28 & 0.686 @xmath22 0.007 & 0.052 @xmath22 0.007 & 0.901 @xmath22 0.005 & 0.029 @xmath22 0.004 & 0.38 & 0.56 & 0.001 & 0.014 & 0.001 + @xmath15 & 1.634 @xmath22 0.011 & 0.075 @xmath22 0.011 & 2.091 @xmath22 0.009 & 0.051 @xmath22 0.006 & 0.39 & 0.68 & 0.001 & 0.001 & 0.001 + @xmath16 & 0.949 @xmath22 0.006 & 0.041 @xmath22 0.006 & 1.190 @xmath22 0.006 & 0.030 @xmath22 0.005 & 0.39 & 0.73 & 0.001 & 0.001 & 0.001 + + @xmath27 & 0.019 @xmath22 0.017 & 0.084 @xmath22 0.011 & 0.446 @xmath22 0.030 & 0.128 @xmath22 0.017 & 0.49 & 1.52 & 0.001 & 0.041 & 0.001 + @xmath56 & 0.693 @xmath22 0.020 & 0.095 @xmath22 0.013 & 1.355 @xmath22 0.035 & 0.160 @xmath22 0.021 & 0.49 & 1.68 & 0.001 & 0.011 & 0.001 + @xmath57 & 1.624 @xmath22 0.023 & 0.114 @xmath22 0.014 & 2.505 @xmath22 0.044 & 0.196 @xmath22 0.022 & 0.49 & 1.72 & 0.001 & 0.005 & 0.001 + @xmath28 & 0.707 @xmath22 0.010 & 0.046 @xmath22 0.008 & 0.904 @xmath22 0.010 & 0.045 @xmath22 0.008 & 0.51 & 0.98 & 0.001 & 0.039 & 0.001 + @xmath15 & 1.642 @xmath22 0.012 & 0.058 @xmath22 0.006 & 2.056 @xmath22 0.017 & 0.079 @xmath22 0.009 & 0.49 & 1.36 & 0.001 & 0.002 & 0.001 + @xmath16 & 0.931 @xmath22 0.007 & 0.038 @xmath22 0.009 & 1.148 @xmath22 0.013 & 0.052 @xmath22 0.009 & 0.49 & 1.37 & 0.001 & 0.017 & 0.001 + [ tbl - gmm ] figure [ fig9_cont ] maps the spatial distributions of gc candidates ( left panel ) and foreground / background objects ( right ) . the color - filled contours show the distribution of surface number density . the black crosses mark the positions of the optical centers of galaxies ( from top , ngc 1399 , ngc 1387 , ngc 1404 , and ngc 1389 ) . in the left panel , three gc systems are clearly visible around ngc 1399 , ngc 1404 , and ngc 1387 . the number of gcs around ngc 1389 is too small , so we omit the galaxy from the following analysis . in the right panel , the distribution of foreground / background objects appears fairly uniform . the slight overdensity in the region between ngc 1399 and ngc 1387 is likely due to the presence of a distant galaxy cluster , given that many objects in this region are bluer than 0.3 in @xmath46 . another overdense region is right on the center of ngc 1399 , which indicates that gcs in highly dense regions can be misclassified as foreground / background objects in our gc selection procedure , as mentioned in section 3 . note that the residuals of the mosaic ccd gaps are discernible as a vertical feature in the middle of the panel . it is noteworthy that the gc distribution around ngc 1399 shows interesting features . on the one hand , there is an overabundance of gcs in the outer region of ngc 1399 toward ngc 1404 and ngc 1387 . this was also pointed out by @xcite , and it may indicate interactions of ngc 1399 with ngc 1404 and ngc 1387 in the recent past ( @xcite ) . on the other hand , a @xmath505 displacement is found between ngc 1399 s optical center and the center of the gc distribution in the inner region of the galaxy as denoted by a red contour in the density map . @xcite reported an asymmetric x - ray halo for ngc 1399 , consisting of three components with different centers ; a central component , a galactic component centered 1@xmath12 southwest of ngc 1399 , and a cluster component centered @xmath556 northeast of the galaxy . while the center of the gc distribution does not coincide with the x - ray centers , the asymmetric distribution of the gcs may suggest that the ngc 1399 is not yet dynamically relaxed and may be undergoing merger events . figure [ fig10_rden ] presents the surface number densities against the galactocentric radius for the three galaxies . in order to examine the radial extent of each gc system , we calculated the surface gc number density as a function of the distance from the galaxy center . we first set up a series of radial bins ( annulus ) , each containing approximately equal numbers of gcs ( @xmath5100 gcs for ngc 1399 ; @xmath540 gcs for ngc 1404 ; @xmath525 gcs for ngc 1387 ) . the number of gcs in each bin was then corrected using the corresponding completeness function of the bin . we finally divided the corrected numbers of gcs by the area of annuli to get surface densities . the surface density profile of each galaxy is well described by de vaucouleurs @xmath58 fits ( dashed lines ) . we determine the limiting radii of gc systems of ngc 1404 ( 34 ) and ngc 1387 ( 37 ) as the radial points where the density profiles begin to depart from de vaucouleurs law . since the gc system of ngc 1399 is extended ( @xmath545 ; @xcite ) well beyond the field of view ( @xmath59 ) of our mosaic observations , we regard the entire gcs in the images as ngc 1399 s gc system , except for the regions inside the limiting radii of ngc 1404 and ngc 1387 . figure [ fig11_cmd ] presents the cmds for all point - sources . magnitudes and colors were corrected for the galactic extinction , as mentioned in section 2.3 . the black and gray dots represent the gc candidates and foreground / background objects , respectively . the dashed lines indicate the limiting magnitude of 50% completeness . the tilt of the limiting magnitude lines results from lower detection efficiency of red gcs than blue gcs in the @xmath0 band . the @xmath46 colors corresponding to gcs , background galaxies , and foreground stars can be determined both by the @xmath46 versus @xmath47 diagram ( the bottom panel of figure [ fig6_twocol ] ) and by the @xmath4@xmath46 cmd ( the left panel of figure [ fig11_cmd ] ) . gcs are placed at @xmath60 , while most background galaxies are bluer at @xmath46 @xmath61 @xmath62 and foreground stars are at @xmath63 and 1.1 close to both ends of the gc range . the @xmath4@xmath44 cmd ( right panel ) shows that many of the background galaxies have similar colors to the gcs , making it difficult to discriminate gcs from galaxies in this color . the foreground star sequences , particularly on the red side , are more clearly separated from gcs than in @xmath46 . this comparison corroborates that the use of various color combinations including @xmath0-band colors provides a powerful tool for discriminating gcs from other objects . it is interesting to note that the cmds do not show bimodality for gcs brighter than @xmath64 . the disappearance was previously reported for ngc 1399 ( @xcite ) , a few galaxies in the virgo cluster ( @xcite ) , and other giant galaxies ( @xcite ) . the unimodal distribution is often explained by merging of blue and red sequences at the highest luminosities as a result of the blue tilt phenomenon @xcite , which is a trend for the blue gcs to have redder colors at higher luminosities . however , @xcite and @xcite have reported that there is no blue tilt in the gc system of ngc 1399 . interestingly , our data reveal that the number of blue gcs rapidly drops at the brightest magnitudes ( @xmath4 @xmath39 21.5 ) . this suggests that the unimodal distribution of brightest gcs is not the consequence of the blue tilt , but instead owing to the relative scarcity of blue , brightest gcs . figure [ fig12_colhist ] shows three representative color histograms of the entire gc system within the limiting radius in ngc 1399 ( top ) , ngc 1404 ( middle ) , and ngc 1387 ( bottom ) . the gcs fainter than the limiting magnitudes and/or the ones located in the innermost region of each galaxy , where the completeness test is unreliable ( see section 2.3 ) , are taken out from the subsequent analysis . the thick solid lines are smoothed histograms with gaussian kernels . the filled histograms in the top row are for the brightest gcs , as mentioned in the previous section . for ngc 1399 ( top row ) , gcs show bimodal distributions for all colors considered in agreement with previous findings in @xmath11 ( @xcite ) . for ngc 1404 ( middle ) and ngc 1387 ( bottom ) , the dotted histograms are for all gcs within the limiting radii of the galaxies . the solid histograms represent `` corrected '' distributions , for which the contribution of ngc 1399 gcs is statistically subtracted from the original distributions and will be used in the further analysis . ngc 1404 gcs exhibit pronounced bimodality with an evident dip between two groups in all colors , consistent with the previous studies for ngc 1404 s inner gcs by @xcite and @xcite . the clear separation between blue and red gcs for ngc 1387 gcs is also consistent with what was found in @xmath11 by @xcite . blue gcs are much less abundant in ngc 1404 and ngc 1387 than in ngc 1399 . a closer scrutiny of figure [ fig12_colhist ] reveals that the exact morphologies of the gc color distributions change depending on colors considered . the most distinctive case is given by the @xmath46 distribution of ngc 1399 . compared to the @xmath44 and @xmath65 histograms , the @xmath46 distribution exhibits more prominent blue gc peak and have red gcs with weaker peak and larger dispersion in color . the @xmath46 distribution shows spread in colors of red gcs that is nearly a factor of two larger than that of blue gcs , while the @xmath44 and @xmath65 histograms have similar variances for the red and blue gcs . the histograms for ngc 1404 and ngc 1387 also show changes in their shape depending on colors . to be more quantitative in the bimodality analysis , we used the gaussian mixture modeling ( gmm ) code by @xcite . table [ tbl - gmm ] presents the results of the gmm analysis for six color combinations ( @xmath46 , @xmath45 , @xmath66 , @xmath44 , @xmath47 , and @xmath65 ) for the three galaxies . it gives the color index , the mean ( @xmath67 , and the standard deviation ( @xmath68 ) of blue and red gc colors , the red gc fraction ( @xmath69 ) , and the ratio of the standard deviations between blue and red gcs ( @xmath70/@xmath71 ) . the last three columns summarize the probabilities of preferring a unimodal distribution over a bimodal distribution ( @xmath72-values ) derived based on the likelihood ratio test ( lrt ; @xmath73 ) , on the separation of the means relative to their variances ( @xmath74 ) and on the kurtosis of a distribution ( @xmath75 ) . from the lrt is @xmath76 in all colors , indicating the bimodal distributions are preferred over the unimodal ones . the @xmath77 values lead to the same conclusion in all cases . the @xmath78 values support the bimodal distributions in most cases , but favor a unimodal distribution for @xmath46 of ngc 1399 with @xmath79 . ] figure [ fig13_tab4 ] presents the results from the gmm analysis for all six color combinations . to quantify the shape of histograms , we use the red gc fractions ( left ) and the ratios of standard deviations between blue and red gc colors ( right ) . ngc 1399 gc sample ( black solid lines ) does not include gcs around ngc 1404 and ngc 1387 , and the gc sample of ngc 1404 ( red dashed lines ) and ngc 1387 ( blue dotted ) is corrected for ngc 1399 gc contribution . for ngc 1399 , we find that both the numbers and dispersions of blue and red gcs change significantly depending on colors that are used . the trend is more evident when comparing the @xmath0-band colors with the rest . for ngc 1404 and ngc 1387 , the interesting features emerge after the correction for the contribution of ngc 1399 gcs . first , the red gc fractions in all colors remain constant regardless of the colors , as expected in figure [ fig12_colhist ] showing distinct separation between blue and red gcs . it is plausible that the early interactions of the galaxies with ngc 1399 have preferentially left the central , red gcs and later blue gcs have been accreted from outer region of ngc 1399 . second , the color dispersions vary significantly depending on colors , following a similar pattern to the case of ngc 1399 for @xmath16 , @xmath28 , and @xmath15 . for @xmath0-band colors , however , the patterns are different in the sense that they tend to be constant within the errors . the variation of dispersion as a function of colors is not expected if the metallicity - to - color converting relations are straight . instead , the slopes seem different between the blue and red parts of the metallicity color relations and such an effect varies systematically from color to color ( @xcite ) . figure [ fig14_n1399h ] displays the color distributions as a function of galactocentric radius for gcs in ngc 1399 . the subsequent analysis is only valid for ngc 1399 because small number statistics prohibit us making any firm statement on the radial variations of gc colors for the two fainter galaxies after the correction for the ngc 1399 gc contamination . in order to investigate the radial variation of color bimodality , we set up a series of radial bins containing approximately equal number ( @xmath5300 ) of gcs and perform the gmm analysis for each bin . we focus particularly on three main parameters of the gmm analysis that characterize the bimodal distributions ; the number ratio between blue and red gcs , the mean colors of the groups , and their color dispersions . figure [ fig15_n1399r ] shows the variations of color bimodality properties as a function of galactocentric radius . there is an obvious radial trend for the mean colors of both blue and red gcs getting bluer with increasing radius in all colors . the radial color gradient of red gcs is steeper than that of blue gcs . the red gc fraction in each color rapidly decreases with increasing galactocentric radius out to @xmath80 @xmath61 @xmath81 , and it remains fairly constant beyond the point . there is a weak radial trend in gc color dispersions , in that the ratio between dispersions of blue and red gcs increases with radius . we have performed wide - field @xmath30 photometry of gcs in the central region of the fornax cluster of galaxies with the mosaic ii ccd imager on the 4 m blanco telescope at ctio . this is one of the widest and deepest @xmath0-band studies on extragalactic gc systems . the reduction was carried out with the daophot ii / allframe package . using two - color diagrams and the magnitude cut of @xmath82 , a total of 2037 gc candidates were selected among 12,134 objects detected in all four bands . our estimate of the contamination by background galaxies and foreground stars in our sample is in the range of 9%16% . the sample is @xmath0-band limited , and the completeness of our photometry is 50% at @xmath83 according to our artificial star tests . we provide the @xmath30 photometric catalog of the gc candidates online . for the gc systems of ngc 1399 , ngc 1404 , and ngc 1387 , we have investigated the spatial distributions , cmds , and color distributions . for ngc 1399 gcs , the radial variations in the properties of bimodality were examined as well . the main results are summarized as follows . 1 . we find an asymmetric distribution of gc candidates around ngc 1399 . the overabundance of gcs at the southeast side of the inner region and toward the directions of ngc 1404 and ngc 1387 in the outer region suggests that there were recent interactions of ngc 1399 with the galaxies and that ngc 1399 is not yet dynamically relaxed . we specify the radial extent of each gc system at the radius where the surface density profile begins to depart from the de vaucouleurs fit . the limiting radii are @xmath534 for ngc 1404 and @xmath537 for ngc 1387 . the gc system of ngc 1399 is spatially extended beyond the field of view of our observation . we show that the gc systems in the three galaxies exhibit bimodal color distributions in all the colors considered in this study . for the brightest gcs in ngc 1399 , the bimodality becomes weak or disappear depending on colors . we find with gmm that the morphology of the color histograms changes significantly , depending on the colors used . the color dispersion of red gcs is larger than that of blue gcs in @xmath0-band colors by more than a factor of two , while the red and blue gcs show similar dispersions in @xmath44 and @xmath65 . the number ratios between blue and red gcs vary depending on colors used as well . when corrected for the contribution of ngc 1399 gcs , the red gc fractions of ngc 1404 and ngc 1387 remain constant regardless of the colors used . we suggest a possible scenario that could explain the presence of two discrete gc populations in ngc 1404 and ngc 1387 : ( 1 ) the central , red gcs have been preferentially left through the early interactions of the galaxies with ngc 1399 and ( 2 ) later blue gcs have been accreted from outer region of ngc 1399 . we confirmed that the mean colors of blue and red gcs in ngc 1399 show a radial trend of becoming bluer with increasing galactocentric radius . the gradient of red gcs is steeper than that of blue gcs . acknowledges support from international exchange program for university researchers ( nrf-2011 - 013-c00031 ) and from mid - career research program ( no . 2012r1a2a2a01043870 ) through the national research foundation ( nrf ) of korea grant funded by the ministry of education , science and technology ( mest ) , and support by the nrf of korea to the center for galaxy evolution research ( no . 2012 - 8 - 1743 ) and by the korea astronomy and space science institute research fund 2011 and 2012 . this work is partially supported by the kasi - yonsei joint research program ( 20112012 ) for the frontiers of astronomy and space science funded by the korea astronomy and space science institute . this material is based upon work supported by aura through the nsf under aura cooperative agreement ast 0132798 , as amended . bassino , l. p. , faifer , f. r. , forte , j. c. , et al . 2006 , , 451 , 789 bassino , l. p. , richtler , t. , & dirsch , b. 2006 , , 367 , 156 bekki , k. , forbes , d. a. , beasley , m. a. , & couch , w. j. 2003 , , 344 , 1334 blakeslee , j. p. , cho , h. , peng , e. w. , et al . 2012 , , 746 , 88 blakeslee , j. p. , jordn , a. , mei , s. , et al . 2009 , , 694 , 556 bridges , t. j. , hanes , d. a. , & harris , w. e. 1991 , , 101 , 469 cantiello , m. , & blakeslee , j. p. 2007 , , 669 , 982 cardelli , j. a. , clayton , g. c. , & mathis , j. s. 1989 , , 345 , 245 chies - santos , a. l. , larsen , s. s. , cantiello , m. , et al . 2012 , , 539 , a54 chung , c. , yoon , s .- j . , lee , s .- y . , & lee , y .- w . 2012 , , 204 , 3 dirsch , b. , richtler , t. , geisler , d. , et al . 2003 , , 125 , 1908 drinkwater , m. j. , gregg , m. d. , & colless , m. 2001 , , 548 , 139 faifer , f. r. , forte , j. c. , norris , m. a. , et al . 2011 , , 416 , 155 fleming , d. e. b. , harris , w. e. , pritchet , c. j. , & hanes , d. a. 1995 , , 109 , 1044 forbes , d. a. , beasley , m. a. , brodie , j. p. , & kissler - patig , m. 2001 , , 563 , 143 forbes , d. a. , brodie , j. p. , & grillmair , c. j. 1997 , , 113 , 1652 forbes , d. a. , grillmair , c. j. , williger , g. m. , elson , r. a. w. , & brodie , j. p. 1998 , , 293 , 325 forte , j. c. , faifer , f. , & geisler , d. 2005 , , 357 , 56 forte , j. c. , faifer , f. , & geisler , d. 2007 , , 382 , 1947 landolt , a. u. 1992 , , 104 , 340 larsen , s. s. 1999 , , 139 , 393 larsen , s. s. , brodie , j. p. , huchra , j. p. , forbes , d. a. , & grillmair , c. j. 2001 , , 121 , 2974 liu , c. , peng , e. w. , jordn , a. , et al . 2011 , , 728 , 116 masters , k. l. , jordn , a. , ct , p. , et al . 2010 , , 715 , 1419 mieske , s. , jordn , a. , ct , p. , et al . 2006 , , 653 , 193 mieske , s. , jordn , a. , ct , p. , et al . 2010 , , 710 , 1672 monet , d. g. , levine , s. e. , canzian , b. , et al . 2003 , , 125 , 984 muratov , a. l. , & gnedin , o. y. 2010 , , 718 , 1266 paolillo , m. , fabbiano , g. , peres , g. , & kim , d .- w . 2002 , , 565 , 883 richtler , t. , dirsch , b. , gebhardt , k. , et al . 2004 , , 127 , 2094 richtler , t. , grebel , e. k. , domgoergen , h. , hilker , m. , & kissler , m. 1992 , , 264 , 25 robin , a. c. , reyl , c. , derrire , s. , & picaud , s. 2003 , , 409 , 523 schlegel , d. j. , finkbeiner , d. p. , & davis , m. 1998 , , 500 , 525 schuberth , y. , richtler , t. , hilker , m. , et al . 2010 , , 513 , a52 secker , j. 1995 , , 107 , 496 stetson , p. b. 1990 , , 102 , 932 stetson , p. b. 1993 , in iau colloq . 136 : stellar photometry - current techniques and future developments , ed . c. j. butler & i. elliott ( cambridge : cambridge univ . press ) , 291 stetson , p. b. 1994 , , 106 , 250 stetson , p. b. 2000 , , 112 , 925 strader , j. , brodie , j. p. , spitler , l. , & beasley , m. a. 2006 , , 132 , 2333 valdes , f. g. 1998 , in asp conf . ser . 145 , astronomical data analysis software and systems vii , ed . r. albrecht , r. n. hook , & h. a. bushhouse ( san francisco , ca : asp ) , 53 villegas , d. , jordn , a. , peng , e. w. , et al . 2010 , , 717 , 603 wagner , s. , richtler , t. , & hopp , u. 1991 , , 241 , 399 yoon , s .- j . , lee , s .- y . , blakeslee , j. p. , et al . 2011a , , 743 , 150 yoon , s .- j . , sohn , s. , kim , h .- s . 2012 , , submitted yoon , s .- j . , sohn , s. t. , lee , s .- y . , et al . 2011b , , 743 , 149 yoon , s .- j . , yi , s. k. , & lee , y .- w . 2006 , sci , 311 , 1129

the loss of ultracold trapped atoms due to deeply inelastic reactions has previously been taken into account in effective field theories for low - energy atoms by adding local anti - hermitian terms to the effective hamiltonian . here we show that when multi - atom systems are considered , an additional modification is required in the equation governing the density matrix . we define an effective density matrix by tracing over the states containing high - momentum atoms produced by deeply inelastic reactions . we show that it satisfies a lindblad equation , with local lindblad operators determined by the local anti - hermitian terms in the effective hamiltonian . we use the lindblad equation to derive the universal relation for the two - atom inelastic loss rate for fermions with two spin states and the universal relation for the three - atom inelastic loss rate for identical bosons .

1607.08084

the development of technologies to trap and cool neutral atoms has led to the emergence of cold atom physics as a new subfield of atomic physics . the atoms can be cooled to temperatures that are orders of magnitude smaller than the tiny differences between the hyperfine energy levels of the atoms . many of the most important loss processes for ultracold atoms involve deeply inelastic reactions , which produce atoms with kinetic energies much larger than those of the cold trapped atoms . one such process is three - body recombination , in which a collision of three low - energy atoms results in the binding of two of the atoms into a diatomic molecule . if the binding energy of the molecule is large compared to the energy scale of the cold atoms , the momenta of the molecule and the recoiling atom are often large enough for them to escape from the trapping potential for the atoms . ultracold atoms can be described by a local nonrelativistic effective field theory for which the coupling constant is the scattering length @xcite . local effective field theories can be applied most rigorously to a system in which there is an energy gap separating the low - energy particles described explicitly by the effective field theory from the higher - momentum particles @xcite . in a system consisting of low - energy atoms , conservation of energy ensures that a high - momentum atom can only appear through a virtual state that , by the time - energy uncertainty principle , has a short lifetime . during that short lifetime , the high - momentum atom can propagate only over a short distance . its effects on low - energy atoms are therefore restricted to short distances . these effects can be reproduced by local hermitian operators in the hamiltonian for the effective field theory . local effective field theories have also been applied to systems with deeply inelastic reactions that produce particles with momenta too large to be described accurately within the effective field theory . for example , a deeply inelastic three - body recombination event produces a molecule and an atom whose momenta may be outside the domain of validity of the effective theory . the large energy release from the inelastic reaction comes from the conversion of rest energy into kinetic energy . the standard argument for the locality of the effective field theory does not apply . the particles with large momenta produced by the inelastic reaction can propagate over arbitrarily long distances , so their effects on low - energy particles are not obviously restricted to short distances . nevertheless , there are general arguments based on the uncertainty principle that indicate that their effects can be taken into account systematically through local anti - hermitian operators in the effective hamiltonian @xcite . the effective hamiltonian can be expressed as @xmath0 , where @xmath1 and @xmath2 are hermitian . the dynamics of a multi - atom system with deeply inelastic reactions is conveniently described by a density matrix . a system that starts as a pure quantum state with @xmath3 low - energy atoms evolves into a mixed quantum state that is a superposition of a state with @xmath3 low - energy atoms and states with fewer low - energy atoms , as the inelastic reactions shift probability from the low - energy atoms into the high - momentum atoms . an effective density matrix @xmath4 can be defined by tracing the density matrix over states containing high - momentum atoms . naively we might expect the time evolution equation for @xmath4 to be @xmath5 - i \left\{k,\rho\right\}. \label{eq : drhodt - naive}\end{aligned}\ ] ] as we will demonstrate in this paper , the correct evolution equation for the effective density matrix is the _ lindblad equation _ @xcite , which has an additional term . the lindblad equation arises in the quantum information theory of _ open quantum systems_. an open quantum system consists of all the degrees of freedom of both the subsystem of interest and the environment . under special circumstances , the density matrix for the subsystem evolves in time according to the lindblad equation . in the lindblad equation for the density matrix of an effective field theory obtained by integrating out deeply inelastic reactions , the additional lindblad term is local , and it can be deduced from the local anti - hermitian terms in the effective hamiltonian . an open quantum system in which the subsystem of interest is a field theory is called an _ open effective field theory _ @xcite . grozdanov and polonyi have proposed that an open effective field theory for the hydrodynamic modes of a quantum field theory can be used as a framework for deriving dissipative hydrodynamics @xcite . burgess , holman , tasinato , and williams have applied open effective field theory to the super - hubble modes of primordial quantum fluctuations in the early universe @xcite . in the stochastic inflation framework , the master equation is the lindblad equation . since the density matrix for an effective field theory in which deeply inelastic reaction products have been integrated out satisfies the lindblad equation , this system can also be regarded as an open effective field theory @xcite . in this case , the environment consists of the high - momentum particles produced by the deeply inelastic reactions . the paper is organized as follows . in section [ sec : densitymatrix ] , we summarize the basic properties of the density matrix and we introduce the lindblad equation . in section [ sec : effective theory ] , we explain how integrating out deeply inelastic reaction products results in local operators in the effective lagrangian density . we also explain why the effective density matrix obtained by tracing over states that include deeply inelastic reaction products must satisfy the lindblad equation . in section [ sec : atomloss ] , we apply the lindblad equation to the mean number of low - energy atoms . we derive the universal relation for the two - atom inelastic loss rate for fermions with two spin states and the universal relation for the three - atom inelastic loss rate for identical bosons . our results are summarized briefly in section [ sec : summary ] . in an appendix , we demonstrate how the lindblad equation for the density matrix emerges from the diagrammatic analysis of a simple field theory model with a deeply inelastic two - atom scattering reaction . the density matrix can be used to describe pure quantum states , mixed quantum states , and statistical ensembles of quantum states . the dynamics of a multi - atom system with atom losses must be described by the density matrix in order to be able to track the losses in the different @xmath3-atom sectors . to lay the foundation for the rest of the paper , we summarize the key properties of the density matrix and we introduce the lindblad equation . a pure state in a quantum system can be represented by a vector @xmath6 in a complex vector space . the time evolution of the quantum state is described by the schrdinger equation : @xmath7 , where the hamiltonian @xmath1 is a hermitian operator . the time evolution preserves the norm @xmath8 of a state . an arbitrary statistical ensemble of quantum states can be represented by a density matrix @xmath4 . the density matrix has the following basic properties : * it is hermitian : @xmath9 . * it is positive : @xmath10 for all nonzero states @xmath6 . * it can be normalized to have unit trace : @xmath11 . the density matrix can also describe a pure quantum state , in which case @xmath12 . the time evolution of the density matrix is described by the von neumann equation : @xmath13 , \label{evol : schrodinger}\ ] ] which follows from the schrdinger equation for @xmath6 . this evolution equation has the following properties : * it is linear in @xmath4 . * it preserves the trace of @xmath4 . it therefore respects the normalization condition @xmath11 . * it is _ markovian _ : the future is determined by the present only . there is no additional dependence on the past history . the system average of an operator @xmath14 can be expressed as the trace of its product with the density matrix : @xmath15 if the operator has no explicit time dependence , the time evolution of the system average is determined by the evolution equation of @xmath4 in eq . . in the field of quantum information theory , the full system is often separated into a _ subsystem _ of interest and its _ environment _ @xcite . of particular interest is the decoherence of the subsystem due to the effects of the environment . the basis for the quantum states of the full system can be chosen to be direct products of basis states for the subsystem and those for the environment . a density matrix @xmath4 for the subsystem can be obtained from the density matrix @xmath16 for the full system by the partial trace over the states of the environment : @xmath17 the density matrix for the full system evolves according to the von neumann equation in eq . . given the initial density matrix @xmath18 for the subsystem and a specified initial state of the environment , the von neumann equation can in principle be solved to determine @xmath19 at future times . it is possible in principle to construct a self - contained differential equation for @xmath19 , but its evolution is non - markovian @xcite . the time derivative of @xmath20 is determined by the present density matrix and by its history from time 0 to @xmath21 . the previous history is needed to take into account the flow of information between the subsystem and the environment . there are situations in which the time evolution of the subsystem can be described approximately by a self - contained markovian differential equation . the time during which the subsystem is observed must be much larger than the time scale for correlations between the subsystem and the environment . we must also restrict our consideration to the low - frequency behavior of the subsystem , which can be accomplished by smoothing out over times larger than the correlation time scale . the density matrix @xmath4 for the subsystem necessarily satisfies the three basic properties listed before eq . : it is hermitian , it is positive , and it can be normalized . it would be desirable for its time evolution to also be in accord with the three basic properties listed after eq . : it should be linear in @xmath4 , it should preserve the trace of @xmath4 , and it should be markovian . in 1976 , lindblad and , independently , gorini , kossakowski , and sudarshan showed that these conditions together with one additional technical condition require the time evolution equation to have the form @xcite @xmath22 - \frac{i}{2 } \sum_n \left ( l_n^\dagger l_n \rho + \rho l_n^\dagger l_n -2 l_n \rho l_n^\dagger \right ) , \label{evol : lindblad}\ ] ] where @xmath1 is a hermitian operator and the @xmath23 s are an additional set of operators called _ lindblad operators_. the evolution equation in eq . is called the _ lindblad equation_. the linear operator acting on @xmath4 defined by the right side of eq . is called the _ lindbladian_. thus the lindbladian is the generator of the time evolution of the density matrix . the additional technical condition required to derive the lindblad equation is that the linear operator that determines the time evolution of @xmath4 should be completely positive @xcite . for any complex vector space , this linear operator has a natural extension to an operator acting on the tensor product of the quantum - state spaces for the subsystem and the other complex vector space . complete positivity requires the extension of the operator to be positive in the larger space . an accessible derivation of the lindblad equation can be found in lecture notes on quantum information and computation by john preskill @xcite . the lindblad equation in eq . can be expressed in the form @xmath22 - i \left\{k,\rho\right\ } + i \sum_n l_n \rho l_n^\dagger , \label{evol : lindblad-2}\ ] ] where the hermitian operator @xmath2 is @xmath24 comparison with the naive evolution equation in eq . reveals that @xmath25 can be interpreted as a non - hermitian effective hamiltonian for the subsystem . the additional _ lindblad term _ in eq . is necessary to preserve the trace of @xmath4 . in this section , we review how effective field theories for cold atoms are constructed , with particular focus on the treatment of deeply inelastic reactions . we first discuss the construction of the effective lagrangian , and how integrating deeply inelastic reaction products out of a theory results in new local operators in the effective lagrangian density . we then discuss how such a formalism can be used to study multi - particle states by means of an effective density matrix obtained by tracing over states that include deeply inelastic reaction products . has contributions from intermediate states @xmath26 , where @xmath27 is a different hyperfine state . if @xmath27 is a higher hyperfine state , the virtual atoms can propagate only over short distances , so their effects can be mimicked by local @xmath28 operators . if @xmath27 is a lower hyperfine state , the virtual atoms can propagate over long distances , but the imaginary part of the amplitude is still local and can again be mimicked by local @xmath28 operators . , width=377 ] an effective field theory is obtained by removing ( `` integrating out '' ) states from a field theory . the simplest applications involve removing particles with much higher energies . we consider atoms with mass @xmath29 in a specific hyperfine state labelled @xmath30 and with kinetic energies much smaller than the splitting @xmath31 between hyperfine states . as illustrated in fig . [ fig:2to2-elastic ] , there is a contribution to the elastic scattering amplitude from scattering into a virtual pair of particles in a higher hyperfine state labelled @xmath27 , which then rescatter into atoms in the original hyperfine state . the virtual atoms are off their energy shells by large amounts of order @xmath31 . by the time - energy uncertainty principle , the virtual states have short lifetimes of order @xmath32 , during which the @xmath27 atoms can propagate only over short distances much smaller than the wavelengths of the @xmath30 atoms . the effects of the higher hyperfine states can be modeled in the effective field theory by a contact interaction of the form @xmath33 , where @xmath30 is the quantum field that annihilates an atom in the specific hyperfine state @xmath30 . ( we use the same symbol for the label of the hyperfine state and for its quantum field . ) the @xmath28 contact interaction is the leading term in a series of local operators in the effective lagrangian that mimic the effects of the higher hyperfine states to arbitrary precision . this series is obtained at tree - level by taylor expanding the scattering amplitude in the momentum transfer and then writing down terms in the effective lagrangian density that reproduce this expansion : @xmath34 where the coefficients @xmath35 , @xmath36 , are real valued . the individual operators are renormalized when loop corrections are included , but the effective theory is still capable of reproducing the original theory to arbitrary precision provided operators with a sufficient number of gradients are retained . operators with @xmath3 gradients correct the theory at order @xmath37 , where @xmath38 is the momentum transfer and @xmath39 is the momentum scale associated with the energy splitting @xmath31 between the hyperfine states . we have integrated the higher hyperfine states out of the theory by replacing them by the interactions terms in eq . . , where @xmath27 is a lower hyperfine state , to the rate for the deeply inelastic scattering process @xmath40 . , width=566 ] a less obvious opportunity to remove states from the theory arises when atoms in the hyperfine state @xmath30 scatter inelastically into atoms in a lower hyperfine state @xmath27 . such a scattering process is an example of a _ deeply inelastic reaction _ , in which the difference in the rest energies of the initial - state and final - state atoms is converted into large kinetic energies of the final - state atoms . the rate for the deeply inelastic scattering process @xmath40 is related by the optical theorem to the imaginary part of the amplitude for @xmath41 ( see fig . [ fig:2to2-inelastic ] ) . that amplitude is analytic in the momentum transfer @xmath42 , because the nearest non - analyticity in the amplitude is at the threshold energy for the @xmath26 state , which is lower by an amount of order @xmath43 . as a result , we can taylor expand the amplitude in powers of @xmath44 and reproduce this expansion by terms in the effective lagrangian as in eq . . in this case , the coefficients @xmath35 , @xmath36 , are complex valued . we are particularly interested in their imaginary parts , which come from the deeply inelastic scattering reaction . these terms mimic the effects of the lower hyperfine states in the effective theory . it may seem nonintuitive that the effects of inelastic scattering to on - shell particles can be mimicked by local operators , because once an atom in the lower hyperfine state @xmath27 is created , it can propagate over long distances . in fact , as far as low - energy atoms in the hyperfine state @xmath30 are concerned , the inelastic scattering process is quite local . this is because the reaction region in which the low - energy atoms disappear can be reconstructed by tracking the inelastic scattering products back to their origin . the inelastic reaction products have high momenta of order @xmath45 and therefore short wavelengths of order @xmath46 , so they can locate the decay with a resolution of order @xmath46 . therefore the reaction is localized over a region of size @xmath46 , which is very small compared to the wavelengths of the incoming low - energy atoms . some aspects of the effective field theory obtained by integrating out deeply inelastic reactions are most easily understood by considering a deeply inelastic 1-body reaction , namely the decay of an atom . the atom could be a metastable excited state , such as the @xmath47 state of @xmath48he , which was one of the earliest atoms for which bose - einstein condensation was achieved @xcite . the radiative decay of the atom into its ground state is a deeply inelastic reaction . for simplicity , we assume the ground - state atoms interact weakly with the metastable atoms and that , once they are produced , the ground - state atoms quickly escape from the system . the hamiltonian @xmath49 for the effective theory is the sum of the hermitian hamiltonian for the metastable atoms and an anti - hermitian piece @xmath50 . the hermitian operator @xmath2 can be expressed as @xmath51 , where @xmath52 is the decay rate of the metastable atom and @xmath53 is the number operator that counts the metastable atoms : @xmath54 the quantum mechanics of such a theory is unconventional . the effective hamiltonian @xmath49 commutes with the number operator @xmath53 , so the time evolution generated by @xmath49 does _ not _ change the number of atoms in a state . instead the effects of atom decay must be taken into account by transferring the probability carried by the @xmath55atom component to states containing fewer atoms . an @xmath55atom state evolves into a superposition of states with @xmath3 and fewer atoms . the norm of a state containing @xmath3 atoms decreases to zero exponentially with the decay rate @xmath56 . this is the correct result : if the probability for one atom to still be an atom after time @xmath21 is @xmath57 , the probability for @xmath3 atoms to still be @xmath3 atoms is @xmath58 . we typically want more information about where the probability goes . an @xmath55atom state evolves into a superposition of states with @xmath3 , @xmath59 , @xmath60 atoms . in the full theory , the superposition can be described by the density matrix @xmath16 . in the effective theory , we describe the superposition of states with different atom numbers using an _ effective density matrix _ @xmath4 , from which we remove ground - state atoms by tracing over states that include them : @xmath61 the subscript `` deep '' indicates that the partial trace is over states that include deeply inelastic reaction products . this effective density matrix , like @xmath16 , is hermitian , positive , and has unit trace : @xmath62 . these properties follow from the definition in eq . ( [ eq : rho - eff - defn ] ) . naively we might expect the time evolution equation for @xmath4 to be eq . , which reduces to @xmath63 - \frac{i}{2 } \gamma \left\ { n , \rho\right\ } .\ ] ] however this equation does _ not _ conserve the total probability @xmath64 . the correct evolution equation is the lindblad equation : @xmath65 - \frac{i}{2 } \gamma \left\ { n , \rho\right\ } + i \gamma \int d^3{\bm{r}}\,\psi({\bm{r}})\,\rho\,\psi^\dagger({\bm{r } } ) . \label{eq : muon - evol - eq}\ ] ] time evolution preserves the trace of @xmath4 , because the trace of the additional lindblad term cancels the trace of the anticommutator term in eq . . the role of the lindblad term is easily understood if we use the evolution equation to calculate the rate of change of the probability @xmath66 for finding @xmath3 metastable atoms in the system . this probability can be expressed as the partial trace of @xmath4 over states @xmath67 that contain @xmath3 atoms : @xmath68 the partial trace of the evolution equation in eq . gives @xmath69 the partial trace of the commutator term in eq . is 0 , because the operator @xmath1 does not change the atom number . this allows a complete set of @xmath55atom states to be inserted between @xmath1 and @xmath4 . the partial trace of the anticommutator term in eq . gives @xmath70 , which is the rate at which probability leaves the @xmath55atom sector because of the decay of an atom . the partial trace of the lindblad term in eq . gives @xmath71 , which is the rate at which probability enters the @xmath55atom sector from the decay of an atom in the @xmath72atom sector . this expression can be obtained by inserting complete sets of @xmath72atom states on the left and right of @xmath4 in the lindblad term in eq . ( [ eq : muon - evol - eq ] ) and then rearranging the factors to give @xmath73 the number operator @xmath53 , which is given in eq . , was obtained by replacing the sum of @xmath74 between @xmath75 and @xmath76 by the identity operator . the number operator in eq . can be replaced by its eigenvalue @xmath77 . the lindblad term in eq . ( [ eq : muon - evol - eq ] ) is essential to get the correct physical behavior for the time evolution of the total number of metastable atoms . the expectation value of the atom number is @xmath78 we can use eq . to determine the time dependence of @xmath79 : @xmath80 .\ ] ] after shifting the index of the second term on the right side , we obtain @xmath81 this implies that @xmath82 , as expected for a collection of atoms with decay rate @xmath52 . we now consider a more general effective field theory obtained by integrating out the reaction products from deeply inelastic reactions . by the arguments presented in sec . [ sec : locality ] , the effects of the deeply inelastic reactions can be reproduced by local anti - hermitian terms in the effective hamiltonian . the effective hamiltonian can be expressed in the form @xmath83 , where @xmath1 and @xmath2 are both hermitian operators . the operator @xmath2 can be expressed in the form @xmath84 where the local operators @xmath85 annihilate low - energy atoms in configurations corresponding to inelastic reactions . the real constants @xmath86 can be determined by matching low - energy observables in the full theory and in the effective theory . the hamiltonian @xmath87 commutes with the number operator @xmath53 for low - energy atoms , so the time evolution generated by @xmath49 does _ not _ change the number of low - energy atoms in a state . instead the effects of the deeply inelastic reactions must be taken into account by transferring the probability carried by the @xmath55atom component to states containing fewer atoms . the norm of a state containing @xmath3 atoms decreases to zero at a rate that increases with @xmath3 . an @xmath55atom state evolves into a superposition of states with @xmath3 and fewer atoms . we describe this superposition of states using an _ effective density matrix _ @xmath4 defined by tracing over the deeply inelastic decay products as in eq . . more precisely , we trace out any state containing an atom with momentum exceeding the ultraviolet cutoff of the effective field theory . the effective density matrix , like the underlying density matrix , is hermitian , positive , and it has unit trace : @xmath62 . these properties follow from the definition in eq . ( [ eq : rho - eff - defn ] ) . as is familiar in the field of quantum information @xcite , the definition of the effective density matrix @xmath4 as the partial trace in eq . ( [ eq : rho - eff - defn ] ) implies that @xmath20 satisfies a self - contained time evolution equation that is completely determined by @xmath4 and by the full density matrix @xmath88 at some initial time . this evolution equation is however non - markovian : the time derivative @xmath89 is determined not only by @xmath20 but also by @xmath90 at the earlier times @xmath91 . this dependence on the history takes into account the effects of high - momentum atoms that are produced by inelastic reactions and subsequently interact with the low - energy atoms . however the density matrix @xmath20 for the effective theory does not need to reproduce the full density matrix on short time scales of order @xmath92 . it is sufficient to reproduce its effects after time averaging over time intervals much larger than @xmath92 . this time average removes transients associated with high - momentum atoms that can not be described accurately within the effective field theory . it is only after removing these transients that it is possible to have a density matrix @xmath20 that is markovian . thus the proper definition of the effective density matrix requires that the partial trace in eq . ( [ eq : rho - eff - defn ] ) be supplemented by an appropriate time average that makes @xmath20 markovian . naively we might expect the time evolution equation for @xmath4 to be eq . , where @xmath2 is the hermitian operator in eq . . however , this equation does _ not _ conserve the total probability @xmath64 . the correct evolution equation is the lindblad equation : @xmath93- i \left\{k,\rho\right\ } + 2i \sum_i \gamma_i \int \!\!d^3r\ , \phi_i \rho \phi_i^\dagger . \label{eq : lindbladphi}\ ] ] as in eq . , the trace of the additional lindblad term cancels the trace of the anticommutator term . the local operators @xmath85 , which annihilate configurations of low - energy atoms that can undergo a deeply inelastic reaction , are the _ lindblad operators_. in quantum information theory , a lindblad operator is sometimes called a _ quantum jump operator_. in the low - energy effective theory , a lindblad operator produces a jump to a state with a different low - energy atom number . given the form of @xmath2 in eq . , the lindblad equation is a necessary consequence of our physical requirements on the effective density matrix : @xmath4 is hermitian , positive , has unit trace , and it is markovian . an explicit diagrammatic illustration of the emergence of the lindblad equation from tracing over states that include deeply inelastic reaction products is given in appendix [ sec : deductive ] . in order to obtain the lindblad equation in eq . , it is essential that @xmath2 have the structure shown in eq . ( [ eq : k - phi ] ) . a more generic form for this operator is @xmath94 where @xmath95 is a positive hermitian matrix . it is a hermitian matrix , because @xmath2 is hermitian by definition . it is guaranteed to be a positive matrix by the optical theorem , which implies that the @xmath96 matrix satisfies @xmath97 . matrix elements of @xmath2 reproduce the anti - hermitian parts of scattering amplitudes @xmath98 , where @xmath99 and @xmath100 are states in the effective theory that are connected by intermediate deeply inelastic reaction channels . the hermitian parts of these amplitudes are reproduced by @xmath1 . the optical theorem guarantees the positivity of the anti - hermitian parts . the double sum in eq . is easily rewritten in the canonical form in eq . by expanding @xmath95 in terms of outer products of its eigenvectors . in this section , we use the lindblad equation to determine the time evolution of the mean number of low - energy atoms , which can be changed by deeply inelastic reactions . we also derive universal relations for the inelastic two - atom loss rate for fermions with two spin states and for the inelastic three - atom loss rate for identical bosons . we consider an open effective field theory with effective hamiltonian @xmath101 , where @xmath1 and @xmath2 are hermitian . the hermitian operator @xmath2 in the anti - hermitian part can be expressed in terms of local lindblad operators @xmath85 as in eq . . the time evolution of the density matrix @xmath4 for the effective field theory is given by the lindblad equation in eq . . we reconsider an atom number operator @xmath53 that is conserved by the reactions in the effective field theory but is violated by deeply inelastic reactions in the full theory . since lindbladian time evolution preserves the trace of @xmath4 , the system average of @xmath53 can be expressed as @xmath102 the statement that @xmath53 is conserved by the reactions in the effective theory can be expressed as the commutation relation @xmath103=0 . \label{[n , h]}\ ] ] the inelastic reaction responsible for the @xmath104 term in @xmath2 changes the atom number @xmath53 by some integer @xmath105 . this statement can be expressed as a commutation relation : @xmath106=-k_i \phi_i(\bm{r } ) . \label{[n , phi]}\ ] ] we now consider the time evolution of @xmath107 . by taking the time derivative of @xmath108 , inserting the time evolution equation in eq . , and rearranging the terms , we get @xmath109 ) - \frac{1}{\hbar } \sum_i \gamma_i \int \!\ ! d^3r \ , { \rm tr } \big ( \rho ( n\phi_i^\dagger \phi_i + \phi_i^\dagger \phi_i n -2 \phi_i^\dagger n \phi_i ) \big ) . \label{nevol : lindblad}\end{aligned}\ ] ] the first term on the right side vanishes , because @xmath53 commutes with @xmath1 . by using the commutation relation in eq . in the second term on the right side , the rate of change of atom number reduces to @xmath110 the rate of change in @xmath107 has been expressed as a sum of expectation values of the same terms that appear in the expression for @xmath2 in eq . but multiplied by integers @xmath111 . the atoms that are studied in cold atom experiments can form deeply bound diatomic molecules whose binding energies are orders of magnitude larger than the energy scales of the cold trapped atoms . three - body recombination , in which three low - energy atoms collide and two of them bind into a molecule , is a deeply inelastic reaction if the molecule is deeply bound . the momenta of the outgoing molecule and the recoiling atom are often large enough for them to escape from the trapping potential . in a locally homogeneous gas of atoms in thermal equilibrium , the rate of decrease in the local number density @xmath112 of low - energy atoms can be expressed as a rate equation : @xmath113 where @xmath114 is the 3-body recombination event rate constant , which depends on the temperature @xmath96 . bosonic atoms that are all in the same hyperfine state can be described by an effective field theory with a quantum field @xmath30 . at the low temperatures in cold atom experiments , the only relevant interaction between the atoms is s - wave scattering , whose strength is determined by the scattering length @xmath115 . the interaction term in the hamiltonian is @xmath116 this interaction term can be treated using perturbation theory , provided the scattering length @xmath115 is not much larger than the range of the interactions between the atoms , which for ultracold atoms is the van der waals length scale . we now consider a deeply inelastic three - body recombination reaction that produces a diatomic molecule with binding energy much larger than the temperature @xmath96 . its effect on low - energy atoms can be described in the effective field theory by adding to the effective hamiltonian the anti - hermitian term @xmath117 , where the local hermitian operator @xmath2 is @xmath118 and where @xmath119 is a constant that does not depend on the temperature . the expression for @xmath2 in eq . has a single term with lindblad operator @xmath120 and coefficient @xmath121 . the expression for the atom loss rate in eq . thus has a single term with integer @xmath122 . in a locally homogeneous system of atoms , the local version of the loss rate in eq . is @xmath123 the expectation value of @xmath124 is the local number density : @xmath125 . in a thermal gas , the expectation value of @xmath126 is @xmath127 . comparing with eq . , we see that the constant @xmath119 in eq . is the @xmath128 limit of the 3-body recombination event rate constant @xmath114 . its temperature dependence is negligible , because @xmath129 is much smaller than the binding energy of the molecule . in a bose - einstein condensate at zero temperature , the expectation value of @xmath126 can be expressed as @xmath130 , where @xmath131 is the mean field . the atom loss rate is given by eq . except that the prefactor on the right side is smaller than for a thermal gas by a factor of 1/6 . in cold atom experiments , the scattering length @xmath115 can be controlled experimentally by tuning an external magnetic field to near a feshbach resonance . if @xmath115 is much larger than the range of the interactions between the atoms , which for ultracold atoms is the van der waals length scale , the interactions must be treated nonperturbatively . fermionic atoms in two hyperfine states 1 and 2 with scattering length @xmath115 can be described by an effective field theory with quantum fields @xmath132 and @xmath133 . the interaction term in the hamiltonian can be expressed as @xmath134 where @xmath135 is the bare coupling constant . if the ultraviolet cutoff @xmath136 is imposed on the momenta of virtual atoms , the bare coupling constant is @xmath137 if the pair of atoms in the hyperfine states 1 and 2 has a _ spin - relaxation _ scattering channel into a pair atoms in lower hyperfine states 3 and 4 , the optical theorem implies that the scattering length @xmath115 is complex with a small negative imaginary part . the energy release from the spin - relaxation reaction is much larger than the energy scales for ultracold atoms , so this is a deeply inelastic reaction . the high - momentum atoms in the hyperfine states 3 and 4 can be integrated out to get an effective field theory for low - energy atoms in the hyperfine states 1 and 2 only . the interaction term in the hamiltonian is still that in eq . , except that now @xmath138 has an anti - hermitian part because the complex scattering length @xmath115 makes the bare coupling constant @xmath135 in eq . complex . determining the loss rate of atoms due to the deeply inelastic spin - relaxation reaction is not trivial , because the large scattering length makes the problem nonperturbative . however , an exact result for the inelastic two - atom loss rate in any state has been proposed by shina tan @xcite . the rates of decrease in the numbers @xmath139 and @xmath140 of atoms in the two hyperfine states are given by the universal relation @xmath141 where @xmath142 is a property of the system called the _ contact_. the coefficient of @xmath142 , which is proportional to the imaginary part of @xmath143 , is determined by 2-body physics . the contact was first introduced by shina tan in 2005 @xcite . it is the thermodynamic variable conjugate to @xmath143 . the contact @xmath142 is an extensive variable : it is the integral over space of the contact density @xmath144 , which measures the number of 1 - 2 pairs per ( volume)@xmath145 . ( the unusual power of volume arises from an anomalous dimension associated with a non - gaussian renormalization group fixed point @xcite . ) shina tan derived other universal relations involving the contact that hold for any state of the system @xcite . they include the `` adiabatic relation '' that specifies how the free energy of a system is affected by a change in the scattering length @xcite : @xmath146 many other universal relations involving the contact were subsequently derived @xcite . the universal relation for the inelastic two - atom loss rate in eq . ( [ uni2loss ] ) can be derived by expressing the hermitian operator @xmath2 in the effective hamiltonian @xmath147 as @xmath148 where the local lindblad operator is @xmath149 the operator product @xmath150 is singular : its matrix elements diverge linearly with the cutoff @xmath136 . the multiplication by @xmath135 , which according to eq . decreases as @xmath151 for large @xmath136 , makes @xmath152 a finite operator whose matrix elements have well - behaved limits as @xmath153 . in a spin - relaxation scattering reaction , the decreases in the atom numbers @xmath139 and @xmath140 are both @xmath154 . according to eq . , the rates of decrease in their system averages are therefore @xmath155 the universal loss rate in eq . ( [ uni2loss ] ) then follows from the identification of the contact density as @xcite @xmath156 this identification can be verified by using the adiabatic relation in eq . . the only dependence of the hamiltonian on the scattering length @xmath115 is in the interaction term in the hamiltonian density @xmath157 in eq . . if the tiny imaginary part of the scattering length is neglected , the derivative of the hermitian part @xmath1 of the effective hamiltonian with respect to @xmath143 is @xmath158 by the feynman - hellman theorem , the system average of the left side of this equation is the left side of the adiabatic relation in eq . . with the identification of the contact density in eq . , the system average of the right side of eq . is the right side of the adiabatic relation in eq . this completes the derivation of the inelastic two - atom loss rate in eq . ( [ uni2loss ] ) and the adiabatic relation in eq . . in ref . @xcite , the authors presented an incorrect derivation of the universal relation for the inelastic two - atom loss rate that suggests that there are additional terms on the right side of eq . . they assumed the density matrix @xmath4 satisfies the naive time evolution equation in eq . . this equation implies that the number of atoms is conserved , but that the probability @xmath159 decreases with time . in ref . @xcite , the authors assumed that the mean atom number @xmath107 was given by @xmath160 . the resulting expression for the time derivative of @xmath107 can be expressed as a double integral over space of an expression involving the product of the number density operator and the hamiltonian density operator . upon applying the operator product expansion to the product of these operators , they obtained an expansion for @xmath161 in terms of increasingly higher - dimension operators with the leading term given by eq . . the derivation of the universal relation using the lindblad equation makes it clear that there are no additional terms beyond the contact term in eq . ( [ uni2loss ] ) . bosonic atoms that are all in the same hyperfine state can be described by a nonrelativistic quantum field theory with a quantum field @xmath30 @xcite . if the scattering length @xmath115 is much larger than the range of interactions of the atoms , two - atom interactions and three - atom interactions must both be treated nonperturbatively . the resulting quantum field theory is characterized by a discrete scaling symmetry in which lengths and time are multiplied by integer multiples of @xmath162 and @xmath163 , respectively , where @xmath164 and @xmath165 is a universal number . three - atom interactions are determined by a parameter @xmath166 with dimensions of momentum , upon which physical observables can only depend log - periodically . if @xmath115 is positive , two - atom interactions produce a weakly bound diatomic molecule called the _ shallow dimer_. regardless of the sign of @xmath115 , three - atom interactions produce a sequence of weakly bound triatomic molecules called _ efimov trimers _ @xcite . in ref . @xcite , several universal relations were derived that hold for any system consisting of identical bosons . the universal relations relate various properties of the system to two extensive thermodynamic variables : the 2-body contact @xmath167 , which is the analog of the contact for fermions with two spin states , and the 3-body contact @xmath168 . the atoms that are studied in cold atom experiments form deeply bound diatomic molecules ( _ deep dimers _ ) whose binding energies are orders of magnitude larger than the energy scales of the cold trapped atoms . the deep dimers provide various pathways for deeply inelastic reactions that result in the loss of three atoms . they include ( a ) three - body recombination , in which three low - energy atoms collide and two of them bind into a deep dimer , ( b ) dimer relaxation , in which a shallow dimer and an atom scatter into a deep dimer and an atom , and ( c ) efimov trimer decay into a deep dimer and an atom . in ref . @xcite , a universal relation for the three - atom inelastic loss rate was presented but not derived . in this section , we use the lindblad equation to sketch the derivation of this relation . to derive the universal relation for the three - atom inelastic loss rate , we follow the formalism laid out in ref . @xcite . in the absence of deep dimers , the interaction term in the hamiltonian can be written as @xmath169 since this effective field theory has ultraviolet divergences , it must be regularized . the bare coupling constants @xmath170 and @xmath171 depend on an ultraviolet momentum cutoff @xmath136 : @xmath172 where @xmath115 is the two - body scattering length of the bosons and @xmath166 is a three - body parameter introduced in ref the three - body parameter @xmath166 can be determined from any three - body datum , such as the binding energy of an efimov trimer or the atom - dimer scattering length . ( see ref . @xcite for explicit relations ) . the numerical prefactor @xmath173 in eq . depends on the ultraviolet cutoff prescription and has the value @xmath174 if @xmath136 is a sharp momentum cutoff @xcite . three - atom losses from deeply inelastic reactions involving deep dimers can be taken into account by adding an imaginary part to the coupling constant @xmath171 in the hamiltonian in eq . . the resulting anti - hermitian part of the hamiltonian is @xmath117 , where @xmath2 is @xmath175 physical observables can be expressed particularly conveniently in terms of @xmath166 and an additional real three - body loss parameter @xmath176 @xcite defined by analytically continuing the three - body coupling constant @xmath171 in eq . to complex values using the substitution @xmath177 inserting @xmath122 into the expression in eq . for the three - atom inelastic loss rate , we obtain the universal relation @xmath178 where @xmath168 is the three - body contact , which can be expressed as @xmath179 the prefactor @xmath180 depends on the ultraviolet cutoff @xcite . since its precise form is not very insightful , it will not be given here . the three - body contact can also be defined in terms of the expectation value of the logarithmic derivative with respect to @xmath166 of the hermitian part of the effective hamiltonian in eq . at fixed @xmath115 @xcite : @xmath181 the universal relation in eq . , with the factor @xmath182 approximated by @xmath183 , was previously given in ref . it was also given previously in ref . @xcite for the special case of an efimov trimer state . in this paper , we have shown that integrating out the high - momentum atoms produced by deeply inelastic reactions produces an open effective field theory in which the time evolution of the effective density matrix is given by the lindblad equation . the effective density matrix can be obtained from the density matrix of the full theory by taking a partial trace over states that include high - momentum atoms and then carrying out an appropriate time average to eliminate high - frequency transients . the lindblad operators are local , and they can be deduced from the anti - hermitian terms in the effective hamiltonian . the lindblad terms in the evolution equation are essential to get the correct evolution equation for the mean atom number . we used the lindblad equation to present the first correct derivation of the universal relation for the two - atom inelastic loss rate for fermionic atoms with two spin states that interact through a large scattering length . we also used the lindblad equation to present the first derivation of the universal relation for the three - atom inelastic loss rate for identical bosonic atoms with a large scattering length . the lindblad equation has many other applications to atom loss processes in cold atom physics . an obvious extension is to heteronuclear systems , for which there are two types of efimov trimers . due to the smaller discrete scaling factor associated with one type of efimov trimer , atom loss processes in such systems have recently been of great theoretical and experimental interest @xcite . the lindblad equation could be applied to fermionic atoms with two spin states on the upper branch of the two - atom spectrum at scattering lengths for which three - body recombination into the shallow dimer is a deeply inelastic reaction @xcite . it could also be applied to losses of dipolar atoms from three - body recombination into deep dimers @xcite . e.b . and h .- w.h . acknowledge useful discussions with shina tan . the research of e.b . was supported in part by the department of energy under grant de - fg02 - 05er15715 , by the national science foundation under grant phy-1310862 , and by the simons foundation . the research of h .- w.h . was supported in part by the bmbf under contract 05p15rdfn1 , and by the deutsche forschungsgemeinschaft ( sfb 1245 ) . the research of g.p.l . was supported in part by the national science foundation . in this section , we demonstrate how the lindblad structure of the evolution equation for the density matrix emerges from a diagrammatic analysis of a simple field theory model with a deeply inelastic 2-body reaction . to reduce visual clutter , we introduce compact notation for the integral over space and for the integral over a momentum : @xmath184 we also set @xmath185 . we consider a quantum field theory with two fields @xmath30 and @xmath27 . we refer to the particles annihilated by these field operators as @xmath30 atoms and @xmath27 atoms , respectively . the field operators satisfy canonical commutation relations if the atoms are bosons and canonical anticommutation relations if they are fermions . the hamiltonian for the full theory has the form @xmath186 the free - field terms in the hamiltonian are where @xmath29 is the mass of the atoms . we have chosen the rest energies of a @xmath30 atom and a @xmath27 atom to be zero and @xmath188 , respectively . the on - shell energies of atoms with momentum @xmath189 are the interaction hamiltonian includes a term that allows a pair of @xmath30 atoms to scatter into a pair of @xmath27 atoms : @xmath191 where the ellipses denote further local interaction terms , such as @xmath192 and @xmath193 . their precise form will not be needed in the following discussion . the leading contribution to the transition amplitude for the process @xmath194 from inelastic reactions with a @xmath195 intermediate state comes from the imaginary part of the one - loop diagram in fig . [ fig : scattering ] . the energy release in the reaction @xmath196 is @xmath197 , and the corresponding momentum scale is @xmath198 . we are interested in systems consisting of @xmath30 atoms whose momenta are small compared to @xmath199 . the reaction @xmath196 is therefore a deeply inelastic reaction . we refer to the momentum scale @xmath199 as _ high momentum_. because of the large energy release , the deeply inelastic scattering process @xmath196 is effectively local and instantaneous . it takes place over a spatial region of size @xmath202 and during a time interval of length @xmath203 . we proceed to show how this locality can be exploited to remove high - momentum @xmath27 atoms from the theory and construct a low - energy effective field theory for @xmath30 atoms . we do this first for two @xmath30 atoms in this subsection and then for a system containing many @xmath30 atoms in subsection [ sec : effrho ] . for simplicity , we assume the coupling @xmath35 is small and we work to leading order in @xmath35 . the leading contribution to the @xmath194 scattering amplitude from a @xmath195 intermediate state is given by the diagram in fig . [ fig : scattering ] . using time - ordered perturbation theory ( or using feynman perturbation theory and integrating by contours over the loop energies ) , the off - shell scattering amplitude is @xmath204 where @xmath205 and @xmath189 are the total energy and the total momentum of the two incoming @xmath30 atoms , respectively . for convenience , we consider @xmath194 scattering in the center - of - mass frame . in this reference frame , the incoming @xmath30 atoms have momenta @xmath206 and energies @xmath207 . the on - shell scattering amplitude is therefore @xmath208 . the total cross section for @xmath196 scattering can be obtained using the optical theorem by evaluating the imaginary part of @xmath96 on the energy shell . the real part of the scattering amplitude in eq . is ultraviolet divergent . the divergence can be canceled by a renormalization of the coupling constant for the @xmath196 interaction . after renormalization , the integral over @xmath42 in eq . is dominated by high momenta of order @xmath199 . consequently we can expand the on - shell scattering amplitude @xmath208 in powers of @xmath209 . we are primarily interested in constructing an effective hamiltonian that takes into account the leading term @xmath210 . successively higher powers of @xmath209 could be taken into account through successively higher - order gradient terms in the effective hamiltonian . similarly , successively higher powers of @xmath211 in the expansion of @xmath96 about the on - shell energy could be taken into account through successively higher - order time - derivative terms in the effective lagrangian . in the sector with only two @xmath30 atoms , the leading effect of the scattering amplitude @xmath212 can be taken into account by adding a local term to the free hamiltonian @xmath213 . the resulting effective hamiltonian is @xmath214 this equation defines the hermitian operators @xmath1 and @xmath2 . the real part of @xmath210 , which is ultraviolet - divergent , can be cancelled by renormalizing the coupling constant for the @xmath215 interaction term contained in the ellipses in eq . ( [ eq : hint ] ) . the anti - hermitian part @xmath50 of the effective hamiltonian comes from the imaginary part of @xmath210 , which is @xmath216 the locality of the inelastic scattering process , which implies @xmath217 , allows us to simplify the contributions to correlators involving the operator @xmath218 from its annihilation of @xmath27 atoms that come from @xmath219 scattering . for example , the correlator @xmath221 illustrated in fig . [ fig : scattering - simp ] can be simplified as follows : @xmath222 where @xmath96 is given in eq . ( [ psi - selfenergy ] ) . generally , we can make the following replacements in such situations : replacing the free hamiltonian @xmath213 by the effective hamiltonian @xmath49 is all that is needed to analyze the impact of inelastic scattering processes on states with only two @xmath30 atoms . analyzing multi-@xmath30 states is more complicated , however , because a system that is described initially by a state with @xmath53 @xmath30 atoms evolves into a superposition of states with @xmath53 , @xmath224 , @xmath225 , @xmath30 atoms . the state with two @xmath30 atoms also evolves into a superposition , but there are only two states , @xmath226 and @xmath227 , and we do not care about the second one . for @xmath228 , we need the density matrix @xmath229 to track the superposition of states containing different numbers of @xmath30 atoms over time . where @xmath232 and @xmath233 are the vacuum states annihilated by @xmath234 and @xmath235 , respectively . the full density matrix @xmath16 can be expanded in the basis of direct product states . its time evolution is given by @xmath236 where @xmath237 is the full hamiltonian in eq . ( [ eq : hamiltonian ] ) . we can define an effective density matrix @xmath20 by tracing over the @xmath27 states : this operator acts only on @xmath30 states . a convenient basis for the effective density matrix consists of outer products of the @xmath30 states defined in eq . of the form @xmath239 . the time derivative of the effective density matrix can be obtained by differentiating eq . and using the time dependence of the full density matrix in eq . : @xmath240 the contributions from the kinetic terms in the full hamiltonian in eq . ( [ eq : hamiltonian ] ) are simple : the first equation holds because @xmath213 does not act on @xmath27 states . the second equation holds because @xmath242 depends only on @xmath27 fields . holds for any operator @xmath243 constructed out of the field @xmath27 and any operator @xmath244 . this can be verified by expressing the partial trace as a sum over a complete set of @xmath27 states and inserting a complete set of @xmath27 states between @xmath243 and @xmath245 . ] the evolution equation reduces to @xmath246 + \mathrm{tr}_\phi\big(h_{\rm int}\,\rho_{\mathrm{full}}- \rho_{\mathrm{full}}\,h_{\rm int } \big).\ ] ] since there are two terms in the interaction hamiltonian in eq . , there are four contributions to the partial trace in eq . . we first consider the contribution to the partial trace of @xmath247 from the first term in @xmath248 in eq . ( [ eq : hint ] ) . this contribution is dominated by terms in which @xmath249 annihilates @xmath27 atoms generated ( through interactions ) by the @xmath30-sector of @xmath16 , leading to correlators like eq . ( [ eq : phi2psid ] ) . we can therefore use the substitution in eq . ( [ eq : phi^2approx ] ) to obtain @xmath250 \longrightarrow -\tfrac{1}{4}t(0,0)\int_{\bm{r } } \psi^{\dagger 2}(\bm{r})\psi^2(\bm{r } ) \ ; \rho.\end{aligned}\ ] ] similarly , we can use the substitution in eq . ( [ eq : phid^2approx ] ) to obtain the contribution to the partial trace of @xmath251 from the second term in @xmath248 : @xmath252 \longrightarrow -\tfrac{1}{4}t^*(0,0)\int_{\bm{r } } \rho \ ; \psi^{\dagger 2}(\bm{r})\psi^2(\bm{r}).\end{aligned}\ ] ] the contributions to the sum of the partial traces in eqs . and from the real part of @xmath210 can be added to the term @xmath253 $ ] in eq . to get @xmath254 $ ] , where @xmath1 is the hermitian part of the effective hamiltonian defined in eq . ( [ eq : heff ] ) . the contributions to the sum of the partial traces in eqs . and from the imaginary part of @xmath210 gives @xmath255 , where @xmath50 is the anti - hermitian part of the effective hamiltonian defined in eq . ( [ eq : heff ] ) . note that we are making a key physical assumption about @xmath16 when we use the substitution in eq . ( [ eq : phi^2approx ] ) to replace @xmath256 by @xmath257 . as we indicated above , this substitution is valid provided @xmath249 annihilates @xmath27 atoms produced by the @xmath30-sector of @xmath16 . in principle , it is also possible for @xmath249 to annihilate @xmath27 atoms from the @xmath27-sector of @xmath16 . we assume that such contributions can be ignored because the probability for finding two @xmath27 atoms at the same space - time point is vanishingly small ( and therefore the probability of an inverse reaction , @xmath258 , is negligible ) . this is the case , for example , if @xmath16 describes a situation in which all @xmath27 atoms are produced by inelastic @xmath219 reactions and , once produced , they either escape from the system or they interact so weakly with the low - energy @xmath30 atoms that they decouple . the contribution to the partial trace of @xmath247 from the second term in the interaction hamiltonian @xmath248 in eq . is @xmath259 = \mathrm{tr}_\phi\left[\tfrac{1}{2}g\int_{\bm{r}}\psi^2(\bm{r})\rho_{\mathrm{full}}\phi^{\dagger 2}(\bm{r})\right].\end{aligned}\ ] ] in the second expression , the factor of @xmath260 has been moved to the right side of @xmath16 . to verify this equality , we begin by inserting the definition of the partial trace on the left side of eq . : @xmath261 the @xmath262 factors remove @xmath27 atoms from the bras on the left side of this equation . taking into account the symmetry under permutations of the integration variables , their effect can be taken into account by the substitution @xmath263 the kets on the right side of eq . can be expressed as @xmath264 making the substitutions in eqs . ( [ eq : brasub ] ) and and using the delta functions to integrate over @xmath265 and @xmath266 , we obtain the expression on the right side of eq . . upon making the substitution in eq . , we obtain @xmath267 \longrightarrow -\tfrac{1}{4}t^*(0,0 ) \int_{\bm{r } } \psi^2(\bm{r})\,\rho\ , \psi^{\dagger 2}(\bm{r}).\end{aligned}\ ] ] similarly , the contribution to the partial trace of @xmath251 from the second term in @xmath248 in eq . ( [ eq : hint ] ) is @xmath268 \longrightarrow -\tfrac{1}{4}t(0,0)\int_{\bm{r } } \psi^2(\bm{r})\,\rho\,\psi^{\dagger 2}(\bm{r}).\end{aligned}\ ] ] the real part of @xmath210 cancels in the difference between the contributions in eqs . and . the contributions from the imaginary part of @xmath210 to the partial trace in eq . is proportional to the integral over space of the operator @xmath269 . adding the four contributions in eqs . , , , and to the partial trace in eq . , we see that the evolution equation for the effective density matrix is @xmath270 - \frac{i}{4}\,{\rm im}t(0,0 ) \int_{\bm{r } } \big [ \psi^{\dagger2}\psi^2(\bm{r})\ , \rho + \rho\ , \psi^{\dagger2}\psi^2(\bm{r } ) -2\ , \psi^2(\bm{r})\ , \rho\ , \psi^{\dagger2}(\bm{r } ) \big],\end{aligned}\ ] ] where @xmath271 is given in eq . . this has the standard form of the lindblad equation . the last term removes @xmath30 atoms two at a time to account for their disappearance due to inelastic scattering into pairs of high - momentum @xmath27 atoms . e. braaten and h .- w . hammer , universality in few - body systems with large scattering length , phys . rept . * 428 * , 259 ( 2006 ) [ cond - mat/0410417 ] . h. georgi , effective field theory , ann . nucl . part . sci . * 43 * , 209 ( 1993 ) . kaplan , effective field theories , nucl - th/9506035 . manohar , effective field theories , lect . notes phys . * 479 * , 311 ( 1997 ) [ hep - ph/9606222 ] . r. shankar , effective field theory in condensed matter physics , in _ conceptual foundations of quantum field theory _ , ed . cao ( cambridge university press , 1999 ) [ cond - mat/9703210 ] . burgess , introduction to effective field theory , ann . nucl . part . sci . * 57 * , 329 ( 2007 ) [ hep - th/0701053 ] . e. braaten , h .- w . hammer , and g.p . lepage , open effective field theories from deeply inelastic reactions , arxiv:1607.02939 . g. lindblad , on the generators of quantum dynamical semigroups , commun . math . phys . * 48 * , 119 ( 1976 ) . v. gorini , a. kossakowski , and e.c.g . sudarshan , completely positive dynamical semigroups of @xmath53-level systems , j. math . * 17 * , 821 ( 1976 ) . s. grozdanov and j. polonyi , viscosity and dissipative hydrodynamics from effective field theory , phys . d * 91 * , 105031 ( 2015 ) [ arxiv:1305.3670 ] . burgess , r. holman , g. tasinato , and m. williams , eft beyond the horizon : stochastic inflation and how primordial quantum fluctuations go classical , jhep * 1503 * , 090 ( 2015 ) [ arxiv:1408.5002 ] . burgess , r. holman , and g. tasinato , open efts , ir effects and late - time resummations : systematic corrections in stochastic inflation , jhep * 1601 * , 153 ( 2016 ) [ arxiv:1512.00169 ] . j. preskill , lecture notes for physics 229 : quantum information and computation ( unpublished ) , chapter 3 . f. pereira dos santos , j lonard , ju . wang , c.j . barrelet , f. perales , e. rasel , c.s . unnikrishnan , m. leduc , and c. cohen - tannoudji , bose - einstein condensation of metastable helium , phys . lett . * 86 * , 3459 ( 2001 ) [ cond - mat/0103387 ] . e. braaten , universal relations for fermions with large scattering length , lect . notes phys . * 836 * , 193 ( 2012 ) [ arxiv:1008.2922 ] . e. braaten and l. platter , exact relations for a strongly interacting fermi gas from the operator product expansion , phys . lett . * 100 * , 205301 ( 2008 ) [ arxiv:0803.1125 ] . e. braaten and h .- w . hammer , universal relation for the inelastic two - body loss rate , j. phys . b * 46 * , 215203 ( 2013 ) [ arxiv:1302.5617 ] . e. braaten , d. kang , and l. platter , universal relations for identical bosons from 3-body physics , phys . lett . * 106 * , 153005 ( 2011 ) [ arxiv:1101.2854 ] . d. h. smith , e. braaten , d. kang , and l. platter , two - body and three - body contacts for identical bosons near unitarity , phys . lett . * 112 * , 110402 ( 2014 ) [ arxiv:1309.6922 ] . bedaque , h .- w . hammer , and u. van kolck , renormalization of the three - body system with short range interactions , phys . lett . * 82 * , 463 ( 1999 ) [ nucl - th/9809025 ] . e. braaten and h .- w . hammer , enhanced dimer deactivation in an atomic / molecular bec , phys . rev . a * 70 * , 042706 ( 2004 ) [ cond - mat/0303249 ] . f. werner and y. castin , general relations for quantum gases in two and three dimensions . ii . bosons and mixtures , phys . a * 86 * , 053633 ( 2012 ) [ arxiv:1210.1784 ] . k. helfrich , h .- w . hammer , and d.s . petrov , three - body problem in heteronuclear mixtures with resonant interspecies interaction , phys . a * 81 * , 042715 ( 2010 ) [ arxiv:1001.4371 ] . tung , k. jimenez - garcia , j. johansen , c.v . parker , and c. chin , geometric scaling of efimov states in a @xmath272li@xmath273cs mixture , phys . lett . * 113 * , 240402 ( 2014 ) [ arxiv:1402.5943 ] . r. pires , j. ulmanis , s. hfner , m. repp , a. arias , e.d . kuhnle , and m. weidemller , observation of efimov resonances in a mixture with extreme mass imbalance , phys . lett . * 112 * , 250404 ( 2014 ) [ arxiv:1403.7246 ] .

variability in the infrared emission from disks around pre - main sequence stars over the course of days to weeks appears to be common , but the physical cause of the changes in disk structure are not constrained . here we present coordinated monitoring of one young cluster with the _ spitzer _ and _ chandra _ space telescopes aimed at studying the physical source of the variability . in fall 2011 we obtained ten epochs of chandra acis photometry over a period of 30 days with a roughly 3 day cadence contemporaneous with 20 epochs of spitzer [ 3.6],[4.5 ] photometry over 40 days with a roughly 2 day cadence of the ic 348 cluster . this cadence allows us to search for week to month long responses of the infrared emission to changes in the high - energy flux . we find no strong evidence for a direct link between the x - ray and infrared variability on these timescales among 39 cluster members with circumstellar disks . there is no significant correlation between the shape of the infrared and x - ray light curves , or between the size of the x - ray and infrared variability . among the stars with an x - ray flare none showed evidence of a correlated change in the infrared photometry on timescales of days to weeks following the flare . this lack of connection implies that x - ray heating of the planet forming region of the disk is not significant , although we can not rule out rapid or instantaneous changes in infrared emission .

1407.4985

the inner au around a pre - main sequence star is a complex region . dust and gas flow inward through the disk , possibly accompanied by massive planets built in the outer disk ( e.g. * ? ? ? close to the star the dust will sublimate , creating an inner wall at a few tenths of an au @xcite . the gas will flow within this radius @xcite until it is loaded on the large dipolar stellar magnetic field and lifted out of the midplane . once locked onto the stellar field this material free - falls onto the star , creating a shock as it strikes the stellar surface and heating a small spot to well above the photospheric temperature @xcite . add in the possibility of a significant wind being driven outward from close to the star ( e.g. * ? ? ? * ) , and one can begin to understand the wide range of physical processes at work close to the star . synoptic observations point toward even more complexity . models ( e.g. * ? ? ? * ) and observations ( e.g. * ? ? ? * ) of accretion variability indicate that the flow of material onto the star fluctuates on short timescales . hot and cold spots dot the stellar surface leading to rapid optical variability ( e.g. * ? ? ? large scale disk instabilities lead to outbursts lasting years to decades ( e.g. * ? ? ? * ; * ? ? ? * ) and frequent fluctuations at a wide range of infrared wavelengths indicate structural changes at or near the sublimation edge of the circumstellar disk on weekly timescales ( e.g. * ? ? ? * ; * ? ? ? * ; * ? ? ? the frequency of moderate x - ray flares , believed to arise from magnetic reconnection events in the coronae @xcite , indicate that the magnetic field is not only complex but constantly varying . even single epoch measurements of the magnetic field find that not all young stars have strictly dipolar magnetic fields ; some are dominated by the octopolar or quadropolar components close to the stellar surface @xcite . here we attempt to take advantage of the infrared fluctuations to study the disk structure in more detail . by searching for correlations between the infrared variability and other forms of variability ( x - ray , accretion , etc . ) we can directly study the influence of various factors that may set the structure of the dusty disk . recent observations have found that the occurrence rate of infrared variability increases with x - ray luminosity @xcite , suggestive of some sort of connection between the two . enhanced x - ray emission was observed following the @xmath0few magnitude outburst of the fu ori object v1647 ori @xcite , suggesting a direct connection between the change in disk structure and enhanced x - ray emission during these large accretion bursts . x - rays are also likely an important contributor to the clearing of the disk by photo - evaporation within the first few million years @xcite . high - energy x - ray emission may influence the structure of the disk directly through heating and ionization @xcite or may serve as an indirect tracer of rearrangements of the magnetic field ( e.g. * ? ? ? previous theoretical studies have focused mainly on the gas , where the x - ray illumination is most strongly felt @xcite but it may penetrate into the dust layer and heat and/or ionize the dust . the frequent flaring of x - ray emission , with factors of 10 increase in x - ray emission a common occurrence @xcite , means that this illumination of the dust is constantly , and strongly , variable . @xcite find that roughly half of the stars in taurus exhibit x - ray variability , while @xcite find that solar - type stars in orion spend @xmath030% of their time in an elevated state . this x - ray variability is a mix of rotational modulation @xcite and flares @xcite with larger fluctuations among stars with disks @xcite . young stellar objects have been found to have very strong magnetic fields @xcite that likely interact with the disk , although this interaction has been difficult to observe directly . the structure of the magnetic field can be very complex ( e.g. * ? ? ? * ; * ? ? ? * ) and even among well studied stars it can be hard to characterize observationally ( e.g. * ? ? ? * ) . given the complex , but prominent , nature of the magnetic field , studying an indirect tracer like x - ray emission may shed light on its structure and how these structural changes affect the planet forming region in a large sample of young stellar objects . contemporaneous changes in x - ray flux and infrared emission from the disk would point towards either a direct influence of the x - ray photons on the disk , or changes in magnetic field structure that affect both the high - energy flux and the emission from the disk . @xcite report a difference in the relative size of x - ray fluctuations between stars with disks and diskless systems . @xcite did not find a direct correlation between x - ray and near - infrared variability in the coronet cluster , but this survey was limited to observations over the course of only one week of a small handful of objects . the diversity in magnetic field and circumstellar disk properties among young stellar objects suggests that a small sample may not be able to capture the full range of interactions . by extending this type of analysis to a larger sample ( @xmath0100 vs 10 objects ) on longer timescales ( 40 days vs one week ) at a wavelength that has less contamination from stellar photospheric emission ( 3 - 5 vs. 1 - 2 ) we can broadly address the interactions of the dusty disk and the magnetic field / x - ray emission . to do this we obtained coordinated chandra and spitzer observations of the ic 348 cluster . this 2 - 3 myr old cluster contains @xmath0300 known cluster members , 50% of which contain an infrared excess indicative of circumstellar material @xcite . in section 2 we present our observations , in section 3 we search for correlated variability and in section 4 we discuss the implications of these results . the field was observed by @xmath1 10 times between oct 17 , 2011 and nov 17 , 2011 . the nominal 4-chip acis - i array was used in `` very faint '' mode . each observation was about 10 ks and centered at 03:44:31.50 + 32:08:33.70 ( j2000 . ) roll was left nominal so not all sources were in the field of view for all exposures ( see table [ xray_log ] for full details ) . the data were processed through the standard ciao pipeline at the chandra x - ray center , using their software version d8.4 . this version of the pipeline automatically incorporated a noise correction for low energy events . background was nominal and non - variable . since the focus was on the bright sources which may vary significantly , sources were identified in the individual frames . to identify point sources , photons with energies below 300 ev and above 8.0 kev were filtered out from this merged event list . this excluded energies which generally lack a stellar contribution . by filtering the data as described , contributions from hard , non - stellar sources such as x - ray binaries and agn are attenuated , as is noise . a monochromatic exposure map was generated in the standard way using an energy of 1.49 kev which is a reasonable match to the expected peak energy of the stellar sources and the @xmath1 mirror transmission . the ciao tool wavdetect was then run on a series of flux corrected images binned by 1 , 2 and 4 pixels . the thresholds were set to limit false detections to about 1 per 100 detections . the output resulted in the detection of between 119 -139 sources ( mean 135.5 ) at each source position an extraction ellipse was calculated following @xcite updated for the appropriate roll . this provided an extraction ellipse containing 95% of the source flux within the 0.3 - 8 kev range . for each of the sources , a background ellipse was identified . the background was an annular ellipse with the same center , eccentricity , and rotation as the source . the outer radius was 6 times the radius of the source and the inner radius was 3 times larger than the source . from this region any nearby sources were subtracted with ellipses 3 times the size of the source ellipse . the net counts were calculated by subtracting the background counts ( corrected for area ) and multiplying the result by 1.053 to correct for the use of a 95% encircled energy radius . we also repeatedly observed ic 348 with the _ spitzer _ space telescope using the irac instrument during the fall 2011 observing window ( pid60160 ) . the observing strategy was identical to that described in @xcite , and here we provide a brief summary . both [ 3.6 ] and [ 4.5 ] photometry were obtained simultaneously over the course of 20 epochs from oct 15 , 2011 to nov 23 , 2011 ( table [ obs_log ] ) . the cadence was chosen to trace the daily to weekly fluctuations in infrared emission with one observation every two days . the cluster was mapped in a 5 by 5 grid , with each grid point separated by 360 " and the total field centered at 03:44:20 + 32:03:01 . since the fields of view of the [ 3.6 ] and [ 4.5 ] arrays do not completely overlap , our final map covered by both bands was not a square and instead had a total area of roughly 0.4@xmath2 by 0.25@xmath2 . images were taken in hdr mode with a frame time of 12 seconds and 3 cycles at each position . the irac data reduction pipeline is described in @xcite and @xcite with updates appropriate for the warm spitzer mission described in gutermuth et al . in prep . as a nearby ( @xmath0320 pc ) , young ( 2 - 3 myr ) , relatively compact cluster ic 348 has proven to be a valuable laboratory for studying young stellar object variability @xcite . with over 300 cluster members , only one of which is earlier than a0 , this region of low mass star formation contains a wide variety of circumstellar structures from pre - natal envelopes to disks with many - au wide gaps @xcite . in the infrared , almost 60% of the sources with an infrared excess are variable at [ 3.6 ] , [ 4.5 ] @xcite , while 89% of the pre - main sequence stars brighter than i=14 are variable in the optical @xcite on timescales of weeks to months . the frequency of these fluctuations indicates that the majority of pre - main sequence stars experience variability on their stellar surface , probed by the optical data , and within an au of the star , probed by the infrared data . in our search for correlated x - ray and infrared variability , we focus on cluster members that have the most complete infrared and x - ray time - series data . cluster members , and their stellar properties ( l@xmath3,t@xmath4 , etc . ) , are drawn from @xcite as compiled by @xcite . there are a total of 133 cluster members detected on more than one x - ray epoch , with 107 detected on more than four epochs . we restrict ourselves to these 107 stars because they have enough epochs for studying variability . while we include x - ray fluxes from previous studies of ic 348 @xcite , our analysis focuses on the data that are contemporaneous with the infrared data . as a sub - sample of the cluster , the x - ray detected sources are incomplete below l=0.3l@xmath5 and t@xmath6=3600 k ( fig [ demographics ] ) . within ic 348 , the x - ray luminosity is typically 3 to 4 orders of magnitude lower than the bolometric luminosity @xcite . x - ray emission is known to be brighter among sources without active accretion @xcite and in our sample this will present itself as a detection efficiency that is a function of the infrared sed slope ( defined as the logarithmic slope of @xmath7 versus @xmath8 over the 3 - 8 range ) since the presence of a circumstellar disk as measured with the infrared excess is correlated with accretion . using the dereddened flux , the sed slope varies from @xmath9 for sources without a disk to @xmath10 for sources with an envelope and we use @xmath11 as the boundary between a source with or without a circumstellar disk . @xcite find that within ic 348 the median x - ray luminosity among stars with disks is @xmath00.3 dex lower than comparable stars without disks . figure [ demographics ] shows the distribution of sed slope for x - ray detections compared with the entire sample , demonstrating our inefficient sensitivity to sources with disks . for the majority of our analysis we restrict ourselves to the sources with a disk . the presence of infrared variability is assessed using the techniques described in @xcite . in short , we use the reduced chi - square statistic for each band , and well as the stetson index s , which is sensitive to correlated variability in the two irac bands . the use of the stetson index does require us to restrict the analysis to sources detected in both bands , but the added sensitivity of this index compared to the reduced chi - square offsets the reduction in sample size . a star is marked as an infrared variable if @xmath12)>3 $ ] or @xmath13)>3 $ ] or [email protected] . based on prior analysis , we are sensitive to @xmath15 of fluctuations down to 0.04 mag for sources brighter than 14th magnitude @xcite ; this sensitivity limit is relatively independent of source brightness because most of the uncertainty is due to systematic effects for these bright targets @xcite . while spitzer monitoring data is available from multiple observing campaigns , we focus on the contemporaneous fall 2011 data when assessing variability since we are mainly concerned with the behavior during the same time frame as any x - ray fluctuations . the vast majority of stars variable in one season are also variable in other seasons , and our results would not significantly change if we included the entire dataset . since the 3 - 5@xmath16 emission probed by this photometry is dominated by the inner edge of the dusty disk @xcite , we are not sensitive to the dynamics of the regions of the disk beyond this point . given that we are mainly concerned with the change in x - ray flux , rather than its overall level , we do not convert photon fluxes to luminosities , which requires assumptions about the distance to the cluster and the underlying x - ray spectrum . while the arrival time of each photon is recorded , we instead use the photon flux averaged over each 10ks observing run , with each epoch comprising one point on our light curve . we chose to do this to increase sensitivity at the expense of knowledge about the x - ray variability on very short timescales . we restrict ourselves to sources detected on at least four x - ray epochs . photon fluxes range from 1e-6 to 1e-4 photons / s/@xmath17 , or 1 to 1200 net photons per 10ks block . we can roughly estimate the x - ray luminosities by assuming an average photon energy of 1.5kev , to convert from photon flux to energy flux , and a distance of 320pc ; we find luminosities range from @xmath010@xmath18 to @xmath010@xmath19 erg / sec with a typical luminosity of @xmath010@xmath20erg / sec similar to previous studies @xcite . our final sample consists of 39 stars ( table [ var_table ] ) with circumstellar disks with sufficient x - ray photometry to examine variability , all of which were detected in the infrared . figure [ lc_examples ] show examples of light curves for stars with and without x - ray and infrared variability to demonstrate the fidelity of our data . our sensitivity to infrared variability is fairly uniform over our entire sample because the uncertainties are dominated by systematics for the vast majority of sources . this is not the case for the x - ray emission whose uncertainties are dominated by poisson statistics and scale with the square root of the flux . this leads to substantially nonuniform sensitivities throughout our sample . as with the infrared flux we use the reduced chi - square to select variable x - ray emission , using @xmath21 as the boundary between variables and non - variables this picks out variations that are significant relative to the noise , which varies substantially between the brightest and the faintest sources . as a result , the size of the fluctuations selected as variable will strongly depend on the mean x - ray flux . on average this @xmath22 boundary corresponds to a factor of 2 increase in the flux relative to the mean with fluctuations reaching close to a factor of 10 . despite the non - continuous nature of our observations and the use of different selection criteria for selecting variables as compared to these previous studies , we pick out stars with similar fluctuations . @xcite find that among stars with disks , on timescales of a week , fluctuations are typically a factor of 3 , similar to our results . overall our sample consists of 39 sources with ( 1 ) detections on four or more x - ray epochs , ( 2 ) detections in both irac bands on all epochs , ( 3 ) an infrared excess indicative of a disk . among these 39 stars , 31 have significant fluctuations in the infrared , while 28 have detectable x - ray variability . our observing cadence was designed to probe long term effects of x - ray fluctuations on disk emission , at the expense of knowledge about any short , instantaneous changes in disk emission . while very rapid changes in infrared emission have been observed @xcite , week to month long variations make up a substantial fraction of the infrared variability @xcite . the time between an x - ray epoch and the nearest infrared observation ranges from 0.1 to 2.1 days ( 12 - 180 ks ) and we are completely insensitive to changes in infrared emission on timescale shorter than this . we instead focus on responses that cover more than 2 infrared epochs ( @xmath14300ks = 4 days ) . this caveat should be kept in mind when interpreting our results , especially as they apply to physical connections between x - ray and infrared variability . we first look at the 28 stars with disks that are variable in both the x - ray and infrared to see if the light curves themselves are correlated . this is assessed using the kendall s tau and spearman s rho statistics . both of these coefficients serve as non - parametric tests of whether increases or decreases in the x - ray are mirrored in infrared . we find no significant correlation between the x - ray and infrared light curves suggesting that there is no strong correlation between the infrared and x - ray variability . we also find that the size of the infrared fluctuations is not correlated with the size of the x - ray fluctuations among these variable stars . @xcite find correlated variations in optical flux and soft , but not hard , x - ray flux of ctts in ngc 2264 between observations separated by two weeks , suggesting that there is a difference between the behavior of the soft and hard x - ray emission . we examine how the median x - ray photon energy changes over the course of the light curve to see if it changes in concert with the infrared emission . if the soft x - ray emission tracks the infrared emission while the hard x - ray flux is constant , then the median photon energy will decrease as the infrared emission increases . given that we typically only detect 24 photons per source per 10ks block we use the median flux instead of just the soft x - ray photons to reduce the errors and maximize our sensitivity to fluctuations . using both the kendall s tau and spearman s rho metrics , we do not find a correlation between the median photon energy and the infrared flux in any of our disk - bearing targets . we also find no significant correlation in the size of the infrared variability and the size of the variability in median photon energy . among the two brightest x - ray sources , lrll 2 and 6 , for which we measure @xmath0400 photons per 10 ks , while there is some evidence for a change in median photon energy this change does not track with the infrared emission . the kendall s tau and spearman s rho statistics are effective at selecting correlated variations when there is no time - delay between the infrared and x - ray fluctuations and if the light curve is monotonic . @xcite find that roughly a tenth of the stars in orion with known optical periods have detectable x - ray periods and any corresponding , but phase shifted , variability in the infrared would be missed by the previous metrics . to account for a time delay between the two light curves we employ the cross - correlation function , defined as : @xmath23[y(t_i-\tau)-\bar{y}]}{\sigma_x\sigma_y}\ ] ] where @xmath24 and @xmath25 represent the x - ray and infrared light curves , @xmath26 , @xmath27 , @xmath28 , @xmath29 are the respective means and standard deviations of the two light curves and @xmath30 is the time lag between the two light curves @xcite . this effectively shifts the infrared light curves by @xmath30 days and calculates how well the two light curves match after this shift . a positive value of @xmath30 implies that the ir light curve leads the x - ray light curve , while a negative lag corresponds to the x - ray light curve leading the ir light curve . such an effect might be expected if there is an appreciable time delay between the occurrence of an x - ray fluctuation and the response of the disk or if the x - ray and infrared flux show a rotational modulation that is out of phase . given that the data are not continuously sampled , we interpolate the more heavily sampled infrared light curve onto the x - ray light curve in our analysis . we assess the uncertainty in the ccf at different lags by computing the ccf of a particular x - ray light curve with infrared light curves from 100 random cluster members , and use the dispersion in the value of the ccf at each lag as its uncertainty . figure [ ccf ] shows our results for lrl 5 and 58 as examples . when examining all 28 cluster members with variability in both the x - ray and infrared , we find no evidence of a strong cross - correlation signal between a lag of -20 and 20 days , with our strongest sensitivity to lags between -10 and 10 days . again there is no strong evidence for a direct connection between the infrared and x - ray variability . the ccf , while effective at picking out light curves with a time lag , breaks down in the presence of flares . @xcite find that roughly a quarter of stars within taurus exhibit flares , with these flaring sources making up half of the variable sample . while these outlier points within a light curve can easily disrupt any underlying correlation , they could also be a source of changes in infrared emission if the flare is powerful enough . we define a source as having a flare if it is variable in the x - ray ( @xmath31 ) and it has one epoch whose flux is greater than three times larger than the median flux . we can also look for x - ray flares within a single 10 ks epoch among the brightest sources since each photon is tracked when taking x - ray observation . only one source ( lrll 36 ) shows evidence for an x - ray flare during one of these epochs , and this star has been included in our analysis . there are a total of 13 cluster members with an x - ray flare , all but one of which is variable in the infrared . in none of these stars do we see strong evidence for an extended change in the ir flux following the x - ray flare . figure [ flare ] shows the x - ray and infrared light curves of one star with an x - ray flare demonstrating the lack of response in the infrared light curve . if the infrared flux is changing before the flare , it continues along its course after the flare . we examine in detail the change in ir flux associated with an x - ray flare by calculating the change in [ 3.6 ] and [ 4.5 ] magnitude between the two epochs ( i.e. about 2 days ) surrounding the x - ray flare and comparing this to all consecutive epoch magnitude changes in the infrared light curve . figure [ flare ] shows an example of this analysis . this allows us to quantify the infrared fluctuation closest to the x - ray flare and how it compares to the fluctuations seen at other times . in none of the light curves do we see evidence that the ir fluctuation near the x - ray flare is substantially different than what is seen in any other part of the light curve . this is consistent with our visual inspection which found no evidence for a change in infrared flux in response to an x - ray flare . again we note that given our observing cadence we are severely limited in assessing the immediate response of the infrared flux to an x - ray flare . the time between an x - ray epoch and an infrared epoch ranges from 0.1 to 2.1 days ( 12 - 180ks ) . the majority of x - ray flares have decay times less than 60ks with an average decay time of 10ks @xcite and almost all x - ray flares would have decayed away by the time the system was observed in the infrared . ysos typically exhibit one flare per star per 500ks with the time between bright flares ( l@xmath32erg / sec ) a factor of a few longer than this @xcite . @xcite find similar results for solar - type stars , with roughly one flare per star per 650ks . a theory for the week to month long infrared variability that relies solely on x - ray flares would require the infrared response to last at least @xmath0500ks(=5.8days ) to extend the infrared variability throughout the quiescent period between flares . even with our limited cadence we can exclude these types of large , long infrared responses . we find no direct connection between x - ray fluctuations and infrared variability . there is no clear correlation between the shapes of the infrared and x - ray light curves and there are no large , long - term changes in infrared emission following x - ray flares . this is consistent with the lack of correlated x - ray / infrared variability seen in the coronet cluster @xcite . given the cadence of our observations we can not comment on short ( less than a few days ) , immediate responses of the disk to the changing x - ray flux ; we can only constrain the type of interactions that lead to sustained week - long fluctuations . this caveat should be kept in mind in the context of the different physical situations that we discuss below . interpreting this result depends on how x - ray variability could be physically connected to changes in disk structure . our [ 3.6],[4.5 ] infrared observations are dominated by emission from the inner edge of the disk , which is made up of a slightly rounded wall of dust a few tenths of an au from the star at the boundary where the dust becomes hot enough to sublimate @xcite . there are three main possibilities connecting x - ray variability to changes in the structure of this wall , which we will consider in turn : ( 1 ) heating of dust by x - ray photons , ( 2 ) ionization of dust by x - ray photons , ( 3 ) interactions between the circumstellar disk and the hot plasma created during the flare _ x - ray heating : _ a rapid increase in the stellar flux illuminating the disk would lead to an almost instantaneous change in the temperature of the disk surface layers @xcite , causing the infrared emission to increase . our observations show no evidence for a long - term change in infrared emission in response to changes in x - ray flux , suggesting that heating by x - rays is not an important factor in setting the dust temperature . this is not surprising since models predict that much of the x - ray flux will be absorbed by gas in the tenuous upper layers of the disk @xcite and dust emission models are generally able to reproduce circumstellar disk emission in a wide range of stars without including illumination from stellar x - rays @xcite . in ic 348 the ratio of x - ray luminosity to bolometric luminosity ranges from 10@xmath33 @xcite again suggesting that x - ray heating does not contribute significantly to the dust temperature . _ x - ray ionization _ as with heating , an increase in x - ray flux may increase the ionization fraction of the dust through the increased collisions with electrons and ions @xcite , leading to observable dynamical effects due to interactions between this ionized material and the magnetic field . even if x - ray heating is minimal , the increased depth at which x - ray ionization can penetrate the disk @xcite leaves open another avenue for x - ray photons to change the dust structure . @xcite provide detailed models of such a scenario and find that a large flare ( l@xmath34erg / s ) that lasts for a day ( = 86ks ) can substantially change the height of the inner disk . this effect was predicted to decay over the course of a week , leading to extended perturbations to the infrared emission . @xcite find that eight out of 32 flares have l@xmath32erg / s while @xcite find similar numbers for the occurrence of large flares . in our sample we find no strong evidence for an extended response to an x - ray although we do not observe any flares as large as employed in these models and can not comment on any immediate response to such outbursts . _ high temperature plasma _ x - ray flares are associated with magnetic reconnection events injecting magnetic energy into the upper reaches of the stellar atmosphere @xcite . this energy creates a @xmath0100 mk plasma that expands into the surrounding coronal loop , injecting thermal energy into the area surrounding the star , decaying on a timescale that is proportional to the size of the magnetic loop @xcite . based on detailed modeling of the decay time and emission properties of x - ray flares in orion favata et al . find that 9 out of 22 flares reach larger than 5r@xmath3 and could potentially interact with the disk in systems with very close in dust . @xcite model a different type of flare that originates at the interface between the disk and the stellar magnetic field that provides a direct interaction between the flare and the disk structure . in this model the disk height can change by a factor of two in response to a l@xmath35erg / s sized flare . we see no evidence for long - term reactions to x - ray flares suggesting that these types of interactions do not drive the majority of the observed infrared variability . as with the disked sources , in the diskless stars we see no evidence for a correlation between the x - ray and infrared variability . without excess emission , the infrared flux in the diskless stars is tracing the stellar flux rather than the dust in the inner disk . stellar variability in diskless stars is dominated by the rotation of star spots , which may be located at sites of amplified magnetic field strength . we see no response in the infrared light curve following x - ray flares , similar to the systems with protoplanetary disks . analysis with the cross - correlation function as well as kendall s tau and spearman s rho statistics , confirms the lack of similarity in light curve shapes . compared to stars with disks , diskless stars show weaker fluctuations in the infrared @xcite and in the x - ray @xcite . previous studies have found that while x - ray and optical periods are sometimes similar @xcite there is often not a direct correlation between the x - ray and optical fluctuations @xcite . @xcite conclude that even though x - ray production is connected to the presence of a magnetically active , spotted surface , as evidenced by the connection between optical variability and x - ray luminosity , the regions of x - ray production and the star spots are not spatially coincident . we find no evidence for an increased frequency of infrared variability among stars with x - ray variability . we see no correlation in the shape of the light curves in stars that exhibit variability in both bands , which is especially true in the stars with moderate x - ray flares for which we see no evidence for a change in infrared flux in response to the flare . we put limits on the relation between x - ray variability and the week to month long infrared variability of dust at the inner edge of the disk that is common among young stellar objects . based on the observational evidence , the influence of x - ray heating on the temperature of the dust close to the star is relatively minor , which in turn has implications for the overall structure of the inner edge of the disk . below z / [email protected] the densities are high enough that the gas and dust temperature are highly coupled through collisions @xcite , allowing the dust temperature to set the gas temperature . the gas temperature in turn sets the pressure scale height of the disk , which is the basis of the vertical structure of the disk in a region where planet formation is expected to occur . if the x - rays can not influence the dust thermal structure , then they will not influence the overall structure in this area of the disk . compared to fuv and euv photons , x - rays have a higher penetration depth because the opacity of the gas decreases towards higher energies and if x - ray photons can not reach the dusty layers , then fuv and euv photons contribute even less to the dust temperature . the dust temperature is instead set by a combination of viscous dissipation and irradiation by stellar optical emission @xcite , at least until the disk is significantly depleted @xcite . beyond @xmath01 au , which is not probed by our infrared photometry , the x - rays may still have a strong influence on disk structure through e.g. mri turbulence @xcite . in terms of constraining the origin of the long - term infrared fluctuations , we can rule out the influence of x - ray ionization as well as interactions between the x - ray emitting plasma and the dust at the sublimation front close to the star from explaining the majority of the observed infrared variability . these factors may come into play on short timescales for a small handful of stars , or at levels below our detection limits , but generally they are not important in setting disk structure . more likely possibilities include perturbations related to accretion heating @xcite , mri driven turbulence lifting dust out of the midplane @xcite or perturbations from a companion @xcite with observations pointing toward accretion bursts , variable obscuration of the central star and changes in the inner disk structure @xcite . we would like to thank the anonymous referee for their substantive comments that greatly improved this paper . the scientific results reported in this article are based in part on observations made by the chandra x - ray observatory . support for this work was provided by the national aeronautics and space administration through chandra award number g02 - 13028a issued by the chandra x - ray observatory center , which is operated by the smithsonian astrophysical observatory for and on behalf of the national aeronautics space administration under contract nas8 - 03060 . this work is based in part 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 . support for the work was provided by nasa through an award issued by jpl / caltech . rg gratefully acknowledges funding support from nasa adap grants nnx11ad14 g and nnx13af08 g and caltech / jpl awards 1373081 , 142329 and 1440160 in support of spitzer space telescope observing programs . alencar , s.h.p . , et al . 2012 , , 541 , 116 alexander , f. & preibisch , t. 2012 , , 539 , 64 aresu , g. , kamp , i. , meijerink , r. , woitke , p. , thi , w .- f . , & spaans , m. 2011 , , 526 , 163 ataiee , s. , pinilla , p. , zsom , a. , dullemond , c.p . , dominik , c. , & ghanbari , j. 2013 , , 553 , l3 benz , a.o . & gdel , m. 2010 , , 48 , 241 borucki , w.j . , et al . 2011 , , 736 , 19 bouvier , j. , et al . 2007 , , 463 , 1017 caramazza , m. , flaccomio , e. , micela , g. , reale , f. , wolk , s.j . , & feigelson , e.d . 2007 , , 471 , 645 cody , a.m. , et al . 2014 , , 147 , 82 dalessio , p. , cant , j. , calvet , n. , & lizano , s. 1998 , , 500 , 411 dalessio , p. , calvet , n. , nartmann , l. , franco - hernndez , r. , & servn , h. 2006 , , 638 , 314 chiang , e.i . & goldreich , p. 1997 , , 490 , 368 cohen , r.e . , herbst , w. , williams , e.c . 2004 , , 127 , 1602 donati , j .- f . & landstreet , j.d . 2009 , , 47 , 333 donati , j .- f . , 2012 , , 425 , 2948 donati , j .- f . , 2011 , , 417 , 1747 dupree , a.k . , 2012 , , 750 , 73 eisner , j. , graham , j.r . , akeson , r.l . , & najita , j. 2009 , , 692 , 309 eisner , j. , et al . 2010 , , 718 , 774 espaillat , c. , et al . 2013 , , 762 , 62 fang , m. , kim , j.s . , van boekel , r. , sicilia - aguilar , a. , henning , t. , & flaherty , k. 2013 , , 207 , 5 favata , f. , flaccomio , e. , reale , f. , micela , g. , sciortino , s. , shang , h. , stassun , k.g . , & feigelson , e.d . 2005 , , 160 , 469 feigelson , e. , townsley , l. , gudel , m. & stassun , k. 2007 , in protostars and planets v , ed . reipurth , b. , jewitt , d. , keil , k. ( tucson , az : univ . arizona press ) , 313 flaccomio , e. , et al . 2003 , , 582 , 398 flaccomio , e. , et al . 2007 , , 160 , 450 flaccomio , e. , micela , g. , favata , f. , & alencar , s.p.h . 2010 , , 516 , 8 flaccomio , e. , micela , g. , & sciortino , s. 2012 , , 548 , 85 flaherty , k.m . , muzerolle , j. , rieke , g. , gutermuth , r. , balog , z. , herbst , w. , & megeath , s.t . 2013 , , 145 , 66 fragner , m.m . , & nelson , r.p . 2010 , , 511 , 77 forbirch , j. , et al . 2007 , , 464 , 1003 gaskell , c.m . & peterson , b.m . 1987 , , 65,1 getman , k.v . , broos , p.s . , salter , d.m . , garmire , g.p . & hogerheijde , m.r . 2011 , , 730 , 6 glassgold , a.e . , najita , j. , & igea , j. 2004 , , 615 , 972 gregory , s.g . , donati , j .- f . , morin , j. , hussain , g.a.j . , mayne , n.j . , hillenbrand , l.a . , & jardine , m. 2012 , , 755 , 97 gutermuth , r.a . , megeath , s.t . , myers , p.c . , allen , l.e . , pipher , j.l . , & fazio , g.g . 2009 , , 184 , 18 goodson , a.p . & winglee , r.m . 1999 , , 524 , 159 hamaguchi , k. , grosso , n. , kastner , j.h . , weintraub , d.a . , & richmond , m. 2010 , , 714 , 16 hillenbrand , l.a . , et al . 2013 , , 145 , 59 igea , j. , & glassgold , a.e . 1999 , , 518 , 848 ingleby , l. , et al . 2013 , , 767 , 112 johns - krull , c.m . 2007 , , 664 , 975 kama , m. , min , m. , & dominik , c. 2009 , , 506 , 1199 ke , t.t . , huang , h. , & lin , d.n.c . 2012 , , 745 , 60 kulkarni , a.k . & romanova m.m . 2008 , , 386 , 673 lada , c.j . , et al . 2006 , , 131 , 1574 lin , d.n.c . , bodenheimer , p. , & richardson , d.c . 1996 , nature , 380 , 606 luhman , k.l . , stauffer , j.r . , muench , a.a . , rieke , g.h . , lada , e.a . , bouvier , j. , lada , c.j . 2003 , , 593 , 1093 luhman , k.l . , lada , e.a . , muench , a.a . , & elston , r.j . 2005 , , 618 , 810 mcclure , m.k . , et al . 2013 , , 769 , 73 megeath , s.t . , et al . 2012 , , 144 , 192 monnier , j.d . , 2005 , , 624 , 832 morales - caldern , m. , et al . 2011 , , 733 , 50 morales - caldern , m. , et al . 2012 , , 753 , 149 muench , a.a . , lada , c.j . , luhman , k.l . , muzerolle , j. , & young , e. 2007 , , 134 , 411 nagel , e. , espaillat , c. , dalessio , p. , & calvet , n. 2012 , , 747 , 139 orlando , s. , reale , f. , peres , g. , & mgnone , a. 2011 , 415 , 3380 owen , j.e . , ercolano , b. , clarke , c.j . , & alexander , r.d . 2010 , , 401 , 1415 preibisch , t. , et al . 2005 , , 160 , 401 preibisch , t. & zinnecker , h. 2002 , , 123 , 1613 schad , t.a . , penn , m.j . , & lin , h. 2013 arxiv:/1303.4463 skinner , s.l . & gdel , m. 2013 , , 765 , 3 stassun , k.g . , van den berg , m. & feigelson , e. 2007 , , 660 , 704 stauffer , j. , et al . 2014 , , 147 , 83 stelzer , b. & neuhuser , r. 2001 , , 377 , 538 stelzer , b. et al . 2007 , 468 , 463 stlzer , b. , preibisch , t. , alexander , f. , mucciarelli , p. , flaccomio , e. , micela , g. & sciortino , s. 2012 , , 537 , 135 szulgyi , j. , pascucci , i. , brahm , p. apai , d. , bouwman , j. , & mor , a. 2012 , , 759 , 47 tielens , a.g.g.m . 2005 , the physics and chemistry of the interstellar medium ( cambridge , uk : cambridge university press ) turner , n.j . , carballido , a. , & sano , t. 2010 , , 708 , 188 valenti , j.a . & johns - krull , c.m . 2004 , , 292 , 619 wolk , s.j . , harnden , f.r . , jr . , flaccomio , e. , micela , g. , favata , f. , shang , h. , feigelson , e.d . 2005 , , 160 , 423 wolk , s.j . , spitzbart , b.d . , bourke , t.l . , & alves , j. 2006 , , 132 , 1100 wolk , s.j . , rice , t.s . , & aspin , c. 2013 , , 773 , 145 zanni , c. , & ferreira , j. 2013 , , 550 , 99 zhu , z. , hartmann , l. , gammie , c.f . , book , l.g . , simon , j.b . , & engelhard , e. 2010 , , 713 , 1134 ccccc 13425 & 2011 - 10 - 17 06:11:07 & 55851.3 & 9.91 & 118.1 + 13426 & 2011 - 10 - 19 22:05:11 & 55853.9 & 9.91 & 119.5 + 13427 & 2011 - 10 - 22 13:31:32 & 55856.5 & 9.91 & 121.1 + 13429 & 2011 - 10 - 28 09:52:52 & 55862.4 & 10.41 & 125.6 + 13430 & 2011 - 10 - 31 06:54:51 & 55865.3 & 9.91 & 128.3 + 13431 & 2011 - 11 - 03 09:45:09 & 55868.4 & 9.92 & 131.9 + 13432 & 2011 - 11 - 06 18:22:08 & 55871.8 & 9.91 & 136.7 + 13433 & 2011 - 11 - 10 01:35:24 & 55875.0 & 10.79 & 142.9 + 13434 & 2011 - 11 - 13 06:48:37 & 55878.3 & 9.91 & 150.6 + 13428 & 2011 - 11 - 17 05:34:13 & 55882.2 & 9.22 & 163.2 + ccc 42463744 & 2011 - 10 - 15 18:22:53 & 55849.8 + 42464000 & 2011 - 10 - 18 09:44:53 & 55852.4 + 42464256 & 2011 - 10 - 20 00:45:16 & 55854.0 + 42464512 & 2011 - 10 - 22 09:56:54 & 55856.4 + 42464768 & 2011 - 10 - 24 05:31:16 & 55858.2 + 42465024 & 2011 - 10 - 26 13:08:49 & 55860.6 + 42465280 & 2011 - 10 - 27 19:42:07 & 55861.8 + 42465536 & 2011 - 10 - 30 10:18:35 & 55864.4 + 42465792 & 2011 - 11 - 1 04:52:37 & 55866.2 + 42466048 & 2011 - 11 - 3 01:27:30 & 55868.1 + 42466304 & 2011 - 11 - 4 19:14:03 & 55869.8 + 42466560 & 2011 - 11 - 6 10:30:23 & 55871.5 + 42466816 & 2011 - 11 - 8 20:29:04 & 55873.9 + 42467072 & 2011 - 11 - 10 20:15:18 & 55875.9 + 42467328 & 2011 - 11 - 12 19:38:16 & 55877.8 + 42467584 & 2011 - 11 - 14 22:39:14 & 55880.0 + 42467840 & 2011 - 11 - 17 00:21:17 & 55882.0 + 42468096 & 2011 - 11 - 18 19:53:33 & 55883.8 + 42468352 & 2011 - 11 - 21 09:56:45 & 55886.4 + 42468608 & 2011 - 11 - 23 03:08:30 & 55888.1 + ccccccccc 2 & 03:44:35.36 & 32:10:04.6 & y & y & n & 8970 & 137 & -1.47 + 5 & 03:44:26.03 & 32:04:30.4 & y & y & n & 5520 & 9.90 & -1.24 + 6 & 03:44:36.94 & 32:06:45.4 & y & y & n & 5830 & 17.0 & -2.08 + 15 & 03:44:44.73 & 32:04:02.6 & y & y & n & 3778 & 1.90 & -1.40 + 21 & 03:44:56.15 & 32:09:15.5 & y & y & y & 5250 & 3.90 & -1.98 + 31 & 03:44:18.16 & 32:04:57.0 & y & y & y & 5945 & 9.60 & -1.85 + 32 & 03:44:37.89 & 32:08:04.2 & y & y & n & 4060 & 1.40 & -1.56 + 36 & 03:44:38.47 & 32:07:35.7 & y & y & y & 4205 & 1.50 & -0.73 + 37 & 03:44:37.99 & 32:03:29.8 & y & y & y & 4205 & 0.99 & -1.03 + 41 & 03:44:21.61 & 32:10:37.6 & y & y & y & 4060 & 0.790 & -1.72 + 55 & 03:44:31.36 & 32:00:14.7 & y & y & y & 3778 & 0.680 & -1.19 + 58 & 03:44:38.55 & 32:08:00.7 & y & y & n & 3669 & 0.720 & -1.73 + 61 & 03:44:22.29 & 32:05:42.7 & y & y & y & 3955 & 0.540 & -1.45 + 65 & 03:44:33.98 & 32:08:54.1 & y & y & n & 3850 & 0.560 & -2.41 + 67 & 03:43:44.62 & 32:08:17.9 & y & y & n & 3741 & 0.480 & -2.04 + 72 & 03:44:22.57 & 32:01:53.7 & y & y & y & 3488 & 0.510 & -2.23 + 82 & 03:44:37.41 & 32:06:11.7 & y & y & n & 4060 & 0.510 & -2.43 + 83 & 03:44:37.41 & 32:09:00.9 & y & y & y & 3705 & 0.510 & -0.84 + 90 & 03:44:33.31 & 32:09:39.6 & y & y & y & 3560 & 0.410 & -2.39 + 91 & 03:44:39.21 & 32:09:44.7 & y & y & y & 3560 & 0.390 & -1.53 + 97 & 03:44:25.56 & 32:06:17.0 & y & y & y & 3524 & 0.540 & -1.75 + 110 & 03:44:37.40 & 32:12:24.3 & y & y & y & 3560 & 0.340 & -1.59 + 116 & 03:44:21.56 & 32:10:17.4 & y & y & y & 3632 & 0.290 & -2.44 + 9024 & 03:44:35.37 & 32:07:36.2 & y & y & n & ... & ... & -1.12 + 71 & 03:44:32.58 & 32:08:55.8 & y & n & n & 3415 & 0.470 & -2.01 + 75 & 03:44:43.78 & 32:10:30.6 & y & n & n & 3669 & 0.620 & -0.53 + 103 & 03:44:44.59 & 32:08:12.7 & y & n & n & 3560 & 0.380 & -1.16 + 113 & 03:44:37.19 & 32:09:16.1 & y & n & n & 4205 & 0.320 & -1.99 + 186 & 03:44:46.33 & 32:11:16.8 & y & n & n & 3560 & 0.310 & -2.07 + 203 & 03:44:18.10 & 32:10:53.5 & y & n & n & 3741 & 0.021 & -1.42 + 1872 & 03:44:43.31 & 32:01:31.6 & y & n & n & ... & ... & 0.43 + 88 & 03:44:32.77 & 32:09:15.8 & n & y & n & 3379 & 0.360 & -2.39 + 108 & 03:44:38.70 & 32:08:56.7 & n & y & n & 3379 & 0.240 & -2.44 + 146 & 03:44:42.63 & 32:06:19.5 & n & y & n & 3705 & 0.190 & -2.48 + 226 & 03:44:31.42 & 32:11:29.4 & n & y & y & 3091 & 0.078 & -1.33 + 52 & 03:44:43.53 & 32:07:43.0 & n & n & n & 3705 & 0.920 & -2.42 + 139 & 03:44:25.31 & 32:10:12.7 & n & n & n & 3161 & 0.200 & -1.22 + 141 & 03:44:30.54 & 32:06:29.7 & n & n & n & 3778 & 0.350 & -2.32 + 1401 & 03:44:54.69 & 32:04:40.3 & n & n & n & ... & ... & 0.10 + ccccccccc 29 & 03:44:31.53 & 32:08:45.0 & y & y & y & 4900 & 2.10 & -2.88 + 45 & 03:44:24.29 & 32:10:19.4 & y & y & y & 4350 & 0.920 & -2.72 + 48 & 03:44:34.88 & 32:06:33.6 & y & y & y & 4278 & 0.990 & -2.70 + 66 & 03:44:28.47 & 32:07:22.4 & y & y & y & 4132 & 0.710 & -2.98 + 69 & 03:44:27.02 & 32:04:43.6 & y & y & n & 3705 & 0.460 & -2.95 + 86 & 03:44:27.88 & 32:07:31.6 & y & y & n & 3560 & 0.460 & -2.73 + 92 & 03:44:32.67 & 32:06:46.5 & y & y & y & 3488 & 0.390 & -2.87 + 125 & 03:44:21.66 & 32:06:24.8 & y & y & n & 3451 & 0.290 & -2.64 + 62 & 03:44:26.63 & 32:03:58.3 & y & n & n & 3161 & 0.390 & -2.88 + 105 & 03:44:11.26 & 32:06:12.1 & y & n & n & 3850 & 0.390 & -2.88 + 170 & 03:44:28.42 & 32:11:22.5 & y & n & n & 3415 & 0.190 & -2.77 + 9 & 03:44:29.17 & 32:09:18.3 & n & y & n & 5520 & 11.00 & -2.86 + 16 & 03:44:32.74 & 32:08:37.5 & n & y & n & 5700 & 0.750 & -2.72 + 23 & 03:44:38.72 & 32:08:42.0 & n & y & n & 4730 & 4.10 & -2.96 + 47 & 03:44:55.51 & 32:09:32.5 & n & y & n & 5250 & 1.90 & -2.89 + 50 & 03:44:55.63 & 32:09:20.2 & n & y & y & 4590 & 1.40 & -2.87 + 53 & 03:44:16.43 & 32:09:55.2 & n & y & y & 5250 & 1.40 & -2.80 + 74 & 03:44:34.27 & 32:10:49.7 & n & y & n & 3560 & 0.620 & -2.92 + 79 & 03:45:01.52 & 32:10:51.5 & n & y & n & 5250 & 0.940 & -2.81 + 142 & 03:43:56.20 & 32:08:36.3 & n & y & n & 3850 & 0.330 & -2.91 + 154 & 03:44:37.79 & 32:12:18.2 & n & y & n & 3198 & 0.190 & -2.79 + 171 & 03:44:44.85 & 32:11:05.8 & n & y & n & 3451 & 0.170 & -2.75 + 174 & 03:44:04.11 & 32:07:17.1 & n & y & n & 3632 & 0.180 & -2.86 + 178 & 03:44:48.83 & 32:13:22.1 & n & y & y & 3451 & 0.130 & -2.60 + 184 & 03:44:53.76 & 32:06:52.2 & n & y & y & 3270 & 0.180 & -2.96 + 188 & 03:44:56.12 & 32:05:56.7 & n & y & y & 3451 & 0.130 & -2.73 + 4 & 03:44:31.19 & 32:06:22.1 & n & n & n & 7200 & 34.00 & -2.72 + 10 & 03:44:24.66 & 32:10:15.0 & n & n & n & 6890 & 13.0 & -2.89 + 30 & 03:44:19.13 & 32:09:31.4 & n & n & n & 7200 & 6.50 & -2.84 + 33 & 03:44:32.59 & 32:08:42.5 & n & n & n & 3488 & 0.870 & -2.58 + 35 & 03:44:39.26 & 32:07:35.5 & n & n & n & 4730 & 3.60 & -2.92 + 38 & 03:44:23.99 & 32:11:00.0 & n & n & n & 6030 & 3.10 & -2.77 + 56 & 03:44:05.00 & 32:09:53.8 & n & n & n & 4660 & 0.930 & -2.87 + 59 & 03:44:40.13 & 32:11:34.3 & n & n & n & 4900 & 1.400 & -2.76 + 98 & 03:44:38.62 & 32:05:06.5 & n & n & n & 3270 & 0.330 & -2.59 + 151 & 03:44:34.83 & 32:11:18.0 & n & n & n & 3560 & 0.260 & -2.83 + 163 & 03:44:35.45 & 32:08:56.3 & n & n & n & 3091 & 0.140 & -2.67 + = 3600k and l@xmath3=0.3l@xmath5 , and are more sensitive to sources without disks ( @xmath9 ) than sources with disks.[demographics],title="fig : " ] = 3600k and l@xmath3=0.3l@xmath5 , and are more sensitive to sources without disks ( @xmath9 ) than sources with disks.[demographics],title="fig : " ] = 3600k and l@xmath3=0.3l@xmath5 , and are more sensitive to sources without disks ( @xmath9 ) than sources with disks.[demographics],title="fig : " ] per sec . points to the left of the vertical dashed line are derived from earlier chandra observations of ic 348 @xcite . the infrared light curves are shown in the corresponding bottom panels . error bars for [ 3.6 ] ( squares ) and [ 4.5 ] ( crosses ) photometry are shown as the left and right vertical lines . * upper left : * lrll 31 , a source that is variable in both x - ray and infrared . * upper right : * lrll 103 , a star that is variable in the infrared , but not in the x - ray . * lower left : * lrll 108 , a star that is variable in the x - ray but not the infrared . * lower right : * lrll 141 , a star that is not variable in either the x - ray or infrared . light curves for all stars with disks are included in figure [ all_lc].[lc_examples],title="fig : " ] per sec . points to the left of the vertical dashed line are derived from earlier chandra observations of ic 348 @xcite . the infrared light curves are shown in the corresponding bottom panels . error bars for [ 3.6 ] ( squares ) and [ 4.5 ] ( crosses ) photometry are shown as the left and right vertical lines . * upper left : * lrll 31 , a source that is variable in both x - ray and infrared . * upper right : * lrll 103 , a star that is variable in the infrared , but not in the x - ray . * lower left : * lrll 108 , a star that is variable in the x - ray but not the infrared . * lower right : * lrll 141 , a star that is not variable in either the x - ray or infrared . light curves for all stars with disks are included in figure [ all_lc].[lc_examples],title="fig : " ] per sec . points to the left of the vertical dashed line are derived from earlier chandra observations of ic 348 @xcite . the infrared light curves are shown in the corresponding bottom panels . error bars for [ 3.6 ] ( squares ) and [ 4.5 ] ( crosses ) photometry are shown as the left and right vertical lines . * upper left : * lrll 31 , a source that is variable in both x - ray and infrared . * upper right : * lrll 103 , a star that is variable in the infrared , but not in the x - ray . * lower left : * lrll 108 , a star that is variable in the x - ray but not the infrared . * lower right : * lrll 141 , a star that is not variable in either the x - ray or infrared . light curves for all stars with disks are included in figure [ all_lc].[lc_examples],title="fig : " ] per sec . points to the left of the vertical dashed line are derived from earlier chandra observations of ic 348 @xcite . the infrared light curves are shown in the corresponding bottom panels . error bars for [ 3.6 ] ( squares ) and [ 4.5 ] ( crosses ) photometry are shown as the left and right vertical lines . * upper left : * lrll 31 , a source that is variable in both x - ray and infrared . * upper right : * lrll 103 , a star that is variable in the infrared , but not in the x - ray . * lower left : * lrll 108 , a star that is variable in the x - ray but not the infrared . * lower right : * lrll 141 , a star that is not variable in either the x - ray or infrared . light curves for all stars with disks are included in figure [ all_lc].[lc_examples],title="fig : " ] and [ 4.5 ] are shown as vertical tick marks ( left and right respectively ) to the left of dotted line . while the star is variable in the infrared , its variability does not seem to change in response to the x - ray flare . ( right ) distribution of epoch - to - epoch changes in [ 3.6 ] ( squares ) and [ 4.5 ] ( crosses ) magnitude for lrll 90 . the change experienced between the two epochs surrounding the x - ray flare are shown as vertical lines . the ir flux variation experienced due to the x - ray flare is not significantly different than that seen during other epochs . a similar behavior is seen in the other stars that have an x - ray flare , indicating that the observed x - ray flares do not drive large changes in infrared flux.[flare],title="fig : " ] and [ 4.5 ] are shown as vertical tick marks ( left and right respectively ) to the left of dotted line . while the star is variable in the infrared , its variability does not seem to change in response to the x - ray flare . ( right ) distribution of epoch - to - epoch changes in [ 3.6 ] ( squares ) and [ 4.5 ] ( crosses ) magnitude for lrll 90 . the change experienced between the two epochs surrounding the x - ray flare are shown as vertical lines . the ir flux variation experienced due to the x - ray flare is not significantly different than that seen during other epochs . a similar behavior is seen in the other stars that have an x - ray flare , indicating that the observed x - ray flares do not drive large changes in infrared flux.[flare],title="fig : " ]

i investigate the baryonic tully - fisher relation for a sample of galaxies with extended 21 cm rotation curves spanning the range . a variety of scalings of the stellar mass - to - light ratio are considered . for each prescription for , i give fits of the form . presumably , the prescription that comes closest to the correct value will minimize the scatter in the relation . the fit with minimum scatter has and . this relation holds over five decades in mass . galaxy color , stellar fraction , and are correlated with each other and with , in the sense that more massive galaxies tend to be more evolved . there is a systematic dependence of the degree of maximality of disks on surface brightness . high surface brightness galaxies typically have of the maximum disk value , while low surface brightness galaxies typically attain of this amount .

astro-ph0506750

the tully - fisher relation ( tully & fisher 1977 ) has played an important role in establishing the extragalactic distance scale ( e.g. , pierce & tully 1988 ; sakai et al . 2000 ) . in this context , it has been treated as a simple and convenient empirical relation between luminosity and line - width . the reason why it works is also important , particularly to our understanding of galaxies as physical objects and how they formed ( e.g. , eisentstein & loeb 1996 ; mcgaugh & de blok 1998a ; courteau & rix 1999 ; van den bosch 2000 ; navarro & steinmetz 2000a , b ) . now that the issue of the distance scale is widely considered to be settled , we can hope to place an absolute scale on galaxy mass as well as luminosity . the physical basis of the tully - fisher relation is widely presumed to be a relation between a galaxy s total mass and rotation velocity ( e.g. , freeman 1999 ) . luminosity is a proxy , being proportional to stellar mass , which in turn depends on the total mass . mcgaugh et al . ( 2000 ) found that a more fundamental relationship between the baryonic mass and rotation velocity does indeed exist , provided that both stellar and gas mass are considered ( milgrom & braun 1998 ) . the relation resulting from the sum of stellar and gas mass is referred to as the baryonic tully - fisher ( btf ) relation . the btf has also been investigated by bell & de jong ( 2001 ) , verheijen ( 2001 ) , gurovich et al . ( 2004 ) , and pfenniger & revaz ( 2005 ) . these efforts find broadly similar results , in that there is such a relation . however , details of the relation differ . mcgaugh et al . ( 2000 ) found a steep slope ( @xmath8 ) from a sample dominated by galaxies with @xmath9 or @xmath10-band photometry and rotation velocities estimated from the line - width @xmath11 . verheijen ( 2001 ) found much the same from @xmath12-band photometry and flat rotation velocities measured from resolved 21 cm cubes . both these authors assumed a constant value of the stellar mass - to - light ratio for all galaxies . using verheijen s data and a grid of stellar population models to refine the estimate of stellar mass , bell & de jong ( 2001 ) found a somewhat shallower slope ( @xmath13 ) . gurovich et al . ( 2004 ) find a break in the relation , with low mass galaxies following a steeper slope . there is an equal variety in the physical interpretations . mcgaugh et al . ( 2000 ) argue that the regularity of the btf implies that , after the observed stars and gas are accounted for , no further comparably massive reservoirs of baryons are likely to exist in disk galaxies . pfenniger & revaz ( 2005 ) use the same data to argue the opposite case : the scatter is somewhat reduced if there are dark baryons weighing several times the observed gas mass . in the context of @xmath14cdm , one would expect still more dark baryons in order to match the universal baryon fraction ( e.g. , mo & mao 2004 ) , though these need not be associated with the disk . the situation at present remains confused . the purpose of this paper is to provide the best empirical btf relation possible with the currently available data . much depends on the value of the stellar mass . i consider the effects on the btf of varying the stellar mass over a broad range . mass - to - light ratios are scaled by several recipes : as a fraction of maximum disk ; with respect to stellar population synthesis models ; and by the mass discrepancy acceleration ( mdacc ) relation ( mcgaugh 2004 ) . this treats mond ( milgrom 1983 ) as a purely phenomenological prescription . a grid of btf fits are given that covers essentially any plausible choice of stellar mass - to - light ratio . the scatter of the relation varies with this choice , and an optimal choice that minimizes the scatter in the btf relation is clear . lccccccccc ugc 2885 & 300 & 30.8 & 5.0&22.0 & 13.0 & [ 0.47 ] & 2.6 & 0.8 & 1.5 + ngc 2841 & 287 & 32.3 & 1.7&21.1 & 4.6 & 0.74 & 7.9 & 2.2 & 3.8 + ngc 5533 & 250 & 19.0 & 3.0&23.0 & 11.4 & 0.77 & 27.0 & 2.5 & 3.4 + ngc 6674 & 242 & 18.0 & 3.9&22.5 & 8.3 & 0.57 & 18.0 & 1.1 & 2.6 + ngc 3992 & 242 & 15.3 & 0.92&20.4 & 4.1 & 0.72 & 5.8 & 2.1 & 4.9 + ngc 7331 & 232 & 13.3 & 1.1&21.5 & 4.5 & 0.63 & 5.8 & 1.4 & 2.5 + ngc 3953 & 223 & 7.9 & 0.27&20.6 & 3.9 & 0.71 & 2.7 & 2.0 & 2.7 + ngc 5907 & 214 & 9.7 & 1.1&20.7 & 4.0 & 0.78 & 4.7 & 2.6 & 3.9 + ngc 2998 & 213 & 8.3 & 3.0&20.3 & 5.4 & 0.45 & 2.8 & 0.7 & 1.2 + ngc 801 & 208 & 10.0 & 2.9&21.9 & 12.0 & 0.61 & 8.4 & 1.3 & 1.4 + ngc 5371 & 208 & 11.5 & 1.0&21.4 & 7.9 & 0.65 & 3.4 & 1.6 & 1.6 + ngc 5033 & 195 & 8.8 & 0.93&23.0 & 5.8 & 0.55 & 4.6 & 1.0 & 4.6 + ngc 3893 & 188 & 4.20 & 0.56&20.3 & 2.3 & [ 0.56 ] & 2.6 & 1.1 & 2.0 + ngc 4157 & 185 & 4.83 & 0.79&21.3 & 5.0 & 0.66 & 3.1 & 1.6 & 2.4 + ngc 2903 & 185 & 5.5 & 0.31&20.5 & 2.0 & 0.55 & 3.6 & 1.0 & 3.6 + ngc 4217 & 178 & 4.25 & 0.25&21.4 & 4.2 & 0.77 & 2.4 & 2.5 & 2.2 + ngc 4013 & 177 & 4.55 & 0.29&21.1 & 3.5 & 0.83 & 4.0 & 3.2 & 3.1 + ngc 3521 & 175 & 6.5 & 0.63&20.5 & 2.4 & 0.68 & 3.8 & 1.8 & 2.7 + ngc 4088 & 173 & 3.30 & 0.79&20.9 & 3.1 & 0.51 & 1.7 & 0.9 & 1.1 + ngc 3877 & 167 & 3.35 & 0.14&20.1 & 3.0 & 0.68 & 2.4 & 1.8 & 1.7 + ngc 4100 & 164 & 4.32 & 0.30&20.1 & 2.8 & 0.63 & 2.9 & 1.4 & 2.4 + ngc 3949 & 164 & 1.39 & 0.33&19.6 & 1.7 & 0.39 & 1.0 & 0.6 & 0.8 + ngc 3726 & 162 & 2.62 & 0.62&21.6 & 5.5 & 0.45 & 1.8 & 0.7 & 1.0 + ngc 6946 & 160 & 2.7 & 2.7&21.7 & 5.6 & 0.40 & 1.1 & 0.6 & 0.5 + ngc 4051 & 159 & 3.03 & 0.26&21.2 & 4.2 & 0.62 & 1.9 & 1.4 & 1.2 + ngc 3198 & 156 & 2.3 & 0.63&21.6 & 2.6 & 0.43 & 3.1 & 0.6 & 2.6 + ngc 2683 & 155 & 3.5 & 0.05&21.1 & 1.2 & 0.65 & 7.0 & 1.6 & 5.8 + ngc 3917 & 135 & 1.4 & 0.18&22.1 & 2.9 & 0.60 & 2.4 & 1.3 & 1.3 + ngc 4085 & 134 & 1.0 & 0.13&20.0 & 1.6 & 0.47 & 1.4 & 0.8 & 1.2 + ngc 2403 & 134 & 1.1 & 0.47&21.4 & 2.1 & 0.39 & 1.5 & 0.6 & 1.4 + ngc 3972 & 134 & 1.0 & 0.12&20.6 & 2.1 & 0.55 & 2.1 & 1.0 & 1.5 + ugc 128 & 131 & 0.57 & 0.91&24.2 & 9.2 & 0.60 & 5.0 & 1.3 & 1.1 + ngc 4010 & 128 & 0.86 & 0.27&22.5 & 2.9 & [ 0.54 ] & 2.1 & 1.0 & 1.4 + f568v1 & 124 & 0.66 & 0.34&23.3 & 3.2 & 0.57 & 5.4 & 1.1 & 3.0 + ngc 3769 & 122 & 0.80 & 0.53&21.1 & 1.7 & [ 0.64 ] & 1.4 & 1.5 & 1.2 + ngc 6503 & 121 & 0.83 & 0.24&21.9 & 1.7 & 0.57 & 1.7 & 1.1 & 1.7 + f5683 & 120 & 0.44 & 0.39&23.1 & 4.0 & 0.61 & 4.3 & 1.3 & 1.3 + ngc 4183 & 112 & 0.59 & 0.34&22.8 & 3.4 & 0.39 & 1.8 & 0.6 & 0.7 + f563v2 & 111 & 0.55 & 0.32&22.1 & 2.1 & 0.51 & 3.6 & 0.9 & 1.8 + f5631 & 111 & 0.40 & 0.39&23.6 & 4.3 & 0.64 & 9.0 & 1.5 & 3.0 + ngc 1003 & 110 & 0.30 & 0.82&21.6 & 1.9 & 0.55 & 0.5 & 1.0 & 0.2 + ugc 6917 & 110 & 0.54 & 0.20&22.9 & 3.4 & [ 0.53 ] & 3.5 & 1.0 & 1.4 + ugc 6930 & 110 & 0.42 & 0.31&22.3 & 3.0 & [ 0.59 ] & 2.1 & 1.2 & 0.8 + m 33 & 107 & 0.48 & 0.13&21.1 & 1.7 & 0.55 & 1.2 & 1.0 & 0.6 + ugc 6983 & 107 & 0.57 & 0.29&23.0 & 3.6 & [ 0.45 ] & 4.3 & 0.7 & 1.7 + ngc 247 & 107 & 0.40 & 0.13&23.4 & 2.9 & 0.54 & 3.0 & 1.0 & 1.1 + ngc 7793 & 100 & 0.41 & 0.10&20.3 & 1.1 & 0.63 & 1.9 & 1.4 & 1.2 + ngc 300 & 90 & 0.22 & 0.13&22.2 & 2.1 & 0.58 & 1.4 & 1.2 & 0.7 + ngc 5585 & 90 & 0.12 & 0.25&21.9 & 1.4 & 0.46 & 0.8 & 0.7 & 0.5 + ngc 55 & 86 & 0.10 & 0.13&21.5 & 1.6 & 0.54 & 1.0 & 1.0 & 0.2 + ugc 6667 & 86 & 0.25 & 0.08&23.8 & 2.8 & [ 0.65 ] & 2.5 & 1.5 & 1.0 + ugc 2259 & 86 & 0.22 & 0.05&22.3 & 1.3 & & 3.9 & & 2.1 + ugc 6446 & 82 & 0.12 & 0.30&23.1 & 3.1&[0.39 ] & 2.2 & 0.5 & 0.5 + ugc 6818 & 73 & 0.04 & 0.10&23.5 & 1.9 & [ 0.43 ] & 0.7 & 0.6 & 0.2 + ngc 1560 & 72 & 0.034 & 0.098 & 23.2 & 1.3 & 0.57 & 4.1 & 1.1 & 1.0 + ic 2574 & 66 & 0.010 & 0.067 & 23.4 & 2.2 & 0.42 & 0.6 & 0.6 & 0.1 + ddo 170 & 64 & 0.024 & 0.061 & 24.1 & 1.3 & & 15.0 & & 1.5 + ngc 3109 & 62 & 0.005 & 0.068 & 23.1 & 1.6 & [ 0.47 ] & 1.0 & 0.8 & 0.1 + ddo 154 & 56 & 0.004 & 0.045 & 23.2 & 0.5 & 0.32 & 2.0 & 0.4 & 0.1 + ddo 168 & 54 & 0.005 & 0.032 & 23.4 & 0.9 & 0.32 & 0.6 & 0.4 & 0.2 + mcgaugh et al . ( 2000 ) have already described the btf over a large range of velocity and mass . this large dynamic range is critical to constraining the slope of the relation , and also its absolute normalization since low mass galaxies are frequently gas rich and insensitive to assumptions about @xmath1 . few other samples cover such a large dynamic range ( e.g. , gurovich et al . 2004 ) . one thing that stands to be improved is the accuracy of the data . the data used here are from the sample of sanders & mcgaugh ( 2002 ) , as trimmed for accuracy by mcgaugh ( 2004 ) . the reader is referred to these papers , and references therein , for a further description of the sample . these galaxies all have extended 21 cm maps from which the flat rotation velocity @xmath15 is measured . the use of @xmath15 provides a considerable improvement in accuracy over line - width measurements ( verheijen 2001 ) . the result is a much cleaner btf relation that is not afflicted by possible corrections to line - widths for turbulent motion . turbulent corrections have the potential to affect the slope of the btf by systematically adjusting the line - width inferred rotation velocities of low mass galaxies . in addition to resolved atomic gas measurements , all galaxies have detailed surface photometry from which stellar masses can be estimated . the sample is identical to that in mcgaugh ( 2004 ) , with one exception . after re - examining the data , ngc 2915 has been replaced by ugc 6818 . though the individual data points for ngc 2915 are quite precise , this galaxy does not have a well defined @xmath15 , the quantity of interest for the btf . the relative distance uncertainty for ngc 2915 is also uncomfortably large ( meurer , mackie , & carignan 1994 ; karachentsev et al . in contrast , ugc 6818 only barely missed the cut imposed in mcgaugh ( 2004 ) , and has more robust global measurements . having done this exercise , i can see where further improvements could be made for individual galaxies , but these are generally very minor . these are all the galaxies that are available with an obtainably high standard of accuracy . the data are given in table 1 . column 1 gives the name of the galaxy . column 2 gives @xmath15 in @xmath16 . these are the fit values given by sanders & mcgaugh ( 2002 ) which are the average of the outer points . as a test , i have remeasured @xmath15 by eye for the sub - sample of verheijen & sancisi ( 2001 ) . these agree to within a few @xmath16 with the values given by verheijen ( 2001 ) and with those tabulated here . @xmath15 is easily and robustly measured , provided only that the rotation curve is extended enough . column 3 gives the stellar mass for the mass - to - light ratio from column 10 , and column 4 gives the gas mass ( both in units of @xmath17 ) . column 5 gives the central surface brightness of the disk in @xmath18 @xmath19 , and column six the exponential disk scale length . the @xmath18-band is the only band - pass in common to all galaxies ; consistent results are found in the subset with @xmath12-band data ( sanders & verheijen 1998 ; verheijen 2001 ; mcgaugh 2004 ) . column 7 gives the @xmath20 color obtained from the original source given in sanders & mcgaugh ( 2002 ) if available , or from ned if not . colors measured with ccds are given preference ; if these are not available , rc3 values are used . @xmath20 colors in square brackets are inferred from other measured colors through stellar population models . it was most often the case that @xmath21 had been measured instead of @xmath20 . the colors are used to infer the mass - to - light ratio of the stellar population from said models . several possible mass - to - light ratios are given in the last three columns : maximum disk in column 8 , stellar population synthesis in column 9 , and mdacc in column 10 ( mcgaugh 2004 ) . the btf is expressed as @xmath22 where @xmath23 is the normalization and @xmath24 the slope . note that since mass is used here rather than magnitudes , the slope @xmath25 in traditional magnitude units would be @xmath26 . @xmath15 is the measured rotation velocity in the flat part of the rotation curve , and @xmath5 represents all measured baryonic mass . it includes both stars and gas . most of these galaxies are disk dominated , but bulge mass is included where present ( see sanders 1996 ) . the gas mass is that measured in neutral hydrogen , corrected for helium and metals . other gas phases are presumed to be negligible . in effect , the molecular gas mass is subsumed into the stellar mass ( see discussion in mcgaugh 2004 ) . by the same token , any net internal extinction is also subsumed by @xmath1 . this can not be a large effect , given the consistency between @xmath18 and @xmath12 bands ( sanders & verheijen 1998 ) . the baryonic mass of a galaxy is therefor @xmath27 the stellar mass is given by @xmath28 where @xmath1 is the mass - to - light ratio of the stellar population . 1 illustrates the various tully - fisher relations that can be constructed from the data in table 1 : the luminosity - rotation velocity relation , @xmath29-@xmath15 ; this converted into stellar mass , @xmath30-@xmath15 ; that with gas only ( no stars ) , @xmath31-@xmath15 ; and the btf , @xmath5-@xmath15 . the choice of @xmath1 for each galaxy is critical . in fig . 1 , the behavior in the @xmath30-@xmath15 plane is markedly different from that in @xmath29-@xmath15 . use of the mdacc mass - to - light ratio substantially reduces the scatter when mapping from @xmath29 to @xmath30 . in addition , the break at @xmath32 noted by mcgaugh et al . ( 2000 ) becomes apparent in @xmath30-@xmath15 , though it is not visible in @xmath29-@xmath15 . constant mass - to - light ratios were used in mcgaugh et al . ( 2000 ) , but the break was apparent there because of the large number of very low mass , gas dominated dwarfs . the data quality restrictions imposed here exclude those objects . simply applying a constant @xmath1 to the galaxies here would preclude the discovery of the break , since this would just be a shift in the scale of @xmath29-@xmath15 . yet a specific , well - defined method for determining @xmath1 recovers the break without input about its existence . i explore three choices for the stellar mass - to - light ratio : maximum disk , @xmath33 ; stellar population synthesis models , @xmath34 ; and that from the mdacc , @xmath35 . for each of these choices , i construct a grid of @xmath1 scaled from each by a constant factor : @xmath36 , @xmath37 , or @xmath38 : @xmath39 this provides a prescription for @xmath1 that is specified by method and scaling factor . 2 illustrates the btf relations stemming from various choices . a few details concerning each method are worth noting here . maximum disk is the highest mass - to - light ratio consistent with but not exceeding the rotation curve data . the inner shape of rotation curves are often quite consistent with the shape predicted by the observed baryons ( van albada & sancisi 1986 ; selwood 1999 ; palunas & williams 2000 ) , leading some to argue that disks must be nearly maximal . i consider the range @xmath40 . i adopt the maximum disk value given by the original source for the data for each galaxy ( see sanders & mcgaugh 2002 ) . caution should be exercised in interpreting @xmath33 . some authors leave room for a dark halo , so that even `` maximum '' disk may provide only @xmath41 of the observed velocity at the peak of the disk contribution ( sackett 1997 ) . others leave zero room , fitting disk - only models as far out as possible ( e.g. , palunas & williams 2000 ) . this is a fairly subtle distinction in rotation curve decompositions , but does make a noticeable difference in @xmath1 . for example , the average @xmath10-band mass - to - light ratio of the palunas & williams ( 2000 ) sample is @xmath42 . since @xmath43 , if we scaled this down from a 100% to 84% contribution , the mean would be @xmath44 ( see also barnes , sellwood , & kosowsky 2004 ) . in general , there is no uniform definition of maximum disk . the data originate from a wide variety of sources , so there is no guarantee as to how maximal maximum disk is . moreover , things can go both ways . for low surface brightness disks , very small differences in the rotation curve can lead to large changes in the inferred value of @xmath33 ( swaters et al . 2000 ; mcgaugh , rubin , & de blok 2001 ) . for these galaxies and some of the more extreme dwarf galaxies , i have re - assessed the value of maximum disk . in spite of these caveats , the original value appears sensible for most of the galaxies in table 1 . there are a few exceptions . for the ursa major data ( verheijen & sancisi 2001 ) , which comprises a substantial plurality of the data in table 1 , a `` soft '' limit was imposed on the value of @xmath33 so that @xmath15 of the fitted pseudo - isothermal halo did not diverge to absurdly large values ( verheijen 1997 ) . this can easily happen for galaxies that are well described by maximum disk and for which the data are not very extended in radius . there is little clear need for a halo in such cases ( palunas & williams 2000 ) , so @xmath15 is poorly constrained and tends towards large values since only the rising portion of the contribution of the dark matter is seen . unfortunately , imposing this `` soft '' constraint can sometimes lead to @xmath33 which is very sub - maximal . this is a subtle point which only became apparent because @xmath45 for four of the galaxies in table 1 ( sanders & verheijen 1998 ) . this is a mathematical impossibility , so i have set @xmath46 in these cases . stellar population models have advanced to the point where they give plausible estimates of the mass - to - light ratio , even for the composite stellar populations of spiral galaxies . they are not yet perfect of course , but do provide a decent choice for estimating @xmath1 ( bell & de jong 2001 ; portinari et al . 2004 ) . here i employ the models of bell et al . ( 2003 ) to estimate the stellar mass - to - light ratio from the observed color : @xmath47 ( from their table 7 ) . the @xmath20 color is given precedence in estimating @xmath34 . when @xmath20 is not available , whatever color is available is used . the most common substitute is @xmath21 , which is expected to be nearly as well correlated with @xmath1 as @xmath20 ( bell & de jong 2001 ) . the same model ( from table 7 of bell et al . 2003 ) is used to estimate @xmath20 ( the bracketed colors in table 1 ) , but @xmath34 is based on the observed color . credible color information could not be located for ugc 2259 and ddo 170 , the only pieces missing from table 1 . mcgaugh ( 2004 ) used the detailed shapes of the rotation curves of the galaxies in table 1 to show that there is an empirical relation between acceleration and the amplitude of the mass discrepancy ( essentially the ratio of dark to baryonic mass ) . this relation holds at every point along a resolved rotation curve for any non - zero choice of mass - to - light ratio . the scatter about this relation depends on this choice ; @xmath35 is determined by minimizing the scatter with respect to the mean _ local _ relation . it is interesting to see here how this @xmath1 estimator fares with the _ global _ btf . the mdacc is a purely empirical relation . it is mathematically equivalent to mond : the @xmath35 are the same as the mond best fit values ( begeman , broeils , & sanders 1991 ; sanders 1996 ; sanders & verheijen 1998 ; de blok & mcgaugh 1998 ) . here we must make the distinction between mond as a fundamental theory ( with its associated difficulties ) , and as a successful recipe for fitting rotation curves . only the latter is required . indeed , the mdacc is the local analog of the btf , and had mond never been invented , we would perhaps already have recognized the mdacc purely as an empirical relation ( sancisi 2003 ) . just as the tully - fisher relation can be used empirically to estimate distances without understanding its physical basis , so too can the mdacc be utilized to estimate stellar mass - to - light ratios without prejudice concerning its theoretical basis . equations ( 4 - 6 ) are used to estimate the stellar mass for the various scalings . for each scaling , a grid of 10 choices of the scaling constant are made : @xmath48 to 1.0 in steps of 0.1 relative to maximum disk , and for @xmath37 and @xmath38 values ranging from 0.2 to 2.0 in steps of 0.2 . the maximum disk scaling obviously can not exceed unity , while for the other scalings there is no reason not to consider values larger than one . for example , @xmath49 would simply imply an imf heavier than assumed in the nominal population model which has been adopted . however , these values should not exceed maximum disk , and are not allowed to do so . if the choice of @xmath37 or @xmath38 exceeds the maximum disk value for a particular galaxy , the maximum disk value is used instead . half of the sample has saturated at maximum by @xmath50 , so larger values are not considered . the btf ( logarithm of equation 1 ) is fit . ] to the data in table 1 for each set of choices for @xmath1 . every galaxy carries equal weight in the fits . since the data have been selected to be of high quality , the residual uncertainties are likely to be dominated by systematic effects ( such as the precise distance to each galaxy ) rather than factors internal to the data , though these obviously matter as well . in any case , the choice of mass - to - light ratio completely dominates the results . lccccc @xmath51 & + 1.0 & 2.52 & 0.37 & 3.48 & 0.17 & 0.239 + + @xmath31 & + 0.0 & 4.67 & 0.48 & 2.28 & 0.48 & 0.313 + + @xmath36 & + 0.1 & 3.24 & 0.37 & 3.09 & 0.37 & 0.244 + 0.2 & 2.76 & 0.35 & 3.38 & 0.35 & 0.230 + 0.3 & 2.50 & 0.35 & 3.55 & 0.35 & 0.224 + 0.4 & 2.34 & 0.34 & 3.67 & 0.34 & 0.222 + 0.5 & 2.23 & 0.34 & 3.75 & 0.34 & 0.220 + 0.6 & 2.15 & 0.34 & 3.81 & 0.34 & 0.220 + 0.7 & 2.09 & 0.34 & 3.87 & 0.34 & 0.220 + 0.8 & 2.05 & 0.34 & 3.91 & 0.34 & 0.220 + 0.9 & 2.01 & 0.34 & 3.94 & 0.34 & 0.220 + 1.0 & 1.98 & 0.34 & 3.97 & 0.34 & 0.221 + + @xmath37 & + 0.2 & 3.81 & 0.33 & 2.81 & 0.15 & 0.204 + 0.4 & 3.44 & 0.30 & 3.04 & 0.14 & 0.186 + 0.6 & 3.23 & 0.29 & 3.19 & 0.14 & 0.183 + 0.8 & 3.08 & 0.29 & 3.29 & 0.13 & 0.183 + 1.0 & 2.98 & 0.29 & 3.37 & 0.13 & 0.186 + 1.2 & 2.87 & 0.28 & 3.44 & 0.13 & 0.188 + 1.4 & 2.75 & 0.27 & 3.52 & 0.13 & 0.191 + 1.6 & 2.65 & 0.27 & 3.58 & 0.13 & 0.194 + 1.8 & 2.54 & 0.26 & 3.64 & 0.12 & 0.197 + 2.0 & 2.46 & 0.26 & 3.69 & 0.12 & 0.200 + + @xmath38 & + 0.2 & 3.05 & 0.25 & 3.18 & 0.25 & 0.165 + 0.4 & 2.46 & 0.20 & 3.53 & 0.20 & 0.129 + 0.6 & 2.11 & 0.17 & 3.74 & 0.17 & 0.112 + 0.8 & 1.87 & 0.16 & 3.89 & 0.16 & 0.103 + 1.0 & 1.70 & 0.15 & 4.00 & 0.15 & 0.098 + 1.2 & 1.60 & 0.15 & 4.07 & 0.15 & 0.100 + 1.4 & 1.58 & 0.16 & 4.10 & 0.16 & 0.107 + 1.6 & 1.60 & 0.18 & 4.10 & 0.18 & 0.118 + 1.8 & 1.61 & 0.20 & 4.10 & 0.20 & 0.129 + 2.0 & 1.64 & 0.22 & 4.10 & 0.22 & 0.140 + the results of fitting the btf are given in table 2 . the intercept @xmath52 and slope @xmath24 are recorded , together with the formal uncertainties in each ( @xmath53 and @xmath54 ) . also given is the scatter of the data about each relation , @xmath55 . for reference , the input data , in the form of the normal luminosity - based tully - fisher relation is given . it has a scatter @xmath56 , which is equivalent to 0.6 mag . ( base ten logarithms are used throughout ) . note also that the limit @xmath57 is equivalent to the gas - only relation , so is given only once . the results summarized in table 2 are illustrated in figs . 3 and 4 . as the assumed mass - to - light ratio increases , the slope @xmath24 gradually increases while the zero point @xmath23 decreases to compensate . the lowest mass galaxies in table 1 are dominated by gaseous rather than stellar mass , so the btf tends to pivot about them . this can be seen by eye in fig . 2 , where the low end of the relation hardly budges while the more massive , star dominated galaxies move up and down with @xmath1 . these low mass , gas - rich galaxies provide a critical anchor point for the absolute calibration of the btf since their location in this diagram is insensitive to the choice of @xmath1 . the scatter @xmath55 about each fit is shown in fig . the maximum disk scaling has scatter comparable to the scatter in the input luminosities . the use of color information with population synthesis models provides a small increment of improvement in the scatter , as it should if the models succeed in improving the estimate of @xmath1 over a constant value for all objects . as stellar mass is rendered unimportant for very small @xmath36 and @xmath37 the scatter starts to increase . the limit of zero stellar mass , with gas only , obviously makes for an inadequate btf , as one would expect . irrespective of the physics underlying the btf , we expect that the prescription for @xmath1 that comes closest to the correct value will minimize the scatter about it . there may be some intrinsic scatter , but if we can get @xmath1 right , there should be no scatter left due to it . the @xmath35 prescription gives less scatter than either @xmath33 or @xmath34 . there is clear variation in the scatter with @xmath38 , with a well defined minimum at @xmath58 . the same mass - to - light ratio chosen to minimize the residuals from the local mdacc also minimizes the scatter in the global btf . this was already apparent in fig . 6 of mcgaugh ( 2004 ) , and is quantitatively confirmed here . it is rare in extragalactic astronomy that we find a correlation as strong as the btf ( @xmath59 for @xmath58 ) . indeed , the btf is so sharply defined that it has a kurtosis of 1.5 . it is difficult to avoid the conclusion that the choice @xmath58 does effectively give the correct @xmath1 . this provides an absolute calibration of the btf : @xmath60 with @xmath5 in @xmath61 and @xmath15 in @xmath16 . indeed , the scatter in this btf relation is so small that it could be used to estimate @xmath1 with nearly as great accuracy as the full mdacc , but with considerably less information . to apply the latter , we require a well resolved rotation curve and hi surface density map , and detailed surface photometry . the btf requires only global quantities : @xmath29 , @xmath31 from a single dish 21 cm observation , and an estimate of @xmath15 ( preferably from a rotation curve which is sufficiently resolved to perceive the flat part , though @xmath62 would be an adequate substitute , with some penalty in accuracy . ) this btf is consistent with the relation first reported by mcgaugh et al . the slope is identical , but the normalization is somewhat larger . this change is not particularly significant ( @xmath63 ) . it is largely due to the improvement in data quality and the use of @xmath15 rather than @xmath64 . the line - width @xmath11 is systematically larger than @xmath65 , which is more closely approximated by @xmath66 ( broeils 1992 ; verheijen 1997 ) . thus we should expect some increase in @xmath23 simply from the change of circular velocity measures . indeed , it is reassuring that the relation found here is so closely consistent with the previous version . mcgaugh et al . ( 2000 ) chose @xmath1 by a different method , taking a constant value for all galaxies motivated by population models . more importantly , the data are largely independent : the btf of mcgaugh et al . ( 2000 ) included 110 galaxies from bothun et al . ( 1985 ) , 14 from matthews , van driel , & gallagher ( 1998 ) , and 65 from eder & schombert ( 2000 ) that are completely distinct from and independent of the data in table 1 . the agreement of such diverse and independent data sets provides strong confirmation of the basic empirical result . the btf is by far the strongest correlation present in the data , but it is not the only one . there are a few others worth noting . some of these are shown in fig . many others are not shown to avoid redundancy : anything that is correlated with disk mass is also correlated with @xmath15 through the btf . correlation coefficients for the quantities appearing in fig . 5 are given in table 3 . one interesting quantity that can be constructed from the data in table 1 is the fraction of the total observed baryonic mass in the form of stars : @xmath67 . the variation of @xmath68 with disk mass basically shows the turn - away of the stellar mass tf from the btf as one goes from star to gas domination . there is lots of scatter in the @xmath68-@xmath5 diagram which is not reflected in the btf . this reiterates that what matters is mass , not the form it is in . lcccc @xmath15 & 0.99 & 0.78 & 0.77 & 0.53 + @xmath5 & & 0.76 & 0.76 & 0.53 + @xmath68 & & & 0.84 & 0.54 + @xmath1 & & & & 0.64 + together with the mass - to - light ratio and color , @xmath68 is an indicator of the evolutionary state of a galaxy ( mcgaugh & de blok 1997 , 1998a ; schombert , mcgaugh , & eder 2001 ) . as evolution proceeds and a galaxy converts its initial gas into stars , the stellar fraction increases . as old stars accumulate , the mean color reddens , and the mass - to - light ratio increases . one therefor expects these quantities to be related . the evolutionary quantities @xmath68 , @xmath1 , and color do indeed correlate as expected . not only does the mass - to - light ratio increase as the stellar population reddens , so too does @xmath1 climb as @xmath68 increases . indeed , the @xmath1-@xmath68 correlation is the next strongest in table 3 after the btf . in this sample , the evolutionary quantities are correlated with baryonic mass . they are also correlated with @xmath15 , as shown previously by sanders ( 1996 ) . this must follow , given the btf . indeed , figures constructed with @xmath15 instead of @xmath5 look so similar that they are difficult to distinguish . one curious feature of the correlations with disk mass is that all of the evolutionary indicators suggest that more massive disks are typically more evolved . this is opposite what one might naively expect from a hierarchical galaxy formation picture . in such a scenario , small galaxies form first so should have evolved the most . that the opposite is true seems more consistent with monolithic or even top - down galaxy formation . similar results have been found recently at high redshift ( e.g. , juneau et al . 2005 ; treu et al . 2005 ) , where the tendency for massive galaxies to be the most evolved is called `` downsizing . '' if galaxies do form hierarchically , the observed trend suggests that the mechanisms which regulate a galaxy s post - formation evolution dominate over the formation epoch in determining its present evolutionary status . this would appear to be a continuous function of mass , given the continuity in fig . 5 . it does not appear to be as simple as a single epoch of reionization suppressing galaxy formation at a characteristic mass scale . one word of caution is that while these data span a broad range of galaxy properties , they do not constitute a complete volume limited sample . the details of these correlations may well change as other data are added . nevertheless , it seems unlikely that the trend apparent in this figure could be reversed : massive galaxies still in the early stages of evolution seem to be very rare . it is extremely difficult to obtain a sample extending to very low mass which is complete in any meaningful sense . since an hi map is a prerequisite for membership in this sample , it seems likely that there are some low mass , high @xmath68 galaxies which are not represented here . so while there are some clear correlations with disk mass _ in this sample _ , there is no guarantee that precisely these correlations will hold for all galaxies . there is more hope that correlations between the evolutionary parameters @xmath68 , @xmath1 , and @xmath20 will hold , if only because they make so much sense . these cautions do not apply to the btf . there is no hint of deviation from it : the residuals are small and uncorrelated with any measured parameter . there is no reason to suspect that any rotating galaxy deviates from the btf even those that should deviate do not ( mcgaugh & de blok 1998a ; verheijen 2001 ) . the btf appears to be a fundamental relation . the remarkable consistency of the mdacc mass - to - light ratios with stellar population synthesis models has been noted previously ( sanders 1996 ; mcgaugh & de blok 1998b ; sanders & mcgaugh 2002 ; mcgaugh 2004 ) . i have added more color information to the data in table 1 than has previously been available , and this point remains true . indeed , there are now enough data that we can perform a fit in the same manner as done for population synthesis models . following the format of bell & de jong ( 2001 ) , the data are fit to a relation of the form @xmath69 portinari et al . ( 2004 ) note that while their @xmath18-band model @xmath1 are linearly correlated for @xmath70 , they show a break at this point . @xmath1 turns down to lower values for bluer colors than predicted by extrapolation of the line fit to redder colors . i therefor consider two fits to the data : one covering data of all colors , and the other restricted to @xmath70 the relations from the stellar population synthesis models of bell et al . ( 2003 ) and portinari et al . ( 2004 ) are compared with the fits to the data in table 4 . the relation of bell et al . ( 2003 ) is that used to estimate @xmath34 ( equation 7 ) . note the close agreement between the two models : these are virtually indistinguishable in fig . the zero points @xmath71 of the models are dominated by the choice of imf , which is fortuitously close ( kroupa & weidner 2003 ) . lccl bell et al . & @xmath72 & 1.737 & scaled salpeter imf + portinari et al . & @xmath73 & 1.69 & kroupa imf + fit to the data & @xmath74 & 1.83 & @xmath70 + fit to the data & @xmath75 & 2.31 & all colors + there is close agreement between the parameters @xmath71 and @xmath76 fit to the data and those predicted by the models . the mdacc mass - to - light ratios , determined by dynamical methods completely independent of the population models , show precisely the same trends . a slight offset in normalization is apparent , but its formal significance is low . it is tempting to interpret this offset as the molecular gas which has been subsumed into @xmath35 . if the mass of gas in the molecular phase is typically 10% to 20% of that in stars , the offset between models and data would be reconciled . the break at @xmath77 in the models of portinari et al . ( 2004 ) is apparent in the data . fig . 5 shows ( as the dashed line ) the model slope for bluer colors ( from their fig . 28 ) , normalized to match the fit to the data at the break point . this is entirely consistent with the downward trend in the data . the scatter in @xmath1 is also as expected , being larger in @xmath18 than in @xmath12 ( sanders & mcgaugh 2002 ; mcgaugh 2004 ) . this is true also above and below the break point at @xmath78 . below this point , one expects a tremendous amount of scatter from the rapid evolution of young populations . here the scatter is enormous : @xmath79 . above this color , one expects variation in @xmath1 to settle down , as the effects of individual star formation events are moderated by the accumulation of mass from previous generations . for these redder colors , the scatter is @xmath80 . in sum , the @xmath58 mass - to - light ratios are optimal not only in terms of minimizing the scatter in the btf and the mdacc ( from which they come ) , but also in terms of our expectations for stellar populations . indeed , it is hard to imagine better agreement with independent population models to which no fit has been made . moreover , the scatter in the dynamical relations is so small for these high quality data that @xmath1 estimated from the btf itself ( equation 7 ) are nearly indistinguishable from @xmath35 from the mdacc . either empirical method can be employed with unprecedented precision . the value of the slope @xmath24 of the btf is somewhat controversial , being of considerable physical importance . the nominal expectation of cdm is that @xmath81 ( navarro & steinmetz 2000a , b ) , and courteau et al . ( 2003 ) argue that this is consistent with their @xmath10-band data . the @xmath10-band is a good but not perfect indicator of mass , and the hst calibrated distances of sakai et al . ( 2000 ) give an @xmath10-band slope of 4 . this emphasizes the need for a good estimate of @xmath5 and not just @xmath29 . mcgaugh et al . ( 2000 ) found @xmath4 from data and @xmath1 estimates independent of those used here . the key aspect of that study was the large number of very gas rich , low rotation velocity galaxies which tied down the low mass end of the btf . verheijen ( 2001 ) obtained the same result . bell & de jong ( 2001 ) generated stellar population models in an attempt to improve the estimate of @xmath1 . applied to the data of verheijen , they found a btf with @xmath82 . i have , in effect , repeated this analysis with more data and updated models , and find very much the same result : for @xmath83 , @xmath84 ( table 2 ) . it is interesting that while the models and data are in good agreement ( 4.2 ) , use of @xmath34 gives somewhat shallower @xmath24 than @xmath35 or @xmath33(fig . this is attributable in part to the slight difference in normalization ( @xmath71 in table 4 ) . this can not be the entire reason for the difference , as increasing @xmath37 to 2 only increases @xmath24 to 3.7 . the shortcoming of @xmath34 estimated from a single color is that galaxies become beads on a string in the @xmath1-@xmath85 diagram : all the data collapse to fall on the dotted line in fig 5 . the inability to estimate deviations from the mean relation and thereby include a realistic estimate of the scatter seems to result in a bias - band @xmath1-color relation , which became flatter between bell & de jong ( 2001 ) and bell et al . ( 2003 ) . ] in the determination of the slope . relative to the population predicted @xmath34 , @xmath35 is skewed to high values for red colors and to low values for blue colors . one expects such a skew relative to a single straight line fit ( portinari et al . 2004 ) ; it causes a systematic difference in the slope in spite of the close agreement seen in fig . 5 and table 4 . the critical issue for constraining the slope is the dynamic range of the data . as a further test of the slope of the btf , i have sought out galaxies that extend the relation to lower rotation velocities . the slowest rotator in table 1 has @xmath86 . i have searched the literature for galaxies with slower rotation velocities that are of adequate quality to make a useful comparison . there are many slow rotators already in the sample of mcgaugh et al . ( 2000 ) , but there the rotation velocity estimate was based on a line - width ( eder & schombert 2000 ) . here i require that there be a resolved measurement adequate for estimating @xmath15 . table 5 contains the objects found that met these criteria . many of these are due to the recent excellent work of begum & changular ( 2003 , 2004a , b ) which extends down to objects approaching the globular cluster mass scale . in table 5 i give the data necessary for the btf from the data given in each reference cited there . i have estimated @xmath15 myself from the published data . in addition to the best estimate , i give a generous range of uncertainty . in many of these cases , there is a substantial correction for asymptotic drift . the lower limit of @xmath15 takes the lower end of the published errors ignoring the asymptotic drift correction . the upper limit takes the upper end of the experimental errors and includes the correction . this procedure results in a very broad error estimate : the uncertainties listed in table 5 are much larger than @xmath63 . lcccc eso215g?009 & @xmath87 & 23 & 714 & 1 + ugc 11583 & @xmath88 & 119 & 36 & 2,3 + ngc 3741 & @xmath89 & 25 & 224 & 4 + wlm & @xmath90 & 31 & 65 & 5 + kk98 251 & @xmath91 & 12 & 98 & 3 + gr 8 & @xmath92 & 5 & 14 & 6 + cam b & @xmath93 & 3.5 & 6.6 & 7 + ddo 210 & @xmath94 & 0.9 & 3.6 & 8 + for the stellar and gas mass , i take the value given by the original authors . @xmath30 has often been estimated using the models of bell & de jong ( 2001 ) , so these correspond roughly to @xmath95 estimates . rather than an uncertainty , i consider the full range of possible stellar masses , from zero to maximum disk . these are the vertical lines in fig . the extrapolation of the btf fit to the data in table 1 does an excellent job of predicting the data for these lower mass galaxies . the objects in table 5 follow the slope @xmath4 down to unprecedented low velocity . the btf remains valid over five decades in mass . 6 also illustrates why the slope has been difficult to constrain . the lower limit of other studies ( @xmath96 , @xmath97 : e.g. , bell & de jong 2001 ; courteau et al . 2003 ) is shown in the left hand panel . the data considered there extend over only a small fraction of the range studied here ( down to @xmath98 with the extreme dwarfs in table 5 ) . this lack of dynamic range probably dominates all other factors ( such as the mass determination method or the velocity estimator used ) in constraining the slope . indeed , if we were to truncate the data in table 1 at the same level , we would fail to perceive the break - point in the stellar mass tully - fisher relation . such a sample would be dominated by star - dominated galaxies , and fail to provide the constraint on @xmath1 which follows when gas - dominated objects are included . the steep slope of the btf should come as no surprise , as it is completely consistent with the results of mcgaugh et al . that study made use many low mass galaxies , 19 of which have @xmath99 . those objects are completely independent of the galaxies discussed here . the new , more accurate data in table 5 simply return the same result with less scatter . other workers investigating low mass galaxies have also inferred the need for a steep slope ( e.g. , gurovich et al . 2004 ; pizagno et al . 2004 ) . formally , a slope as shallow as @xmath81 deviates from the optimal btf by @xmath100 . this can be made less by changing @xmath1 , but only at the price of degrading the correlation and the the many consistency checks on @xmath1 . in order to recover a slope as shallow as @xmath81 , one requires @xmath101 or @xmath102 ( table 2 ) . such absurdly sub - maximal disks would fall considerably short of the mass which is directly observed in stars locally . consideration of the extreme dwarfs renders it even more difficult to reconcile a shallow slope with the data . while it is possible , at least in principle , to move massive galaxies down in mass by reducing their mass - to - light ratios in order to accomodate a shallow slope , it is not possible to move the extreme dwarfs very far up . even taking maximum disk in those cases makes little difference to the slope , and causes the curious situation that the imf in these low mass galaxies must be systematically heavier than that in giant galaxies . it is thus very difficult to reconcile a shallow ( @xmath81 ) slope for the btf with the data . one application of the result here is to quantify the degree to which galaxy disks are maximal . there is considerable debate as to whether high surface brightness disks are maximal ( e.g. , sellwood 1999 ; courteau & rix 1999 ) . there would seem little doubt that low surface brightness disks are dark matter dominated ( de blok & mcgaugh 1997 , 2001 ) , but an argument for maximal disks can be made even in these objects ( fuchs 2003 ) . it is therefor of considerable interest to investigate how maximal disks are , and how disk maximality varies with disk properties . for the mdacc mass - to - light ratios favored here , the fraction of maximum disk in each case is @xmath103 this is plotted against disk mass and surface density in fig . there is only a weak correlation of @xmath104 with disk mass ( @xmath105 ) which depends heavily on rather few points at low mass ( cf . persic & salucci 1988 ) . dynamical arguments stemming from the adherence of low surface brightness galaxies to the tully - fisher relation ( zwaan et al . 1995 ; sprayberry et al . 1995 ; hoffman et al . 1996 ) suggest that disk maximality @xmath104 should correlate with surface brightness ( tully & verheijen 1997 ; mcgaugh & de blok 1998a ; zavala et al . 2003 ) . we can improve on this by using the mass - to - light ratios @xmath35 to convert the observed central surface brightness into the central surface mass density of stars : @xmath106 as anticipated , there is a good correlation between @xmath104 and @xmath107 ( @xmath108 ) . a fit to the data in fig . 7 gives @xmath109 in terms of the more directly observable central surface brightness , this translates to @xmath110 . there is considerably greater scatter about this latter relation . irrespective of how we frame the relation , or what mass - to - light ratio prescription we prefer , it seems inevitable that the disk contribution must decline systematically as surface density declines . low surface brightness disks are inevitably dark matter dominated . in contrast , high surface density disks contribute a non - negligible fraction of the total mass at small radii for plausible @xmath1 . remarkably , this leaves no residual signature in the tully - fisher relation ( mcgaugh & de blok 1998a ; courteau & rix 1999 ) in spite of the generally modest radius at which rotation curves achieve @xmath15 . statistics of these data , divided into quartiles by @xmath107 , are given in table 6 . the typical @xmath111 in the highest surface density quartile . @xmath104 can not exceed unity , and is projected to saturate at @xmath112 . this is comparable to the highest surface brightness disks that exist ( marshall 2004 ) . in the lowest surface density quartile , the typical fraction of maximum disk drops to @xmath113 . this confirms and quantifies the well - known result that low surface brightness disks are sub - maximal ( de blok & mcgaugh 1997 ) . there is no empirical indication that we have reached a lower limit in @xmath104 . disks lower in surface brightness than the most extreme considered here do exist , though they have yet to be observed with sufficient accuracy to be included here . ccccc 1 & 15 & 23.31 & 21 & 0.25 + 2 & 15 & 22.31 & 99 & 0.48 + 3 & 15 & 21.52 & 295 & 0.74 + 4 & 15 & 20.41 & 964 & 0.78 + the meaning of @xmath104 for high surface brightness galaxies is subject to the caveats discussed in 3.1 . in particular , the effective definition of maximum disk typically accounts for 84% rather than 100% of the velocity at the peak of the disk contribution ( sackett 1997 ) . if the disks in the highest quartile have 78% of this mass , then they contribute 74% of the velocity at 2.2 scale lengths ( @xmath43 ) . this is not very different from the sub - maximal contribution of 63% advocated by bottema ( 1993 ) and is certainly within the galaxy - to - galaxy scatter . kregel , van der kruit & freeman ( 2005 ) find a slightly lower mean velocity contribution , but their sample is dominated by intermediate surface brightness galaxies , so such a result is consistent with the trend apparent in fig . 7 . the results in the literature seem broadly consistent , bearing in mind that a good deal depends on what is really meant by `` maximum disk . '' the most important point here is that the degree of maximality of a disk depends systematically upon its surface density . there is no magic value of @xmath104that is a fixed fraction for all disks ( bottema 1997 ; de blok & mcgaugh 1996 ) . i have explored the baryonic tully - fisher relation for many choices of stellar mass - to - light ratios using a sample of high quality data spanning a large dynamic range in mass . i provide fits to the btf for each @xmath1 . there is a particular choice , based on the minimization of the scatter in the local mass - discrepancy acceleration relation ( the mdacc : mcgaugh 2004 ) , that also minimizes the scatter in the btf . this optimal btf is @xmath60 with @xmath15 in @xmath16 and the total baryonic mass of a galaxy in @xmath61 . this provides a remarkably precise method of estimating the mass of rotating galaxies by observation of a single global observable , the level at which the rotation curve becomes flat . the form in which the baryonic mass resides , stars or gas , makes no difference to the btf . only the sum matters . this strongly suggests that the btf is a fundamental relation between rotation velocity and baryonic mass . it further implies that there is no other large reservoir of baryons which matter to the sum : the stars and gas observed in spiral galaxies account for essentially all of the baryonic mass therein . the mass - to - light ratios determined for the optimal btf are in exceptionally good agreement with stellar population synthesis models . this consistency , together with the agreement between local and global empirical relations connecting baryonic mass to the observed dynamics , implies that the baryonic mass is well determined . this would appear to solve the long standing problem of the uncertainty in the mass of stellar disks . using these robust stellar mass estimates , i have examined a variety of evolutionary measures . the stellar fraction , mass - to - light ratio , and color all correlate with each other as one would expect . little evolved galaxies with low @xmath68 tend to have blue colors and low @xmath1 ; more evolved galaxies have higher stellar fractions , redder colors , and higher mass - to - light ratios . these quantities are also correlated with disk mass and rotation velocity : more massive disks tend to be more evolved . the degree to which disks are maximal varies systematically with stellar surface density . high surface brightness galaxies tend to be more nearly maximal , typically with @xmath114 of the maximum disk value at @xmath115 . low surface brightness galaxies are sub - maximal , with @xmath116 of maximum disk at @xmath117 . there is considerable variation from galaxy to galaxy . in no case is the stellar mass completely negligible at small radii , a fact that is important to mass models and constraints on the inner slope of the halo mass distribution ( core or cusp ) . the work of ssm is supported in part by nsf grant ast0206078 and nasa grant nag513108 . ssm is grateful for conversations on this subject with many people , especially greg bothun , bob sanders , brent tully , and rob swaters . ssm thanks the astronomy department of case western reserve university for its hospitality during a sabbatical visit when much of this work was done . barnes , e.i . , sellwood , j.a . , & kosowsky , a. 2004 , , 128 , 2724 begeman , k. g. , broeils , a. h. , & sanders , r. h. 1991 , mnras , 249 , 523 begum , a. , & chengalur , j.n . 2003 , , 409 , 879 begum , a. , & chengalur , j.n . 2004a , , 424 , 509 begum , a. , & chengalur , j.n . 2004b , , 413 , 525 begum , a. , chengalur , j.n . , & hopp , u. 2003 , new astronomy , 8 , 267 begum , a. , chengalur , j.n . , & karachentsev , i.d . 2005 , , 433 , l1 bell , e. f. & de jong , r. s. 2001 , , 550 , 212 bell , e. f. , mcintosh , d. h. , katz , n. , & weinberg , m. d. 2003 , , 149 , 289 bothun , g.d . , aaronson , m. , schommer , b. , mould , j. , huchra , j. ; sullivan , w.t . , iii 1985 , , 57 , 423 bottema , r. 1993 , , 275 , 16 bottema , r. 1997 , , 328 , 517 broeils , a. h. 1992 , ph.d . thesis , university of groningen courteau , s. , macarthur , l.a . , dekel , a. , van den bosch , f. , mcintosh , d.h . , & dale , d. 2003 , astro - ph/0310440 courteau , s. & rix , h .- w . 1999 , , 513 , 561 de blok , w.j.g . , & mcgaugh , s.s . 1996 , , 469 , l89 de blok , w.j.g . , & mcgaugh , s.s . 1997 , , 290 , 533 de blok , w.j.g . , & mcgaugh , s.s . 1998 , , 508 , 132 de blok , w.j.g . , mcgaugh , s.s . , & rubin , v.c . 2001 , , 122 , 2396 eder , j. , & schombert , j. m. 2000 , , 131 , 47 eisenstein , d. j. & loeb , a. 1996 , , 459 , 432 freeman , k.c . 1999 , in the low surface brightness universe , iau col . 171 , asp conference series 170 , 3 fuchs , b. 2003 , , 284 , 719 gurovich , s. , mcgaugh , s.s . , freeman , k.c . , jerjen , h. , staveley - smith , l. , & de blok , w.j.g . 2004 , pasa , 21 , 412 hoffman , g. l. , salpeter , e. e. , farhat , b. , roos , t. , williams , h. , & helou , g. 1996 , , 105 , 269 jackson , d.c . , skillman , e.d . , cannon , j.m . , & cote , s. 2004 , , 128 , 1219 juneau , s. , et al . 2005 , , 619 , l135 karachentsev , i. d. , et al . 2003 , , 398 , 479 kregel , m. , van der kruit , p.c . , & freeman , k.c . 2005 , , in press ( astro - ph/0501503 ) kroupa , p. & weidner , c. 2003 , , 598 , 1076 marshall , j. 2004 , ph.d . thesis , university of maryland matthews , l.d . , van driel , w. , & gallagher , j.s . 1998 , aj , 116 , 1169 mcgaugh , s.s . 2004 , , 609 , 652 mcgaugh , s.s . , & de blok , w.j.g . 1998a , , 499 , 41 mcgaugh , s.s . , & de blok , w.j.g . 1998b , , 499 , 66 mcgaugh , s.s . , rubin , v.c . , & de blok , w.j.g . 2001 , , 122 , 2381 mcgaugh , s. s. , schombert , j. m. , bothun , g. d. , & de blok , w. j. g. 2000 , , 533 , l99 milgrom , m. 1983 , , 270 , 371 milgrom , m. & braun , e. 1988 , apj , 334 , 130 mo , h. j. , & mao , s. 2004 , , 353 , 829 meurer , g. r. , mackie , g. , & carignan , c. 1994 , , 107 , 2021 navarro , j. f. & steinmetz , m. 2000a , , 528 , 607 navarro , j. f. & steinmetz , m. 2000b , , 538 , 477 palunas , p. & williams , t. b. 2000 , , 120 , 2884 persic , m. , & salucci , p. 1988 , , 234 , 131 pfenniger , d. , & revaz , y. 2005 , , in press ( astro - ph/0409621 ) pierce , m. j. , & tully , r. b. 1988 , , 330 , 579 pizagno , j. , blanton , m.r . , weinberg , d.h . , bahcall , n.a . , & brinkmann , j. 2004 , astro - ph/0410672 portinari , l. , sommer - larsen , j. , & tantalo , r. 2004 , , 347 , 691 sackett , p.d . 1997 , , 483,103 sakai , s. , et al . 2000 , , 529 , 698 sancisi , r. 2003 , iau symposium 220 , 192 sanders , r. h. 1996 , , 473 , 117 sanders , r.h . , & mcgaugh , s.s . 2002 , , 40 , 263 sanders , r. h. & verheijen , m. a. w. 1998 , , 503 , 97 schombert , j.m . , mcgaugh , s.s . , & eder , j. 2001 , , 121 , 2420 sellwood , j. a. 1999 , asp conf . ser . 182 : galaxy dynamics - a rutgers symposium , 351 sprayberry , d. , bernstein , g. m. , impey , c. d. , & bothun , g. d. 1995 , , 438 , 72 steinmetz , m. , & navarro , j.f . 1999 , , 513 , 555 swaters , r.a . , madore , b.f . , & trewhella , m. 2000 , , 2000 , , 531 , l107 treu , t. , ellis , r. s. , liao , t. x. , & van dokkum , p. g. 2005 , , 622 , l5 tully , r. b. & fisher , j. r. 1977 , , 54 , 661 tully , r. b. & verheijen , m. a. w. 1997 , , 484 , 145 van albada , t. s. & sancisi , r. 1986 , royal society of london philosophical transactions series a , 320 , 447 van den bosch , f. c. 2000 , , 530 , 177 verheijen , m. a. w. 1997 , ph.d . thesis , university of groningen verheijen , m. a. w. 2001 , , 563 , 694 verheijen , m. a. w. & sancisi , r. 2001 , , 370 , 765 warren , b.e . , jerjen , h. , & koribalski , b.s . 2004 , , 128 , 1152 zavala , j. , avila - reese , v. , hernndez - toledo , h. , & firmani , c. 2003 , , 412 , 633 zwaan , m.a . , van der hulst , j.m . , de blok , w.j.g . , & mcgaugh , s.s . 1995 , , 273 , l35

we recalculate , within the bfkl approach and at the next - to - leading order , the jet vertex relevant for the production of mueller - navelet jets in proton collisions and of forward jets in dis . we consider both processes with incoming quark and gluon . the starting point is the definition of quark and gluon impact factors in the bfkl approach . following this procedure we show explicitly that all infrared divergences cancel when renormalized parton densities are considered . we compare our results for the vertex with the former calculation of refs . . * the next - to - leading order jet vertex for mueller - navelet and forward jets revisited * f. caporale , d.yu . ivanov , b. murdaca , a. papa , a. perri .6 cm _ dipartimento di fisica , universit della calabria , _ _ and istituto nazionale di fisica nucleare , gruppo collegato di cosenza , _ _ i-87036 arcavacata di rende , cosenza , italy _ .2 cm _ sobolev institute of mathematics and novosibirsk state university , _ _ 630090 novosibirsk , russia _

1112.3752

the mueller - navelet jet production process @xcite was suggested as an ideal tool to study the regge limit of perturbative quantum chromodynamics ( qcd ) in proton - proton ( or proton - antiproton ) collisions . it is an inclusive process @xmath5 where two hard jets @xmath6 and @xmath7 are produced ( the transverse momenta of jets are much larger than the qcd scale , @xmath8 ) in the kinematical region where the jets are separated by a large interval of rapidity , @xmath9 . this regime requires large center of mass energy @xmath10 of the proton collisions , @xmath11 , since @xmath12 . it can be studied experimentally at modern high energy hadron colliders , lhc and tevatron . the bfkl approach @xcite is the most suitable framework for the theoretical description of the high - energy limit of hard or semi - hard processes . at high @xmath10 , large logarithms of the energy compensate the small coupling and must be resummed at all orders of the perturbative series . the bfkl approach provides a systematic way to perform the resummation in the leading logarithmic approximation ( lla ) , which means resummation of all terms @xmath13 , and in the next - to - leading logarithmic approximation ( nla ) , which means resummation of all terms @xmath14 . in the bfkl approach , both in the lla and in the nla , the high - energy scattering amplitudes are expressed by a suitable factorization of a process - independent part , the green s function for the interaction of two reggeized gluons , and process - dependent terms , the so - called impact factors of the colliding particles ( see , for instance , @xcite ) . the green s function is determined through the bfkl equation , whose kernel is known at the next - to - leading order ( nlo ) @xcite . the impact factors of the colliding particle are a necessary ingredient for the complete description of a process in the bfkl approach and , therefore , to get a contact with phenomenology . the only impact factors calculated so far with nlo accuracy are those for colliding quark and gluons @xcite , for forward jet production @xcite , for the @xmath15 transition @xcite , and for the @xmath16 to light vector meson transition at leading twist @xcite . the d0 collaboration at tevatron @xcite observed power - like rise of the mueller - navelet jet cross section with energy , though the d0 results revealed an even stronger rise than predicted by lla bfkl calculations . besides the cross section measurement , it was suggested to study a less inclusive observable , such as the decorrelation of jets in the relative azimuthal angle . the d0 experiment @xcite found less decorrelation than predicted by lla calculations @xcite . important improvements were made toward the description of the process within nla accuracy . effects related with qcd running coupling were studied in @xcite . in @xcite the full nlo green s function was implemented , but the jets impact factors were taken into account at the leading order only . recently the results of a complete nla analysis of the process ( [ process ] ) were reported @xcite , which incorporates nlo corrections to both the bfkl green s function and the jets impact factors , calculated earlier in @xcite . the authors of @xcite found that for kinematics typical of the lhc experiment the effect of nlo corrections to the jet impact factors is very important , of the same order as the one obtained from the nlo correction to green s function . this observation is similar to one obtained earlier in the nla analysis of the diffractive double @xmath17-electroproduction @xcite . another important conclusion of @xcite is that the results for mueller - navelet jet observables obtained within complete bfkl nla analysis appeared to be very close to the one calculated in the conventional collinear factorization at the nlo , with the only exception of the ratio between the azimuthal angular moments @xmath18 . in our opinion it would be important to have an independent calculation of mueller - navelet jet observables in nla . the aim of the present paper is the calculation of nlo correction to the jet impact factor in order to have an independent check of the results of @xcite . in many technical steps we follow closely the method used in @xcite , but we will take advantage of starting from the general definition for the impact factors at nlo , see @xcite , which allows us to come to the results more shortly . the paper is organized as follows . in the next section we will present the factorization structure of the cross section , recall the definition of bfkl impact factor and discuss the treatment of the divergences arising in the calculation ; in section 3 we give the derivation of the quark contribution to the impact factor ; section 4 is devoted to the calculation of the gluon part ; finally , in section 5 we summarize our results and make a comparison to the ones of @xcite . the state of the jets can be described completely by their rapidities , @xmath19 , and transverse momenta , @xmath20 and @xmath21 . we denote the azimuthal angles of the produced jets as @xmath22 and @xmath23 . it is convenient to define the sudakov decomposition for the jets momenta . for a jet in the fragmentation region of the proton with momentum @xmath24 , one has @xmath25 where we assume @xmath26 neglecting the proton mass and the longitudinal fraction @xmath27 is related to the jet rapidity in the center of mass system by @xmath28 in qcd collinear factorization the cross section of the process reads @xmath29 where the @xmath30 indices specify parton types ( quarks @xmath31 ; antiquarks @xmath32 ; or gluon @xmath33 ) , @xmath34 denotes the initial proton parton density function ( pdf ) , the longitudinal fractions of the partons involved in the hard subprocess are @xmath35 , @xmath36 is the factorization scale and @xmath37 is the partonic cross section for the production of jets , and @xmath38 is the energy of parton - parton collision . at lowest order each jet is represented by a single parton having high transverse momentum , and the partonic subprocess is given by an elementary two - to - two scattering . in the discussed mueller - navelet kinematics the higher order contributions to the partonic cross section have to be resummed using bfkl approach . such resummation at nla accuracy depends on the details of jet definition ( jet algorithm ) and will be specified later . a convenient starting point for our discussion is the case of inclusive forward parton - parton scattering , considered in @xmath39 dimensions to regularize the appearing divergences . due to the optical theorem , the cross section is related to the imaginary part of the forward parton - parton scattering amplitude , @xmath40 which is given in the bfkl approach at nla by @xmath41 where the green s function obeys the bfkl equation @xmath42 and @xmath43 , @xmath44 are the parton impact factors calculated separately for the cases of massless quark and gluon in @xcite . here @xmath45 , @xmath46 are the transverse momenta of the reggeized gluons , the energy scale parameter @xmath47 is arbitrary , the amplitude , indeed , does not depend on its choice within nla accuracy . in this subsection we review the definition of the impact factors in nlo . both the kernel of the equation for the green s function and the parton impact factors can be expressed in terms of the gluon regge trajectory , @xmath48 and the effective vertices for the reggeon - parton interaction . is the proton momentum , @xmath49 is the fraction of proton momentum carried by the parton and @xmath50 is the transverse momentum of the incoming reggeized gluon . ] to be more specific , we will give below the formulae for the case of forward quark impact factor . we start with the lo , where the quark impact factors are given by @xmath51^ * \:d\rho_a\ ; , \label{eq : a7}\ ] ] where @xmath50 is the reggeon transverse momentum , and @xmath52 denotes the reggeon - quark vertices in the lo or born approximation . the sum @xmath53 is over all intermediate states @xmath54 which contribute to the @xmath55 transition . the phase space element @xmath56 of a state @xmath54 , consisting of particles with momenta @xmath57 , is ( @xmath58 is initial quark momentum ) @xmath59 while the remaining integration in ( [ eq : a7 ] ) is over the squared invariant mass of the state @xmath54 , @xmath60 in the lo the only intermediate state which contributes is a one - quark state , @xmath61 . the integration in eq . ( [ eq : a7 ] ) with the known reggeon - quark vertices @xmath62 is trivial and the quark impact factor reads @xmath63 where @xmath33 is qcd coupling , @xmath64 , @xmath65 is the number of qcd colors . in the nlo the expression ( [ eq : a7 ] ) for the quark impact factor has to be changed in two ways . first one has to take into account the radiative corrections to the vertices , @xmath66 secondly , in the sum over @xmath53 in ( [ eq : a7 ] ) , we have to include more complicated states which appear in the next order of perturbative theory . for the quark impact factor this is a state with an additional gluon , @xmath67 . however , the integral over @xmath68 becomes divergent when an extra gluon appears in the final state . the divergence arises because the gluon may be emitted not only in the fragmentation region of initial quark , but also in the central rapidity region . the contribution of the central region must be subtracted from the impact factor , since it is to be assigned in the bfkl approach to the green s function . therefore the result for the forward quark impact factor , see @xcite and fig . [ fig : if ] , reads @xmath69^*\:d\rho _ a\:\theta(s_\lambda -m_a^2 ) & & \nonumber \\ -\frac{1}{2}\int d^{d-2 } k\frac{\vec q^{\,\ , 2}}{\vec k^{\ , 2 } } \phi_{q}^{(0)}(\vec k)\mathcal{k}_r^{(0)}(\vec k,\vec q\ , ) \ln\left ( \frac{s_{\lambda}^2}{(\vec k-\vec q\,)^2 s_0}\right ) \ ; , & & \end{aligned}\ ] ] where the intermediate parameter @xmath70 should go to infinity . the second term in the r.h.s . of ( [ eq : a19 ] ) is the subtraction of the gluon emission in the central rapidity region . the dependence on @xmath71 vanishes because of the cancellation between the first and second terms . @xmath72 is the part of lo bfkl kernel related to real gluon production @xmath73 the factor in ( [ eq : a19 ] ) which involves the regge trajectory arises from the change of energy scale ( @xmath74 ) in the vertices @xmath75 . the trajectory function @xmath76 can be taken here in the one - loop approximation ( @xmath77 ) , @xmath78 in the eqs . ( [ eq : a7 ] ) and ( [ eq : a19 ] ) we suppress for shortness the color indices ( for the explicit form of the vertices see @xcite ) . the gluon impact factor @xmath79 is defined similarly . in the gluon case only the single - gluon intermediate state contributes in the lo , @xmath80 , which results in @xmath81 here @xmath82 and @xmath83 . whereas in nlo additional two - gluon , @xmath84 , and quark - antiquark , @xmath85 , intermediate states have to be taken into account in the calculation of the gluon impact factor . similarly to the parton - parton scattering ( [ pp - scat ] ) one can represent the resummed jet cross section in the form @xmath86 where we introduce jet impact factors differential with respect to the variables parameterizing the jet phase space , @xmath87 following @xcite we consider our process in the frame of a generic and infrared safe jet algorithm . in practice , this is done by introducing into the integration over the partonic phase space a suitably defined function which identifies the jet momentum with the momentum of one parton or with the sum of the two or more parton momenta when the jet is originated from the a multi - parton intermediate state . in our accuracy the jet can be formed by one parton in lo and by one or two partons when the process is considered in nlo . in the simplest case , the jet momentum is identified with the momentum of the parton in the intermediate state @xmath88 by the following jet function @xcite : @xmath89 in the more complicated case when the jet originates from a state of two partons with momenta @xmath90 and @xmath91 , we need another function @xmath92 , whose explicit form is specific for the chosen jet algorithm . an example of jet selection function in the case of the _ cone algorithm _ is the following @xcite : @xmath93 -\left [ \frac{|\vec k_1|+|\vec k_2|}{\max(|\vec k_1| , & + & s_j^{(2)}(\vec k_1 ; x\beta_1)\theta\left ( \left[\delta y^2+\delta\phi^2 \right ] -\left [ \frac{|\vec k_1|+|\vec k_2| } { \max(|\vec k_1|,|\vec k_2| ) } r_{\mbox{\scriptsize{cone } } } \right]^2 \right ) \\ & + & s_j^{(2)}(\vec k_1+\vec k_2 ; x)\theta\left ( \left [ \frac{|\vec k_1|+|\vec k_2| } { \max(|\vec k_1|,|\vec k_2| ) } r_{\mbox{\scriptsize{cone } } } \right]^2 - \left [ \delta y^2+\delta\phi^2 \right ] \right ) \ ; , \nonumber\end{aligned}\ ] ] where the sudakov decomposition of the parton momenta @xmath94 is used , with @xmath95 and @xmath96 , owing to momentum conservation in the partonic subprocess . @xmath97 in ( [ s_3 ] ) is the cone - size parameter , @xmath98 and @xmath99 are the difference of rapidity and azimuthal angle in the two parton state , respectively : @xmath100 the three terms in @xmath101 represent the case in which the jet is formed by the parton @xmath91 or the parton @xmath90 or both , respectively . in the generic case , the following relations for the jet function must be fulfilled in order the jet algorithm be infrared safe : @xmath102 such reduction of @xmath103 is required in order that the singular contributions generated by the real emission be proportional to the lowest order cross section . these contributions are canceled with the soft and collinear singularities arising from the virtual corrections and the collinear counterterms coming from the pdfs renormalization . besides , we assume that the jet selection function @xmath92 is symmetric under the exchange of the final state parton kinematic variables , @xmath104 and @xmath105 , @xmath106 the collinear counterterms appear due to the replacement of the bare pdfs by the renormalized physical quantities which obey dglap evolution equations , in the @xmath107 factorization scheme : @xmath108 \ ; , & \nonumber\\ & f_g(x)=f_g(x,\mu_f)-\frac{\alpha_s}{2\pi}\left(\frac{1}{\hat \varepsilon } + \ln\frac{\mu_f^2}{\mu^2}\right ) \int\limits^1_x\frac{dz}{z}\left[p_{gq}(z)f_q(\frac{x}{z},\mu_f ) + p_{gg}(z)f_g(\frac{x}{z},\mu_f)\right]\ ; , \label{dglappdfs}\end{aligned}\ ] ] where @xmath109 and the dglap splitting functions are : @xmath110\ ; , \\ \label{pqg } p_{qg}(z)&=&t_r\left [ z^2+(1-z)^2\right]\ ; , \;\;\ ; \mbox{with } \ t_r=\frac{1}{2}\ ; , \\ \label{pgg } p_{gg}(z)&=&2c_a\left ( \frac{z}{(1-z)_+}+\frac{(1-z)}{z } + z(1-z)\right)+\frac{(11c_a-4n_ft_r)}{6}\delta(1-z ) \;,\\ \label{pgq } p_{gq}(z)&=&c_f\frac{\left [ 1+(1-z)^2\right]}{z}\;;\end{aligned}\ ] ] here the plus - prescription is defined by @xmath111 the other counterterm is related with qcd charge renormalization , in the @xmath107 scheme : @xmath112\;. \label{charge - ren}\ ] ] having the results for the lowest order parton impact factors ( [ eq : a77 ] ) and ( [ eq : a77a ] ) , we get the jet impact factor at the lo level as @xmath113 given as the sum of the gluon and all possible quark and antiquark pdfs contributions . substituting here the bare qcd coupling and bare pdfs by the renormalized ones , we obtain the following expressions for the counterterms : @xmath114 @xmath115 for the charge renormalization , and @xmath116 @xmath117 \label{c.count.t}\ ] ] for the collinear counterterm . the latter can be rewritten in the form @xmath118 @xmath119\ ; . \label{c.count.t.a}\ ] ] besides , we present the expression for the bfkl counterterm which , in accordance to the second line of eq . ( [ eq : a19 ] ) , provides the subtraction of the gluon radiation in the central rapidity region : @xmath120 now we have all the necessary ingredients to perform our calculation of the nlo corrections to the jet impact factor . as a starting point for our consideration we will use the results of @xcite for the partonic amplitudes obtained in the calculation of partonic impact factors , introducing there the appropriate jet functions : @xmath121 for the amplitudes with one - parton state in the case of one - loop virtual corrections and @xmath122 for the cases with two partons in the final state ( real emission ) , in order to define the corresponding contribution to jet cross sections . for shortness we will present intermediate results for @xmath123 structures defined always as @xmath124 we will consider separately the subprocesses initiated by the quark and the gluon pdfs and denote @xmath125 we start with the case of incoming quark ( see fig . [ fig : quark ] ) . is the proton momentum , @xmath126 is the fraction of proton momentum carried by the quark , @xmath127 is the longitudinal jet momentum , @xmath128 is the transverse jet momentum and @xmath50 is the transverse momentum of the incoming reggeized gluon . ] virtual corrections are the same as in the case of the inclusive quark impact factor @xcite : @xmath129}{\varepsilon\ , ( 4\pi)^\varepsilon}\frac{\gamma^2(1+\varepsilon)}{\gamma(1 + 2\varepsilon ) } \left(\vec q^{\,\ , 2}\right)^{\varepsilon } \int_0 ^ 1 dx \sum_{a = q , \bar q } f_{a}(x ) \ , s_j^{(2)}(\vec q\,;x)\ ] ] @xmath130 @xmath131 here and in what follows we put the arbitrary scale of dimensional regularization equal to unity , @xmath132 . we expand ( [ qvirt ] ) in @xmath133 and get @xmath129}{\varepsilon \ , ( 4\pi)^\varepsilon}\frac{\gamma^2(1+\varepsilon)}{\gamma(1 + 2\varepsilon ) } \left(\vec q^{\,\ , 2}\right)^{\varepsilon } \int_0 ^ 1 dx \sum_{a = q , \bar q } f_{a}(x ) \ , s_j^{(2)}(\vec q\,;x)\ ] ] @xmath134 @xmath135 + { \cal o}(\varepsilon ) \ , . \label{qvirt - ex}\ ] ] for the incoming quark case , real corrections originate from the quark - gluon intermediate state . we denote the momentum of the gluon by @xmath88 , then the momentum of the quark is @xmath136 ; the longitudinal fraction of the gluon momentum is denoted by @xmath137 . thus , the real contribution has the form @xcite @xmath138 where @xmath139 the low limit in the @xmath140-integration appears due to the restriction on the invariant mass of intermediate state , which enters the definition ( [ eq : a19 ] ) of nlo impact factor . since @xmath141 and assuming @xmath142 , we obtain that @xmath143 . we consider separately the term proportional to @xmath144 and to @xmath145 . the @xmath144-term is not singular for @xmath146 , therefore the limit @xmath147 , or @xmath148 , can be safely taken . we get @xmath149 in order to isolate all divergences , it is convenient to perform the change of variable @xmath150 and to present the integral in the form @xmath151 \vec l , \beta \vec l , x\beta ; x\right)\;. \nonumber\end{aligned}\ ] ] the soft divergence appears for @xmath152 ; in this region we can introduce the counterterm @xmath153 s_j^{(2)}\left(\vec q ; x\right)\;,\nonumber\end{aligned}\ ] ] which equals @xmath154}{\varepsilon \,(4\pi)^\varepsilon } \frac{\gamma^2(1+\varepsilon)}{\gamma(1 + 2\varepsilon)}\left(\vec q^{\,\ , 2 } \right)^{\varepsilon}\int_0 ^ 1 dx \sum_{a = q , \bar q } f_{a}(x ) \ , s_j^{(2)}(\vec q\,;x)\ ; . \label{qreal - soft}\ ] ] collinear divergences arise for @xmath155 and for @xmath156 ; in these regions we can isolate the two following counterterms : @xmath157 \int_0 ^ 1dx \sum_{a = q , \bar q } f_{a}(x ) s_j^{(2)}(\vec q\,;x)\;,\end{aligned}\ ] ] @xmath158\;. \nonumber\end{aligned}\ ] ] in both these expressions we have introduced an arbitrary cutoff parameter @xmath159 and subtracted the soft divergence . after a simple calculation we obtain @xmath160}{\varepsilon \,(4\pi)^\varepsilon } \frac{\gamma^2(1+\varepsilon)}{\gamma(1 + 2\varepsilon ) } \left(\lambda^{2}\right)^{\varepsilon } \int_0 ^ 1dx \sum_{a = q , \bar q } f_{a}(x ) s_j^{(2)}(\vec q\,;x ) \nonumber \\ & \times & c_f \left[-\frac{3}{2}+4\varepsilon \right]+\mathcal{o}(\varepsilon ) \;.\end{aligned}\ ] ] the term @xmath161 can be rewritten in the following form : @xmath162}{\varepsilon \,(4\pi)^\varepsilon } \frac{\gamma^2(1+\varepsilon)}{\gamma(1 + 2\varepsilon ) } \left(\lambda^{2}\right)^{\varepsilon } \int_0 ^ 1 dx \sum_{a = q , \bar q } f_{a}(x ) \left\lbrace -\frac{3}{2}c_fs_j^{(2)}(\vec q\,;x ) \right . \nonumber & + & \int_0 ^ 1 d\beta\left [ p_{qq}(\beta)+\ , 2\varepsilon ( 1+\beta^2 ) \left(\frac{\ln(1-\beta)}{1-\beta}\right)_+c_f + \varepsilon\ , c_f(1-\beta)\right ] \\ & \times&\left . s_j^{(2)}(\vec q\ , ; x\beta)\right\rbrace + \mathcal{o}(\varepsilon)\;,\end{aligned}\ ] ] where we performed the change of variable @xmath163 , used the plus - prescription ( [ plus ] ) and the expansion @xmath164 finally , we can define the term @xmath165 which can be calculated at @xmath166 . we remark that @xmath167 and @xmath168 depend on the cutoff @xmath159 , but in the total expression @xmath169 this dependence disappears . the part proportional to @xmath145 in the r.h.s . of eq . ( [ real ] ) reads @xmath170 the collinear singularity appears at @xmath171 ; in this region we can introduce the counterterm @xmath172}{\varepsilon \,(4\pi)^\varepsilon } \frac{\gamma^2(1+\varepsilon)}{\gamma(1 + 2\varepsilon ) } \left(\lambda^{2}\right)^{\varepsilon } \\ \nonumber & \times & \int_0 ^ 1 dx \sum_{a = q , \bar q } f_{a}(x ) \int_0 ^ 1\ , d\beta \ , \left[\frac{c_a}{c_f}p_{gq}(\beta)+ \ , \varepsilon \ , c_a \ , \beta \ , \right ] \ , s_j^{(2)}(\vec q\,;x\beta ) + \mathcal{o}(\varepsilon ) \;,\end{aligned}\ ] ] where @xmath173 has been set equal to zero since the expression is finite in the @xmath174 limit and , again , the cutoff @xmath159 was introduced . another singularity appears when @xmath152 , actually at any value of gluon transverse momentum @xmath175 . in this region @xmath176 and it is convenient to introduce the counterterm @xmath177 } { \vec k^ { 2}(\vec q-\vec k)^2 } s_j^{(2)}\left(\vec q-\vec k ; x\right)\ ; , \label{softca}\end{aligned}\ ] ] where the averaging over the relative angle between the vectors @xmath175 and @xmath178 has been performed . the integration over @xmath140 gives the following result for the counterterm : @xmath179 the finite part of the real corrections proportional to @xmath145 is therefore defined by @xmath180 when the quark part of bfkl counterterm , given in ( [ counterterm ] ) , @xmath181 is combined with @xmath182 given in ( [ softcaa ] ) , we see that the dependence on @xmath183 disappears , as expected , and we get @xmath184 we collect first the contributions given in ( [ qvirt - ex ] ) , ( [ qreal - soft ] ) , ( [ coll1a ] ) , ( [ coll2a ] ) and ( [ collc_a ] ) : @xmath185}{\varepsilon \,(4\pi)^\varepsilon } \frac{\gamma^2(1+\varepsilon)}{\gamma(1 + 2\varepsilon ) } \int_0 ^ 1 dx\ , \sum_{a = q , \bar q } f_{a}(x ) \\ \nonumber & \times&\left\lbrace \left [ \left(\vec q^{\,\ , 2}\right)^{\varepsilon } \left(\frac{n_f}{3 } -c_a\ln\left(\frac{s_0 } { \vec q^{\ : 2}}\right ) - \frac{11 c_a}{6}\right)\,+\ , \varepsilon\left(c_a\left ( \frac{85}{18 } + \frac{\pi^2}{2}\right ) -\frac{5}{9}n_f\right.\right.\right.\\ \nonumber & + & \left.\left . c_f\left ( 3\ln\frac{\vec q^{\ : 2}}{\lambda^2}-4\right)\right)\right ] \ s^{(2)}_j(\vec q\,;x ) \\ \nonumber & + & \left(\lambda^{2}\right)^{\varepsilon } \int_0 ^ 1 d\beta\left[p_{qq}(\beta ) + \frac{c_a}{c_f}p_{gq}(\beta)\right ] \ s^{(2)}_j(\vec q\,;x\beta ) \\ \nonumber & + & \left . \varepsilon \int_0 ^ 1 d\beta\left[2(1+\beta^2)\left(\frac{\ln(1-\beta)}{1-\beta } \right)_+c_f + c_f(1-\beta ) + c_a\beta\right]\ s^{(2)}_j(\vec q\,;x\beta)\ , \right\rbrace+\mathcal{o}(\varepsilon)\;.\end{aligned}\ ] ] then , we collect the finite contributions , given in eqs . ( [ finitec_f ] ) and ( [ finitec_a ] ) , transforming them to the form used in @xcite : @xmath186 @xmath187 @xmath188 @xmath189\ ] ] @xmath190 @xmath191 @xmath192 -\frac{2 \ , \vec q^{\ : 2}\theta[(1-\beta)|\vec q - \vec k|-\beta |\vec k| ] } { \beta ( \vec q-\vec k)^2}s_j^{(2)}(\vec k;x ) \right\}\right ] + \mathcal{o}(\varepsilon)\ , .\ ] ] besides , we define @xmath193 given in eq . ( [ softca+c ] ) . another contribution originates from the collinear and charge renormalization counterterms , see eqs . ( [ charge.count.t ] ) and ( [ c.count.t.a ] ) , @xmath194}{\varepsilon \,(4\pi)^\varepsilon } \int_0 ^ 1 dx\ , \sum_{a = q , \bar q } f_{a}(x ) \left[\left(\mu_r^2\right)^\varepsilon \left(\frac{11c_a}{6}-\frac{n_f}{3 } \right ) s_j^{(2)}\left(\vec q\,;x\right ) \right.\ ] ] @xmath195 \ s^{(2)}_j(\vec q\,;x\beta)\right ] \ ; . \label{c.t .- q}\ ] ] finally , the quark part of the jet impact factor is given by the sum of the above four contributions and can be presented as the sum of two terms : @xmath196 @xmath197}{\varepsilon \,(4\pi)^\varepsilon } \frac{\gamma^2(1+\varepsilon)}{\gamma(1 + 2\varepsilon ) } s^{(2)}_j(\vec q\,;x ) \right ] \ ; .\ ] ] and @xmath198 @xmath199 \ s^{(2)}_j(\vec q\,;x)\ ] ] @xmath200 \,\ln\frac{\lambda^2}{\mu_f^2 } \ , \ s^{(2)}_j(\vec q\,;x\beta)\ ] ] @xmath201\ s^{(2)}_j(\vec q\,;x\beta)\ , \right\rbrace\;.\ ] ] is the proton momentum , @xmath126 is the fraction of proton momentum carried by the gluon , @xmath127 is the longitudinal jet momentum , @xmath128 is the transverse jet momentum and @xmath50 is the transverse momentum of the incoming reggeized gluon . ] we consider now the case of incoming gluon ( see fig . [ fig : gluon ] ) . virtual corrections are the same as in the case of the inclusive gluon impact factor @xcite : @xmath202}{\varepsilon\ , ( 4\pi)^\varepsilon } \frac{\gamma^2(1+\varepsilon)}{\gamma(1 + 2\varepsilon ) } \left(\vec q^{\,\ , 2}\right)^{\varepsilon}\ , \int_0 ^ 1 dx \ , \frac{c_a}{c_f}\ , f_g(x ) \ , s_j^{(2)}(\vec q\,;x ) \\ \nonumber & \times&\left[c_a\ln\left(\frac{s_0}{\vec q^{\ : 2}}\right ) + c_a\left(\frac{2}{\varepsilon}-\frac{11 + 9\varepsilon}{2(1 + 2\varepsilon ) ( 3 + 2\varepsilon)}\right.\right . \\ \nonumber & + & \left.\frac{n_f}{c_a}\frac{(1+\varepsilon)(2+\varepsilon)-1 } { ( 1+\varepsilon)(1 + 2\varepsilon)(3 + 2\varepsilon)}+\psi(1 ) + \psi(1-\varepsilon)-2\psi(1+\varepsilon)\right ) \\ \nonumber & + & \left . c_a\frac{\varepsilon}{(1+\varepsilon)(1 + 2\varepsilon ) ( 3 + 2\varepsilon)}\left(1+\varepsilon-\frac{n_f}{c_a}\right ) \frac{1}{(1+\varepsilon ) } \right ] \;.\end{aligned}\ ] ] the @xmath133-expansion has the form @xmath203}{\varepsilon\ , ( 4\pi)^\varepsilon } \frac{\gamma^2(1+\varepsilon)}{\gamma(1 + 2\varepsilon ) } \left(\vec q^{\,\ , 2}\right)^{\varepsilon}\ , \int_0 ^ 1 dx \ , \frac{c_a}{c_f}\ , f_g(x ) \ , s_j^{(2)}(\vec q\,;x ) \nonumber\\ \nonumber & \times&\left[\ , c_a\left ( \ln\left(\frac{s_0}{\vec q^{\ : 2}}\right ) + \frac{2}{\varepsilon}-\frac{11}{6}\right ) + \frac{n_f}{3}\right . \\ & + & \left . \varepsilon \left\ { \ , c_a\left(\frac{67}{18 } -\frac{\pi^2}{2}\right)-\frac{5}{9}n_f \right\ } \right ] + \mathcal{o}(\varepsilon ) \ ; .\end{aligned}\ ] ] in the nlo gluon impact factor real corrections come from intermediate states of two particles , which can be quark - antiquark or gluon - gluon . in this subsection we consider the former . the real contribution from the quark - antiquark case is @xcite : @xmath205 s_j^{(3)}(\vec q-\vec k,\vec k , x\beta;x)\;,\end{aligned}\ ] ] with @xmath206 below we discuss separately the first and the second contributions in the r.h.s . of eq . ( [ real - qq ] ) , which we denote @xmath207 and @xmath208 . the first contribution is @xmath209 s_j^{(3)}(\vec q-\vec k,\vec k , x\beta;x)\;.\end{aligned}\ ] ] here we have collinear divergences for @xmath210 and @xmath211 . the contribution in these kinematical regions is the same , as can be easily seen after the changes of variables @xmath212 and @xmath213 , since @xmath214 and taking into account the property ( [ symms3 ] ) that the @xmath92 jet selection function has to possess . therefore we can write @xmath215 and isolate the collinearly divergent part given by @xmath216}{\varepsilon\ , ( 4\pi)^\varepsilon } \frac{\gamma^2(1+\varepsilon)}{\gamma(1 + 2\varepsilon ) } \left(\lambda^{2}\right)^{\varepsilon}\,\int_0 ^ 1 dx \ , f_g(x ) \\ \nonumber & \times&\int_0 ^ 1 d\beta \ , \left [ p_{qg}(\beta ) + \varepsilon\beta(1-\beta)\right ] \,s_j^{(2)}(\vec q;x\beta ) + \mathcal{o}(\varepsilon ) \;,\end{aligned}\ ] ] where we have introduced , as before , the cutoff parameter @xmath159 . the finite part is therefore defined by @xmath217 where one can take @xmath218 limit and get @xmath219 @xmath220+\mathcal{o}(\varepsilon ) \ ; . \label{finitegc_fa}\ ] ] the second contribution in ( [ real - qq ] ) is @xmath221 here the collinear divergence appears for @xmath222 and the integral in this region can be identified with @xmath223}{\varepsilon\ , ( 4\pi)^\varepsilon } \frac{\gamma^2(1+\varepsilon)}{\gamma(1 + 2\varepsilon ) } \left(\lambda^{2}\right)^{\varepsilon}\ , \left(\frac{1}{3}+\frac{\varepsilon}{6}\right ) \nonumber \\ & \times & \int_0 ^ 1 dx \ , \frac{c_a}{c_f } \ , f_g(x ) s_j^{(2)}(\vec q\,;x ) + \mathcal{o}(\varepsilon)\;.\nonumber\end{aligned}\ ] ] then , the finite part can be written as @xmath224 after the change of variable @xmath212 in ( [ gg.ca ] ) and ( [ collc_ag ] ) we have @xmath225 + \mathcal{o}(\varepsilon ) \;. \nonumber\end{aligned}\ ] ] the real contribution from the gluon - gluon case is @xcite : @xmath227 where @xmath228 we note that here the lower integration limit in @xmath140 is @xmath229 , whereas the upper limit is @xmath230 . this comes from the @xmath231 function in the impact factor definition ( [ eq : a19 ] ) , which restricts the radiation of either of the two gluons into the central region of rapidity . using the symmetry of the integrand under the change of variables describing the two gluons , @xmath232 and @xmath233 ( thanks to the symmetry property ( [ symms3 ] ) of the jet function ) , we get @xmath234 in this form the upper limit of @xmath140 integration can be put to unity . in @xmath235 the lower integration limit @xmath173 can be put equal to zero . then , after the change of variable @xmath150 , we obtain @xmath236 s_j^{(3)}(\vec q-\beta\vec l,\beta\vec l , x\beta;x)\;.\end{aligned}\ ] ] in this expression one has both soft and collinear divergences . the soft divergence can be isolated in the counterterm @xmath237 s_j^{(2)}(\vec q\,;x ) \;,\end{aligned}\ ] ] which equals @xmath238 } { \varepsilon\ , ( 4\pi)^\varepsilon } \frac{\gamma^2(1+\varepsilon)}{\gamma(1 + 2\varepsilon ) } \left(\vec q^{\,\ , 2}\right)^{\varepsilon}\ , \frac{2\ , c_a}{\varepsilon } \int_0 ^ 1 dx \ , \frac{c_a}{c_f } \ , f_g(x)\ , s_j^{(2)}(\vec q\,;x ) \ ; .\end{aligned}\ ] ] after the subtraction of the soft divergence , collinear divergences still appear for @xmath156 and @xmath239 and can be isolated by the following two counterterms : @xmath240 and @xmath241 these counterterms equal @xmath242 } { \varepsilon\ , ( 4\pi)^\varepsilon } \frac{\gamma^2(1+\varepsilon)}{\gamma(1 + 2\varepsilon ) } \left(\lambda^2\right)^{\varepsilon}\ , \int_0 ^ 1 dx \ , \frac{c_a}{c_f } \ , f_g(x ) \ , s_j^{(2)}(\vec q\,;x ) \nonumber \\ & \times & c_a\ , \left(-\frac{11}{6}+\ , \varepsilon\ , \frac{67}{18}\,\right ) + \mathcal{o}(\varepsilon ) \ , , \end{aligned}\ ] ] @xmath243}{\varepsilon\ , ( 4\pi)^\varepsilon } \frac{\gamma^2(1+\varepsilon)}{\gamma(1 + 2\varepsilon ) } \left(\lambda^2\right)^{\varepsilon}\ , \int_0 ^ 1 dx \ , \frac{c_a}{c_f } \ , f_g(x ) \int_0 ^ 1 d\beta ( 1-\beta)p(1-\beta ) \\ \nonumber & \times & 2 \,c_a \left[\frac{1}{(1-\beta)_+}+2\varepsilon \left(\frac{\ln(1-\beta ) } { 1-\beta}\right)_+\,\right ] \\ & \times & s_j^{(2)}(\vec q\,;x\beta)+\mathcal{o}(\varepsilon ) \;,\end{aligned}\ ] ] where to obtain the last equation we made the change of variable @xmath163 and expanded the term @xmath244 . the finite part is therefore defined by @xmath245 the @xmath246 term , defined in ( [ real - gg ] ) , has a collinear divergence for @xmath247 . it can be isolated in the following integral : @xmath248}{\varepsilon\ , ( 4\pi)^\varepsilon } \frac{\gamma^2(1+\varepsilon)}{\gamma(1 + 2\varepsilon ) } \left(\lambda^2\right)^{\varepsilon}\ , \int_0 ^ 1 dx \ , \frac{c_a}{c_f } \ , f_g(x ) \nonumber \\ \nonumber & \times & \int_0 ^ 1 d\beta \ , 2 \ , c_a\ , p(\beta ) \ , s_j^{(2)}(\vec q\,;x\beta ) + \mathcal{o}(\varepsilon ) \ ; , \nonumber\end{aligned}\ ] ] where @xmath173 has been put equal to zero thanks to the property of lowest order jet function ( [ jetf0 ] ) . another singularity appears when @xmath152 ; in this region we can isolate the term @xmath249 } { \vec k^ { 2 } ( \vec k-\vec q\,)^2 } \ s^{(2)}_j(\vec q-\vec k;x ) \nonumber \\ & & = \ , \frac{c_a } { ( 4\pi)^{\varepsilon } } \int_0 ^ 1 dx \ , \frac{c_a}{c_f } \ , f_g(x ) \int \frac{d^{d-2}\vec k}{\pi^{1+\varepsilon } } \frac{\vec q^{\ : 2}}{\vec k^ { 2}(\vec k-\vec q\,)^2 } \ln\frac{s^2_\lambda}{\vec k^2(|\vec k|+|\vec q-\vec k|)^2 } \ s^{(2)}_j(\vec q-\vec k;x)\nonumber\;.\end{aligned}\ ] ] the finite part of @xmath246 is therefore defined by @xmath250 when the gluon part of bfkl counterterm , given in ( [ counterterm ] ) , is combined with @xmath251 , given in ( [ beta_1 g ] ) , we see that the dependence on @xmath183 disappears and we obtain @xmath252 we collect first the contributions which contain singularities given in ( [ virtualg - e ] ) , ( [ collc_fg ] ) , ( [ collc_ag ] ) , ( [ softaa ] ) , ( [ colla1a ] ) , ( [ colla2a ] ) and ( [ collb ] ) and get @xmath253}{\varepsilon\ , ( 4\pi)^\varepsilon } \frac{\gamma^2(1+\varepsilon)}{\gamma(1 + 2\varepsilon)}\ , \int_0 ^ 1 dx \ , \frac{c_a}{c_f}\ , f_g(x ) \nonumber \\ & \times&\left\lbrace \left[\left(\vec q^{\ : 2}\right)^\varepsilon \left ( \frac{11 c_a}{6}-\frac{n_f}{3 } -c_a\ln\left(\frac{s_0}{\vec q^{\ : 2}}\right ) \right ) -2\left(\lambda^2\right)^\varepsilon \left ( \frac{11 c_a}{6}-\frac{n_f } { 3}\right ) \right . \right.\nonumber \\ & + & \left.\varepsilon\left(\frac{\pi^2 c_a}{2}+\frac{13 n_f}{18}\right)\right ] \ s_j^{(2)}(\vec q\,;x)\nonumber \\ & + & \left(\lambda^{2}\right)^\varepsilon\int_0 ^ 1 d\beta \left(p_{gg}(\beta)+2 n_f\frac{c_f}{c_a}p_{qg}(\beta)\right ) s_j^{(2)}(\vec q\,;x\beta ) \nonumber \\ & + & 2 \varepsilon \int_0 ^ 1d\beta \left [ n_f \frac{c_f}{c_a } ( 1-\beta)\beta \right.\nonumber \\ & + & \left.\left . \ , 2 \ , c_a \left ( \frac{\ln(1-\beta)}{1-\beta } \right)_+(1-\beta)p(1-\beta ) \right]s_j^{(2)}(\vec q\,;x\beta ) \right\}+\mathcal{o}(\varepsilon)\;. \nonumber\end{aligned}\ ] ] then , we collect the contributions given in ( [ finitegc_fa ] ) , ( [ finitegc_aa ] ) , ( [ finitea ] ) , ( [ finiteb ] ) , and get ( transforming to the form used in @xcite ) @xmath254 @xmath255 @xmath256 @xmath257 @xmath258 @xmath259\ ] ] @xmath260 \int \frac{d^{2 } \vec l}{\pi \vec l^2}\right.\ ] ] @xmath261 @xmath262 @xmath263 @xmath264 } { \vec k^2(\vec q-\vec k)^2}\ , s_j^{(2)}(\vec k;x)\right\rbrace\right]\;.\ ] ] besides , we define @xmath265 given in eq . ( [ softca+c_gluon ] ) . another contribution originates from the collinear and charge renormalization counterterms , see eqs . ( [ charge.count.t ] ) and ( [ c.count.t.a ] ) , @xmath266 } { \varepsilon \,(4\pi)^\varepsilon } \int_0 ^ 1 dx\ , f_g(x ) \left[\left(\mu_r^2\right)^\varepsilon \left(\frac{11c_a}{6}-\frac{n_f}{3 } \right)\ , \frac{c_a}{c_f } \ , s_j^{(2)}\left(\vec q\,;x\right ) \right.\ ] ] @xmath267 \ s^{(2)}_j(\vec q\,;x\beta)\right ] \ ; . \label{c.t .- g}\ ] ] finally , the gluon part of the jet impact factor is given by the sum of the above four contributions and can be presented as a sum of two terms : @xmath268 @xmath269}{\varepsilon \,(4\pi)^\varepsilon } \frac{\gamma^2(1+\varepsilon)}{\gamma(1 + 2\varepsilon ) } s^{(2)}_j(\vec q\,;x ) \right]\ ] ] and @xmath270 @xmath271 \ s^{(2)}_j(\vec q\,;x)\ ] ] @xmath272 \,\ln\frac{\lambda^2}{\mu_f^2 } \ , \ s^{(2)}_j(\vec q\,;x\beta)\ ] ] @xmath273 \,c_a + \,2\ , n_f\,\frac{c_f}{c_a } \beta(1-\beta)\right]\ s^{(2)}_j(\vec q\,;x\beta)\ , \right\rbrace\;.\ ] ] we have recalculated the jet vertices for the cases of quark and gluon in the initial state , first found in the papers by bartels _ et al . _ our approach is different since the starting point of our calculation is the known general expression for nlo bfkl impact factors , given in ref . @xcite , applied to the special case of partons in the initial state . nevertheless , in many technical steps we followed closely the derivation of refs . @xcite . we checked our result by replacing everywhere the jet selection functions @xmath274 and @xmath92 by 1 and performing all integrations : we got back to known results for the nlo quark and gluon impact factors @xcite . let us discuss now the infrared finiteness of the obtained result for the jet impact factor . the nlo correction to the jet vertex ( impact factor ) has the form @xmath275 where each part is the sum of the quark and gluon contributions , @xmath276 given in eqs . ( [ singular2 ] ) , ( [ non - singular2 ] ) and in eqs . ( [ singular2 g ] ) , ( [ non - singular2 g ] ) , respectively . @xmath277 and @xmath278 are manifestly finite . for @xmath279 we have @xmath280 @xmath281 @xmath269}{\varepsilon \,(4\pi)^\varepsilon } \frac{\gamma^2(1+\varepsilon)}{\gamma(1 + 2\varepsilon ) } s^{(2)}_j(\vec q\,;x ) \right ] \ , .\ ] ] having the explicit form of the lowest order jet function ( [ jetf0 ] ) , it is easy to see that the integration of @xmath279 over @xmath50 with any function , regular at @xmath282 , will give some finite result . in particular , the finite result will be obtained after the convolution of @xmath279 with bfkl green s function , see eq . ( [ crx ] ) , which is required for the calculation of the jet cross section . this may look somewhat surprising , but , in fact , it is not so since the impact factor is not , strictly speaking , a physical observable . the divergence in ( [ singular2end ] ) arises from virtual corrections and , precisely , from the factor @xmath283 entering the definition of the impact factor . in the computation of physical impact factors this divergence is cancelled by the one arising from the integration in the first term of eq . ( [ singular2end ] ) , which is related with real emission . in the calculation of the jet vertex the @xmath50 integration is `` opened '' and , therefore , there is no way to get the divergence needed to balance the one arising from virtual corrections . however , in the construction of any physical cross section , the jet vertex is to be convoluted with the bfkl green s function , which implies the integration over the reggeon transverse momentum @xmath50 . in our approach the energy scale @xmath47 remains untouched and need not be fixed at any definite scale . the dependence on @xmath47 will disappear in the next - to - leading logarithmic approximation in any physical cross section in which jet vertices are used . however , the dependence on this energy scale will survive in terms beyond this approximation and will provide a parameter to be optimized with the method adopted in refs . @xcite . in order to compare our results with the ones of @xcite , we need to perform the transition from the standard bfkl scheme with arbitrary energy scale @xmath47 to the one used in @xcite , where the scale of energy depends on the reggeon momentum . the change to the scheme where the energy scale @xmath47 is replaced to any factorizable scale @xmath284 leads to the following modification of each impact factor ( @xmath285 ) , see @xcite , @xmath286 where @xmath287 and @xmath288 are the lowest order impact factor and bfkl kernel . therefore changing from @xmath47 to @xmath289 we obtain the following replacement in our result for the jet impact factor : @xmath290 @xmath291 \ ; .\ ] ] note that @xmath292 is not singular at @xmath293 and , therefore , it can be calculated at @xmath294 . such contribution to the jet impact factors , @xmath292 , in the considered scheme with @xmath295 produces a completely equivalent effect on the physical jet cross section as the factors @xmath296 and @xmath297 which enter eq . ( 76 ) of @xcite ( see eqs . ( 101 ) , ( 102 ) in @xcite for the definition of @xmath296 , @xmath297 ) . therefore , for the final comparison one needs to consider our results for @xmath277 and @xmath278 ( modulo the appropriate normalization factor ) with the ones given in eq . ( 105 ) of @xcite and eq . ( 67 ) of @xcite for the quark and gluon contributions , respectively . for this purpose we identify , following @xcite , the @xmath159 parameter with the collinear factorization scale @xmath298 . in the gluon contribution we found a complete agreement , whereas in the quark case we point out to the following misprints in the eq . ( 105 ) of @xcite : * in the real @xmath145 term , the expression * @xmath299*@xmath300*@xmath88 * must be replaced by * @xmath299*@xmath301*@xmath88 * , both in the numerator and the denominator ; * @xmath302 in front of the same term has to be replaced by @xmath303 or , similarly , by @xmath304 ; * just after it , @xmath305 is to be interpreted as @xmath300 . exactly these corrections were pointed out first in ref . @xcite . in that paper the results for the quark and gluon contributions to the jet vertices are presented in a form different from the one used in the original calculation of refs . however , one can see after some analysis , that these two forms turn to be completely equivalent , contrary to the impression that the text before eq . ( 3.4 ) in ref . @xcite may induce . recently another work devoted to the calculation of the jet impact factor appeared @xcite . it is an interesting application of the lipatov s effective action method @xcite to the problem in question . within this method , a particular regularization of the longitudinal divergences has been proposed . in the traditional approach , these divergences are regularized by the account of the bfkl counterterm , see eq . ( [ counterterm ] ) . unfortunately more work with the results of @xcite seems still to be done in order to get an independent check of the final results for the jet impact factors . d.i . thanks the dipartimento di fisica delluniversit della calabria and the istituto nazionale di fisica nucleare ( infn ) , gruppo collegato di cosenza , for the warm hospitality and the financial support . this work was also supported in part by the grants rfbr-09 - 02 - 00263 and rfbr-11 - 02 - 00242 . thanks a. sabio vera for a very stimulating conversation . j. bartels , d. colferai and g. p. vacca , eur . j. c * 24 * ( 2002 ) 83 [ arxiv : hep - ph/0112283 ] . j. bartels , d. colferai and g. p. vacca , eur . j. c * 29 * ( 2003 ) 235 [ arxiv : hep - ph/0206290 ] . mueller and h. navelet , nucl . b * 282 * ( 1987 ) 727 . fadin , e.a . kuraev , l.n . lipatov , phys . lett . b * 60 * ( 1975 ) 50 ; e.a . kuraev , l.n . lipatov and v.s . fadin , zh . * 71 * ( 1976 ) 840 [ sov jetp * 44 * ( 1976 ) 443 ] ; * 72 * ( 1977 ) 377 [ * 45 * ( 1977 ) 199 ] ; ya.ya . balitskii and l.n . lipatov , sov . * 28 * ( 1978 ) 822 . m. ciafaloni and g. rodrigo , jhep * 0005 * ( 2000 ) 042 [ hep - ph/0004033 ] . j. bartels , s. gieseke and c. f. qiao , phys . d * 63 * ( 2001 ) 056014 [ _ erratum - ibid . _ * 65 * ( 2002 ) 079902 ] [ arxiv : hep - ph/0009102 ] ; j. bartels , s. gieseke and a. kyrieleis , phys . d * 65 * ( 2002 ) 014006 [ arxiv : hep - ph/0107152 ] ; j. bartels , d. colferai , s. gieseke and a. kyrieleis , phys . d * 66 * ( 2002 ) 094017 [ arxiv : hep - ph/0208130 ] ; j. bartels and a. kyrieleis , phys . rev . d * 70 * ( 2004 ) 114003 [ arxiv : hep - ph/0407051 ] ; v.s . fadin , d.yu . ivanov and m.i . kotsky , phys . * 65 * ( 2002 ) 1513 [ yad . fiz . * 65 * ( 2002 ) 1551 ] [ arxiv : hep - ph/0106099 ] ; nucl . phys . b * 658 * ( 2003 ) 156 [ arxiv : hep - ph/0210406 ] . b. abbott _ et al . _ [ d0 collaboration ] , phys . * 84 * ( 2000 ) 5722 [ arxiv : hep - ex/9912032 ] . s. abachi _ et al . _ [ d0 collaboration ] , phys . * 77 * ( 1996 ) 595 [ arxiv : hep - ex/9603010 ] . v. del duca and c. r. schmidt , phys . d * 49 * ( 1994 ) 4510 [ arxiv : hep - ph/9311290 ] . stirling , nucl . b * 423 * ( 1994 ) 56 [ arxiv : hep - ph/9401266 ] . l.h . orr and w.j . stirling , phys . d * 56 * ( 1997 ) 5875 [ arxiv : hep - ph/9706529 ] . j. kwiecinski , a.d . martin , l. motyka and j. outhwaite , phys . lett . b * 514 * ( 2001 ) 355 [ arxiv : hep - ph/0105039 ] . a. sabio vera , nucl . b * 746 * , 1 ( 2006 ) [ hep - ph/0602250 ] . a. sabio vera and f. schwennsen , nucl . b * 776 * ( 2007 ) 170 [ arxiv : hep - ph/0702158 ] . c. marquet and c. royon , phys . d * 79 * ( 2009 ) 034028 [ arxiv:0704.3409 [ hep - ph ] ] . d. colferai , f. schwennsen , l. szymanowski , s. wallon , jhep * 1012 * ( 2010 ) 026 [ arxiv:1002.1365 [ hep - ph ] ] . d.yu . ivanov and a. papa , nucl . b * 732 * ( 2006 ) 183 [ arxiv : hep - ph/0508162 ] ; eur . j. c * 49 * ( 2007 ) 947 [ arxiv : hep - ph/0610042 ] ; f. caporale , a. papa and a. sabio vera , eur . j. c * 53 * ( 2008 ) 525 [ arxiv:0707.4100 [ hep - ph ] ] .

the dynamic orchestration of starbirth activity in the starburst - ringed galaxy m94 ( ngc 4736 ) is investigated using images from the ultraviolet imaging telescope ( fuv - band ) , hubble space telescope ( nuv - band ) , kitt peak 0.9-m telescope ( h , r , and i bands ) , and palomar 5-m telescope ( b - band ) , along with spectra from the international ultraviolet explorer and lick 1-m telescopes . the wide - field uit image shows fuv emission from ( a ) an elongated nucleus , ( b ) a diffuse inner disk , where h is observed in _ absorption _ , ( c ) a bright inner ring of h ii regions at the perimeter of the inner disk ( r = 48 = 1.1 kpc ) , and ( d ) two 500-pc size knots of hot stars exterior to the ring on diametrically opposite sides of the nucleus ( r = 130 = 2.9 kpc ) . the hst / foc image resolves the nuv emission from the nuclear region into a bright core and a faint 20-long `` mini - bar '' at a position angle of 30 deg . optical and iue spectroscopy of the nucleus and diffuse inner disk indicates a yr - old stellar population from low - level starbirth activity blended with some liner activity . analysis of the h , fuv , nuv , b , r , and i - band emission along with other observed tracers of stars and gas in m94 indicates that most of the star formation is being orchestrated via ring - bar dynamics involving the nuclear mini - bar , inner ring , oval disk , and outer ring . the inner starburst ring and bi - symmetric knots at intermediate radius , in particular , argue for bar - mediated resonances as the primary drivers of evolution in m94 at the present epoch . similar processes may be governing the evolution of the `` core - dominated '' galaxies that have been observed at high redshift . the gravitationally - lensed `` pretzel galaxy '' ( 0024 + 1654 ) at a redshift of.5 provides an important precedent in this regard .

astro-ph0011101

star - forming rings or `` pseudorings '' are common to early and intermediate - type spiral galaxies ( cf . athanassoula & bosma 1985 ; buta , purcell , & crocker 1995 ; buta & combes 1996 ) , including our own milky way galaxy ( cf . gusten & mezger 1982 ; clemens , sanders , & scoville 1988 ; waller 1990a ) . such ring - like accumulations of gas and associated starbirth activity may have helped to build the inner parts of many primeval disk galaxies ( friedli & benz 1995 ) , as exemplified by the recent discovery of a starburst - ringed galaxy at @xmath4 ( colley , tyson , & turner 1996 ; tyson et al . the formation and maintenance of these starburst rings are often attributed to orbital resonances with rotating bar or `` oval '' asymmetries in the stellar disks ( cf . combes 1994 , byrd et al . 1994 ; combes et al . 1995 ; buta & combes 1996 and references therein ) . however , other dynamical mechanisms including gravitational instabilities ( elmegreen 1992 , 1994 ; kenney & jogee 1997 ) , outward propagating star formation ( walker , lebofsky , & rieke 1988 ; waller , gurwell , & tamura 1992 ) , and even radially - driven pileups from nuclear outbursts ( waller et al . 1992 ; tenorio - tagle et al . 1997 ) may play significant roles in orchestrating some of the starburst rings that are observed . as the closest early - type spiral galaxy of low inclination , m94 ( ngc 4736 ) has received concentrated attention from both observers and theorists . this ( r)sa(r)ab - type galaxy ( de vaucouleurs et al . 1991 ) is noted for its inner ring of ongoing starburst activity ( r @xmath5 45@xmath1 ) , oval stellar distribution at intermediate radius ( r @xmath5 220@xmath1 ) ( cf . mulder & van driel 1993 ; mulder 1995 ; mollenhoff , matthias , & gerhard 1995 ) , and outer stellar ring near its de vaucouleurs radius ( r@xmath6 @xmath5 330@xmath1 ) . * figure 1 * ( extracted from the digital sky survey using the skyview utility [ mcglynn , scollick , & white 1996 ] ) shows the outermost portions of m94 , highlighting the oval disk and outer pseudoring . the inner starburst ring is a prominent source of h@xmath0 , h i , and co emission ( smith et al . 1991 ; mulder & van driel 1993 ; gerin , casoli , & combes 1991 ) . the discovery of compact thermal & nonthermal radio sources in the ring ( duric & dittmar 1988 ) indicates the presence of dense h ii regions and young snrs . the ring s velocity field can be described by a combination of circular rotation with velocities of order 200 km / s and residual non - circular motions of order 15 km / s ( mulder 1995 ) to 25 km / s ( buta 1988 ) , depending on the adopted inclination and major axis position angle . interior to the ring , the bright bulge and inner disk show twisted isophotes at red and near - ir wavelengths , indicative of a weak bar - like distortion ( beckman et al . 1991 ; shaw et al . 1993 ; mollenhoff et al . 1995 ) . fir and co observations interior to the ring reveal a rich ism with gas surface densities exceeding that of the ring ( smith & harvey 1994 ; garman & young 1986 ; gerin et al . 1991 ; wong & blitz 2000 ) . optical spectroscopy of the nuclear region yields liner - type emission lines along with absorption lines from the circumnuclear stellar population , consistent with an early main - sequence stellar turnoff ( a4a7 ) and corresponding age of @xmath2500 myr ( pritchett 1977 ; keel 1983 ; taniguchi et al . . further support for a young central population comes from nir spectroscopy which shows deep co absorption bands from red giant and asymptotic giant branch stars of similar age ( walker et al . these authors have proposed an outward propagating mode of star formation , whereby ngc 253 , m82 , m94 , and m31 represent increasingly evolved versions of the same starbursting sequence . although the kinematics of the ring show very little evidence for outward _ expanding motions _ ( contrary to prior claims of bulk expansion [ van der kruit 1974 ; 1976 ] ) , they also do not preclude a scenario for _ radially propagating _ star formation . other investigators have modeled the inner and outer rings in terms of _ resonant dynamics _ mediated by bar or `` oval '' potentials interior to the rings ( gerin et al . 1991 ; shaw et al . 1993 ; mollenhoff et al . 1995 ; mulder & combes 1996 ) , with the observed non - circular motions resulting from dispersion orbits near the lindblad resonances ( buta 1988 ) . in this paper , we present and discuss new observational clues to the dynamical mechanisms governing the star formation in m94 . the ultraviolet images obtained by the ultraviolet imaging telescope , in particular , reveal hitherto unrecognized patterns of recent star formation whose presence lends further support to the hypothesis of galaxy evolution via bar - mediated resonances . the various imaging and spectroscopic observations and reductions are described in section 2 . the resulting fuv , nuv , h@xmath0 , r , and i - band emission morphologies are presented and compared in section 3 . radial intensity profiles and other photometric comparisons are discussed in section 4 . uv and optical spectroscopy of the inner disk and nucleus is presented in section 5 . kinematic properties and inferred dynamical scenarios are considered in section 6 . our summary of the wavelength - dependent morphological , spectro - photometric , and dynamical properties of m94 appears in section 7 , wherein evolutionary implications are discussed . in the following sections , we assume a distance to m94 of 4.6 ( 75/h@xmath7 ) mpc , based on the galaxy s recession velocity of 345 km / s with respect to the local group ( sandage & tammann 1981 ) . the corresponding spatial scale is 22.3 pc / arcsec . unresolved sources imaged by the hst / foc provide additional constraints on the distance , as discussed in section 4 . we adopt a nominal inclination of 40@xmath8 and major - axis position angle of 120@xmath8 , while recognizing that both of these quantities may vary significantly with radius and with measuring technique ( e.g. morphological vs. kinematic determinations ) ( bosma , van der hulst , & sullivan 1977 ; buta 1988 ; mulder & van driel 1993 ; mulder 1995 ; mollenhoff et al . 1995 ; wong & blitz 2000 ) . a log of the ultraviolet and visible imaging is presented in * table 1*. a listing of complementary uv and visible spectra is shown in * table 2*. the ultraviolet imaging telescope ( uit ) imaged m94 in the far - ultraviolet ( @xmath9 = 1521 , @xmath10 = 354 ) on 1995 march 12 as part of the 16-day spacelab/_astro-2 _ mission aboard the space shuttle endeavour . this wide - field telescope images 40-arcmin fields of view at @xmath23 arcsec resolution . in the case of m94 , the uit image represents the only extant uv image of the entire galaxy ( see * figure 2a * ( plate xxx ) ) . the 1040-sec exposure was obtained with a dual - stage image intensifier with csi photocathodes and was recorded on carbon - backed iiao kodak film . after processing of the film and scanning of the emulsion , the resulting digitized `` density image '' was fog - subtracted , flat - fielded , linearized to `` exposure units , '' and calibrated to flux units using iue observations of standard stars ( cf . stecher et al . 1992 , 1997 ; waller et al . 1995 and references therein ) . correction for image distortion produced by the magnetically focused image intensifiers was carried out according to the procedures described by greason et al . the resulting corrections amounted to a few arcsec in the field center ( which includes all of m94 s fuv emission ) increasing to 1020 arcsec near the edge of the 40 arcmin field of view . astrometry was tied to 10 compact knots evident in both the fuv and b - band images ( see next subsection ) . positions in the resulting distortion - corrected image are good to @xmath23 arcsec , and the spatial resolution is of similar magnitude . the hubble space telescope s faint object camera ( foc ) imaged the center of m94 in the near - ultraviolet ( @xmath9 = 2300 , @xmath10 = 500 ) on 18 , july 1993 before the optical repair mission as part of a uv imaging survey of 110 large nearby galaxies ( maoz et al . 1995 , 1996 ) . after standard stsci pipeline processing , the 596 sec nuv exposure has a @xmath11 field of view at 0.022@xmath1/pixel ( see * figure 2a * ( plate xxx ) ) . the spherical aberration of hst s primary mirror resulted in a point - spread function ( psf ) featuring a sharp core with fwhm @xmath5 0.05@xmath1 and about 15% of the total light surrounded by an extensive halo of several arcsec radius containing most of the energy . following maoz et al . ( 1996 ) , our calibration of the detected foc counts into flux densities assume a conversion of @xmath12 erg s@xmath13 @xmath14 @xmath13/count s@xmath13 , while noting that the photflam conversion in the image header is @xmath15 erg s@xmath13 @xmath14 @xmath13/count s@xmath13 . flux uncertainties are estimated at @xmath25% over large areas increasing to @xmath220% for compact sources ( maoz et al . 1995 , 1996 , and references therein ) . a wide - field ( @xmath16 ) b - band image was obtained with the palomar 5-m telescope and tek3 ccd camera ( 1024 @xmath17 1024 pix ) on 1994 february 11 under hazy skies . this 600-sec exposure is saturated in the central 2 arcmin , but contains high s / n detections of the oval disk and parts of the outer ring . astrometry of this image is tied to the positions of several foreground stars , as measured on a corresponding image in the digitized sky survey ( stsci 1994 ) . groundbased h@xmath0 , r , and i - band images of m94 were obtained with the now - retired kpno 0.9-m telescope and rca-1 ccd camera ( 508 @xmath17 316 pix ) on 17 , february 1986 . these images have @xmath18 fields of view at 0.86@xmath1/pix . sky conditions varied from photometric to hazy , yielding psfs of about 2@xmath1 ( fwhm ) . astrometry is tied to 5 foreground stars that are common to the kpno ( h@xmath0 , r , i - band ) and palomar ( b - band ) images . for calibration purposes , spectrophotometric standard stars ( bd26@xmath82606 & hd84937 ) were imaged before and after the target imaging . subtraction of the red continuum from the h@xmath0-band image was carried out with the r - band image according to the formulations in waller ( 1990b ) , whereby corrections were made for the @xmath238% [ n ii]@xmath196548 , 6584 contribution to the total h@xmath0 @xmath20 [ n ii ] line emission and resulting @xmath215% contamination of the h@xmath0 image after transmission by the 36 bandwidth h@xmath0 filter . a pure red - continuum image was also produced , by removing the contaminating h@xmath0 + [ n ii ] line emission from the r - band image ( waller 1990b ) ( see * figure 2a * ( plate xxx ) ) . photometry of the h@xmath0 emission from the @xmath21 field yields a total flux of 9.9 @xmath17 10@xmath22 erg @xmath14 s@xmath13 , 14% higher than that determined by kennicutt & kent ( 1983 ) within a 7@xmath23-diameter aperture ( after correcting for a 38% [ n ii ] contamination within their 100 bandwidth ) . the starburst ring ( r = 30@xmath1 @xmath24 60@xmath1 ) is measured to have f(h@xmath0 ) = 9.6 @xmath17 10@xmath22 erg @xmath14 s@xmath13 , which is nearly 1.8 times higher than that obtained by smith et al . ( 1991 ) from the 75 bandwidth image of pogge ( 1989 ) . some of these discrepancies can be attributed to the varying bandpasses and corresponding uncertainties in the [ n ii ] emission being transmitted , vagaries in the continuum subtraction , problematic h@xmath0 _ absorption _ produced by a - type stars in the inner disk ( see section 4.2 ) , and final calibration ( smith et al . because of hazy conditions during the i - band exposure , calibration of the i - band image was done by bootstrapping to the i - band radial intensity profiles resulting from the photoelectric photometry of munoz - tunon et al . ( 1989 ) and the @xmath25 color profiles of beckman et al . ( 1991 ) . fuv ( 1200 to 2000 ) and nuv ( 2000 to 3200 ) spectra of m94 s inner disk were obtained from the international ultraviolet explorer ( iue ) archive . the iue data were accessed via the iue data analysis center at nasa / gsfc and are representative of the new spectral image processing system ( newsips ) . * table 2 * lists the image numbers , dates , exposure times , nominal positions and roll angles of the low - resolution fuv ( swp ) and nuv ( lwp & lwr ) iue spectra . here , swp , lwp , and lwr respectively refer to iue s short - wavelength prime , long - wavelength prime , and long - wavelength redundant cameras . * figure 3 * shows the iue apertures ( @xmath26 ) on grey - scale images of the fuv emission , where the mapping is based on the nominal positions and roll angles of the fuv ( swp ) and nuv ( lwp & lwr ) observations . although the accuracy of the nominal positions is of order @xmath2710@xmath1 , * figure 3 * indicates that the apertures were most likely sampling the inner disk rather than the starburst ring . * table 2 * notes which apertures are filled with disk ( d ) and/or nuclear ( n ) emission , based on visual inspection of the overlays . the limiting spectral resolution of these data is about 6 . groundbased visible spectra ( 3800 to 7500 ) were obtained from the lick 1-m nickel telescope and image dissecting spectrograph ( ids ) . the summed spectrum of the central region ( 8.1 arcsec circular aperture ) is a mean - flux average of 64 min in blue and red grating settings , with 16 min in an intermediate setting to ensure the overlap area was well calibrated . the spectrum has been rebinned to 2.5- pixels from the original , which still oversamples the resolution of @xmath2810 fwhm . we note that adjacent pixels in the ids spectrum are not statistically independent , yielding detections similar to those of a non - centroiding photon counter which spreads single photons across several output pixels . we complement this composite spectrum with analysis of the image - dissector scanner data presented by keel ( 1983 ) and discuss a new high - resolution spectrum , obtained with the 2.5-m isaac newton telescope ( int ) on la palma , using the intermediate - dispersion spectrograph with image photon - counting system ( ipcs ) detector . the 1 " slit was oriented approximately along the major axis ( pa 135@xmath29 ) for this 1000-second exposure . there is useful signal over 45 spatial increments of 0.6@xmath1 each . we consider equivalent widths of the lines along with spectral slopes and discontinuities of the continuum emission . * figure 2a * ( plate xxxx ) shows the dramatically different morphologies that are detected at ultraviolet and visible wavelengths . the inner disk and bulge component that is so prominent at r - band completely disappears in the fuv image . the oval disk at intermediate radius also has no fuv counterpart . instead , the fuv image is characterized by ( a ) an extended and elongated nucleus , ( b ) a diffuse inner disk , ( c ) a bright inner ring at the perimeter of the inner disk ( r = 48@xmath1 = 1.1 kpc ) , and ( d ) two 500-pc size knots exterior to the ring ( r = 130@xmath1 = 2.9 kpc ) . the insert in * figure 2a * ( plate xxxx ) contains the hst / foc pre - costar image of the nuclear region ( maoz et al . 1995 , 1996 ) . this near - uv image shows a marginally - resolved nucleus ( fwhm @xmath5 0.1@xmath1 ) embedded in a bright core of @xmath21@xmath1 ( 22 pc ) diameter along with a faint `` mini - bar '' that has a total length of 20@xmath1 ( 450 pc ) at a position angle of approximately 30@xmath8 . low - level ripples in the emission ( which are included in the estimate of the bar s total length ) are evident 2@xmath1 , 5@xmath1 and 9@xmath1 to the sw of the nucleus . maoz et al . ( 1995 ) attribute these features to bow shocks or tidal arms resulting from a recent merger event . however , we see that the ripples overlap with the larger concentric arcs that are evident in the uit / fuv image at projected radii of @xmath29@xmath1 and 15@xmath1 ( see next subsection ) . similar ripples are also present in an archival hst / wfpc2 v - band image of this region with connections to larger spiral arcs . to the north of the nucleus are two point sources whose flux densities are consistent with those of single b - type supergiant stars ( see section 4 ) . the uit / fuv emission from the inner disk includes concentric fuv arcs to the sw of the nucleus which can be traced for @xmath2@xmath30 radians ( @xmath2700 pc ) . these arcs are of low contrast and , because of the noise characteristics of the uit imaging , are not amenable to typical contrast enhancement techniques ( e.g. median filtering ) . however , they do show up as enhancements in the radial distribution of intensities ( see * figure 2b * ) . typical spacing between the arcs is 9@xmath1 ( 200 pc ) , with additional features at 15@xmath1 and 34@xmath1 . the arcs themselves appear to show some substructure at the limits of resolution . contrary to a merging scenario ( maoz et al . 1995 ) , which would seem to be precluded by the lack of significant tidal effects beyond the inner few hundred parsecs , these features more likely indicate an orbital dynamic at work in the inner disk . at an interpolated orbital velocity of 140 km / s and corresponding shear rate of -1360 km@xmath13 kpc@xmath31 ( see section 6 ) , such features could have been differentially swept out over a timescale of @xmath220100 myr . this estimated timespan is consistent with a population of late b - type stars whose main - sequence lifetimes are of similar duration . the uit / fuv image of the inner disk also shows a brighter arc to the north , just inside the starburst ring at a radius of 40@xmath1 ( see * figure 2b * ) , along with widespread diffuse emission at a level amounting to @xmath215% the mean surface brightness of the ring . the nature of these resolved and unresolved fuv components is uncertain , although some clues can be obtained from the longer wavelength imagery . at h@xmath0 the inner disk appears _ in absorption _ with respect to the continuum emission from the underlying population of stars . the strongest absorbers at h@xmath0 are the photospheric atmospheres of b and a - type stars , whose temperatures are sufficiently cool for hydrogen to remain neutral and sufficiently warm for significant population of the h balmer ( n=2 ) electronic energy level ( cf . mihalas 1978 ) . in the uit / fuv image , the diffuse light from the inner disk probably arises from these same stars . indeed , _ the uit / fuv image represents the first view of this young stellar population , unconfused by the longer - wavelength emission from the older inner disk and bulge components_. * figure 4 * compares a spatially - filtered r - band image with an @xmath25 color image of the inner disk . the spatially - filtered r - band image ( * fig . 4a * ) was created by median smoothing the r - band image over a @xmath32 window and then subtracting the smoothed image from the original . the resulting fine - scale structure includes the nuclear `` mini - bar '' previously noted by mollenhoff et al . ( 1995 ) and mulder ( 1995 ) , along with a `` dark '' spiral arc of diminished emissivity to the west and other flocculent spiral structures associated with the starburst ring . the archival hst / wfpc2 v - band image of this region resolves the flocculent structure into spiral dust lanes of high pitch angle which cross through the nearly circular ring . the @xmath25 color image ( * fig . 4b * ) shows reddening along the western spiral arc seen in ( * fig . 4a * ) and in another arc to the ne , along with blue knots in the starburst ring . _ none of the fine - scale structures and reddened features in the optical inner disk have counterparts in the uit / fuv image . _ therefore , neither emitting nor absorbing structures , nor reddening dust lanes in the _ old disk _ can be defining the observed concentric fuv arcs to the sw and ne . we infer that the fuv arcs are recently - generated structures of emitting b and a - type stars and/or scattering clouds of low dust optical depth . we further speculate that the emitting and/or scattering fuv sources , being relatively young , are distributed in a much thinner disk than that associated with the optical fine - scale structures and reddened features , thus explaining the morphological differences between these two wavelength regimes . further insights on the stellar content of the inner disk can be gained from the iue spectroscopy presented in section 6 . the inner starburst ring in m94 is the single dominant emission feature in the fuv , h@xmath0 , and radio continuum bands . rectification of the galaxy to its nominal face - on orientation ( pa = 120@xmath8 @xmath33 [ mulder & van driel 1993 ] ) shows that the ring is almost perfectly circular , with a mean radius of 47@xmath1 ( 1.1 kpc ) and fwhm of 21@xmath1 ( 0.49 kpc ) . the latter measurement is based on doubling the measured hwhm of the emission _ beyond _ the mean radius , thereby excluding the diffuse contribution interior to the ring . adopting an inclination of 35@xmath8 , as in the dispersion orbit model of buta ( 1988 ) , would yield a finite but small ellipticity of 0.063 , with a 16@xmath8 offset between the kinematic line of nodes and projected major axis . we are not able to further constrain the ring s orientation , and so shall continue to assume an inclination of 40@xmath8 , based on the h i study of mulder & van driel ( 1993 ) . a comparison of the fuv and h@xmath0 morphologies reveals strong similarities . * figure 5a * shows the h@xmath0 emission , and * figure 5b * shows the ratio of h@xmath0 and fuv intensities in the inner disk and ring . except for the spiral - like ridges of enhanced h@xmath0 emission to the wnw and ese , very little coherent variation in this ratio is evident in the ring . in particular , the intensity ratios do not show any radial displacement in the fuv emission relative to that at h@xmath0 . from this lack of radial structuring in the h@xmath0/fuv ratio , we can conclude that _ little evidence is found for outward or inward propagating starburst activity _ , as explained below . a morphological comparison of h@xmath0 and fuv emitting regions provides a useful means of tracking sequential patterns of star formation in disk galaxies . the h@xmath0 emission is dominated by the most massive and short - lived stars ( m @xmath34 20 m@xmath35 ) , while the fuv emission arises mostly from less massive , longer- lived stars ( 20 @xmath28 m / m@xmath35 @xmath28 2.5 ) . for an evolving star cluster with an imf typical of local star - forming regions , the h@xmath0 emission typically reaches a maximum within a million years of the star - forming episode , and quickly decays with an e - folding timescale of about 3 myr . by contrast , the fuv emission reaches a maximum at about 3 myr ( due to the onset of b - type supergiants ) , decaying to 1/e in yet another 3 myr . the h@xmath0/fuv ratio is seen to decrease by 2 orders of magnitude after 10 myr ( e.g. hill et al . subsequently , the h@xmath0 emission will vanish , while the fuv declines slowly for the next few 100 myrs . in the outward propagating starburst scenario , the h@xmath0 emission would concentrate where the propagating wavefront is located on the outer perimeter of the fuv - emitting ring . inward propagating scenarios would have the h@xmath0 emission interior to the fuv ring . spatial displacements of h@xmath0 and fuv emission have been observed across the spiral arms of m51 ( oconnell 1997 ; petit et al . 1996 ) , m74 ( marcum et al . 1997 ) , and ngc 4258 ( courtes et al . 1993 ) . these displacements have been interpreted as the result of spiral density waves concentrating gas along the insides of the spiral arms and subsequent migration of the evolving clusters past the spiral wave fronts . given the resolution of our images ( @xmath2100 pc ) and characteristic timescale between h@xmath0 and fuv maxima ( 3 myr ) , any residual propagation of starburst activity would have to proceed at a speed less than 35 km / s to avoid detection . implications of this propagation speed limit are discussed in section 7 . the arm - like enhancements in the h@xmath0/fuv ratio correspond to similar features in a recent co mapping with the bima interferometer ( wong et al . these arm - like extensions away from the ring have further analogues at b - band , where a complex spiral pattern is evident ( see next subsection ) . one of the most remarkable aspects of m94 s fuv emission morphology is the pair of knots on diametrically opposite sides of the nucleus . both knots have projected radii of approximately 130@xmath1 ( 2.9 kpc ) . a line drawn through the knots intersects the nucleus at a position angle of 105@xmath8 , intermediate between the pas of the inner starburst ring ( 127@xmath8 ) and the oval disk ( 95@xmath8 ) ( mollenhoff et al . the two knots are roughly 20@xmath1 ( 500 pc ) in size , with the eastern knot showing the more complex structure . only the eastern knot shows significantly in our continuum - subtracted h@xmath0 image ( * figure 5 * ) . other much smaller knots of fuv emission appear at low s / n to the sw , ssw , ene , and nne . the sw , ssw , and ene knots have faint counterparts in the light of h@xmath0 , while all of these features have counterparts at b - band ( see below ) . * figure 6a * and * figure 6b * respectively show the b - band image before and after spatial filtering . the filtering in this case involves median smoothing over a @xmath36 window and then dividing the original image by this smoothed image . median normalized spatial filtering _ reveals fine - scale structure over a wide range of surface brightnesses ( e.g. waller et al . 1998 ) . here , it highlights the starburst ring and bisymmetric knots as significant enhancements above the disk at b - band . the filtered image also reveals many flocculent spiral arms outside of the ring , whose relation with the knots is somewhat ambiguous . the western knot appears part of a dominant arm with a `` shingled '' morphology connecting to the ring in the northeast . the eastern knot appears to be associated with several arms , including one which would be the symmetric counterpart to the arm that links the western knot to the ring . the marginally detected fuv knots to the sw , ssw , ene , and nne all show enhanced b - band emission from associations of massive stars among the myriad spiral arms . in summary , the bi - symmetric fuv knots have b - band counterparts which appear to be part of the complex spiral arm structure at these radii . their symmetric prominence on opposite sides of the nucleus , however , requires a dynamical explanation that is spatially more specific than that of the arms . similarly , the starburst ring can be regarded as an especially bright , tightly wound component of the overall spiral pattern , whose prominence also indicates special dynamical circumstances . bi - symmetric knots or plumes have been noted in other ringed - barred galaxies , including ngc 1433 ( buta 1986 ) , ngc 7020 ( buta 1990 ) , ngc 7098 ( buta 1995 ) , and ic 4214 ( buta & combes 1996 ) . as discussed in these papers , the symmetric features are likely tracing dynamical resonances that are connected with the general ring - bar phenomenon . from the uit and hst / foc images , circular aperture photometry of the nucleus out to a radius of 5@xmath1 yields @xmath37 = 14.7 mag and @xmath38 = 13.5 mag , respectively . the resulting @xmath39 color of 1.2 mag is significantly redder than those derived from iue spectra of the inner disk ( see next section ) . between 5@xmath1 10@xmath1 radii , the non - nuclear emission is significantly bluer than those derived using iue , with @xmath40 . the latter colors indicate the presence of late b and hotter stars , depending on the amount of reddening ( fanelli et al . estimates of the reddening have been derived from the visible colors ( smith et al . 1991 ; see also next subsection ) , visible spectra and subsequent modelling of the stellar populations ( taniguchi et al . 1996 ; pritchett 1977 ) . the resulting estimates of visual extinction are 0.51.0 mag in the nuclear region , corresponding to uv color excesses of @xmath41 = 0.10.2 mag . therefore , the corrected @xmath42 color of the nucleus could be close to 1.0 mag , or the equivalent of a stellar population with a late a - type stellar cutoff ( fanelli et al . 1992 ) . the non - nuclear point sources in the hst / foc near - uv image of the nuclear region provide helpful checks on the distance to m94 . for the brightest point source , maoz et al . ( 1996 ) lists a monochromatic flux of @xmath43(2300 ) = 2.6 @xmath17 10@xmath44 erg s@xmath13 @xmath14 @xmath13 ( m@xmath45(2300 ) = 17.9 mag ) . this source is unresolved at a resolution of 0.1@xmath1 and corresponding linear scale of @xmath462.2 pc at the assumed distance . although an extremely compact star cluster can not be ruled out , this exceptional uv source is most likely dominated by a single hot supergiant star . in the absence of extinction , a b0 - 2 ia - o supergiant star has @xmath47 @xmath5 @xmath4810.7 mag ( fanelli et al . 1992 ) , which would imply a distance modulus of ( @xmath49 ) = 28.6 mag and corresponding distance of 5.2 mpc . assuming nuclear color excesses of @xmath50 = 0.150.27 mag ( taniguchi et al . 1996 ) and galactic - type extinction law , the corresponding near - uv extinctions would be @xmath51 @xmath5 12 mag , and the revised distances would be 3.32.1 mpc , respectively . the adopted distance of 4.6 mpc can thus be regarded as a reasonable estimate , given the uncertainties in stellar spectral type and extinction . the second source , 1-arcsec to the e , is 6 times ( 1.9 mags ) fainter consistent with it being an a0 - 2 ia - o supergiant at this distance , if subject to the same amount of extinction . were the galaxy significantly closer , more abundant oba - type giant and main - sequence stars would be resolved which is not the case . * figure 7 * shows radial profiles of the fuv surface brightness and cumulative flux . these profiles were derived from annular - averaged photometry using elliptical annuli consistent with the adopted position angle ( 120@xmath8 ) and inclination ( 40@xmath8 ) . since galactic extinction is negligible towards m94 ( de vaucouleurs et al . 1991 ) , no correction was made . the resulting surface brightness profile ( * fig . 7a * ) is dominated by light from the nucleus , inner disk , starburst ring , and an exponentially - declining disk that is punctuated by an enhancement from the bi - symmetric knots at @xmath52 . the exponentially declining component between 50 and 80-arcsec radii has an e - folding scalelength of only @xmath53 ( 223 pc ) , similar to that found at r and i - bands ( cf . munoz - tunon et al . the fuv emission beyond the bi - symmetric knots also shows a steep decline with an estimated scalelength of @xmath54 ( 446 pc ) , significantly shorter than the 70@xmath1 scalelength measured at r and i - bands in the outer disk ( munoz - tunon et al . 1989 ) . the cumulative flux profile ( * fig . 7b * ) indicates that the half - light radius matches that of the starburst ring , and that more than 80% of the total fuv emission is contained within a radius of @xmath55 ( 1.3 kpc ) . the total fuv flux from m94 is f@xmath45(1520 ) = 6.02 @xmath17 10@xmath56 erg @xmath14 s@xmath13 @xmath13 , corresponding to m(fuv ) = 9.45 mag , or m(fuv ) = @xmath57 mag at the adopted distance of 4.6 mpc . the corresponding fuv luminosity of 1.5 @xmath17 10@xmath58 erg s@xmath13 @xmath13 is the photometric equivalent of 2 @xmath17 10@xmath59 orion nebulae ( bohlin et al . 1982 ) or about 90 30-doradus regions ( as measured on an uit / fuv image of 30 dor out to a radius of 5 arcmin [ 67 pc ] ) , before correcting for the extinction in these sources . adopting a salpeter imf with lower and upper mass limits of 0.1 and 100 m@xmath35 , respectively , yields an uncorrected star formation rate of 0.15 m@xmath35/yr . this sfr estimate assumes continuous star formation and includes the strong contribution of b supergiant stars to the overall luminosity . an overall visual extinction of 1 mag would increase the global luminosity and inferred sfr by about a factor of 11 ( hill et al . 1997 ) . at this rate , it would have taken @xmath210 gyrs to transform the dynamical mass of 1.6 @xmath17 10@xmath60 m@xmath35 that is present within @xmath55 of the nucleus(garman & young 1986 ) into the dominant stellar component that we see today . as a check on these starbirth estimates , we note that the measured h@xmath0 flux of 9.6 @xmath17 10@xmath22 erg @xmath14 s@xmath13 converts to a luminosity of l(h@xmath0 ) = 2.45 @xmath17 10@xmath61 erg / s , or only 5600 equivalent orion nebulae . the origin of this discrepancy is deferred to the next subsection . assuming case b recombination and multiplying the h@xmath0 luminosity by @xmath62 yields a photoionization rate of n@xmath63 = 1.8 @xmath17 10@xmath64 photons / sec , and corresponding star formation rate of 0.22 m@xmath35/yr . the similarity of the fuv and h@xmath0-based sfrs , _ before correcting for extinction _ , suggests either that insignificant obscuration is present in the photometrically dominant starburst ring , or that significant obscuration exists with other mechanisms making up for the greater attenuation of the fuv emission relative to the longer - wavelength h@xmath0 emission . such mechanisms include ( 1 ) an excess contribution of non - ionizing b - type stars to the total fuv emission , ( 2 ) a reduction of the total h@xmath0 emission due to h@xmath0 absorption by the atmospheres of b and a - type stars , and ( 3 ) the absorption of euv photons before they ionize the gas and induce h@xmath0 emission . likely euv absorbers include the nebular dust associated with the h ii regions as well as the various metal species in the o - type stellar atmospheres themselves ( cf . hill et al . 1997 ; waller , parker , & malumuth 1996 ) . we conclude in the next subsection that modest obscuration plus non - ionizing contributions to the total fuv emission best explain the observed levels of h@xmath0 and fuv emission . stellar absorption at h@xmath0 is probably less than a 10% effect overall , given h@xmath0 emission equivalent widths of order 100 in the ring , and stellar absorption equivalent widths peaking at less than 10 . the global ( total ) @xmath65 color of m94 is 1.21 mag , which is characteristic of early - type disk galaxies with circumnuclear starburst activity ( e.g. ngc 1068 , ngc 3351 [ waller et al . the @xmath25 colors shown in * figure 4b * do not vary as much as those found by beckman et al . ( 1991 ) in their photometric study . bootstrapping our measured intensity ratio in the nuclear region to a color of @xmath25 = 0.45 mag as reported by beckman et al . ( 1991 ) , we obtain colors that range from @xmath25 = 0.3 mag in the rings starburst knots to @xmath25 = 0.5 mag in the arcs interior to the ring . beckman et al . ( 1991 ) obtain much bluer colors of order 0.1 mag near the ring . we attribute this discrepancy to our having used the h@xmath0-band image to remove contaminating line emission from the r - band image ( waller 1990b ) , thereby reducing the red flux from the knots in the ring by 20% . comparison of the contaminated and uncontaminated r - band images confirm our attribution . if the reddened arcs are caused by dust , the corresponding color excess relative to that of the starburst knots would amount to @xmath66 mag , or the equivalent visual extinction of roughly 0.6 mag in excess of the knots . allowing for bluer stellar populations in the knots could reduce the estimate of excess extinction in the arcs to negligible levels , while inclusion of some internal reddening in the knots would increase the total estimated extinction by @xmath281 mag . such low estimates for the extinction are similar to the spectroscopic results obtained by pritchett ( 1977 ) and by taniguchi et al . ( 1996 ) in their studies of the nuclear region , where color excesses of 0.27 mag and 0.15 mag , corresponding to @xmath67 mag and 0.48 mag , were respectively obtained . photometry of the fuv and h@xmath0 emission from the starburst ring ( between projected radii of 30@xmath1 and 60@xmath1 ) yields a mean h@xmath0/fuv flux ratio ( in equivalent width units ) of 22 . by way of comparison , we note that this flux ratio is somewhat lower than that found in the orion nebula ( 38 [bohlin et al . 1982 ] ) , within the range of flux ratios evident in m33 s giant h ii regions ( 19 to 69 [ parker , waller , & malumuth 1996 ] ) , and close to the mean of m51 s wide - ranging ghrs ( @xmath68 [ hill et al . 1997 ] ) . a model of steady - state star formation with a salpeter - type imf yields a somewhat lower ratio of 16 , while a modeled 2 myr - old starburst would match our result ( hill et al . 1995 [ see their fig . such a short burst timescale is probably untenable , thus indicating some need to correct our fuv flux for extinction . for 1 mag of visual extinction , the resulting `` corrected '' h@xmath0/fuv flux ratio would be lower by a factor of @xmath25 , leading to a modeled burst age of 6 myr close to the age limit of the ionizing o stars . based on these considerations alone , we surmise that the fuv and h@xmath0 emission from the starburst ring are subject to no more than 1 mag of visual extinction ( on average ) . we also can conclude that the similarity in uncorrected sfrs based on the global fuv and h@xmath0 fluxes is probably coincidental . after correction for 0.51.0 mag of visual extinction , the h@xmath0 flux would be increased by a factor of 1.42.1 , while the fuv emission would increase by a factor of 3.311 . in other words , _ the fuv emission is tracing an additional non - ionizing population i component in the inner disk that is missed at h@xmath0 and , after correction for extinction , is contributing to the overall calculated rate of star formation_. the bi - symmetric fuv knots to the west and east of the ring have fluxes of @xmath69 and @xmath70 erg @xmath14 s@xmath13 @xmath13 , respectively . this translates to the equivalent of 100 and 127 orion nebulae , respectively . only the eastern knot shows significant h@xmath0 emission , with a flux of 2.57 @xmath17 10@xmath56 erg @xmath14 s@xmath13 and hence an h@xmath0/fuv flux ratio of 67 about twice that of the orion nebula thus indicating greater fuv obscuration than is present in orion or a hotter ionizing cluster . examination of the spatially - filtered b - band image ( * figure 6b * ) reveals dark spiral - arm features at the position of the eastern knot which may correspond to obscuring dust clouds . the western knot shows similar structures to one side . the lack of significant h@xmath0 emission from this knot is difficult to explain with dust , and is most likely due to the presence of an older ( @xmath71 myr ) , non - ionizing stellar population . * figure 8a * shows an average of 5 iue / swp spectra of the extranuclear emission from the inner disk . based on the aperture mapping shown in * figure 3a * some emission from the nucleus may be present , but at low levels . the wavelengths of commonly observed fuv absorption and emission lines are indicated on the spectrum for comparison . although the averaged spectrum is of generally low s / n , several features can be identified . of these , the most prominent are the absorption blends of s ii ( 1250 - 1259 ) and si ii ( 1260,1265 ) , the p - cygni profile of s iv ( 1394,1443 ) , the absorption blends of fe iii ( 1601 - 1611 ) , al iii ( 1600 - 1612 ) , c ii ( 1720 - 1722 ) and al ii ( 1719 - 1725 ) , and part of the broad absorption complex of fe iii ( 1891 - 1988 ) . the low - ionization lines , in particular , are most characteristic of b and a - type stars . the strength of the blueshifted s iv absorption feature relative to that of c iv ( 1550 ) along with the strong si ii absorptions at 1260,1265 and 1527,1533 indicate a composite spectral type later than b3 but earlier than b8 ( fanelli et al . 1992 ; kinney et al . 1993 ; walborn , parker , & nichols 1995 ) . the absorptions at 1470 , 1780 , and 1790 remain unidentified . in emission , there is some evidence for c iv ( 1548 - 1551 ) , he ii ( 1640 ) , n iii ] ( 1730,1750 ) , and c iii ] ( 1909 ) much of which may be the result of liner activity being scattered by the ism in the inner disk . the c iii ] ( 1909 ) line emission may be responsible for filling in part of the broad stellar absorption complex of fe iii ( 1891 - 1988 ) . * figure 8b * shows an average of 2 iue / lwp spectra of the disk . the most prominent spectral features are the absorption lines of fe ii ( 2609 , 2750 ) , mg ii ( 2800 ) , and mg i ( 2852 ) . comparison with stellar spectra from the iue spectral atlas ( wu et al . 1991 ) shows that mg ii ( 2800 ) is unusually weak relative to mg i ( 2852 ) ; the line ratios in this spectral range being consistent with light dominated by late g - type stars , based on this cursory comparison . however , the spectrum is too blue and the amplitude of the 2800 break too small for this to be the whole story . comparison with the nuclear spectra ( iue lwr 12221,12238 as listed in table 2 ) shows evidence for a spectrally smooth blue component off the nucleus proper . also , the mg ii 2800 break is roughly 50% smaller than that evident in the nuclear spectrum ( see below ) . all the absorption features are broader in the off - nuclear spectrum , because the light there almost uniformly fills the aperture ( so centering shifts are not an issue ) . the off - nuclear spectrum shown here is not only bluer but shows a diminished 2800 break amplitude . measured as a ratio of flux f@xmath45 above and below 2800 , the ratio is 2.5 at nucleus and 1.8 away from it . using the spectral - break indices developed by fanelli et al . ( 1992 ) , we obtain for the off - nucleus averaged spectrum ( mg ii 2800 mg i 2852 ) = 0.31 mag , ( 2609/2660 ) = 0.45 mag , and ( 2828/2921 ) = 0.31 mag all of which are consistent with a @xmath72 color of 0.20.3 mag , or the equivalent of a late a - type main sequence star or late a to early f - type giant / supergiant star . the off - nuclear continuum is likewise flatter in the swp range , with little trace of the emission - like features around 1900 seen in the nuclear spectrum ( see kinney et al . 1993 ) . these properties can be accounted for , if recent star formation ( few times @xmath73 years ago ) contributes relatively more light off the nucleus than on it , consistent with an aging burst that is spatially more extended than the centrally - concentrated stars of the older disk and bulge . * figure 8c * shows the spectrum of optical emission from the central d = 8.1 arcsec , as obtained with the lick 1-m telescope and ids . absorption lines include the ca ii k&h lines at 3934 and 3968 , h@xmath74 ( 4101 ) , ch g - band ( 4300 ) , h@xmath75 ( 4340 ) , he ii ( 4686 ) , h@xmath76 ( 4860 ) , mg i ( 5170 ) , fe i ( 5270 ) , na d ( 5890,5896 ) , h@xmath0 ( 6563 ) , and atmospheric absorption bands of o@xmath77 ( 6867 ) and h@xmath77o ( 7186 ) . emission lines are restricted to [ n ii ] ( 6584 ) and [ s ii ] ( 6731 ) . the h@xmath76 equivalent width of about 5.4 is huge for an old population typical of spiral bulges , even before any correction for line emission . there has been star formation in this region not long ago . the 4000- break is also suppressed , indicating the effective age is much younger than for typical early - spiral bulges . following the dressler and shectman ( 1987 ) definition ( ratio of f@xmath78 between 3950 - 4050 and 3750 - 3850 ) , the observed break has f@xmath78 = 1.6 , whereas values of 1.92.0 are more usual . from the ids / ipcs on the 2.5-m isaac newton telescope , we see that the nucleus ( inner 1 " or so ) is bluer at optical wavelengths than its immediate surroundings . misalignment in the ipcs detector is less than 0.1 pixel , so this is not an obvious instrumental effect . this nuclear blueing effect is substantial , amounting to about 30% over the 44005200 range ( more or less equivalent to @xmath79 mags , where the mean ( @xmath80 ) color of the bulge is 0.91 mags , which is already bluer than ordinary sa / sb bulges [ keel & weedman 1978 ] ) . outside this area , the color along the slit is quite constant out to 15@xmath1 from the nucleus , where the signal begins to die out . this effect has no observed counterpart in the iue spectra , because of their courser spatial resolution . the prominent h@xmath76 absorption line in the nuclear int / ipcs spectrum has fwhm 1820 , consistent with values seen in mid - to - late a stars . thus , the data are consistent with seeing the main - sequence turnoff near this spectral type , rather than the supergiant dominance of a much younger population giving superficially similar spectral features ( as seen in the starbursting nucleus of ngc 4569 [ keel 1996 ] ) . the optical spectra confirm our imaging result that h@xmath0 is in absorption with respect to the continuum from the underlying population of a and b - type stars . by contrast , the neighboring line of [ n ii ] ( 6584 ) is seen in emission in the lick / ids spectrum . as previously noted , h@xmath76 is also dominated by photospheric absorption . the ids / ipcs spectrum shows some h@xmath75 emission at least at the nucleus , that is not swallowed by the stellar absorption line . * figure 9a * shows the rotation curve based on the h i observations of mulder and van driel ( 1993 ) , where key morphological features are noted . this curve is qualitatively similar to that obtained from a recent interferometric mapping of co ( wong & blitz 2000 ) but is on - average 12% lower , due to the adoption of a 40@xmath8 inclination compared to their 35@xmath8 . * figure 9b * shows the corresponding radial profiles of h i and h@xmath77 gas , the latter being derived from the co observations of gerin et al . ( 1991 ) . the molecular gas component clearly dominates the inner disk s ism and may continue to prevail at higher radii , where the co emission has yet to be measured . the plotted extrapolation beyond the last reliable measurement of co emission is intended to provide an _ upper limit _ on the total gas and thus a lower limit on the q stability index ( see below ) . the interferometric measurements by wong & blitz ( 2000 ) yield even lower extrapolated surface densities of gas consistent with our extrapolated upper limit . moreover , fir measurements with iras and the kao ( smith et al . 1991 ) reveal negligible dust emission beyond 60 arcsec radius further corroborating the gas upper limit used here . the h i profile , though of lower amplitude , shows enhancements at the radii of the starburst ring and the bi - symmetric knots . * figure 9c * shows the corresponding radial profile of the gravitational stability index ( q ) . here , we have considered the simplest case of the gaseous stability , without any coupling with the stellar component , such that @xmath81 where the epicyclic frequency @xmath82 is closely linked to the rotation curve v(r ) through @xmath83 and where @xmath84 is the gas velocity dispersion a quantity which is not well constrained but is probably of order 510 km / s . here , we leave @xmath84 as an unknown . the resulting radial profile of q/@xmath84 out to the limits of the co observations shows a shallow minimum at the radius of the starburst ring . although this minimum is too broad and shallow to explain the more discrete and prominent starburst ring , it may help to explain the relatively young ( @xmath85 yr ) stellar population pervading the inner disk . * figure 10 * shows the orbital resonance diagram that results from the rotation curve along with the radii of key morphological features . a single wave pattern speed of 35 km s@xmath13 kpc@xmath13 ( where d = 4.6 mpc ) would place the nuclear bar inside the inner - inner lindblad resonance ( iilr ) , the starburst ring between the iilr & outer inner lindblad resonance ( oilr ) , the bisymmetric knots at the 4:1 `` ultra - harmonic '' resonance , the oval disk terminating at co - rotation ( cr ) , and the outer pseudo - ring at the outer lindblad resonance ( olr ) . alternatively , a higher pattern speed of 56 km s@xmath13 kpc@xmath13 ( as modeled by mulder & combes ( 1996 ) ) would move the starburst ring just outside the oilr , the bisymmetric knots to the cr radius , and the edge of the oval disk to the olr . this latter model , however , fails to account for the outer ring without invoking additional patterns . further support for the proposed sequence of resonances comes from specific ratios of the corresponding radii ( cf . buta 1986 ) . given a flat rotation curve , the modeled ratio of olr and uhr radii is r(olr)/r(uhr ) = 2.6 . if these olr and uhr resonances are respectively traced by the outer pseudoring and bi - symmetric knots , then we obtain a ratio of radii equaling 2.54 closely matching the modeled ratio . the theoretical ratio of olr and cr radii , being r(olr)/r(cr ) = 1.7 , is also well matched by the observed relative dimensions of the outer ring and oval disk , where a ratio of 1.5 is obtained . admittedly , morphological tracers and rotation curves are insufficient to discriminate between these and other possible kinematic patterns ( cf . buta & combes 1996 ) . further progress on constraining the resonant dynamics in m94 will require analysis of the complete velocity field in the disk ( cf . westpfahl & adler 1996 ; canzian & allen 1997 ) . a complete h i mapping with the vla has been made recently , whose spectral and spatial resolution would be sufficient to derive a complete velocity field and its associated resonant states ( d. westpfahl , private communication ) . until such an analysis is carried out , we think that the resonant state diagrammed in * figure 10 * best explains the most features with the fewest conditions . through uv - optical imaging and spectroscopy , we have found new evidence for bar - mediated resonances as the primary drivers of evolution in m94 at the present epoch . our observational results include evidence for ( 2 . ) an inner disk with diffuse fuv emission in concentric arcs that do not match the fine - scale structures or reddened structures at visible wavelengths . since @xmath86 is observed in absorption here , the uit / fuv image represents the first view of this non - ionizing but relatively young disk component . uv - optical colors and spectral indices in the nucleus and inner disk that indicate b and a - type stars in the presence of modest extinction ( a@xmath87 @xmath281 mag ) along with some liner activity from the nucleus itself . ( 4 . ) a 2.2 kpc diameter starbursting ring at the perimeter of the inner disk that is bright at fuv , h@xmath0 , and radio - continuum wavelengths . the level of starbirth activity in this inner ring rivals the levels observed in starbursting irregular galaxies such as ngc 1569 and ngc 4449 . the inferred star formation rate within the ring and inner disk amounts to 1.5 m@xmath35 yr@xmath13 sufficient to build up the stellar mass of the inner disk and bulge in @xmath210@xmath60 yr . two 500-pc size fuv - emitting knots exterior to the ring on diametrically opposite sides of the nucleus . the bisymmetric knots and starburst ring appear to be especially prominent parts of a complex spiral arm structure , as revealed in a spatially - filtered b - band image . the starburst ring , bi - symmetric knots , oval disk , and outer pseudo - ring as signposts of resonant dynamics in the disk of m94 . more specifically , the radii of these features match those of various orbital resonances , given a pattern speed of 35 km s@xmath13 kpc@xmath13 at our adopted distance and inclination . these orbital resonances are most likely driven by some combination of the nuclear mini - bar and oval distortion in the disk . ( 8 . ) a shallow minimum of gravitational stability at the radius of the starburst ring that extends inward into the inner disk . although too broad to explain the discrete starburst ring , the shallow minimum may help to explain the 10@xmath88-yr old stellar population interior to the ring . although we can set a limit on the speed of outward or inward propagating star formation in the ring , we can not preclude the existence of such a mode . at a propagation speed of 35 km / s , a wave initiated in the nucleus could traverse the inner disk to the radius of the current starburst ring in only 31 myr . therefore , it is possible that the @xmath89-yr old stellar population detected in the inner disk is the result of such an outward propagating wave . the striking difference in emission morphologies at fuv and red wavelengths provides further support for the starburst ring being a transient phenomenon which does not persist at any one radius for very long . either these resonant phenomena come and go , as the oval distortions undergo secular evolution , or their operating radii migrate in response to other dynamical influences on the stars and gas ( cf . combes 1994 ; combes et al . 1995 ; friedli & benz 1995 ) . otherwise , one must invoke strong radial inflows of stars from the starburst ring to populate the inner disk and bulge , a feat requiring unusual circumstances e.g. mergers . the results reported herein may have important implications with regard to observations of the most distant observable galaxies . at redshifts of 15 , the 2-kpc diameter starburst ring in m94 would subtend angles of only ( 0.7@xmath1 1.0@xmath1)h@xmath7/75 in an einstein - de sitter universe ( q@xmath7 = 1/2 ) and ( 0.3@xmath1 0.2@xmath1)h@xmath7/75 in an open ( milne ) universe ( q@xmath7 = 0 ) ( cf . narlikar 1983 ) . the uv - bright nuclear rings evident in ngc 1097 , ngc 1317 , ngc 1433 , ngc 1512 , ngc 2997 , ngc 4321 , and ngc 5248 ( maoz et al . 1995 ; maoz et al . 1996 ; kuchinsky et al . 2000 ; marcum et al . 2000 ) would subtend even smaller angles at the same redshifts . moreover , nuclear rings tend to have higher fuv surface brightnesses than their larger counterparts the inner ring in m94 being a remarkable exception . therefore , some of the `` core - halo '' morphologies that are evident at high - redshift in the restframe fuv ( cf . giavalisco et al . 1997 ) may , in fact be marginally - resolved representations of galaxies with starburst rings in their centers . gravitationally - lensed galaxies are fortuitously magnified , enabling resolutions of their structure at high s / n . an important precedent in this regard is the gravitationally - lensed `` pretzel galaxy '' which lies behind the galaxy cluster 0024 + 1654 at an estimated redshift of 1.2 1.8 ( colley et al . 1996 ; tyson et al . 1997 ) . detailed reconstructions of the multiply - lensed galaxy show a clear annular morphology on a scale of several kpc . if m94 and other nearby ringed galaxies can be used as current - epoch analogues , the `` pretzel galaxy '' and perhaps other marginally - resolved `` core - halo '' galaxies at high redshift may represent youthful inner disks and bulges growing under the organizing influence of oval or bar asymmetries(friedli & benz 1995 ; waller et al . conversely , if evidence for starburst rings at high redshift proves to be sparse , then massive inner disks featuring ring - bar dynamics have yet to form in most systems , or starbursting bulges are masking their presence . we thank david adler , gene byrd , francoise combes , daniel friedli , and david westpfahl for generously providing consultation on the dynamics of ringed - barred spiral galaxies . whw is grateful to john huchra and the oir division of the harvard - smithsonian center for astrophysics for their kind hospitality during his visiting appointment at the cfa . whw also thanks eric murphy and christine winslow , tufts undergraduates who helped craft some of the graphics . uit research is funded through the spacelab office at nasa headquarters under project number 440 - 51 . we are deeply grateful to the crew of sts-67 and the many people who helped make the _ astro-2 _ mission a success . whw acknowledges partial support from nasa s astrophysics data program ( 071 - 96adp ) . giavalisco , m. , et al . 1997 , in the ultraviolet universe at low and high redshift : probing the progress of galaxy evolution , eds . w. h. waller , m. n. fanelli , j. e. hollis , & a. c. danks ( new york : aip press ) , p. 221 marcum , p. m , et al . 1997 , in the ultraviolet universe at low and high redshift : probing the progress of galaxy evolution , eds . w. h. waller , m. n. fanelli , j. e. hollis , & a. c. danks ( new york : aip press ) , p. 88 oconnell , r. w. 1997 , in the ultraviolet universe at low and high redshift : probing the progress of galaxy evolution , eds . w. h. waller , m. n. fanelli , j. e. hollis , & a. c. danks ( new york : aip press ) , p. 11 tyson , j. a. , kochanski , g. , & dellantonio , i. 1997 , in the ultraviolet universe at low and high redshift : probing the progress of galaxy evolution , eds . w. h. waller , m. n. fanelli , j. e. hollis , & a. c. danks ( new york : aip press ) , p. 204 waller , w. h. , et al . 1997 , in the ultraviolet universe at low and high redshift : probing the progress of galaxy evolution , eds . w. h. waller , m. n. fanelli , j. e. hollis , & a. c. danks ( new york : aip press ) , p. 39 waller , w. h. , varosi , f. , boulanger , f. , & digel , s. w. 1998 , in new horizons from multi - wavelength sky surveys , iau symp . 179 , eds . b. j. mclean , d. a. golombek , j. e. hayes , & h. e. payne b. lasker & m. postman ( dordrecht : kluwer ) , p. 194 , ( astro - ph/9612233 )

in some estimation problems , especially in applications dealing with information theory , signal processing and biology , theory provides us with additional information allowing us to restrict the parameter space to a finite number of points . in this case , we speak of discrete parameter models . even though the problem is quite old and has interesting connections with testing and model selection , asymptotic theory for these models has hardly ever been studied . therefore , we discuss consistency , asymptotic distribution theory , information inequalities and their relations with efficiency and superefficiency for a general class of-estimators . .

1207.5653

sometimes , especially in applications dealing with signal processing and biology , theory provides us with some additional information allowing us to restrict the parameter space to a finite number of points ; in these cases , we speak of _ discrete parameter models_. statistical inference when the parameter space is reduced to a lattice was first considered by hammersley @xcite in a seminal paper . however , since the author was motivated by the measurement of the mean weight of insulin , he focused mainly on the case of a gaussian distribution with known variance and unknown integer mean ( see @xcite , page 192 ) ; this case was further developed by khan @xcite . the poisson case also met some attention in the literature and was dealt with by hammersley ( @xcite , page 199 ) and others @xcite . previous works have shown that the rate of convergence of @xmath0-estimators is often exponential [ @xcite , @xcite ] . general treatments of admissibility and related topics are in @xcite ( see also the book @xcite ) ; special cases have been dealt with in @xcite ( page 424 , for the case of a translation integral parameter and of integral data under the quadratic loss ) , @xcite ( for the case of the gaussian distribution ) and @xcite ( for the case of the discrete uniform distribution ) . other papers dealing with optimality in discrete parameter spaces are @xcite . optimality of estimation under a discrete parameter space was also considered by vajda @xcite in a nonorthodox setting inspired by rnyi s theory of random search . other aspects that have been studied are bayesian encompassing @xcite , construction of confidence intervals ( @xcite , pages 224225 ) , comparison of statistical experiments ( @xcite , @xcite , section 2.2 ) , sufficiency and minimal sufficiency @xcite and best prediction @xcite . moreover , in the estimation of complex statistical models ( see @xcite , @xcite , chapter 4 ) and in the calculation of efficiency rates ( see @xcite ) , approximating a general parameter space by a sequence of finite sets has proved to be a valuable tool . a few papers showed the practical importance of discrete parameter models in signal processing , automatic control and information theory and derived some bounds on the performance of the estimators ( see @xcite ) . more recently , the topic has received new interest in the information theory literature ( see @xcite , and the review paper @xcite ) , in stochastic integer programming ( see @xcite ) , and in geodesy ( see , e.g. , @xcite , section 5 ) . however , no general formula for the convergence rate has ever been obtained , no optimality proof under generic conditions has been provided and no general discussion of efficiency and superefficiency in discrete parameter models has appeared in the literature . in the present paper , we provide a full answer to these problems in the case of discrete parameter models for samples of i.i.d . ( independent and identically distributed ) random variables . therefore , after introducing some examples of discrete parameter models in section [ sect - examples ] , in section [ sect - m - estimators ] we investigate the properties of a class of @xmath0-estimators . in particular , in section [ sect - consistencyofmleandbe ] , we derive some conditions for strong consistency ; then , in section [ sect - asymptoticdistributionofthemle ] , we calculate an asymptotic approximation of the distribution of the estimator and we establish its convergence rate . these results are specialized to the case of the maximum likelihood estimator ( ) and extended to bayes estimators in section [ sect - mleandbayesestimators ] . in section [ sect - optimalityandefficiencyofthemle ] , we derive upper bounds for the convergence rate in the standard and in the minimax contexts , and we discuss the relations between information inequalities , efficiency and superefficiency . in particular , we prove that estimators of discrete parameters have uncommon efficiency properties . indeed , under the zero one loss function , no estimator is efficient in the class of consistent estimators for any value of @xmath1 ( @xmath2 being here the true value of the parameter ) and no estimator attains the information inequality we derive . but the still has some appealing properties since it is minimax efficient and attains the minimax information inequality bound . the following examples are intended to show the relevance of discrete parameter spaces in applied and theoretical statistics . in particular , they show that the results in the following sections solve some long - standing problems in statistics , optimization , information theory and signal processing . we recall that a _ statistical model _ is a collection of probability measures @xmath3 where @xmath4 is the _ parameter space_. @xmath4 is a subset of a euclidean or of a more abstract space . we consider tumor transplantability in mice . for a certain type of mating , the probability of a tumor `` taking '' when transplanted from the grandparents to the offspring is equal to @xmath5 where @xmath6 is an integer equal to the number of genes determining transplantability . for another type of mating , the probability is @xmath7 . we aim at estimating @xmath6 knowing that @xmath8 transplants take out of @xmath9 the likelihood is given by @xmath10 in this case the parameter space is discrete and the maximum likelihood estimator can be shown to be @xmath11 $ ] where @xmath12 $ ] is the integer nearest to @xmath13 ( see @xcite , page 236 ) . consider a random variable @xmath14 distributed according to an exponential family where the natural parameter @xmath6 is restricted to a lattice @xmath15 , for fixed @xmath2 and @xmath16 ( see @xcite , page 759 ) . the case of a gaussian distribution has been considered in @xcite ( page 192 ) and @xcite , the poisson case in @xcite ( page 199 ) , @xcite . in particular , @xcite uses the gaussian model to estimate the molecular weight of insulin , assumed to be an integer ( however , see the remarks of tweedie in the discussion of the same paper ) . we consider the optimization problem of the form @xmath17 , where @xmath18 is an integral functional , @xmath19 is the mean under probability @xmath20,@xmath21 is a real - valued function of two variables @xmath13 and @xmath22 , @xmath23 is a random variable having probability distribution @xmath20 and @xmath24 is a finite set . we approximate this problem through the sample average function @xmath25 and the associated problem @xmath26 . see @xcite for some theoretical results and a discussion of the stochastic knapsack problem and @xcite for an up - to - date bibliography . in many applied cases , the requirement that the true model generating the data corresponds to a point belonging to the parameter space appears to be too strong and unlikely . moreover , the objective is often to recover a model reproducing some stylized facts from the original data . in these cases , approximation of a continuous parameter space with a finite number of points allows for obtaining such a model under weaker assumptions . this situation arises , for example , in signal processing and automatic control applications @xcite and is reminiscent of some related statistical techniques , such as the _ discretization _ device of le cam ( @xcite , section 6.3 ) , or the _ sieve estimation _ of grenander ( @xcite ; see also @xcite , remark 5 ) . in information theory , discrete parameter models are quite common , and their estimation is a generalization of binary hypothesis testing that goes under the names of @xmath27-_ary _ _ hypotheses _ ( or _ multihypothesis _ ) _ testing _ , _ classification _ or _ detection _ ( see the examples in @xcite ) . consider a received waveform @xmath28 described by the equation @xmath29 for @xmath30 , where @xmath31 is a deterministic signal , @xmath32 is an additive gaussian white noise and @xmath33 is the noise intensity . the set of possible signals is restricted to a finite number of alternatives , say @xmath34 : the chosen signal is usually the one that maximizes the log - likelihood of the sample , or an alternative criterion function . for example , if the log - likelihood of the process based on the observation window @xmath35 $ ] is used , we have @xmath36.\end{aligned}\ ] ] much more complex cases can be dealt with ; see @xcite for an introduction . in this section , we consider an estimator obtained by maximizing an objective function of the form@xmath37 in what follows , we allow for misspecification . note that the expression @xmath0-estimator stands for _ maximum likelihood type estimator _ , in the spirit of huber @xcite , and not for _ maximum _ ( or _ extremum _ ) _ estimator _ ( see , e.g. , @xcite , page 2114 ) . in the case of a discrete parameter space , uniform convergence reduces to pointwise convergence . therefore , @xmath0-estimators are strongly consistent under less stringent conditions than in the standard case ; in particular , no condition is needed on the continuity or differentiability of the objective function . the following assumption is used in order to prove consistency in the case of i.i.d . replications : 1 . the data @xmath38 are realizations of i.i.d . @xmath39-valued random variables having probability measure @xmath40 . + the estimator @xmath41 is obtained by maximizing over the set @xmath42 , of finite cardinality , the objective function @xmath43 the function @xmath44 is @xmath45-measurable for each @xmath46 and satisfies the @xmath47-domination condition@xmath48 for every @xmath46 , where @xmath49 denotes the expectation taken under the true probability measure @xmath40 . + moreover , @xmath2 is the point of @xmath4 maximizing @xmath50 and @xmath2 is globally identified ( see @xcite , section 2.2 ) . \(i ) the assumption of a finite parameter space seems restrictive with respect to the more general assumption of @xmath4 being countable ( see , e.g. , @xcite ) . however , a1 is compatible with the convex hull of @xmath4 being compact , as in standard asymptotic theory . indeed , the cases analyzed in @xcite have convex likelihood functions and this is a well - known substitute for compactness of @xmath4 ( see @xcite , page 2133 ; see @xcite , for consistency with neither convexity nor compactness ) . moreover , the restriction to finite parameter spaces seems to be necessary to derive the asymptotic approximation to the distribution of @xmath0-estimators . the relative position of the points of @xmath4 is unimportant and the choice of @xmath2 as the maximizer is arbitrary and is made only for practical purposes . note that @xmath2 has no link with @xmath40 apart from being the pseudo - true value of @xmath51 with respect to @xmath40 on the parameter space @xmath4 ( see , e.g. , @xcite , volume 1 , page 14 ) . [ pr - consistencyofthemle ] under assumption , the @xmath0-estimator @xmath41 is a @xmath40-strongly consistent estimator of @xmath2 and is @xmath52-measurable . a similar result of consistency for discrete parameter spaces has been provided by @xcite ( page 446 ) , by @xcite ( pages 325333 ) , by @xcite(pages 12931294 ) as an application of the shannon mcmillan breiman theorem of information theory , by @xcite ( section 2.1 ) as a preliminary result of his work on partial likelihood , and by @xcite ( page 96 , section 7.1.6 ) . for a discrete parameter space , the finite sample distribution of the @xmath0-estimator @xmath41 is a discrete distribution converging to a dirac mass concentrated at @xmath2 . since the determination of an asymptotic approximation to this distribution is an interesting and open problem , we derive in this section upper and lower bounds and asymptotic estimates for probabilities of the form @xmath53 . to simplify the following discussion , we introduce the processes : @xmath54_{j=0,\ldots , j , j\neq i},\vspace*{2pt}\cr \mathbf{x}_{k}\triangleq\mathbf{x}_{k}^{(0)}\vspace*{2pt}\cr \phantom{\mathbf{x}_{k}}=[\ln q(y_{k};\theta _ { 0})-\ln q(y_{k};\theta_{j})]_{j=1,\ldots , j } , } \\ \eqntext{i=1,\ldots , j,}\end{aligned}\ ] ] the probability of the estimator @xmath41 taking on the value @xmath55 can be written as @xmath56 \\[-8pt ] \nonumber & = & \mathbb{p}_{0}\biggl(\sum _ { k=1}^{n}\mathbf{x}_{k}^{(i)}\in\operatorname{int } \mathbb{r}_{+}^{j}\biggr).\end{aligned}\ ] ] the only approaches that have been successful in our experience are large deviations ( in logarithmic and exact form ) and saddlepoint approximations . note that we could have defined the probability in ( [ eq - probabilityforthetaiinldform ] ) as @xmath57 or through any other combination of equality and inequality signs ; this introduces some arbitrariness in the distribution of @xmath41 . however , we will give some conditions ( see proposition [ pr - convergencerateofmle2 ] ) under which this difference is asymptotically irrelevant . section [ sect - apreliminaryresult ] introduces definitions and assumptions and discusses a preliminary result . in sec - tion [ sect - largedeviationsasymptotics ] we derive some results on the asymptotic behavior of @xmath58 using large deviations principles ( ldp ) . then , we provide some refinements of the previous expressions using the theory of exact asymptotics for large deviations , with special reference to the case @xmath59 . at last , section [ sect - saddlepointapproximation ] derives saddlepoint approximations for probabilities of the form ( [ eq - probabilityforthetaiinldform ] ) . as concerns the distribution of the @xmath0-estimator @xmath41 , we shall need some concepts and functions derived from large deviations theory ( see @xcite ) ; we recall that the processes @xmath60 , @xmath61 and @xmath62 have been introduced in ( [ eqlikelihoodratioprocesses ] ) . then , for @xmath63 , we define the moment generating functions @xmath64}\bigr]\\ & = & \mathbb{e}_{0}\bigl[e^{\bolds{\lambda}^{\mathsf{t}}\mathbf{x}^{(i)}}\bigr],\end{aligned}\ ] ] the logarithmic moment generating functions @xmath65}\bigr]\\ & = & \ln\mathbb{e}_{0}\bigl[e^{\bolds{\lambda}^{\mathsf{t}}\mathbf{x}^{(i)}}\bigr],\end{aligned}\ ] ] and the cramr transforms @xmath66,\ ] ] where @xmath67 is the scalar product . note that , in what follows , @xmath68 , @xmath69 and @xmath70 are respectively shortcuts for @xmath71 , @xmath72 and @xmath73 . moreover , for a function @xmath74 , we will need the definition of the _ effective domain _ of @xmath75 , @xmath76 . the following assumptions will be used to approximate the distribution of @xmath41 . 1 . there exists a @xmath77 such that , for any @xmath78 , we have @xmath79^{\eta}<+\infty\quad \forall j , k=0,\ldots , j.\ ] ] @xmath80 in what follows , this assumption could be replaced by a condition as in @xcite ( assumptions h1 and h2 ) . 1 . @xmath81 is _ steep _ , that is , @xmath82 whenever @xmath83 is a sequence in @xmath84 converging to a boundary point of @xmath85 . under assumptions , and , @xmath86 is _ essentially smooth _ ( see , e.g. , @xcite , page 44 ) . a sufficient condition for and essential smoothness is openness of @xmath87 ( see @xcite , page 905 , and @xcite , pages 505506 ) . 1 . @xmath88 , where @xmath89 is the closure of the convex hull of the support of the law of @xmath90 . we will also need the following lemma showing the equivalence between assumption and the so - called _ cramr condition _ @xmath91 , for any @xmath92 . [ lm - cramerconditionandeta - int]under assumption , the following conditions are equivalent : assumption holds ; @xmath91 , for any @xmath92 . as concerns the saddlepoint approximation of section [ sect - saddlepointapproximation ] , we need the following assumption : 1 . the inequality @xmath93\biggr|\\ & & \quad<(1-\delta)\cdot \biggl|\mathbb{e}_{0}\biggl[\prod_{j=0,\ldots , j , j\neq i}\biggl(\frac{q(y;\theta _ { i})}{q(y;\theta_{j})}\biggr)^{u_{j}}\biggr]\biggr|\\ & & \quad<\infty\end{aligned}\ ] ] holds for @xmath94 , @xmath77 and @xmath95 ( @xmath96 denotes the imaginary unit ) . in this section we consider large deviations asymptotics . wenote that , in what follows , @xmath97 stands for@xmath98^{c}\ } $ ] . [ pr - convergencerateofmle2 ] for @xmath99 , under assumption , the following result holds : @xmath100 where @xmath101 is a function such that@xmath102 . under assumptions and : @xmath103 where @xmath104 is a function such that@xmath105 . under assumptions , , and : @xmath106 [ pr - convergencerateofmle]under assumption , the following inequality holds : @xmath107 where @xmath108 is the finite cardinality of the set@xmath109 and @xmath101 is a function such that @xmath102 . under assumptions and : @xmath110 where @xmath104 is a function such that@xmath105 . the proposition allows us to obtain an upper bound on the bias of the @xmath0-estimator , @xmath111 . a better description of the asymptotic behavior of the probability @xmath53 could be obtained , under some additional conditions , from the study of the neighborhood of the contact point between the set @xmath112 and the level sets of the cramr transform @xmath113 . we leave the topic for future work . here we just remark the following brackets on the convergence rate . [ pr - neyconvergencerateofmle ] under assumptions , , and , for sufficiently large @xmath114 , the following result holds : @xmath115 for @xmath99 and for some @xmath116 . when @xmath59 , a more precise convergence rate can be obtained under the following assumption : 1 . when @xmath117 , there is a positive value @xmath118 such that @xmath119 . moreover , the law of @xmath120 is nonlattice ( see @xcite , page 110 ) . [ pr - exactasymptoticsinr ] @xmath80under assumptions , , , and , with @xmath121 and @xmath59 , we have @xmath122 a refinement of the previous asymptotic rates can be obtained using results in @xcite . in this section we consider a different kind of approximation of the probabilities @xmath53 . [ th - saddlepointapproximation ] @xmath123under assumptions , and , for @xmath124 , it is possible to choose @xmath125 such that , for every @xmath126 $ ] , @xmath127 and @xmath128\biggr)\\ & & { } \cdot\bigl[e_{s-3}\bigl(\mathbf{u},\operatorname{int } \mathbb{r}_{+}^{j}\ominus \mathbb{e}_{0}\mathbf{x}^{(i)}\bigr)\\ & & \hspace*{9pt}\phantom{\bigl[}{}+\delta\bigl(\mathbf{u},\operatorname{int } \mathbb { r}_{+}^{j}\ominus\mathbb{e}_{0}\mathbf{x}^{(i)}\bigr)\bigr],\end{aligned}\ ] ] where @xmath129\,\mathrm{d}\mathbf{y},\\ & & q_{\ell\mathbf{u}}(\mathbf{x } ) \\ & & \quad = \sum_{m=1}^{\ell}\frac{1}{m!}{\sum}^{\ast}{\sum}^{\ast\ast}\biggl(\frac{\kappa_{\nu_{1}n}\cdots\kappa_{\nu _ { m}n}}{\nu_{1}!\cdots\nu_{m}!}\biggr)\\ & & \qquad{}\cdot h_{i_{1}}(x_{1})\cdots h_{i_{d}}(x_{d}),\\ & & \bigl|\delta\bigl(\mathbf{u},\operatorname{int } \mathbb{r}_{+}^{j}\ominus\mathbb { e}_{0}\mathbf{x}^{(i)}\bigr)\bigr| \\ & & \quad\leq c\cdot n^{-{(s-2)}/{2}}\end{aligned}\ ] ] and @xmath130 , @xmath131 , @xmath132 , @xmath133 , @xmath134 is the usual hermite chebyshev polynomial of degree @xmath0 , @xmath135 denotes the sum over all @xmath0-tuples of positive integers @xmath136 satisfying @xmath137 , @xmath138 denotes the sum over all @xmath0-tuples @xmath139 with @xmath140 , satisfying @xmath141 , and @xmath142 . note that @xmath143 depends on @xmath125 through the cumulants calculated at @xmath125 . the main question that this theorem leaves open is the choice of the point @xmath125 . usually this point is chosen as a solution @xmath144 of @xmath145 ; this corresponds to a saddlepoint in @xmath146 . @xcite ( section 6 ) and @xcite ( page 480 ) give some conditions for @xmath59 ; @xcite ( page 23 ) and @xcite ( page 153 ) give conditions for general @xmath147 . @xcite suggests that the most common solution is to choose @xmath148 and @xmath144 ( @xmath148 belonging to the boundary of @xmath149 $ ] and @xmath144 solving @xmath145 ) , such that for every @xmath150 $ ] , @xmath151 . this is the same as a dominating point in @xcite ; therefore , , and , for sufficiently large @xmath114 , imply the existence of this point for any @xmath152.=1 in this section , we show how the previous results can be applied to the and bayes estimators under the zero one loss function . the is defined by @xmath153.\end{aligned}\ ] ] this corresponds to the _ minimum - error - probability estimate _ of @xcite and to the _ bayesian estimator _ of @xcite . on the other hand , using the prior densities given by @xmath154 for @xmath46 , the posterior densities of the bayesian estimator are given by @xmath155 the bayes estimator relative to zero one loss @xmath156 ( see section [ sect - riskfunctions ] for a definition ) is the mode of the posterior distribution and is given by @xmath157 \\[-8pt ] \nonumber & = & \arg\max_{\theta\in\theta}\biggl[\frac{1}{n}\sum_{i=1}^{n}\ln f_{y_{i}}(y_{i};\theta)+\frac{\ln\pi(\theta)}{n}\biggr].\\end{aligned}\ ] ] note that the coincides with the bayes estimator corresponding to the uniform distribution @xmath158 for any @xmath46 . assumption can be replaced by the ones ( where assumptions and entail that the likelihood function is asymptotically maximized at @xmath2 only):=-1 1 . the parametric statistical model @xmath159 is formed by a set of probability measures on a measurable space @xmath160 indexed by a parameter @xmath6 ranging over a parameter space @xmath161 , of finite cardinality . let @xmath39 be a measurable space and @xmath162 a positive @xmath33-finite measure defined on @xmath39 such that , for every @xmath46 , @xmath163 is equivalent to @xmath162 ; the densities @xmath164 are @xmath45-measurable for each @xmath46 . + the data @xmath38 are i.i.d . realizations from the probability measure @xmath40 . 2 . the log density satisfies the @xmath47-dominationcondition @xmath165 , for @xmath166,where @xmath49 denotes the expectation taken under the true probability measure @xmath40 . @xmath2 is the point of @xmath4 maximizing @xmath167 and is globally identified . in order to obtain the consistency of bayes estimators , we need the following assumption on the behavior of the prior distribution : 1 . the prior distribution verifies @xmath168 for any @xmath46 . proposition [ pr - consistencyofthemle ] holds for the under assumptions , and , while for bayes estimators is required , too . note that , under correct specification ( i.e. , when the true parameter value belongs to @xmath4 ) , a standard wald s argument ( see , e.g. , lemma 2.2 in @xcite , page 2124 ) shows that @xmath169 is maximized for @xmath170 . as concerns the distribution of the , we have to consider the case in which @xmath171 is given by @xmath172 , @xmath173 by the log - likelihood function @xmath174 , and @xmath61 and @xmath62 by the log - likelihood processes : @xmath175_{j=0,\ldots , j , j\neq i},\vspace*{2pt}\cr \mathbf{x}_{k}\triangleq[\ln f_{y_{k}}(y_{k};\theta_{0})-\ln f_{y_{k}}(y_{k};\theta_{j})]_{j=1,\ldots , j}.}\ ] ] also @xmath68 and @xmath176 are consequently defined . propositions [ pr - convergencerateofmle2 ] and [ pr - convergencerateofmle ] hold when assumption is replaced by assumptions , and . when the model is correctly specified , it is interesting to stress an interpretation of the moment generating function in discrete parameter models . we note that the moment generating functions can be written as follows : @xmath177}\bigr]\nonumber\\ & = & \int f_{y}(y;\theta_{i})^{\sum_{j=0,\ldots , j , j\neq i}\lambda_{j } } \nonumber \\[-8pt ] \\[-8pt ] \nonumber & & \phantom{\int}{}\cdot \prod_{j=1,\ldots , j , j\neq i}f_{y}(y;\theta_{j})^{-\lambda_{j}}\\ & & \hspace*{44pt}\qquad{}\cdot f_{y}(y;\theta_{0})^{1-\lambda_{0}}\mu(\mathrm{d}y).\nonumber\end{aligned}\ ] ] therefore , in this case , the moment generating function @xmath176 reduces to the so - called hellinger transform @xmath178 ( see @xcite , page 43 ) for a certain linear transformation of @xmath179 in @xmath180 : @xmath181^{\gamma_{j}}\\ & & \quad=\int \biggl[\prod_{j=0}^{j}f_{y}(y;\theta_{j})^{\gamma_{j}}\biggr]\mu(\mathrm{d}y),\quad \sum _ { j=0}^{j}\gamma_{j}=1.\end{aligned}\ ] ] moreover , due to its convexity , @xmath182 is surely finite for @xmath180 belonging to the closed simplex in @xmath183 . proposition [ pr - neyconvergencerateofmle ] holds if assumption is replaced by assumptions , and , and if and hold true . however , assumption is unnecessary ; indeed , the fact that @xmath88 can be proved showing that @xmath184 . this is equivalent to the existence , for @xmath185 , of two sets @xmath186 and @xmath187 of positive @xmath162-measure and included in the support of @xmath188 such that , for @xmath189 and @xmath190 , @xmath191 and @xmath192 . this follows easily noting that these densities have to integrate to @xmath193 , are almost surely ( a.s . ) different according to assumption and have the same support according to assumption . in order to derive the distribution of bayes estimators , we consider equation ( [ eq - definitionofthebayesestimators ] ) and we let @xmath194_{j=0,\ldots , j , j\neq i}$ ] then , we can write @xmath195 and we can use the previous large deviations or saddlepoint formulas , simply changing the set over which the @xmath196 is taken . however , care is needed since both formulas hold under the assumption @xmath197 in the case @xmath59 , the similarity of these formulas with the corresponding ones for a neyman pearson test is striking ; this revives the interpretation of a neyman pearson test as a bayesian estimation problem . therefore , our analysis can be seen as a ( minor ) extension of the theory of hypothesis testing to a larger number of alternatives . in this section , we are interested in the problem of efficiency , with special reference to maximum likelihood and bayes estimators . in what follows , we will suppose that the true parameter value belongs to @xmath4 ; this will be reflected in the probabilities that will be written as @xmath198 . indeed , efficiency statements for misspecified models are quite difficult to interpret . in the statistics literature , efficiency ( or superefficiency ) can be defined comparing the behavior of the estimator with respect to a lower bound or , alternatively , to a class of estimators . in the continuous case , the two concepts almost coincide ( despite superefficiency ) . however , in the discrete case , the two concepts diverge dramatically and we need more care in the derivation of the information inequalities and in the statement of the efficiency properties . an interesting problem concerns the choice ofa measure of efficiency for the in discrete parameter models : in his seminal paper , hammersley @xcite derives a generalization of cramr rao inequality for the variance that is also valid when the parameter space is countable . the same inequality has been derived , in slightly more generality , in @xcite . however , this choice is well - suited only in cases in which the is a good measure of risk , for example , if the limiting distribution of the normalized estimator is normal . following the discussion by lindley in @xcite , we consider a different cost function @xmath199 , whose risk function is given by the probability of missclassification : @xmath200 we also define the _ bayes risk _ ( under the zero one loss function ) associated with a prior distribution @xmath201 on the parameter space @xmath4 . in particular , we consider the bayes risk under the risk function @xmath202 as @xmath203 if @xmath204 we define @xmath205 as the _ average probability of error_. note that this is indeed the measure of error used by @xcite . using the risk function @xmath206 , in section [ sect - informationinequalities ] we derive some information inequalities and we prove in section [ sect - optimalityandefficiency ] some optimality and efficiency results for bayes and estimators . in section [ sect - riskfunctions ] we briefly deal with alternative risk functions . this section contains lower bounds for the previously introduced risk function @xmath206 . in the specific case of discrete parameters , these generalize and unify the lower bounds proposed in @xcite . in the following , first of all , a lower bound is proved and then a minimax version of the same result is obtained . when needed , we will refer to the former as _ chapman robbins lower bound _ ( and to the related efficiency concept as _ chapman robbins _ _ efficiency _ ) since it recalls the lower bound proposed by these two authors in their 1951 paper , and to the latter as _ minimax chapman robbins lower bound_. then , from these results , we derive a lower bound for the bayes risk . the proposition of this section is intended to play the role of cramr rao and chapman robbins lower bounds for the variance . it corresponds essentially to stein s lemma in hypothesis testing . moreover , a version of the same bound for estimators respecting ( [ eq - partiallyconsistentestimator ] ) is provided ; this corresponds to a similar result proposed in @xcite [ pr - bahadurlowerbound]under assumptions and , for a strongly consistent estimator @xmath208 : @xmath209 \\[-8pt ] \nonumber & & \quad \ge \sup_{\theta_{1}\in\theta\setminus\ { \theta _ { 0}\ } } \mathbb{e}_{\mathbb{\theta}_{1}}\ln\biggl(\frac{f_{y}(y;\theta _ { 0})}{f_{y}(y;\theta_{1})}\biggr).\end{aligned}\ ] ] on the other hand , if @xmath210 then @xmath211 [ rm - iroptimalityandcroptimality ] note that this inequality only holds for estimators that are consistent or respect condition ( [ eq - partiallyconsistentestimator ] ) , while the one of proposition [ pr - chapman - robbinslowerbound ] holds for any estimator . proposition [ pr - bahadurlowerbound ] provides an upper bound for the _ inaccuracy rate _ of @xcite : @xmath212 for any @xmath16 small enough ( @xmath213 ) . the following result is a minimax lower bound on the probability of misclassification . it is based on the neyman pearson lemma and chernoff s bound . [ pr - chapman - robbinslowerbound]under assumptions and , for any estimator @xmath214 : @xmath215.\nonumber\end{aligned}\ ] ] [ rm - hallbound ] the previous proposition provides an expression for the _ minimax bahadur risk _ ( also called _ ( minimax ) rate of inaccuracy _ ; see ) analogous to chernoff s bound , thus providing a minimax version of remark [ rm - iroptimalityandcroptimality ] . other methods to derive similar minimax inequalities are fano s inequality and assouad s lemma ( see @xcite , page 220 ) ; however , in the present case they do not allow us to obtain tight bounds , since the usual application of these methods relies on the approximation of the parameter space with a finite set of points @xmath4 whose cardinality increases with @xmath114 . clearly , this can not be done in the present case . using lemma 5.2 in @xcite , it is possible to show that the minimax bound is larger than the classical one . under assumption , the bayes risk @xmath216 under the risk function @xmath206 and the prior @xmath201 respects the equality @xmath217 then , proposition [ pr - chapman - robbinslowerbound ] holds also for the bayes risk : clearly this bound is independent of the prior distribution @xmath201 ( provided it is strictly positive , i.e. , holds ) and also holds for the probability of error @xmath218 . this inequality can be seen as an asymptotic version of the van trees inequality for a different risk function . in this section , we establish some optimality results for the in discrete parameter models . the situation is much more intricate than in regular statistical models under the quadratic loss function , in which efficiency coincides with the attainment of the cramr rao lower bound ( despite superefficiency ) . therefore , we propose the following definition . we denote by @xmath219 the risk function of the estimator @xmath220 evaluated at @xmath2 , and by @xmath221 a class of estimators . the estimator @xmath220 is _ efficient with respect to _ ( w.r.t . ) @xmath221 _ and _ w.r.t . @xmath222 _ at _ @xmath2 if @xmath223 the estimator @xmath220 is _ minimax efficient _ w.r.t . @xmath221 _ and _ w.r.t . @xmath222 if @xmath224 the estimator @xmath220 is _ superefficient _ w.r.t . @xmath221 _ _ and__w.r.t . @xmath222 if for every @xmath225 : @xmath226 for every @xmath1 and there exists at least a value @xmath227 such that the inequality is replaced by a strict inequality for @xmath228 . the estimator @xmath220 is _ asymptotically _ @xmath229-_efficient _ w.r.t . @xmath222 _ at _ @xmath2 if it attains the chapman robbins lower bound of proposition [ pr - bahadurlowerbound ] at @xmath2 [ say @xmath230 in the asymptotic form : @xmath231 the estimator @xmath220 is _ asymptotically minimax _ @xmath232-_efficient @xmath222 if it attains the minimax chapman robbins lower bound of proposition [ pr - chapman - robbinslowerbound ] ( say @xmath233 ) in the asymptotic form : @xmath234 the estimator @xmath220 is _ asymptotically _ @xmath229-_superefficient _ w.r.t . @xmath222 if @xmath235 for every @xmath1 and there exists at least a value @xmath227 such that the inequality is replaced by a strict inequality for @xmath228 . as in remark [ rm - iroptimalityandcroptimality ] , it is easy to see that @xmath236-optimality and @xmath232-efficiency w.r.t . @xmath206 coincide . the efficiency landscape offered by discrete parameter models will be illustrated by example [ exam - integermeanofagaussiansample ] . this shows that , even in the simplest case , that is , the estimation of the integer mean of a gaussian random variable with known variance , the does not attain the lower bound on the missclassification probability but it attains the minimax lower bound . moreover , simple estimators are built that outperform the for certain values of the true parameter value @xmath2 . [ exam - integermeanofagaussiansample]let us consider the estimation of the mean of a gaussian distribution whose variance @xmath237 is known : we suppose that the true mean is @xmath238 , while the parameter space is @xmath239 , where @xmath238 is known . the maximum likelihood estimator @xmath41 takes the value @xmath240 if the sample mean takes on its value in @xmath241 and @xmath238 if it falls in @xmath242 ( the position of @xmath243 is a convention ) . therefore : @xmath244 & = & \int_{-\infty}^{0}\frac { e^{-{(\bar{y}-\alpha)^{2}}/{(2\sigma^{2}/n)}}}{\sqrt{2\pi\sigma ^{2}/n}}\,\mathrm{d}\bar{y}\\ & = & \int_{-\infty}^{-{\sqrt{n}\alpha}/{\sigma } } \frac{e^{-{t^{2}}/{2}}}{\sqrt{2\pi}}\,\mathrm{d}t\\ & = & \phi\biggl(-\frac{\sqrt{n}\alpha}{\sigma}\biggr)\\ & = & \frac{e^{-{n\alpha ^{2}}/{(2\sigma^{2})}}}{\sqrt{2\pi n}}\frac{\sigma}{\alpha}\cdot\biggl(1+o\biggl(\frac{1}{n}\biggr)\biggr),\end{aligned}\ ] ] where we have used problem 1 on page 193 in @xcite . proposition [ pr - exactasymptoticsinr ] allows also for recovering the right convergence rate . indeed , we have @xmath245 on the other hand , the lower bound of proposition [ pr - bahadurlowerbound ] yields @xmath246 and the lower bound of proposition [ pr - chapman - robbinslowerbound ] yields @xmath247 therefore , the asymptotically attains the minimax lower bound but not the classical one . in the following , we will show that estimators can be pointwise more efficient than the ; consider the estimator defined by @xmath248 when @xmath249 , @xmath250 coincides with the @xmath41 . then , the behavior of the estimator is characterized by the probabilities : @xmath251 we have ( weak ) consistency if @xmath252 the risk @xmath253 under @xmath2 is then @xmath254;\ ] ] this can be made smaller than the probability of error of the simply taking @xmath255 , thus implying that the is not pointwise efficient . now , we show that this estimator can not converge faster than the chapman robbins lower bound without losing its consistency . indeed , @xmath256 is smaller than the chapman robbins lower bound if @xmath257 and this is never true under ( [ eq - consistencyconditionsintheexample ] ) . if this estimator is pointwise more efficient than the under @xmath2 , then its risk under @xmath258 is given by @xmath259,\ ] ] and this is greater than for the . this shows that a faster convergence rate can be obtained in some points , the price to pay being a worse convergence rate elsewhere in @xmath4 . in the following section , we show some optimality properties of bayes and estimators . we start with an important and well - known fact . [ pr - minimizationofthebayesrisk]under , , and , the bayes risk @xmath260 ( under the zero one loss function ) associated with a prior distribution @xmath201 is strictly minimized by the posterior mode corresponding to the prior @xmath201 , for any finite @xmath114 . the following proposition shows that the is admissible and minimax efficient under the zero one loss and minimizes the average probability of error . it implies that estimators that are more efficient than the at a certain point @xmath261 are less efficient in at least another point @xmath262 . as a result , estimators can be more efficient than minimax efficient ones only on portions of the parameter space , but are then strictly less efficient elsewhere . [ pr - admissibilityandminimaxefficiencyofmle]under assumptions , and , the is admissible and minimax efficient w.r.t . the class of all estimators and w.r.t . @xmath206 and minimizes the average probability of error @xmath218 . in this subsection , we will show that the does not attain the chapman robbins lower bound in the form of proposition [ pr - bahadurlowerbound ] but that it attains the minimax form of proposition [ pr - chapman - robbinslowerbound ] and that efficiency and minimax efficiency are generally incompatible . therefore , the situation described in example [ exam - integermeanofagaussiansample ] is general , for it is possible to show that the is generally inefficient with respect to the lower bounds exposed in proposition [ pr - bahadurlowerbound ] . [ pr - proofofnon - ir - optimality]under assumptions , and : the is not asymptotically @xmath232-efficient w.r.t . @xmath206 at @xmath2 ; the is asymptotically minimax @xmath232-efficient w.r.t . @xmath206 ; an estimator that is asymptotically @xmath232-efficient w.r.t . @xmath206 at @xmath2 is not asymptotically minimax @xmath232-efficient w.r.t . @xmath206 . the assumption of homogeneity of the probability measures , necessary to derive , can be removed in the proof of along the lines of @xcite . ever since it was discovered by hodges , the problem of superefficiency has been dealt with extensively in regular statistical problems ( see , e.g. , @xcite ) . however , these proofs do not transpose to discrete parameter estimation problems , since they are mostly based on the equivalence of prior probability measures with the lebesgue measure and on properties of bayes estimators that do not hold in this case . moreover , the discussion of the previous sections has shown that , in discrete parameter problems , @xmath229-efficiency and efficiency with respect to a class of estimators do not coincide . the following proposition yields a solution to the superefficiency problem . [ pr - superefficiency]under assumptions , and : no estimator @xmath214 is asymptotically @xmath232-superefficient w.r.t . @xmath206 at @xmath261 ; no estimator @xmath214 is superefficient w.r.t . the and @xmath206 . now we consider in what measure the previous results transpose when changing the risk function . following @xcite , we first consider the quadratic cost function and the corresponding risk function : @xmath263 the cost function @xmath264 has the drawback of weighting in the same way points of the parameter space that lie at different distances with respect to the true value @xmath2 . in many cases , a more general loss function can be considered , as suggested in @xcite ( volume 1 , page 51 ) for multiple tests : @xmath265 where @xmath266 for @xmath267 can be tuned in order to give more or less weight to different points of the parameter space . the risk function is therefore given by the weighted probability of misclassification @xmath268 . it is trivial to remark that @xmath269 and the lower bounds of propositions [ pr - bahadurlowerbound ] and [ pr - chapman - robbinslowerbound ] hold also in this case . the same equalities hold also for @xmath270 . as a result , proposition [ pr - proofofnon - ir - optimality ] and proposition [ pr - superefficiency ] apply also to these risk functions . on the other hand , as concerns proposition [ pr - admissibilityandminimaxefficiencyofmle ] and proposition [ pr - superefficiency ] , it is simple to show that with respect to the risk functions @xmath271 and @xmath272 , the results hold only asymptotically ( see @xcite , for asymptotic minimax efficiency of the estimator of the integral mean of a gaussian sample with known variance ) . proof of proposition [ pr - consistencyofthemle ] under , kolmogorov s implies that @xmath40-a.s . @xmath273 , and for @xmath40-a.s . any sequence of realizations , @xmath41 converges to @xmath2 . measurability follows from the fact that the following set belongs to @xmath274 : @xmath275 proof of lemma [ lm - cramerconditionandeta - int ] clearly implies for a certain @xmath276 . on the other hand , suppose that holds ; then , applying recursively hlder inequality : @xmath277\\ & \leq&\sum_{j=0,\ldots , j , j\neq i}\frac{1}{j}\cdot\ln\mathbb { e}_{0}\biggl[\biggl(\frac{q(y;\theta_{i})}{q(y;\theta_{j})}\biggr)^{j\cdot\lambda_{j}}\biggr]\end{aligned}\ ] ] and choosing the @xmath278 s adequately , we get . proof of proposition [ pr - convergencerateofmle2 ] the first two results are straightforward applications of cramr s theorem in @xmath279 ( see , e.g. , @xcite , corollary 6.1.6 , page 253 ) . indeed , it is known that the lower bound holds without any supplementary assumption , while the upper bound requires a cramr condition @xmath280 ; indeed , from lemma [ lm - cramerconditionandeta - int ] , this is equivalent to assumption . then , a full holds : @xmath281 & & \quad \geq-\inf_{\mathbf{y}\in\operatorname{int } \mathbb { r}_{+}^{j}}\sup_{\bolds{\lambda}\in\mathbb{r}^{j}}\bigl\ { \langle\mathbf { y},\bolds{\lambda}\rangle-\lambda^{(i)}(\bolds{\lambda})\bigr\ } , \\[-2pt ] & & \limsup_{n\rightarrow\infty}\frac{1}{n}\ln\mathbb{p}_{0}(\hat{\theta } ^{n}=\theta_{i } ) \\[-2pt ] & & \quad\leq-\inf_{\mathbf{y}\in\mathbb{r}_{+}^{j}}\sup _ { \bolds{\lambda}\in\mathbb{r}^{j}}\bigl\ { \langle\mathbf{y},\bolds { \lambda}\rangle-\lambda^{(i)}(\bolds{\lambda})\bigr\ } .\end{aligned}\ ] ] in order to prove the final result , we have to show that @xmath282 is a @xmath283-_continuity set _ , that is,@xmath284 . it is enough to apply part ( ii ) in lemma on page 903 of @xcite . proof of proposition [ pr - convergencerateofmle ] first of all , we note that @xmath285 . therefore , we can apply large deviations principles , with the candidate rate function @xmath70 ; this is a strictly convex function on @xmath286 globally minimized at @xmath287_{j=1,\ldots , j}.\ ] ] by assumption , @xmath288 is finite and belongs to @xmath289 . from the strict convexity of the level sets of @xmath70 , the set @xmath290 has at most finite cardinality @xmath108 . moreover , since large deviations theory allows us to ignore the part of @xmath97 where @xmath291 , we can replace @xmath292 with a collection of @xmath108 disjoint sets , say @xmath293 , @xmath294 , each of them containing in its interior one and only one of the points of @xmath295 ( see @xcite , page 508 ) : @xmath296 & & \quad=\bigl(1+o(1)\bigr)\cdot\mathbb{p}_{0}\biggl(\sum_{k=1}^{n}\mathbf { x}_{k}\in\operatorname{int } \bigcup_{h=1}^{h}\gamma_{h}\biggr)\\[-2pt ] & & \quad = \bigl(1+o(1)\bigr)\cdot\sum_{h=1}^{h}\mathbb{p}_{0}\biggl(\sum_{k=1}^{n}\mathbf { x}_{k}\in\operatorname{int } \gamma_{h}\biggr).\nonumber\end{aligned}\ ] ] as before , the bounds derive from cramr s theorem in @xmath279 . noting that the contribution of any @xmath293 is the same and recalling ( [ eq - setrepresentationasgammah ] ) , we get the results . proof of proposition [ pr - neyconvergencerateofmle ] the assumptions of the theorem on page 904 of @xcite are easily verified . this shows that a unique dominating point @xmath297 exists and implies , through proposition on page 161 of @xcite ( according to the `` remarks on the hypotheses '' in @xcite , page 905 , the `` lattice '' conditions are not necessary ) , that the stated bracketing of @xmath53 holds . proof of proposition [ pr - exactasymptoticsinr ] @xmath298under assumptions , , and , according to proposition [ pr - convergencerateofmle2 ] we have @xmath299 and we can study the behavior of @xmath300 assumption implies that the conditions of theorem 3.7.4 in @xcite ( page 110 ) are verified , in particular the existence of a positive @xmath301 solution to the equation @xmath302 . from lemma 2.2.5(c ) in @xcite , this implies @xmath303 , and the result follows . proof of theorem [ th - saddlepointapproximation ] we note that the function @xmath304 in @xcite ( page 1117 ) is given by @xmath305\\[1pt ] & = & \ln\mathbb{e}_{0}\exp\bigl[\mathbf{u}\cdot\mathbf{x}^{(i)}\bigr]-\mathbf { u}\cdot\mathbb{e}_{0}\mathbf{x}^{(i)}\\ & = & \lambda^{(i)}(\mathbf{u})-\mathbf{u}\cdot\mathbb{e}_{0}\mathbf{x}^{(i)}.\vspace*{1pt}\end{aligned}\ ] ] therefore , we write the mean @xmath306 and covariance matrix @xmath307 as @xmath308 \mathbf{v}(\mathbf{u } ) & = & \kappa^{\prime\prime}(\mathbf{u})=\frac { \partial^{2}\kappa(\mathbf{u})}{\partial\mathbf{u}^{2}}=\frac{\partial ^{2}\lambda^{(i)}(\mathbf{u})}{\partial\mathbf{u}^{2}}.\vspace*{1pt}\end{aligned}\ ] ] from ( [ eq - probabilityforthetaiinldform ] ) , we have @xmath309 & & \hspace*{-3pt}\quad = \mathbb{p}_{0}\biggl(\sum _ { k=1}^{n}\mathbf{x}_{k}^{(i)}\in\operatorname{int } ( \mathbb{r}_{+}^{j})\biggr)\\[1pt ] & & \hspace*{-3pt}\quad = \mathbb{p}_{0}\biggl\ { \frac{1}{n}\cdot\sum_{k=1}^{n}\bigl(\mathbf { x}_{k}^{(i)}-\mathbb{e}_{0}\mathbf{x}^{(i)}\bigr)\in\operatorname{int } ( \mathbb { r}_{+}^{j})\ominus\mathbb{e}_{0}\mathbf{x}^{(i)}\biggr\ } .\end{aligned}\ ] ] now we verify assumptions ( s.1)(s.4 ) of @xcite . assumption ( s.1 ) is implied by . assumptions ( s.2 ) and ( s.3 ) hold since the random vectors are i.i.d . and nontrivial . at last , ( s.4 ) is implied by ( see , e.g. , @xcite , page 735 ) . since @xmath310 is strictly negative by , @xmath311 does not contain @xmath312 and , according to theorem 1 in @xcite ( page 1118 ) , the result of the theorem follows . proof of proposition [ pr - bahadurlowerbound ] first of all , weprove ( [ eq - bahadurbound - kldivergence ] ) . we suppose that @xmath313 otherwise the inequality is trivial . then , for any @xmath314 , we apply lemma 3.4.7 in @xcite ( page 94 ) with @xmath315 and @xmath316;since @xmath214 is strongly consistent , @xmath317 is ultimately less than any @xmath318 and the bound holds . the second part can be proved as follows . define the sets @xmath319 b_{n}(j ) & = & \biggl\ { \omega\dvtx \frac{1}{n}\ln\biggl(\frac{f_{y}(y;\theta _ { j})}{f_{y}(y;\theta_{0})}\biggr)\\[-2pt ] & & \hspace*{6pt}\le\mathbb{e}_{\mathbb{\theta}_{j}}\ln\biggl(\frac { f_{y}(y;\theta_{j})}{f_{y}(y;\theta_{0})}\biggr)+\varepsilon\biggr\}.\end{aligned}\ ] ] therefore , we have @xmath320 & & \quad=\mathbb{e}_{\theta_{0}}\mathsf{1}\ { \tilde{\theta}^{n}\neq\theta_{0}\ } \\[-1pt ] & & \quad=\mathbb{e}_{\theta_{j}}\frac{f_{y}(y;\theta_{0})}{f_{y}(y;\theta _ { j})}\mathsf{1}\ { \tilde{\theta}^{n}\neq\theta_{0}\ } \\[-1pt ] & & \quad \ge \mathbb{e}_{\theta_{j}}\frac{f_{y}(y;\theta _ { 0})}{f_{y}(y;\theta_{j})}\mathsf{1}\ { a_{n}(j)\ } \\[-1pt ] & & \quad \ge \mathbb{e}_{\theta_{j}}\mathsf{1}\ { a_{n}(j)\ } \mathsf{1}\ { b_{n}(j)\}\\[-1pt ] & & \qquad { } \cdot\exp\biggl\ { -n\cdot\biggl[\mathbb{e}_{\mathbb{\theta}_{j}}\ln\biggl(\frac { f_{y}(y;\theta_{j})}{f_{y}(y;\theta_{0})}\biggr)+\varepsilon\biggr]\biggr\ } \\[-1pt ] & & \quad \ge [ 1-\mathbb{p}_{\theta_{j}}\ { a_{n}^{c}(j)\ } -\mathbb{p}_{\theta _ { j}}\ { b_{n}^{c}(j)\ } ] \\[-1pt ] & & \qquad{}\cdot\exp\biggl\ { -n\cdot\biggl[\mathbb{e}_{\mathbb{\theta } _ { j}}\ln\biggl(\frac{f_{y}(y;\theta_{j})}{f_{y}(y;\theta_{0})}\biggr)+\varepsilon \biggr]\biggr\ } \\[-1pt ] & & \quad \ge [ 1-\mathbb{p}_{\theta_{j}}\ { \tilde{\theta}^{n}\neq\theta _ { j}\ } -\mathbb{p}_{\theta_{j}}\ { b_{n}^{c}(j)\ } ] \\[-1pt ] & & \qquad{}\cdot\exp\biggl\ { -n\cdot \biggl[\mathbb{e}_{\mathbb{\theta}_{j}}\ln\biggl(\frac{f_{y}(y;\theta _ { j})}{f_{y}(y;\theta_{0})}\biggr)+\varepsilon\biggr]\biggr\ } .\end{aligned}\ ] ] this implies : @xmath321 & & \quad \ge -\mathbb{e}_{\mathbb{\theta } _ { j}}\ln\biggl(\frac{f_{y}(y;\theta_{j})}{f_{y}(y;\theta_{0})}\biggr)-\varepsilon\\[-2pt ] & & \qquad { } + \liminf_{n\rightarrow\infty}\frac{1}{n}\ln[1-\mathbb{p}_{\theta _ { j}}\ { \tilde{\theta}^{n}\neq\theta_{j}\ } -\mathbb{p}_{\theta_{j}}\ { b_{n}^{c}(j)\ } ] .\vadjust{\goodbreak}\end{aligned}\ ] ] now , since @xmath322 and@xmath323 , the third term in the right - hand side goes to zero ; since @xmath16 is arbitrary , the result follows . proof of proposition [ pr - chapman - robbinslowerbound ] from the neyman pearson lemma , we have @xmath324 for an arbitrary couple of different alternatives @xmath2 and @xmath258 in @xmath4 . then we can use chernoff s bound ( @xcite , page 93 ) ; the final expression derives from the equality @xmath325 . proof of proposition [ pr - admissibilityandminimaxefficiencyofmle ] in order to prove that the is admissible and minimax we use the bayesian method . using the prior densities given by @xmath326 , the bayes estimator relative to zero one loss @xmath156 coincides with the @xmath41 . therefore , respectively from lemma 2.10 and proposition 6.3 in @xcite , @xmath327 is minimax and admissible . the fact that the minimizes the average probability of error derives from proposition [ pr - minimizationofthebayesrisk ] . proof of proposition [ pr - proofofnon - ir - optimality ] in order to prove the first statement , we apply lemma 2.4 in @xcite ( page 653 ) . clearly @xmath159 is closed in total variation , since it is finite , and is not exponentially convex ; indeed , under assumption , there exist @xmath328 and @xmath329 $ ] , such that the probability measure @xmath330 defined as @xmath331 does not belong to @xmath159 . therefore , from lemma 2.4 in @xcite , there exist @xmath332 such that equation ( 2.12 ) in @xcite holds and , as a consequence of lemma 2.4 in @xcite , the fails to be an inaccuracy rate optimal estimator at least at one of the points @xmath333 . this means that , say for @xmath334 : @xmath335 and this implies that the chapman robbins bound is not attained at @xmath334 . the second statement follows easily from the results of @xcite ( theorem 2 ) on @xmath336 , using equation ( [ eq - asymptoticequivalenceofbayesriskandminimaxproberror ] ) . indeed , the attains the lower bound ( [ eq - chernoffslowerbound ] ) and is therefore asymptotically minimax efficient . if the estimator is asymptotically @xmath232-efficient w.r.t . @xmath206 at @xmath2 , this means that at @xmath2 it is more efficient than the and therefore it has to be less efficient elsewhere ( since from proposition [ pr - admissibilityandminimaxefficiencyofmle ] the minimizes the probability of error ) . therefore , it can not be minimax @xmath232-efficient . proof of proposition [ pr - superefficiency ] for it is enough to follow the proof of proposition [ pr - bahadurlowerbound ] and to reason by contradiction , while is simply another way of stating proposition [ pr - admissibilityandminimaxefficiencyofmle ] . the authors would like to thank lucien birg , mehmet caner , jean - pierre florens , christian gouriroux , christian hess , marc hoffmann , pierre jacob , sren johansen , rasul a. khan , oliver b. linton , christian p. robert , keunkwan ryu , igor vajda and the participants to seminars at universit paris 9 dauphine , crest and institut henri poincar , to esem 2001 in lausanne , to xxxivmes journes de statistique 2002 in bruxelles , to bs / imsc 2004 in barcelona , and to esem 2004 in madrid . all the remaining errors are our responsibility .

we investigate the unification scenario provided by the generalised chaplygin gas model ( a perfect fluid characterized by an equation of state ) . our concerns lie with a possible tension existing between background kinematic tests and those related to the evolution of small perturbations . we analyse data from the observation of the differential age of the universe , type ia supernovae , baryon acoustic oscillations and the position of the first peak of the angular spectrum of the cosmic background radiation . we show that these tests favour negative values of the parameter : we find at the 2 level and that with 85% confidence . these would correspond to negative values of the square speed of sound which are unacceptable from the point of view of structure formation . we discuss a possible solution to this problem , when the generalised chaplygin gas is framed in the modified theory of gravity proposed by rastall . we show that a fluid description within this theory does not serve the purpose , but it is necessary to frame the generalised chaplygin gas in a scalar field theory . finally , we address the standard general relativistic unification picture provided by the generalised chaplygin gas in the case : this is usually considered to be undistinguishable from the standardcdm model , but we show that the evolution of small perturbations , governed by the mszros equation , is indeed different and the formation of sub - horizon gcg matter halos may be importantly affected in comparison with thecdm scenario .

1212.4136

under the assumption of homogeneity and isotropy of the universe at large scales , and that general relativity is the correct theory of gravity , the accelerated expansion @xcite is due to an exotic component , with negative pressure , called dark energy ( de ) . moreover , structures such as galaxies and clusters of galaxies seem to require the existence of another exotic component , dubbed dark matter ( dm ) @xcite , in order to form . de and dm constitute the so - called dark sector of the universe , accounting for about @xmath7 of the total cosmic energy budget . their nature is still a mystery , but they clearly do not fit in the standard model of elementary particles @xcite . one important proposal describing the dark sector of the universe is the unification scenario , where dm and de are considered different manifestations of a unique component . the paradigm of such idea is the chaplygin gas ( gc ) model @xcite , where an equation of state inspired by string theory @xcite allows to obtain a dynamic behaviour mimicking dm in the past and de in current time , as structure formation and the accelerated expansion require . later , the cg has been phenomenologically generalized , leading to the so - called generalized chaplygin gas ( gcg ) @xcite . though the idea is very appealing , the gcg faces crucial drawbacks when the model undergoes observational tests . this has been first indicated in reference @xcite , concerning the behaviour of small perturbations and structure formation , and pointed out again in @xcite . one important aspect is that the so - called kinematic tests ( those based on the background expansion and the calculation of distances ) seem to favour values for the gcg equation of state parameters that imply negative square speed of sound @xcite ( though they do not exclude positive values ) . if this is really the case , up to our knowledge in the literature there are only two ways out : @xmath8 to introduce _ ad hoc _ entropic perturbations @xcite which nullify ( or make positive ) the effective speed of sound or @xmath9 to modify the gravity theory , by supposing e.g. a non - conservative theory of gravity @xcite . our goal in this work is to understand to which extent the results of @xcite are really robust from the point of view of the kinematic tests . this question has already been addressed in @xcite , using the differential age of the universe ( i.e. the @xmath10 test ) and type ia supernova ( snia ) . we extend that analysis by enlarging the number of observational data for the @xmath10 test and by including the position of the first acoustic peak of the cosmic microwave background ( cmb ) spectrum and the baryonic acoustic oscillation ( bao ) tests . we revisit the constraints placed by these datasets , but we do not touch delicate issues such as the calibration problem of the snia data @xcite . our results seem to reveal the robustness of the kinematic tests for the gcg model : we find @xmath11 ( at the 2@xmath3 level ) . therefore , we not only confirm the previous results , but stress that negative values of @xmath1 seem to be favoured over the positive ones ( for comparison , see table 1 of @xcite ) . note that we can not report a real tension `` between background tests and the ones involving perturbations , since at 2@xmath3 we find that positive values of @xmath1 are still allowed ( @xmath12 , whereas in @xcite the authors find @xmath13 , for perturbations ) , but we can speculate how the chaplygin gas could be ' ' saved if such tension would appear in the future , given the fact that many experiments are ongoing and others forthcoming ( such as euclid ) , thus the precision on the observational data is increasing day after day . within the tension scenario , if the gcg is indeed the matter component behind the dark sector , it must get its origin from a different theory of gravity rather than gr . for example , non - standard theories of gravity such as rastall s @xcite , @xmath14 @xcite or horava - lifshitz @xcite could possibly improve the status of the gcg . our discussion in sec . v is devoted to open a new window for gcg based cosmological models where standard gr is replaced by rastall s theory . we also stress that , in general , the accepted idea is that the gcg ( based on gr theory ) works only if its parameter space is reduced to the @xmath6cdm one , in particular if @xmath5 , see eq . ( [ pressure ] ) and eq . ( [ hchap ] ) @xcite . we discuss this issue in sec . we point out that the reduction @xmath5 implies that the gcg does not behave as a pressureless fluid at high redshifts , but as a fluid with a small negative constant pressure . hence the analysis of the growth of sub - horizon gcg dark matter halos during the matter dominated epoch and the `` mszros effect '' could be different from the standard @xmath6cdm picture , possibly leading to a different non - linear clustering pattern at small scales and to a modification of the transfer function . this argument agrees with the conclusions of @xcite . for a opposite point of view , i.e. that the perturbations in the gcg with @xmath5 are the same ( at all perturbative orders ) as in the @xmath6cdm , see @xcite . the paper is organized as follows . in sec . [ sec : succandprob ] , we review briefly the possible tensions within gcg model tests . in sec . [ sec : bgtests ] we describe the observational tests we perform , while in sec . [ sec : anandres ] we carry out the statistical analysis based on such tests . in sec . [ sec : healing ] we investigate a rastall s theory approach which could possibly save the gcg . in sec . [ sec : meszeq ] we discuss the general relativistic limit @xmath15 of the gcg and solve its mszros equation . in sec . [ sec : conc ] we present our conclusions . the gcg model is characterized by the equation of state @xcite @xmath16 where @xmath17 and @xmath1 are free parameters . the original chaplygin gas model , which is somehow connected with the nambu - goto action of string theory , implies @xmath18 @xcite , the case @xmath19 being a phenomenological generalization . integrating the conservation equation for the fluid , @xmath20 leads to the following expression for the gcg density as function of the scale factor : @xmath21^\frac{1}{1 + \alpha}\;,\ ] ] where we have defined the new parameter @xmath22 and present - time quantities are indicated by the subscript @xmath23 . from eq . ( [ rhoc ] ) one can see that the original motivation behind the gcg , i.e. a fluid that evolves from the matter behaviour at early times to a constant density at late times , occurs only if @xmath24 . when one consider a model where the energy content is given by radiation and baryons , besides the chaplygin gas , friedmann s equation becomes @xmath25^\frac{1}{1 + \alpha}\;,\\ \omega_{c0 } & = & 1 - \omega_{b0 } - \omega_{r0}\;,\end{aligned}\ ] ] where @xmath26 is the hubble constant and @xmath27 is the redshift , which is related to the scale factor by @xmath28 . we have also assumed a spatially flat background , according to the recent results of wmap7 @xcite . the hubble parameter today can be expressed as @xmath29 km s@xmath30 mpc@xmath30 , where @xmath31 ( we will adopt the range @xmath32 as a prior for our bayesian analysis ) . we also introduce @xmath33 , for future convenience . the idea of dm - de unification into a single fluid in expression ( [ hchap ] ) , faces difficulties from different points of view . from the theoretical side , if vacuum energy is not responsible for de , it should be explained how it contributes to gravity and therefore how it enters eq . . from the observational point of view , introducing from eq . ( [ pressure ] ) the gcg speed of sound @xmath34 one can see that if background tests favour negative values of the parameter @xmath1 , then @xmath35 is negative . this is not only an undesirable feature of the model , but depending on how much it is negative a `` tension '' between background and perturbative tests may arise . these problems can be alleviated if , for example , positive entropic perturbations @xmath36 are introduced such that the effective speed of sound becomes a positive quantity . however , the introduction of entropic perturbations implies new free parameters : there are some degrees of arbitrariness , and in some cases , such as those of reference @xcite , it is possible only a slight extension to negative @xmath1 . another possibility of achieving negative values of @xmath1 is to implement a scalar version of the gcg model using a non - conservative theory of gravity , like rastall s theory . this has been done in reference @xcite , and in fact negative values of @xmath1 seem still to be favoured at the level of perturbations . this problem concerns not only the confrontation with the power spectrum data , but also the analysis of the anisotropy of the cosmic microwave background radiation ( cmb ) . in references @xcite , the confrontation of the gcg model with the cmb spectrum has been performed , indicating that the most favoured scenario implies @xmath15 , that is , the gcg model reduces to the @xmath6cdm model . this result is one of the reasons for speculating that the gcg model should be be ruled out . there are four main background tests for a cosmological model : 1 . the differential age of old galaxies , given by @xmath10 . the sn ia data . 3 . the position of the first cmb acoustic peak . 4 . the peak position of the baryonic acoustic oscillations ( bao ) . for the differential age data , connected with the evaluation of the age of old galaxies that have evolved passively , leading to values of @xmath10 for specific redshifts , there are 13 observational data @xcite . recently , a compilation of 21 data points has been considered @xcite . the fundamental relation is @xmath37 the value of the hubble parameter today obtained by the hst ( @xmath38 km s@xmath30 mpc@xmath30 ) could also be added to this sample , but this would imply a prior on the final parameter estimation and , therefore , will not be included here . the analysis with the sample of 13 data points for the chaplygin gas model has been made in reference @xcite , leading to the conclusion that slightly negative values for @xmath1 are favoured . in our analysis below we use the set of 21 data points compiled in ref . @xcite . the sn ia test is based on the luminosity distance , given by @xmath39 for the sn ia data , we have two main problems . the first one concerns the choice of the sample . there are many different sn ia data set , obtained with different techniques . in some cases , these different samples may give very different results . the second point is the existence of two different calibration methods : one using cosmology and which takes into account sn with high @xmath27 ( salt2 ) ; the other using astrophysics methods , valid for small @xmath27 ( mlcs2k2 ) @xcite . in some case , the employment of different calibrations can lead to different results also . all this , make the sn ia analysis very delicate . here , however , we use the union sample @xcite , calibrated by the salt2 method . this choice is motivated by looking for a contact with previous results in the literature , including those of reference @xcite . the position of the first peak of the cmb spectrum is a more complex test . it is linked to oscillations of the baryon - photon plasma at the recombination period and is given by @xmath40 a detailed numerical analysis leading to the following fitting formula for parameters of this fundamental quantity @xcite : @xmath41 we will use the following values for the different density parameter : @xmath42 from observation : @xmath43 . the baryonic acoustic oscillations are due to the effect of oscillations in the photon - baryon plasma at the moment of the decoupling , at about @xmath44 ( it is of course the same physics which produces the acoustic peaks structure in the cmb , but now this effect is observed in the baryonic distribution ) . this effect is quantified by the following expression @xcite : @xmath45 in this work we use data from the wigglez dark energy survey @xcite . the gcg behaves as dust in the past . hence , we should identify an effective dm component in the gcg in order to use the above formulas for the analysis . there are different prescriptions in this sense in the literature . we will adopt the decomposition proposed in @xcite , where the effective dm component is given by , @xmath46 we perform a bayesian statistic analysis in comparing the theoretical predictions with the observational data . first , for each dataset , we compute @xmath47 where @xmath48 is the theoretical prediction for the quantity @xmath49 , whereas @xmath50 is the corresponding observational datum , with an error estimation @xmath51 . from the chi - squared , we build the probability distribution function ( pdf ) , as follows : @xmath52 where we assumed the data to be independent and normally distributed ; @xmath53 is a normalization constant . as indicated , the pdf depends on three free parameters : @xmath54 , @xmath1 and @xmath55 , characterizing a three dimensional function . two and one dimensional pdf can be computed integrating ( marginalizing ) on the remaining parameters . as already said , we will compute the @xmath56 separately for each dataset . the total @xmath56 , for the set of the four observational tests , shall be the sum of the separated @xmath56 . accordingly to the exponential form of the pdf , in eq . , the total pdf shall be the product of the single ones . the intervals considered for each of the free parameters are crucial for the final estimations . for @xmath55 , all values @xmath57 are considered . for @xmath54 we consider both delta and constant priors i.e. , for the former prior we fix values for @xmath54 while for the latter we integrate in the interval @xmath58 . for @xmath1 , the situation is more complex , since we can consider two possibilities : @xmath59 , in order to assure a transition from dust in the past to a cosmological constant in the future , or @xmath60 , in order to assure a positive square speed of sound . our goal here is to show that negative values for @xmath1 are preferred . therefore , we focus our analysis on the prior @xmath61 . however , for the sake of completeness , we also carry out a statistical analysis leaving @xmath1 free to vary to arbitrarily negative and positive values . we evaluate the one dimensional pdf for @xmath55 and @xmath1 for the four independent tests listed before and combining all of them . the results are shown in . in we show the one - dimensional probability distributions for @xmath1 and @xmath62 for specific choices of @xmath63 . the values chosen for @xmath64 are shown in the four panels . each panel corresponds to a data set . all the curves have been normalized following @xmath65 except for the h(z ) data , we see that the maximum likelihoods occur for hyper - surfaces with negative values of @xmath1 . under the restriction @xmath61 using @xmath10 ( upper left panel ) , cmb ( upper right panel ) , snia ( lower left panel ) and bao ( lower right panel ) . we have used here delta priors of @xmath54.,title="fig : " ] under the restriction @xmath61 using @xmath10 ( upper left panel ) , cmb ( upper right panel ) , snia ( lower left panel ) and bao ( lower right panel ) . we have used here delta priors of @xmath54.,title="fig : " ] + under the restriction @xmath61 using @xmath10 ( upper left panel ) , cmb ( upper right panel ) , snia ( lower left panel ) and bao ( lower right panel ) . we have used here delta priors of @xmath54.,title="fig : " ] under the restriction @xmath61 using @xmath10 ( upper left panel ) , cmb ( upper right panel ) , snia ( lower left panel ) and bao ( lower right panel ) . we have used here delta priors of @xmath54.,title="fig : " ] under the restriction @xmath61 using @xmath10 ( upper left panel ) , cmb ( upper right panel ) , snia ( lower left panel ) and bao ( lower right panel ) . we have used here delta priors of @xmath54.,title="fig : " ] under the restriction @xmath61 using @xmath10 ( upper left panel ) , cmb ( upper right panel ) , snia ( lower left panel ) and bao ( lower right panel ) . we have used here delta priors of @xmath54.,title="fig : " ] + under the restriction @xmath61 using @xmath10 ( upper left panel ) , cmb ( upper right panel ) , snia ( lower left panel ) and bao ( lower right panel ) . we have used here delta priors of @xmath54.,title="fig : " ] under the restriction @xmath61 using @xmath10 ( upper left panel ) , cmb ( upper right panel ) , snia ( lower left panel ) and bao ( lower right panel ) . we have used here delta priors of @xmath54.,title="fig : " ] for this analysis we left @xmath54 to vary in the range @xmath66 . the final one - dimensional pdf for @xmath1 and @xmath62 are calculated formally as @xmath67 it is crucial to restrict the interval for @xmath1 . if we consider the standard scenario for structure formation , in the gr context without entropic perturbations , it is required that @xmath68 . however , if the only constraint is to impose an acceleration to deceleration transition , then the restriction is @xmath59 , since for @xmath69 the universe accelerates in the past , and it is decelerating today . alternatively , we may leave the value of @xmath1 free in order to test the consistency of the background tests in the context of the gcg model . in what follows , we will consider two cases : @xmath59 and @xmath1 free . for both the cases , the general behaviour follows similar features , that can be summarized as follows . the @xmath10 and the baryonic acoustic oscillations tests predict a maximum for @xmath1 slightly negative , while the position of the first peak indicates a maximum for a small positive value of @xmath1 . for these three tests the pdf decreases as the value @xmath70 is approached . only for the snia test the pdf for @xmath1 can be significant for @xmath69 . it is important to stress that , for the baryonic acoustic oscillations and for the cmb first peak , the pdf becomes essentially zero for @xmath69 . this is due to the decomposition into a `` dark matter '' component employed in eq . ( [ decomposition ] ) : for @xmath69 , the behaviour for `` dark matter '' is reversed , due to the change of sign of the exponent in eq . ( [ decomposition ] ) . we show the results of the analysis in and , for @xmath59 , and in and , for @xmath1 free . the @xmath71 estimation for @xmath1 in the first case is @xmath11 , while in the second case is @xmath72 . this result shows that @xmath59 is a good prior . moreover , we find that @xmath4 with 85% confidence . under the restriction @xmath61 using @xmath10 ( upper left panel ) , cmb ( upper central panel ) , snia ( upper right panel ) , bao ( lower left panel ) and the combination of these four tests ( lower right panel ) . we have used a flat prior for @xmath54 and marginalized over it.,title="fig : " ] under the restriction @xmath61 using @xmath10 ( upper left panel ) , cmb ( upper central panel ) , snia ( upper right panel ) , bao ( lower left panel ) and the combination of these four tests ( lower right panel ) . we have used a flat prior for @xmath54 and marginalized over it.,title="fig : " ] under the restriction @xmath61 using @xmath10 ( upper left panel ) , cmb ( upper central panel ) , snia ( upper right panel ) , bao ( lower left panel ) and the combination of these four tests ( lower right panel ) . we have used a flat prior for @xmath54 and marginalized over it.,title="fig : " ] + under the restriction @xmath61 using @xmath10 ( upper left panel ) , cmb ( upper central panel ) , snia ( upper right panel ) , bao ( lower left panel ) and the combination of these four tests ( lower right panel ) . we have used a flat prior for @xmath54 and marginalized over it.,title="fig : " ] under the restriction @xmath61 using @xmath10 ( upper left panel ) , cmb ( upper central panel ) , snia ( upper right panel ) , bao ( lower left panel ) and the combination of these four tests ( lower right panel ) . we have used a flat prior for @xmath54 and marginalized over it.,title="fig : " ] under the restriction @xmath61 using @xmath10 ( upper left panel ) , cmb ( upper central panel ) , snia ( upper right panel ) , bao ( lower left panel ) and the combination of these four tests ( lower central panel ) . in the lower right panel , we present the contour plots ( 1 , 2 , 3 @xmath3 ) for the total pdf in the ( @xmath1 , @xmath62 ) plane . we have used a flat prior for @xmath54 and marginalized over it.,title="fig : " ] under the restriction @xmath61 using @xmath10 ( upper left panel ) , cmb ( upper central panel ) , snia ( upper right panel ) , bao ( lower left panel ) and the combination of these four tests ( lower central panel ) . in the lower right panel , we present the contour plots ( 1 , 2 , 3 @xmath3 ) for the total pdf in the ( @xmath1 , @xmath62 ) plane . we have used a flat prior for @xmath54 and marginalized over it.,title="fig : " ] under the restriction @xmath61 using @xmath10 ( upper left panel ) , cmb ( upper central panel ) , snia ( upper right panel ) , bao ( lower left panel ) and the combination of these four tests ( lower central panel ) . in the lower right panel , we present the contour plots ( 1 , 2 , 3 @xmath3 ) for the total pdf in the ( @xmath1 , @xmath62 ) plane . we have used a flat prior for @xmath54 and marginalized over it.,title="fig : " ] + under the restriction @xmath61 using @xmath10 ( upper left panel ) , cmb ( upper central panel ) , snia ( upper right panel ) , bao ( lower left panel ) and the combination of these four tests ( lower central panel ) . in the lower right panel , we present the contour plots ( 1 , 2 , 3 @xmath3 ) for the total pdf in the ( @xmath1 , @xmath62 ) plane . we have used a flat prior for @xmath54 and marginalized over it.,title="fig : " ] under the restriction @xmath61 using @xmath10 ( upper left panel ) , cmb ( upper central panel ) , snia ( upper right panel ) , bao ( lower left panel ) and the combination of these four tests ( lower central panel ) . in the lower right panel , we present the contour plots ( 1 , 2 , 3 @xmath3 ) for the total pdf in the ( @xmath1 , @xmath62 ) plane . we have used a flat prior for @xmath54 and marginalized over it.,title="fig : " ] under the restriction @xmath61 using @xmath10 ( upper left panel ) , cmb ( upper central panel ) , snia ( upper right panel ) , bao ( lower left panel ) and the combination of these four tests ( lower central panel ) . in the lower right panel , we present the contour plots ( 1 , 2 , 3 @xmath3 ) for the total pdf in the ( @xmath1 , @xmath62 ) plane . we have used a flat prior for @xmath54 and marginalized over it.,title="fig : " ] , with no restriction , using @xmath10 ( upper left panel ) , cmb ( upper central panel ) , snia ( upper right panel ) , bao ( lower left panel ) and the combination of these four tests ( lower right panel ) . we have used a flat prior for @xmath54 and marginalized over it.,title="fig : " ] , with no restriction , using @xmath10 ( upper left panel ) , cmb ( upper central panel ) , snia ( upper right panel ) , bao ( lower left panel ) and the combination of these four tests ( lower right panel ) . we have used a flat prior for @xmath54 and marginalized over it.,title="fig : " ] , with no restriction , using @xmath10 ( upper left panel ) , cmb ( upper central panel ) , snia ( upper right panel ) , bao ( lower left panel ) and the combination of these four tests ( lower right panel ) . we have used a flat prior for @xmath54 and marginalized over it.,title="fig : " ] + , with no restriction , using @xmath10 ( upper left panel ) , cmb ( upper central panel ) , snia ( upper right panel ) , bao ( lower left panel ) and the combination of these four tests ( lower right panel ) . we have used a flat prior for @xmath54 and marginalized over it.,title="fig : " ] , with no restriction , using @xmath10 ( upper left panel ) , cmb ( upper central panel ) , snia ( upper right panel ) , bao ( lower left panel ) and the combination of these four tests ( lower right panel ) . we have used a flat prior for @xmath54 and marginalized over it.,title="fig : " ] , with no restriction on @xmath1 , using @xmath10 ( upper left panel ) , cmb ( upper central panel ) , snia ( upper right panel ) , bao ( lower left panel ) and the combination of these four tests ( lower central panel ) . in the lower right panel , we present the contour plots ( 1 , 2 , 3 @xmath3 ) for the total pdf in the ( @xmath1 , @xmath62 ) . we have used a flat prior for @xmath54 and marginalized over it.,title="fig : " ] , with no restriction on @xmath1 , using @xmath10 ( upper left panel ) , cmb ( upper central panel ) , snia ( upper right panel ) , bao ( lower left panel ) and the combination of these four tests ( lower central panel ) . in the lower right panel , we present the contour plots ( 1 , 2 , 3 @xmath3 ) for the total pdf in the ( @xmath1 , @xmath62 ) . we have used a flat prior for @xmath54 and marginalized over it.,title="fig : " ] , with no restriction on @xmath1 , using @xmath10 ( upper left panel ) , cmb ( upper central panel ) , snia ( upper right panel ) , bao ( lower left panel ) and the combination of these four tests ( lower central panel ) . in the lower right panel , we present the contour plots ( 1 , 2 , 3 @xmath3 ) for the total pdf in the ( @xmath1 , @xmath62 ) . we have used a flat prior for @xmath54 and marginalized over it.,title="fig : " ] + , with no restriction on @xmath1 , using @xmath10 ( upper left panel ) , cmb ( upper central panel ) , snia ( upper right panel ) , bao ( lower left panel ) and the combination of these four tests ( lower central panel ) . in the lower right panel , we present the contour plots ( 1 , 2 , 3 @xmath3 ) for the total pdf in the ( @xmath1 , @xmath62 ) . we have used a flat prior for @xmath54 and marginalized over it.,title="fig : " ] , with no restriction on @xmath1 , using @xmath10 ( upper left panel ) , cmb ( upper central panel ) , snia ( upper right panel ) , bao ( lower left panel ) and the combination of these four tests ( lower central panel ) . in the lower right panel , we present the contour plots ( 1 , 2 , 3 @xmath3 ) for the total pdf in the ( @xmath1 , @xmath62 ) . we have used a flat prior for @xmath54 and marginalized over it.,title="fig : " ] , with no restriction on @xmath1 , using @xmath10 ( upper left panel ) , cmb ( upper central panel ) , snia ( upper right panel ) , bao ( lower left panel ) and the combination of these four tests ( lower central panel ) . in the lower right panel , we present the contour plots ( 1 , 2 , 3 @xmath3 ) for the total pdf in the ( @xmath1 , @xmath62 ) . we have used a flat prior for @xmath54 and marginalized over it.,title="fig : " ] some ideas on how to avoid a possible `` tension '' and then saving the chaplygin gas appeared recently and concern a modified theory of gravity : rastall s theory @xcite . in the latter the einstein - hilbert action still holds and the modification resides in the conservation law of the matter stress - energy momentum , i.e. @xmath73 where @xmath74 is a parameter and @xmath75 is the ricci scalar curvature . the above modification can be interpreted in various ways . to us the most interesting and significant one is that rastall s idea may be viewed as a kind of semi - classical formulation of quantum phenomena , which we expect to appear when the curvature ( which enters as @xmath75 on the right hand side of eq . ) becomes important . of course , one could choose other scalars rather than @xmath75 in order to represent the riemann tensor , but perhaps @xmath75 is the most natural choice . since eq . must fit into bianchi identities , one can find the following modified einstein equations @xmath76 then , it is clear that , when assuming a flrw background , friedmann equation shall be modified with terms proportional to @xmath77 . we choose , for our discussion in the present section , a single - fluid component with density @xmath78 and pressure @xmath79 . in rastall s theory , friedmann equation becomes @xmath80\;,\ ] ] where note that , differently from gr , now the pressure contributes to the expansion ( via the equation of state parameter @xmath81 ) . moreover , the continuity equation becomes : @xmath82 or , trading the cosmic time for the scale factor , and rearranging the derivatives : @xmath83 clearly , since the energy conservation equation has changed , we can not use again the gcg equation of state and hope to obtain again the same result as in eq . . what we can do is to take advantage of the foregoing analysis and _ assume _ a form for @xmath84 given by eq . . then , we can derive the equation of state and speed of sound of the corresponding fluid in rastall s theory and investigate its stability properties . using @xmath85 , it is easy , by simple inspection of eqs . and to find @xmath86 which is a relation which holds true also for gr . solving eq . for @xmath79 and substituting in the above equation , one finds @xmath87 and for the pressure @xmath88 now , as said , we assume @xmath89^\frac{1}{1 + \alpha}\;,\ ] ] and then it is straightforward to compute density and pressure . in particular , the equation of state parameter and the speed of sound take the form @xmath90-(3 \alpha + 1 ) a^{3 \alpha + 3}+1\right\}+\gamma -1}{\bar{a}\left\{3 \gamma \left[(\alpha + 1 ) a^{3 \alpha + 3}-1\right]-(3 \alpha + 5 ) a^{3 \alpha + 3}+5\right\}+3 \gamma -5}\;,\end{aligned}\ ] ] where we have defined the speed of sound as @xmath91 , i.e. as if it were adiabatic . we can formally invert the function @xmath92 and then obtain a @xmath93 , i.e. a barotropic equation of state . we plot in the above @xmath81 and @xmath35 for the choice @xmath94 and @xmath95 ( the best fit values found in this paper ) and for @xmath77 varying about unity . and of the speed of sound @xmath35 given in eqs . and as functions of the scale factor the parameter @xmath77 has been chosen @xmath96 ( black solid lines ) , @xmath97 ( i.e. the gr limit , red dashed lines ) and @xmath98 ( blue dash - dotted line).,title="fig : " ] and of the speed of sound @xmath35 given in eqs . and as functions of the scale factor . the parameter @xmath77 has been chosen @xmath96 ( black solid lines ) , @xmath97 ( i.e. the gr limit , red dashed lines ) and @xmath98 ( blue dash - dotted line).,title="fig : " ] + in the range chosen for @xmath77 , the speed of sound is monotonically decreasing and , by inspection , positive if @xmath99 . let us investigate briefly the asymptotic : in the far past , for @xmath15 , we have @xmath100 and for @xmath97 we reproduce the known gr result for the gcg . now , in the remote future , for @xmath101 : @xmath102 where again we reproduce correctly the limit @xmath103 when @xmath97 . by construction , the final stage of the evolution is a de sitter one . of course , in all the above calculations we have assumed that @xmath104 . asking for the asymptotic speed of sound of eq . to be positive , we find the following ranges for @xmath77 : @xmath105 for @xmath106 , this amounts to @xmath107 and @xmath108 . we now check the evolution of perturbations . let us start by considering the energy - momentum tensor @xmath109 where @xmath110 . more explicitly , the background components of ( [ emt ] ) are @xmath111 we assume a conformal newtonian gauge line element for scalar perturbations @xmath112\;,\end{aligned}\ ] ] where we introduced the conformal time @xmath113 . the perturbations of the fluid 4-velocity up to first order are given by @xmath114 in the absence of anisotropic stresses the spatial off - diagonal einstein equation implies @xmath115 . assuming the above perturbed metric , we can find the einstein equation for the potential @xmath116 for the rastall s theory , namely @xmath117\;,\\ \label{eij } \phi '' + 3\frac{a'}{a}\phi ' + \left[2\left(\frac{a'}{a}\right ) ' + \left(\frac{a'}{a}\right)^2\right]\phi = 4\pi g a^2\left[\delta p + \frac{\gamma - 1}{2}\left(\delta\rho + \delta\rho - 3\delta p\right)\right]\;,\end{aligned}\ ] ] where the prime denotes derivation wrt the conformal time . introducing the speed of sound ( assuming adiabaticity , i.e. @xmath118 ) : @xmath119\phi = 4\pi g a^2\delta\rho\left(\frac{\gamma - 1}{2 } + \frac{5 - 3\gamma}{2}c_s^2\right)\;.\end{aligned}\ ] ] changing the conformal time for the scale factor and combining the two above equations , we find an equation for the gravitational potential : @xmath120 where the subscript @xmath121 denotes derivation wrt the scale factor , @xmath122 and we have introduced a plane - wave expansion . note the `` rastall factor '' of eq . : @xmath123 since it multiplies the wavenumber @xmath124 , it may be considered as an `` effective '' speed of sound , different from the adiabatic @xmath35 we introduced . in this sense , rastall s theory seems to call into play a sort of `` geometric entropy '' . using eq . in order to reduce the `` rastall factor '' one finds @xmath125 which is exactly the gcg speed of sound in gr ! rastall s theory seems to be able to reproduce the same evolution of perturbations of a fluid in gr which provides a given background expansion . therefore , we may conclude that even framing the gcg in rastall s theory does not save the model from being ruled out due to the behaviour of small perturbations . just to check . we choose as initial conditions @xmath126 and @xmath127 in @xmath128 and let the system evolve with the choice @xmath94 and @xmath129 . in we plot the evolution of @xmath116 and @xmath130 as function of the scale factor for a representative scale @xmath131 h mpc@xmath30 . and @xmath130 as functions of the scale factor . the gcg parameters have been fixed as @xmath132 and @xmath133 . the parameter @xmath77 has been chosen @xmath134 ( black solid lines ) , @xmath97 ( i.e. the gr limit , red dashed lines ) and @xmath135 ( blue dash - dotted line ) . for comparison , the @xmath6cdm lines are depicted as the magenta dotted ones.,title="fig : " ] and @xmath130 as functions of the scale factor . the gcg parameters have been fixed as @xmath132 and @xmath133 . the parameter @xmath77 has been chosen @xmath134 ( black solid lines ) , @xmath97 ( i.e. the gr limit , red dashed lines ) and @xmath135 ( blue dash - dotted line ) . for comparison , the @xmath6cdm lines are depicted as the magenta dotted ones.,title="fig : " ] + clearly , the results are catastrophic . a remaining possibility of salvation is to adopt a scalar field description for the gcg , as done in @xcite . as shown in @xcite ( where the authors do not mention rastall s theory and investigate just the @xmath135 case ) and @xcite in a scalar field approach the effective speed of sound in the rest frame of the field ( i.e. where @xmath136 ) can be written as @xmath137 which evidently vanishes for @xmath135 . since the scalar field description possesses one degree of freedom more than the fluid description ( encoded in the scalar field potential ) the background expansion can be fixed without any drawbacks on the evolution of perturbations , which for @xmath135 may be identical to the one of a pressureless fluid in gr ( e.g. cdm ) . the results of our background tests do not impose negative values for @xmath1 with a high confidence level . this could mean that the gcg has indeed @xmath138 meaning that the dark sector of the universe is actually an ordinary adiabatic fluid with a small , constant and negative pressure . hence , we can argue that at high redshifts the gcg does not look exactly like standard cdm ( @xmath139 ) . our goal in this section is to revisit the issue concerning the equivalence of gcg with @xmath138 and the @xmath6cdm model . obviously both cases produces the same background expansion . however , does this equivalence holds at first order in the perturbations ? we present an analysis of the mszros effect which describes the formation of dark matter structures during the first stages of the matter dominated epoch . we assume a single fluid description , i.e. we neglect the contribution of baryons and radiation . this is a reasonable approximation if we want to track the growth of sub - horizon perturbations during the matter dominated epoch . for the technical details , we follow the analysis of @xcite . for the linear perturbations of ( [ emt ] ) we define the velocity scalar @xmath140 , which is associated with the peculiar velocity by @xmath141 . at first order , the ( 0 - 0 ) component of the einstein equation reads @xmath142 where @xmath143 is the _ total _ density contrast , i.e. @xmath144 . during the time interval between the kinetic decoupling of dm particles from the primordial plasma and the epoch of matter radiation equality ( @xmath145 for the @xmath6cdm model ) the sub - horizon dm perturbations grow only logarithmically with the scale factor . after @xmath146 the dm perturbations obey to @xmath147 . this is the main result behind the so called mszros effect @xcite . we study here an example in which the gcg with @xmath5 behaves differently from the @xmath6cdm . we show that the growth of sub - horizon gcg matter perturbations is different from the standard cdm . in order to obtain a mszros - like equation for the gcg we make use of the covariant conservation of the energy - momentum tensor ( @xmath148 ) . at first - order we find @xmath149 and @xmath150 for the energy and momentum balances of each single component ( which we assume conserving separately ) , respectively . for small scales we can neglect the term @xmath151 in eq . ( [ cons1 ] ) and we also take the sub - horizon limit of the poisson equation eq . ( [ poisson ] ) . together with eq . ( [ cons2 ] ) and with the fact that @xmath152 , we find a mszros - like equation for the gcg with @xmath138 , i.e. @xmath153 : @xmath154a\frac{\,d\delta_{gcg}}{da}+ \nonumber\\ \left[-\frac{3}{2}-\frac{15}{2}w_c+9w_c^2-\frac{3aw_c}{h}\frac{dh}{da}-3a\frac{dw_c}{da}\right]\delta_{gcg}=0\;. \label{small}\end{aligned}\ ] ] the standard equation for cdm in the @xmath6cdm model can be obtained in the same way , but remembering that now @xmath155 since the @xmath6cdm model is basically a two - fluid model . therefore , we have @xmath156a\frac{\,d\delta_{cdm}}{da}-\frac{3}{2}\omega_{cdm}\delta_{cdm}=0\;. \label{smallcdm}\ ] ] identifying the background expansion of the two instances , @xmath6cdm and gcg , one can write @xmath157 and thus @xmath156a\frac{\,d\delta_{cdm}}{da}-\frac{3}{2}(1 + w_c)\delta_{cdm}=0\;. \label{smallcdm2}\ ] ] equations and are clearly different ( note that @xmath84 and @xmath158 have the same evolution ) , therefore @xmath159 and @xmath160 evolve differently . indeed , one can show that @xmath161 . this is also a result of @xcite , see also @xcite . we show in the evolution of a typical sub - horizon scale after the equality ( @xmath162 ) . we assume that both cdm and gcg have the same initial conditions at that time . we set them by generating the power spectrum at that time for the scale @xmath163 h mpc@xmath30 with help of the camb code @xcite . we then solve numerically eq . [ small ] for @xmath164 and @xmath165 . the perturbations in the gcg fluid are strongly suppressed in comparison to the cdm case . this result can also be appreciated in fig . 1 of @xcite . the main conclusion of @xcite is that the @xmath6cdm model and the @xmath5 gcg are indistinguishable . we agree on this at the linear perturbations level since , being the speed of sound vanishing , the evolution of the gravitational potential @xmath166 is determined univocally and it is the same in the two models ( it depends on the background evolution only , which is identical in the two models , by construction ) . this imply that the evolution of @xmath143 is also the same and therefore , if the components conserve independently , the evolution of the density contrasts of baryons and radiation are also respectively identical in the two models . since the latter two are the only observable components , one may conclude that the @xmath6cdm model and the @xmath5 gcg are indistinguishable , from gravity only . no difference can be expected for the e.g. integrated sachs - wolfe effect or for the baryonic large scale structure power spectrum ( which is indeed the one we infer from observation ) . on the other hand , the density contrast @xmath159 seems unable to reach the non - linear regime @xmath167 , see . we wonder if this would not have a dramatic effect on the halo formation . the absence of the latter would be in contrast with the observation of the almost flat velocity curves of galaxies and with lensing experiments . probably , only numerical simulations would provide the correct matter distribution , but semi - analytical studies touching the non - linear regime @xcite could supply additional information about the final fate of the @xmath5 gcg . . the standard pressureless cdm ( @xmath168 ) is shown as the solid line . for the gcg with @xmath138 we show the evolution of @xmath130 for three different values of @xmath62 . left panel : @xmath169 . right panel : @xmath170.,title="fig : " ] . the standard pressureless cdm ( @xmath168 ) is shown as the solid line . for the gcg with @xmath138 we show the evolution of @xmath130 for three different values of @xmath62 . left panel : @xmath169 . right panel : @xmath170.,title="fig : " ] we performed a bayesian analysis of the background behaviour of the gcg model using @xmath10 , snia , cmb and bao datasets . we focused particularly on the parameter @xmath1 , on which a huge literature already exists . our result is that @xmath11 , at the 2@xmath3 level , i.e. negative values of @xmath1 seem to be favoured over the positive ones . indeed @xmath1 is negative with 85% of confidence . the uncertainty is still too large for us to claim that a `` tension '' with perturbative tests ( which constrain @xmath171 at 2@xmath3 ) does exist , but we speculate on the consequences of this occurrence , which is not far from being settled , given the ongoing and forthcoming observational programs which collect more precise data day after day . that kind of inconsistency would immediately rule out the gcg model , since it is unacceptable to obtain two different set of parameters depending on the observational test applied . we assume this to occur and figure out how to possibly save the gcg unification paradigm . we introduce a fluid model in rastall s theory which exactly behaves as the gcg at the background level ( i.e. it provides the same evolution for @xmath10 ) and discuss its perturbative properties . it turns out that small and positive speeds of sound are compatible with @xmath4 . on the other hand , we show that the evolution of perturbations is ruled by an effective speed of sound which is identical to the gcg one in gr , as if rastall s theory introduced a sort of `` geometric entropy '' . therefore , as one should expect , the results are catastrophic , with the density contrast and the gravitational potential growing too fast ( for @xmath4 ) for being in agreement with observation . as a possible salvation , we indicate a scalar field description of the gcg in rastall s theory , where the background expansion and the effective speed of sound can be fixed independently ( and the latter to zero ) . we also address the issue of the @xmath15 limit . it is indistinguishable from the @xmath6cdm model at the linear perturbative regime , but we show that the small negative pressure of the gcg affects the evolution of the gcg density contrast during the matter dominated phase , where it is expected to behave as cdm and provides a different evolution for the sub - horizon matter perturbations . this analysis shows that sub - horizon structures ( e.g. dark halos ) may not form as in the standard @xmath6cdm case . thus , it could be very interesting to perform numerical simulations for matter fluids with a hydrodynamical evolution distinct from the standard pressureless case . this analysis could clarify if the @xmath15 limit of the gcg is really identical to the @xmath6cdm model . this work was supported by cnpq ( brazil ) . hv acknowledges support from the dfg within the research training group 1620 `` models of gravity '' . we would like to thank alejandro aviles and pedro avelino for enlightening remarks and suggestions . 99 a. g. riess _ et al . _ [ supernova search team collaboration ] , `` observational evidence from supernovae for an accelerating universe and a cosmological constant , '' astron . j. * 116 * 1009 ( 1998 ) s. perlmutter _ et al . _ [ supernova cosmology project collaboration ] , `` measurements of omega and lambda from 42 high redshift supernovae , '' astrophys . j. * 517 * 565 ( 1999 ) e. komatsu _ et al . _ [ wmap collaboration ] , `` seven - year wilkinson microwave anisotropy probe ( wmap ) observations : cosmological interpretation , '' astrophys . j. suppl . * 192 * 18 ( 2011 ) r. r. caldwell and m. kamionkowski , `` the physics of cosmic acceleration , '' ann . nucl . part . * 59 * 397 ( 2009 ) g. bertone , d. hooper and j. silk , `` particle dark matter : evidence , candidates and constraints , '' phys . * 405 * 279 ( 2005 ) t. padmanabhan , `` cosmological constant : the weight of the vacuum , '' phys . rept . * 380 * 235 ( 2003 ) a. y. .kamenshchik , u. moschella and v. pasquier , `` an alternative to quintessence , '' phys . b * 511 * 265 ( 2001 ) r. jackiw , `` a particle field theorist s lectures on supersymmetric , nonabelian fluid mechanics and d - branes , '' physics/0010042 . m. c. bento , o. bertolami and a. a. sen , `` generalized chaplygin gas , accelerated expansion and dark energy matter unification , '' phys . rev . d * 66 * 043507 ( 2002 ) n. bilic , g. b. tupper and r. d. viollier , `` unification of dark matter and dark energy : the inhomogeneous chaplygin gas , '' phys . b * 535 * 17 ( 2002 ) h. sandvik , m. tegmark , m. zaldarriaga and i. waga , `` the end of unified dark matter ? , '' phys . d * 69 * 123524 ( 2004 ) l. amendola , f. finelli , c. burigana and d. carturan , `` wmap and the generalized chaplygin gas , '' jcap * 0307 * 005 ( 2003 ) t. barreiro , o. bertolami and p. torres , `` wmap5 constraints on the unified model of dark energy and dark matter , '' phys . d * 78 * 043530 ( 2008 ) v. gorini , a. y. kamenshchik , u. moschella , o. f. piattella and a. a. starobinsky , `` gauge - invariant analysis of perturbations in chaplygin gas unified models of dark matter and dark energy , '' jcap * 0802 * 016 ( 2008 ) o. f. piattella , `` the extreme limit of the generalized chaplygin gas , '' jcap * 1003 * 012 ( 2010 ) j. c. fabris , s. v. b. gonalves , h. e. s. velten and w. zimdahl , `` matter power spectrum for the generalized chaplygin gas model : the newtonian approach , '' phys . d * 78 * 103523 ( 2008 ) j. c. fabris , h. e. s. velten and w. zimdahl , `` matter power spectrum for the generalized chaplygin gas model : the relativistic case , '' phys . d * 81 * 087303 ( 2010 ) r. colistete , jr . and j. c. fabris , `` bayesian analysis of the ( generalized ) chaplygin gas and cosmological constant models using the 157 gold sne ia data , '' class . * 22 * 2813 ( 2005 ) j. c. fabris , p. l. c. de oliveira and h. e. s. velten , `` contraints on unified models for dark matter and dark energy using h(z ) , '' eur . j. c * 71 * 1773 ( 2011 ) r. r. r. reis , i. waga , m. o. calvao and s. e. joras , `` entropy perturbations in quartessence chaplygin models , '' phys . d * 68 * 061302 ( 2003 ) l. amendola , i. waga and f. finelli , `` observational constraints on silent quartessence , '' jcap * 0511 * 009 ( 2005 ) j. c. fabris , t. c. c. guio , m. hamani daouda and o. f. piattella , `` scalar models for the generalized chaplygin gas and the structure formation constraints , '' grav . cosmol . * 17 * 259 ( 2011 ) b. l. lago , m. o. calvao , s. e. joras , r. r. r. reis , i. waga and r. giostri , `` type ia supernova parameter estimation : a comparison of two approaches using current datasets , '' arxiv:1104.2874 [ astro-ph.co ] . p. rastall , `` generalization of the einstein theory , '' phys . d * 6 * 3357 ( 1972 ) c. e. m. batista , m. h. daouda , j. c. fabris , o. f. piattella and d. c. rodrigues , `` rastall cosmology and the @xmath6 cdm model , '' phys . d * 85 * 084008 ( 2012 ) j. c. fabris , m. h. daouda and o. f. piattella , `` note on the evolution of the gravitational potential in rastall scalar field theories , '' phys . b * 711 * 232 ( 2012 ) m. h. daouda , j. c. fabris and o. f. piattella , `` scalar models for the unification of the dark sector , '' aip conf . * 1471 * 57 ( 2012 ) j. c. fabris , o. f. piattella , d. c. rodrigues , c. e. m. batista and m. h. daouda , `` rastall cosmology , '' int . j. mod . * 18 * 67 ( 2012 ) t. p. sotiriou and v. faraoni , `` f(r ) theories of gravity , '' rev . * 82 * 451 ( 2010 ) p. horava , `` quantum gravity at a lifshitz point , '' phys . rev . d * 79 * 084008 ( 2009 ) a. ali , s. dutta , e. n. saridakis and a. a. sen , `` horava - lifshitz cosmology with generalized chaplygin gas , '' gen . grav . * 44 * 657 ( 2012 ) c. -g . park , j. -c . hwang , j. park and h. noh , `` observational constraints on a unified dark matter and dark energy model based on generalized chaplygin gas , '' phys . d * 81 * 063532 ( 2010 ) j. c. fabris , s. v. b. goncalves and r. de sa ribeiro , `` generalized chaplygin gas with @xmath5 and the @xmath6cdm cosmological model , '' gen . grav . * 36 * 211 ( 2004 ) p. p. avelino , l. m. g. beca , j. p. m. de carvalho and c. j. a. p. martins , `` the @xmath6cdm limit of the generalized chaplygin gas scenario , '' jcap * 0309 * 002 ( 2003 ) j. simon , l. verde and r. jimenez , `` constraints on the redshift dependence of the dark energy potential , '' phys . rev . d * 71 * 123001 ( 2005 ) d. stern , r. jimenez , l. verde , m. kamionkowski and s. a. stanford , `` cosmic chronometers : constraining the equation of state of dark energy . i : h(z ) measurements , '' jcap * 1002 * 008 ( 2010 ) r. jimenez , l. verde , t. treu and d. stern , `` constraints on the equation of state of dark energy and the hubble constant from stellar ages and the cmb , '' astrophys . j. * 593 * 622 ( 2003 ) t. -j . zhang and c. ma , `` constraints on the dark side of the universe and observational hubble parameter data , '' adv . astron . * 2010 * 184284 ( 2010 ) c. ma and t. -j . zhang , `` power of observational hubble parameter data : a figure of merit exploration , '' astrophys . j. * 730 * 74 ( 2011 ) m. moresco , a. cimatti , r. jimenez , l. pozzetti , g. zamorani , m. bolzonella , j. dunlop and f. lamareille _ et al . _ , `` improved constraints on the expansion rate of the universe up to @xmath172 from the spectroscopic evolution of cosmic chronometers , '' jcap * 1208 * 006 ( 2012 ) o. farooq , d. mania and b. ratra , `` hubble parameter measurement constraints on dark energy , '' arxiv:1211.4253 [ astro-ph.co ] . r. amanullah , c. lidman , d. rubin , g. aldering , p. astier , k. barbary , m. s. burns and a. conley _ et al . _ , `` spectra and light curves of six type ia supernovae at @xmath173 and the union2 compilation , '' astrophys . j. * 716 * 712 ( 2010 ) m. doran , m. j. lilley , j. schwindt and c. wetterich , `` quintessence and the separation of cmb peaks , '' astrophys . j. * 559 * 501 ( 2001 ) w. hu , m. fukugita , m. zaldarriaga and m. tegmark , `` cmb observables and their cosmological implications , '' astrophys . j. * 549 * 669 ( 2001 ) m. doran , m. lilley and c. wetterich , `` constraining quintessence with the new cmb data , '' phys . b * 528 * 175 ( 2002 ) e. d. reese , j. e. carlstrom , m. joy , j. j. mohr , l. grego and w. l. holzapfel , `` determining the cosmic distance scale from interferometric measurements of the sunyaev - zeldovich effect , '' astrophys . j. * 581 * 53 ( 2002 ) j. lu , l. xu , y. wu and m. liu , `` combined constraints on modified chaplygin gas model from cosmological observed data : markov chain monte carlo approach , '' gen . rel . grav . * 43 * 819 ( 2011 ) d. j. eisenstein _ et al . _ [ sdss collaboration ] , `` detection of the baryon acoustic peak in the large - scale correlation function of sdss luminous red galaxies , '' astrophys . j. * 633 * 560 ( 2005 ) c. blake et al , `` the wigglez dark energy survey : mapping the distance - redshift relation with baryon acoustic oscillations '' mnras , * 418 * , 3 , 1707 ( 2011 ) h. velten and d. j. schwarz , `` constraints on dissipative unified dark matter , '' jcap * 1109 * 016 ( 2011 ) r. a. a. fernandes , j. p. m. de carvalho , a. y. .kamenshchik , u. moschella and a. da silva , `` spherical ` top - hat ' collapse in general chaplygin gas dominated universes , '' phys . rev . d * 85 * 083501 ( 2012 ) p. meszaros , `` the behaviour of point masses in an expanding cosmological substratum , '' astron . astrophys . * 37 * 225 ( 1974 ) c. gao , m. kunz , a. r. liddle and d. parkinson , `` unified dark energy and dark matter from a scalar field different from quintessence , '' phys . rev . d * 81 * 043520 ( 2010 ) p. p. avelino , l. m. g. beca and c. j. a. p. martins , `` linear and nonlinear instabilities in unified dark energy models , '' phys . d * 77 * 063515 ( 2008 ) p. p. avelino , l. m. g. beca and c. j. a. p. martins , `` clustering properties of dynamical dark energy models , '' phys . d * 77 * 101302 ( 2008 ) a. aviles and j. l. cervantes - cota , `` the dark degeneracy and interacting cosmic components , '' phys . d * 84 * 083515 ( 2011 ) [ erratum - ibid . d * 84 * 089905 ( 2011 ) ] a. lewis , a. challinor and a. lasenby , `` efficient computation of cmb anisotropies in closed frw models , '' astrophys . j. * 538 * 473 ( 2000 )

in this paper , an extensive survey of the star forming complex sh2 - 252 has been undertaken with an aim to explore its hidden young stellar population as well as to understand the structure and star formation history for the first time . this complex is composed of five prominent embedded clusters associated with the sub - regions a , c , e , ngc 2175s and teu 136 . we used 2mass - nir and _ spitzer_-irac , mips photometry to identify and classify the young stellar objects ( ysos ) by their infra - red ( ir ) excess emission . using the ir colour - colour criteria , we identified 577 ysos , of which , 163 are class i , 400 are class ii and 14 are transition disk ysos , suggesting a moderately rich number of ysos in this complex . spatial distribution of the candidate ysos shows that they are mostly clustered around the sub - regions in the western half of the complex , suggesting enhanced star formation activity towards its west . using the spectral energy distribution ( sed ) and optical colour - magnitude diagram ( cmd ) based age analyses , we derived probable evolutionary status of the sub - regions of sh2 - 252 . our analysis shows that the region a is the youngest ( 0.5 myr ) , the regions b , c and e are of similar evolutionary stage ( 1 - 2 myr ) and the clusters ngc 2175s and teu 136 are slightly evolved ( 2 - 3 myr ) . morphology of the region in the 1.1 mm map shows a semi - circular shaped molecular shell composed of several clumps and ysos bordering the western ionization front ( if ) of sh2 - 252 . our analyses suggest that next generation star formation is currently under way along this border and that possibly fragmentation of the matter collected during the expansion of the region as one of the major processes responsible for such stars . we observed the densest concentration of ysos ( mostly class i , 0.5 myr ) at the western outskirts of the complex , within a molecular clump associated with water and methanol masers and we suggest that it is indeed a site of cluster formation at a very early evolutionary stage , sandwiched between the two relatively evolved cregions a and b. [ firstpage ] stars : formation stars : premainsequence infrared : ism regions - ism : individual objects : sh2 - 252

1304.5327

in order to understand the interstellar processes which govern star formation , studies of the large scale properties of various star forming regions at different environments are very essential . census of young stellar objects ( ysos ) born in a cloud , their mass , age and spatial distribution , together with mass of the star forming cloud are some of the key data used for the assessment of the time scale and star formation history of a given region . however , identifying the member stars of a given young association is quite difficult . it is well known that ysos show excess emission in infra - red ( ir ) above main sequence ( ms ) photosphere due to thermal emission from their circumstellar material . thus , ysos can be identified by looking for ir excess emission . ysos are often categorized into class 0 , i , ii or iii evolutionary stages @xcite . class 0 objects are deeply embedded protostars that are still experiencing cloud collapse . they are extremely faint at wavelengths shorter than 10 @xmath2 m and have a significant sub - millimeter luminosity . a class i yso is also an object whose emission is dominated by a dense infalling spherical envelope , however these objects are bright in the ir . a class ii yso is characterized by the presence of an optically thick , primordial circumstellar disk , which dominates the star s emission . when the circumstellar disk material becomes optically thin , the star is classified as a class iii star . hence , ir survey of a star forming region can be used as a powerful tool to distinguish the stars with ir excesses from stars without such excesses . the unprecedented sensitivity and mapping capabilities of the _ spitzer _ space telescope provides an excellent platform to survey the star forming regions in the mid - ir and to identify the ir excess sources in the region . furthermore , _ spitzer _ data can also be used to classify ysos at different evolutionary stages , which helps us probing the recent star formation activity of a given region ( e.g. , @xcite ) . massive stars have a profound effect on their natal environment creating wind - blown shells , cavities and hii regions . the immense amount of energy released through their stellar winds and radiation disperse and destroy the remaining molecular gas and likely inhibit further star formation . however , it has also been argued that in some circumstances , the energy input by these massive stars can promote and induce subsequent low mass star formation in the surrounding molecular gas before it disperses ( e.g. , koenig et al . therefore , regions that contain massive stars provide excellent laboratories for studying both high and low mass star formation . identification and characterization of the ysos in star forming complexes hosting massive stars are essential steps to examine the physical processes that govern new generation star formation in such complexes . here we present the results of our multi - wavelength analyses of the star forming complex sh2 - 252 ( @xcite ; @xmath3 = @xmath4 ; @xmath5 = @xmath6 ; [email protected] ; [email protected] ) . sh2 - 252 is an evolved region powered by the central massive ionizing source of o6.5v type , hd 42088 , which is a member of the gemini ob1 association . the dss2-r band image of the region around sh2 - 252 for an area of @xmath0 1.3 @xmath8 1.3 deg@xmath9 is shown in fig . radio observations at 5 ghz detected six extended sources ( sh2 - 252 a to f ) at different spatial locations towards the region , of which , four of them ( a , b , c and e ) are classified as compact ( c ) regions ( @xcite , @xcite ) . the stellar contents of this region were not so well explored until recently . @xcite ( hereafter paper 1 ) unraveled its stellar contents using deep optical , shallow near - infrared ( nir ; 2mass ) , @xmath10 survey along with spectroscopic observations . they estimated the distance ( 2.4 kpc ) , reddening and identified 12 ob stars of spectral type earlier than b8 , 61 @xmath10 emission line sources within the nebula and also 5 prominent embedded clusters associated with the sub - regions a , c , e , ngc 2175s and teu 136 . the age and age spread , initial mass function ( imf ) , @xmath11-band luminosity function ( klf ) etc . of the sub - regions have also been discussed in paper 1 . the main sub - regions of sh2 - 252 such as a , b , c , e , f , ngc 2175s and teu 136 are marked in fig . sh2 - 252 is bounded by an ionization front ( if ) on the west side , while it is density bounded on the east side @xcite . the detailed @xmath12co and @xmath13co maps of the region by @xcite showed that the cloud complex is separated into two fragments by a long rift of little co emission and named them as western cloud fragment ( wcf ) and eastern cloud fragment ( ecf ) , respectively , and are marked in fig . most of the mass of wcf is concentrated in a narrow ridge which borders the if of sh2 - 252 at its west ( @xcite ; @xcite ) . the similarity in the kinematics of these two cloud fragments shows that they belong to the same cloud complex and ecf probably lies at the back edge of the region , though it is projected very close to the ionizing source ( @xcite ; @xcite ) . the most intense co peak of the wcf is located very close to region a , with water and methanol maser emissions ( @xcite ; @xcite ; @xcite ) within its proximity , suggesting recent star formation activity towards this region ( @xcite ; @xcite ; @xcite ) . an extensive overview of the previous studies of sh2 - 252 can be found in paper 1 . this work is the continuation of paper 1 , where we have focused mainly on the characteristics of the ysos in the sh2 - 252 complex for the first time and to constrain the star formation history of the complex . we used the optical , 2mass-@xmath14 , _ spitzer_- irac and mips data sets to identify and classify the ysos of sh2 - 252 region . an attempt has also been made to characterize the physical properties of the ysos based on the spectral energy distribution ( sed ) analyses . section 2 describes various data sets used for the present study . analyses and results including the identification and classification of ysos , their various properties such as ages , masses , seds , spatial distribution etc . have been discussed in section 3 . section 4 deals with discussion on the morphology and star formation scenario of the complex and finally summary of the paper is presented in section 5 . two sets of irac observations in 3.6 , 4.5 , 5.8 and 8.0 @xmath2 m bands ( channels 1 , 2 , 3 and 4 ) were available in the _ spitzer _ space observatory archive program ( program ids : 201 ; 20506 , pis : g. fazio ; j. hester ) which were taken towards wcf and ecf on 2004 october 28 and 2006 march 26 , respectively . the irac data were taken in the high dynamic range mode with three dithers per map position and with integration time of 0.4 sec and 10.4 sec per dither . we downloaded the basic calibrated data images ( versions s18.7.0 and s18.5.0 ) from the _ spitzer _ archive and the raw data were processed and calibrated with the irac pipeline . the final mosaics were created using the mopex pipeline ( version 18.0.1 ) with an image scale of [email protected] per pixel . the total area covered by these two observations is marked in fig . [ area ] by using a dashed box . we performed the point response function ( prf ) fitting method in multiframe mode using the tool apex developed by _ science center on all the _ spitzer_-irac images to extract the magnitudes . the standard prfmap table provided on the _ spitzer _ website were used to fit variable prfs across the image . point sources with peak values more than 5@xmath16 above the background were considered candidate detection . many sources detected in the nebulosity were appeared to be spurious . the spurious sources were identified visually and deleted from the automated detection list . also , some of the sources were not detected by apex automatically . we identified these sources visually and added the coordinates in the user - list mode in apex to extract the magnitudes and thus we made sure that photometry of every genuine source is derived . we have adopted the zero - points for conversion between flux densities and magnitudes to be 280.9 , 179.7 , 115.0 and 64.1 jy in the 3.6 , 4.5 , 5.8 and 8.0 @xmath2 m bands , respectively , following the irac data handbook . there were signs of saturation of bright sources in the long integrated images . the photometric data derived from the short images were replaced for the saturated sources . the two sets of irac observations had substantial overlapping between the regions and we used the average magnitudes of those common sources detected in both the observations . channels 3 and 4 detected far fewer sources than channels 1 and 2 . this deficit exists because channels 1 and 2 are more sensitive than channels 3 and 4 and are also less affected by the bright diffuse emission that dominates the channel 3 and 4 images . the underlying typical stellar photosphere is also intrinsically fainter at channels 3 and 4 than at channels 1 and 2 , which further hinders the ability to detect sources in the longer wavelength irac filters . we finally obtained extraction lists of 24877 , 20591 , 5083 and 3263 sources in the irac 3.6 , 4.5 , 5.8 , and 8.0 @xmath2 m bands , respectively , within the area shown in fig . the irac data of the four band passes were merged , by matching the coordinates using a radial matching tolerance of [email protected] . to ensure good photometric accuracy , we restricted our catalog with sources having uncertainty @xmath17 0.2 mag in all the bands . thus our final irac catalog contains photometry of 31251 sources which were detected in one or more irac bands . out of these , only 1988 sources were detected in all the irac bands . the areas towards wcf and ecf were observed at 24 @xmath2 m using the multi - band imaging photometer for _ spitzer _ ( mips ) on 2006 october 13 and 2008 october 25 ( program ids : 20726 ; 50758 , pis : j. hester ; m. m. drosback ) . we downloaded the bcd images ( version s18.12.0 ) from the _ spitzer _ archive and the final mosaics were created using the mopex pipeline ( version 18.0.1 ) with an image scale of [email protected] per pixel . the total area covered by these observations is marked in fig . the sources were identified visually and their coordinates were given in the user - list mode in apex . to extract the flux , we performed the prf fitting method in the single frame mode . the zero - point value of 7.14 jy from the mips data handbook has been used to convert flux densities to magnitudes . the final catalog contains the 24 @xmath2 m photometry of 646 sources having uncertainty @xmath17 0.2 mag . we cross - correlated our mips-24 @xmath2 m catalog with the source catalog of irac , using a [email protected] matching radius . we obtained 24 @xmath2 m counterparts for 428 sources with irac data in all the bands . the 24 @xmath2 m photometry was mainly used to classify the ysos and to analyze their seds ( see section [ sed ] ) . nir @xmath14 data for point sources within a radius of 30 arcmin around hd 42088 have been obtained from 2mass point source catalog ( psc ) @xcite . to improve the photometric accuracy , we used the data with photometric quality flag ( ph@xmath18qual = aaa ) which gives a s / n @xmath19 10 and a photometric uncertainty @xmath20 0.1 mag . this selection criterion ensures best quality detection in terms of photometry and astrometry as given on the 2mass website . we used 2mass data along with _ spitzer_-irac data to identify and classify the candidate ysos within sh2 - 252 ( see section [ yso_2mass ] ) . the nir and irac sources were bandmerged within @xmath21.2 radial matching tolerance . if more than one 2mass sources satisfied this requirement , then the nearest one was selected . thus we obtained a total of 9097 2mass sources having irac counterpart at least in one band . a colour composite image using the 3.6 , 4.5 and 8.0 @xmath2 m bands of _ spitzer_-irac is shown in fig . [ color - irac ] . the important sub - regions and the ionizing source ( hd 42088 ) of sh2 - 252 are marked in the figure . different irac bands have contribution from various extended emission features . for example , 3.6 @xmath2 m band is dominated by diffuse emission from the 3.3 @xmath2 m pah feature . the 4.5 @xmath2 m band includes the @xmath22 line plus co band head emission and @xmath23 lines , and 5.8 and 8.0 @xmath2 m bands are dominated by strong pah features , although a few sources show extended thermal continuum emissions from warm circumstellar dust in these bands @xcite . pahs are believed to be destroyed in the ionized gas and thought to be excited in the photon dominated region ( pdr ) that is situated at the interface of region and molecular cloud by the absorption of far - uv photons leaking from the region @xcite . therefore pahs are good tracers of the warm pdr that surrounds the region . the most striking feature in fig . [ color - irac ] is the magenta coloured emission that appears almost semi - circular in shape towards the west of sh2 - 252 . the pahs within the pdr re - emit their energy in the 3.6 , 5.8 and 8.0 @xmath2 m bands , and this can be seen as magenta coloured emission in the colour composite image given in fig . [ color - irac ] . altogether , this image shows a bubble - like emission feature around the ionizing source hd 42088 , likely to be created due to the expansion of the region to its surroundings . the bright - rim feature discussed in section [ intro ] ( i.e. , region f ; see also fig . 2 of paper 1 for a close - up view ) coincides well with the pah emission feature seen at the southwest of fig . [ color - irac ] , with a sharp edge pointing towards hd 42088 . this sharp edge indicates the if , that coincides with the molecular ridge seen in the @xmath12co map shown in fig . [ area ] as well as the extinction map shown in fig . [ extmap ] ( see section [ kextmap ] ) . this correlation indicates the interface of the ionized and molecular gas . the correlation of the half - ring structure of pah emission towards the west of sh2 - 252 with the @xmath12co emission and extinction maps indicates that the ionized gas is separated from the molecular cloud by the neutral gas with pah emission . the absence of 8 @xmath2 m emission in the interior of the region can be interpreted as the destruction of pah molecules by the intense uv radiation of the ionizing source . there is also less intense diffuse emission towards the east of sh2 - 252 . however , the ionized gas is bounded more extensively in the western direction than in the eastern direction . it is to be noted that the location of ngc 2175s is devoid of strong pah emission features . interestingly , another bright red emission knot is seen further towards east of ngc 2175s with a bright red point source inside it . a small clustering is evident further east to this point source which coincides with the sub - region teu 136 discussed in paper 1 . there are not many studies on this group of stars in the literature . however , based on our spectral analysis in paper 1 , we concluded that this group is a part of the main sh2 - 252 complex itself . the indication of variable extinction in the sh2 - 252 region has been discussed in paper 1 based on the optical photometric analyses of the bright stars as well as from the distribution of candidate ysos on the @xmath24 colour - colour ( c - c ) diagram . in order to quantify the amount of extinction within each sub - region of sh2 - 252 and to characterize the structure of molecular clouds , we derived the @xmath25 extinction map using the @xmath26 colours of the stars in the 2mass catalog . to improve the quality of the map , the candidate ysos and probable contamination sources ( see section [ yso ] ) were excluded from the list . we used the grid method to determine the mean value of @xmath25 . briefly , we divided the sh2 - 252 region into many grids of size @xmath27 , and the median value of @xmath26 colours of all the sources within each grid was measured . the sources deviating above 3@xmath16 were excluded to calculate the final median colour of each grid . we used the reddening law by @xcite to convert @xmath26 colour in to @xmath25 , using the relation @xmath25 = 1.82 @xmath8 @xmath28 , where , @xmath29 = 0.2 is assumed as an average intrinsic colour of stars in young clusters ( see @xcite ; @xcite ) . since this complex is situated at a distance of 2.4 kpc ( paper 1 ) , the contribution from the foreground field stars would have affected the present extinction measurements . an average foreground extinction of @xmath30= 1.5 mag has been estimated towards this region in paper 1 . therefore , to eliminate the foreground contribution in the extinction measurement , we used only those stars with @xmath25 @xmath31 0.15 mag to generate the extinction map . there were an average of five sources in each grid to calculate the median colour . the final extinction map , having an angular resolution of @xmath32 , which is sensitive down to @xmath30 = 16 mag is shown in fig . [ extmap ] . however , it has to be kept in mind that this map is limited by the sensitivity of the 2mass survey . as a result , the @xmath25 value derived may be underestimated due to the small number of background stars detected in the heavily extincted area of the complex . it is interesting to note that the extinction map derived from the @xmath26 colours shown in fig . [ extmap ] resembles the general distribution of the molecular cloud as outlined by the @xmath12co emission map shown in fig . it traces the presence of molecular cloud towards the sub - regions a , b , c and nicely follows the bright rim feature visible at the west of sh2 - 252 ( i.e. , region f , see fig . 2 of paper 1 ) similar to the ridge of molecular cloud seen in the @xmath12co emission map by @xcite . similarly , the separation between the two cloud fragments wcf and ecf ( see section [ intro ] ) by the long rift of less extinction is also clearly seen in fig . [ extmap ] . the western cloud portion is more extincted as compared to the eastern one . however , isolated extinction complexes are seen in the eastern side as well . it has to be noted that the locations of the clusters ngc 2175s and teu 136 are devoid of heavy reddening . the same trend has already been noticed in paper 1 . the average extinction @xmath30 of the four optically bright members of the cluster ngc 2175s ( see table 3 of paper 1 ) is @xmath0 2.2 mag , which is significantly lower than the extinction towards the western part of the complex . this could be explained by the fact that the strong stellar winds from the four early type sources in ngc 2175s would have removed the dust and gas from its neighborhood or the region could be more evolved than the other parts of the complex . in the ensuing sections , we identify and classify the ysos within sh2 - 252 mainly into class i and class ii categories using their ir colours . we do not classify diskless class iii ysos because they are indistinguishable from field stars based on their ir colours and hence we can not reliably identify them using the current data set . the main limitation of the identification of ysos based on their ir colours is the contamination from various non - stellar sources in irac detection such as , polycyclic aromatic hydrocarbon ( pah ) emitting galaxies , broad - line active galactic nuclei ( agns ) , unresolved knots of shock emission , pah - emission contaminated apertures etc . , which have colours similar to that of ysos and may lead to wrong identification of ysos . @xcite and @xcite introduced the classification criteria for ysos using the ( @xmath33-[4.5]$ ] , @xmath34-[8.0]$ ] ) c - c diagram . these colour criteria since then have been modified by several authors ( e.g. , @xcite , @xcite , @xcite and the references therein ) . we adopted the empirical three - phase scheme introduced by @xcite to identify the ysos with ir excess in this complex . briefly , in phase 1 , after excluding the non - stellar contamination ( see section [ yso_irac ] ) , all four irac bands are used to separate the sources with ir excesses due to disks and envelopes ( class ii and class i , respectively ) from pure photospheres ( class iii / field stars ) based on [ 3.6 ] - [ 5.8 ] and [ 4.5 ] - [ 8.0 ] colours . phase 2 is applied to those sources which lack irac detection at [ 5.8 ] or [ 8.0 ] @xmath2 m , but which have high - quality 2mass detection ( @xmath35 0.1 mag ) in at least @xmath36 and @xmath11 bands . sources are then classified as class ii or class i objects based on their de - reddened ( @xmath37-[3.6])_0 $ ] and @xmath38-[4.5])_0 $ ] ) colours @xcite . in addition , there are a few objects with nir+irac colours of class iiis , but which exhibit a 24 @xmath2 m excess . these objects are included in a new class of ysos i.e. , ` transition disk ' sources . phase 3 re - examines all sources including contaminants and identifies class i objects and transition disk objects on the basis of their mips 24 @xmath2 m photometry . in order to identify the ysos by their irac colours , we used only those sources detected in all the irac bands with photometric uncertainties less than 0.2 magnitude . various non - stellar contaminating sources mentioned above are eliminated by using several criteria described in detail by @xcite . out of the 1988 sources detected in all the irac channels , 444 sources are found to be contaminants . the remaining sample of 1544 sources was used to identify the class i and class ii ysos associated with the region using the irac colour criteria given by @xcite . the above criteria yields 340 ysos with ir excess emission , with 86 and 254 sources having colours consistent with class i and class ii types , respectively . the rest could be class iii / field stars , which we are unable to classify reliably based on their irac colours . however , it is to be noted that in a few cases , highly reddened class ii sources could mimic the colours of class i source @xcite . in fig . [ irac - cc ] ( left panel ) , we have shown the @xmath33-[4.5]$ ] vs @xmath34-[8.0]$ ] c - c diagram for all the uncontaminated irac sources , where the class i and class ii sources are shown as red squares and green triangles , respectively . here we would like to point out that the present yso selection based on the irac- four colour detection is not complete , as many sources falling in the bright nebulous regions were not detected in the 5.8 and 8.0 @xmath2 m bands . the detection of class i and class ii ysos in all the four irac bands is limited mainly by the lower sensitivity and enhanced nebulosity of the [ 5.6 ] and [ 8.0 ] @xmath2 m bands . in order to account for the ysos missing in these bands , we used @xmath36 , @xmath11 , [ 3.6 ] and [ 4.5 ] @xmath2 m data sets and the additional ysos in the complex are identified . @xcite used the dereddened colour criteria to identify such ysos . we used the extinction map ( see fig . [ extmap ] ; section [ kextmap ] ) to deredden the data . to minimize the inclusion of extra - galactic contaminants , we applied a simple brightness limit in the dereddened [ 3.6 ] @xmath2 m photometry i.e. , all the class i ysos must have @xmath33_0 \le$ ] 15 mag and all the class ii ysos must have @xmath33_0 \le$ ] 14.5 mag @xcite . thus , after removing the contaminants and using the @xcite colour criteria , we identified a total of 367 candidate ysos ( 67 class i ; 300 class ii ) with ir excess emission , of which , 211 ( 52 class i and 159 class ii ) are found to be common with the ysos identified using the irac four - colour criteria ( see section [ yso_irac ] ) . finally , we obtained 15 and 141 additional ysos having colours consistent with class i and class ii , respectively using the @xmath41 $ ] and @xmath42 $ ] colour combination . in fig . [ irac - cc ] ( middle panel ) , we have shown the @xmath37-[3.6])_0 $ ] vs @xmath38-[4.5])_0 $ ] c - c diagram for all the uncontaminated sources , where the class i and class ii sources are shown in red squares and green triangles , respectively . in order to identify the transition disk sources , we re - examined the entire catalog of sources having 24 @xmath2 m counterparts and with uncertainty less than 0.2 mag . those sources that were considered photospheric in the previous classification but have significant excess emission at 24 @xmath2 m ( i.e. , [ 5.8]-[24 ] @xmath31 2.5 mag or [ 4.5]-[24 ] @xmath31 2.5 mag ) are typically known as transition disks . they are the class ii sources with significant dust clearing within their inner disks @xcite and we identified 14 such sources within the complex . also , any source which lacks detection in some irac bands yet has bright 24 @xmath2 m photometry are likely to be deeply embedded protostars . we identified 67 such sources and those are included in the class i category in the following analysis . five sources which lacked strong 24 @xmath2 m detection but had been classified as class i ysos in the previous step ( section [ yso_irac ] ) are reclassified as heavily reddened class ii ysos . thus after reclassifying the sources using the 24 @xmath2 m photometry , we have finally identified 577 ysos with ir excess emission in the sh2 - 252 complex . of these , 163 are consistent with class i , 400 are consistent with class ii and 14 are consistent with transition disk ysos . these objects will be used in the following analysis . thus the sh2 - 252 complex seems to have moderately rich number of ysos . in the right panel of fig . [ irac - cc ] we have shown the [ 3.6]-[4.5 ] vs [ 8.0]-[24 ] c - c diagram of all the uncontaminated irac sources having counterparts in 24 @xmath2 m along with the candidate class i , class ii and transition disk sources marked in red squares , green triangles and magenta circles , respectively . to evaluate the completeness of the census of irac detection quantitatively , we plot histograms of the irac sources having error @xmath43 0.2 mag for all the channels and they are shown by using solid lines in fig . [ comp ] . in general , the data can be considered complete till the linear distribution in the histograms which are found to be 16.5 , 16.0 , 14.0 and 13.0 mag for 3.6 , 4.5 , 5.8 and 8.0 @xmath2 m bands , respectively . however , the completeness of the yso census is limited by many factors . bright extended , variable nebulosity in the irac bands can be found across the region , which significantly limits the point source detection in these areas . the yso identification from the 2mass- irac colours is limited by the modest sensitivity of the 2mass survey , while that from the irac - mips colours suffer from the significant saturation in the irac 8 @xmath2 m and mips 24 @xmath2 m images caused by the central luminous sources as well as bright nebulosity . similarly the variable reddening ( see fig . [ extmap ] ) and stellar crowding characteristics across the extend of the region ( see paper 1 ) could also affect the local completeness limit . completeness limits for the _ spitzer _ photometry were also assessed by using a method of inserting artificial stars into the mosaics and then employing our detection algorithms to identify them ( e.g. , @xcite ) . the sub - regions a , b , c , e and f of the sh2 - 252 complex are found to be severely affected by the variable reddening and nebulous background ( paper 1 ) . in order to account for this variable background in the source detection , by following paper 1 , we estimated the completeness of two representative sub - regions i.e. , regions a and c. 90% completeness limits of these two regions have been obtained as 14.0 , 13.5 , 12.5 and 11.5 mag , respectively for 3.6 , 4.5 , 5.8 and 8.0 @xmath2 m bands and are shown by using vertical dashed lines in fig . [ comp ] . the sh2 - 252 complex lies at a distance of 2.4 kpc . the completeness limit of 14.0 mag at 3.6 @xmath2 m corresponds to an approximate stellar mass of 1.0 m@xmath44 for a yso having an age of @xmath0 1 myr @xcite . a more conservative estimate for the 90% completeness of the pms membership of the region can be made from the histogram distribution of candidate ysos shown by using the dashed lines in fig . [ comp ] which seems to be matching with the completeness estimation by artificial star method . by taking variable extinction and stellar crowding characteristics into account , in conclusion , we consider that our yso census of the sh2 - 252 complex is complete down to 1 - 2 m@xmath44 . we constructed the spectral energy distributions ( seds ) using the grid of models and fitting tools of @xcite for characterizing and understanding the nature of ysos in the sh2 - 252 complex . the models were computed by @xcite by using a monte - carlo based 20,000 2-d radiation transfer models from @xcite and by adopting several combinations of central star , disk , in - falling envelope , and bipolar cavity for a reasonably large parameter space . each yso model has seds for 10 viewing angles ( inclinations ) , so the total yso grid consists of 200,000 seds . the sed fitting tool provides the evolutionary stage and physical parameters such as disk mass , disk accretion rate and stellar temperature of ysos and hence is an ideal tool to study the evolutionary status of ysos in star forming regions . of the total ysos identified in sections [ yso_irac ] and [ yso_2mass ] , 183 sources ( 51 class i , 132 class ii ) have counterparts in 24 @xmath2 m and we fit seds to these 183 ysos . the counterparts of these ysos are searched in the optical and nir bands irrespective of their error criteria . of these sources , 54 are found to have counterparts at least in two of the optical bands ( @xmath45 ) along with the @xmath14 , irac and 24 @xmath2 m data . 96 sources have @xmath14 , irac and 24 @xmath2 m data and rest of the sources have irac and 24 @xmath2 m data . thus we have a minimum of 5 and maximum of 13 data points in the wavelength range from 0.36 @xmath2 m to 24 @xmath2 m to construct the seds . the sed fitting tool fits each of the models to the data allowing the distance and external foreground extinction as free parameters . we gave a distance range of 2.1 - 2.7 kpc ( see table 3 of paper 1 ) as the input . an extinction range has been supplied as the input for each source from the @xmath25 value estimated from the @xmath26 colour as discussed in section [ kextmap ] and by allowing an uncertainty of @xmath46 3 mag in the calculated extinction value . we further set photometric uncertainties of 10% for the optical and nir data , 15% for the irac and 24 @xmath2 m data . these values are adopted instead of the formal errors in the catalog in order to fit without any possible bias caused by underestimating the flux uncertainties . we obtained the physical parameters of ysos using the relative probability distribution for the stages of all the ` well - fit ' models . the well - fit models of each source are defined by @xmath47 , where @xmath48 is the goodness - of - fit parameter for the best - fit model and @xmath49 is the number of input observational data points . in fig . [ seds ] , we show the sample seds of class i , class ii and transition disk sources , where the solid black line represents the best - fit and the grey lines are the subsequent well - fits . from the well - fit models for each source derived from the sed fitting tool , we calculated the @xmath50 weighted model parameters such as the stellar mass ( @xmath51 ) , temperature ( @xmath52 ) , stellar age ( @xmath53 ) , mass of the disk ( @xmath54 ) , disk accretion rate ( @xmath55 ) , envelope accretion rate ( @xmath56 ) etc . the error in each parameter is calculated from the standard deviation of all well - fit parameters . the parameters and the corresponding errors of all the ysos are listed in table [ sedpars ] . the ids 1 - 51 and 52 -183 represent class i and class ii sources , respectively . for some of the sources , the errors associated with few parameters ( table [ sedpars ] ) are quite large because we are dealing with a large number of parameter space , with limited number of observational data points . additional observational data points in longer wavelengths would help constrain these parameters more precisely . table [ sedpars ] shows that @xmath0 94% ( 172/183 ) of the sources have an age range between 0.1 to 5 myr . this age distribution is in agreement with the age range discussed in paper 1 using the optical colour magnitude diagram ( cmd ) of the candidate ysos . similarly , masses of the ysos are distributed in a range between 0.4 to 5.3 @xmath57 with a majority of them having masses between 0.5 to 3.0 @xmath57 . this is again in agreement with that of paper 1 . the sed model parameters of ysos , such as the disk accretion rate and envelope accretion rate are considered to be functions of their evolutionary status . stage 0/i , ii and iii ysos have significant infalling envelopes , optically thick disks and optically thin disks , respectively @xcite . in table [ sedpars ] , a majority of the class ii ysos have the disk accretion rate of the order of @xmath0 @xmath58 - 10@xmath59 @xmath60 yr@xmath61 . similarly , the envelope infall rates of most of the class i ysos are @xmath31 10@xmath62 @xmath60 yr@xmath61 , which are relatively higher than that of the class ii ysos , confirming their evolutionary stages . in fig . [ cum ] we have shown the cumulative distribution of class i and class ii ysos as a function of their ages which manifests that class i sources are relatively younger than class ii sources . we have performed the kolmogorov - smirnov ( ks ) test for the age distribution of the class i and class ii sources . the test result shows that the chance of the two populations having been drawn from a same distribution is @xmath63% . it suggests that the age distribution of two populations are different . the distribution of ysos on the @xmath64 cmd is an ideal tool to estimate their approximate ages and masses . the @xmath64 cmd for the candidate class i , class ii and h@xmath65 emission line sources are shown in the fig . 8 of paper 1 , which suggests that majority of the ysos in sh2 - 252 are distributed between the pms isochrones of age 0.1 and 5 myr and evolutionary tracks of masses ranging from 0.5 to 3.0 @xmath57 . this is in agreement with the sed based analysis ( section [ sed ] ) . the age of each yso has been estimated from the cmd as discussed in @xcite and @xcite . briefly , their ages have been derived by comparing their locations on the cmd with pms isochrones of various ages @xcite , after correcting for the distance and reddening of the region . apart from the class i , class ii and h@xmath65 sources , all other sources having age @xmath43 5 myr in the @xmath64 cmd and with @xmath66 0.4 mag were suggested as the candidate pms members of the complex in paper 1 . we also estimated the ages of these additional candidate pms sources as discussed above . there are 80 ysos which have age estimation both from the cmd and sed ( section [ sed ] ) analyses . in fig . [ compare ] we have shown the comparison between the age distribution from cmd analysis with that from sed fitting . squares and triangles represent the class i and class ii ysos , respectively . though the age estimation from both the methods are quite scattered , in general , the distribution can be considered correlated with each other . we have also performed the ks test between the age distribution of ysos from sed and cmd analyses and found that the ages estimated by these two methods are correlated by a factor of @xmath0 81% . hence , we confirm that our age estimation based on both methods is in agreement . we attempted to constrain the evolutionary status of the sub - regions of sh2 - 252 using the sed fitting and cmd analyses . here we would like to caution that although the age estimation using the sed fittings may not be robust , the results agree reasonably well with the ages estimated by us using @xmath64 cmd for a subset of ysos ( section [ cmd ] ) . hence we accept the ages determined by models with reasonable confidence . we estimated the ages of the sub - regions a , c , e , ngc 2175s and teu 136 of sh2 - 252 by using the ages of the candidate ysos lying within their boundaries . the boundaries of these individual regions have been taken from the stellar surface density map . i.e. , [email protected] , [email protected] , [email protected] , [email protected] and [email protected] , respectively for regions a , c , e , ngc 2175s and teu 136 ( paper 1 ) . the histograms of age distribution estimated from the sed and cmd analyses of each sub - region are shown in fig . the median ages of ysos within a , c , e , ngc 2175s and teu 136 have been obtained as @xmath0 0.5 , 1.5 , 1.5 , 2.5 and 2.5 myr , respectively . as evident in the @xmath11-band surface stellar density map ( paper 1 ) , no clustering is apparently seen towards the region b. from the sed analysis of 3 candidate ysos lying within [email protected] radius of region b , a median age of 1 - 2 myr can be assigned for this region . however , the age of sh2 - 252 b should be considered as an approximate value since the statistics is very poor . the above age estimates based on the sed / cmd analyses show that the region a is the youngest among all the sub - regions , whereas regions b , c and e are of similar age and ngc 2175s and teu 136 are found to be slightly evolved than other sub - regions . spatial distribution of ysos in a star - forming complex helps to trace its star formation history . in fig . [ spatial ] , we show the spatial distribution of all the identified candidate ysos in sh2 - 252 ( i.e. , red squares : class i ; green triangles : class ii ; magenta circles : transition disk sources ) overlaid on the irac 4.5 @xmath2 m mosaic image . the 61 @xmath68 emission line sources identified from slitless spectroscopy survey ( paper 1 ) are shown by using blue crosses . the important sub - regions of sh2 - 252 are also marked in the figure . this figure indicates some interesting spatial sequence of ysos within sh2 - 252 . a comparison of this figure with fig . [ color - irac ] clearly indicates that a majority of the candidate ysos identified in sh2 - 252 are preferentially concentrated around the cregions a , c and e , respectively . also , the bright rim feature discussed in section [ intro ] ( i.e. , region f ) , which makes up the western border of the complex contains several number of ysos . there are not many ysos detected towards the region b. the wcf is seen to have more number of ysos when compared to the ecf . overall , the distribution of ysos in the west of sh2 - 252 shows a nice correlation with the semi - circular shaped distribution of molecular gas and the pah emission / region cavity boundaries ( see section [ morphology ] ) . the strong positional coincidence between the ysos and the molecular cloud towards the wcf in sh2 - 252 shows the enhanced star formation activity towards this region as observed in other star forming regions ( e.g. @xcite , @xcite ) . there are a few number of sources scattered towards the east of sh2 - 252 . there are two groups of ysos aligned in a filamentary manner associated with the cloud remnants seen towards the east of cregion e ( see figs . [ area ] and [ extmap ] ) . however , the cluster ngc 2175s does not seem to have significant number of ysos . there seems to be a small clustering of class ii and transition disk ysos towards teu 136 . it is also interesting to note that there is an enhanced concentration of class i ysos towards the east of the sub - region a when compared to other sub - regions of sh2 - 252 . this is an indication that the region a might be at a younger evolutionary stage and we will be iscussing more on this in the ensuing sections . the number of ysos in each class is useful for estimating the relative age of a star forming region ( @xcite ; @xcite ) . since class ii sources are assumed to be more evolved than class i and hence the class i / ii ratio is a function of the evolutionary status of a system @xcite . we used this ratio as a proxy for the relative ages of the sub - regions in sh2 - 252 . the spatial distribution of the candidate ysos given in fig . [ spatial ] shows that most of them are distributed within or in the close proximity of the sub - regions a , c , e and teu 136 . hence we estimated the ratio of class i to class ii ysos for these sub - regions within their radius ( paper 1 ) . the class i / ii ratios have been obtained as 0.54 , 0.24 , 0.36 and 0.13 , respectively for regions a , c , e and teu 136 . this fraction is of almost similar order with the median ratio of class i to class ii sources ( 0.27 ) obtained in nearby clusters @xcite . the various sub - regions within the s254 - 258 complex , a nearby star forming region to sh2 - 252 in the gem ob1 association , also have the class i / ii ratios vary between 0.32 to 0.78 @xcite . thus the class i / ii ratios of the sub - regions of sh2 - 252 can be considered comparable with the other star forming regions . in this scheme , the high concentration of class i ysos towards the region a suggests that the youngest population is located towards the western boundary of the complex . since most of the giant molecular clouds ( gmcs ) in the galaxy are hierarchical in structure , the evolutionary status of the identified ysos helps to get a deep insight about the overall star formation activity of the complex . @xcite studied the sub - region a using nir , sub - mm and radio observations . based on their @xmath11-band surface density map , which is deeper and of better resolution than ours , they found sub - structures within this region . in @xcite , the prominent peak of the stellar density coincides spatially with the central ionizing source of a , which is an o9.5v type star ( paper 1 ) . however , a secondary stellar density peak , which lies @xmath0 50@xmath15 east to the main peak has also been found and its spatial location coincides with the sub - mm peak detected at 850 @xmath2 m @xcite . the authors have attributed this as a possible cold core at an early evolutionary stage than a. similarly , the most intense co peak of the wcf is also located east to the region a with water and methanol maser emissions ( @xcite ; @xcite ; @xcite ) within its proximity , suggesting the recent star formation activity towards the east of a ( @xcite ; @xcite ; @xcite ) . the water and methanol maser emissions associated with 850 @xmath2 m and co peak towards the east of region a suggests that this region is possibly at an early evolutionary stage than a. the high concentration of class i ysos around the region a ( fig . [ spatial ] ) and the large value of class i / ii ratio ( see section [ spatialdist ] ) show that region a is at a young evolutionary stage compared to the other sub - regions of the complex . the average age estimation ( @xmath0 0.5 myr ) based on the sed and cmd analyses also shows that the region a is indeed very young . based on the class i / ii ratio and age histogram analysis , the regions b , c and e are found to be at a similar evolutionary stage and are slightly evolved ( median age @xmath0 1 - 2 myr ) than region a. the presence of a very few number of ysos towards the small cluster ngc 2175s ( see fig . [ spatial ] ) suggests that either the cluster was formed in an early epoch of star formation in sh2 - 252 or it is a case of rapid circumstellar disk evolution of ysos ( e.g. , @xcite ) due to the presence of four early type massive sources at its center ( see table 4 of paper 1 ) . however , the median age estimation based on the cmd analysis gives an approximate age of @xmath0 2 - 3 myr for this cluster . the small cluster teu 136 lying towards the east of sh2 - 252 seems to have more number of class ii ysos than class i. the low value of the class ratio as well as the sed / cmd based age analyses ( median age @xmath0 2 - 3 myr ) show that this cluster is also at a slightly more evolved stage when compared to the sub - regions a , b , c , and e of the sh2 - 252 complex . sh2 - 252 is a large region in the gem ob1 association , excited by the o6.5v star hd 42088 of age 2 - 4 myr . it contains 12 b - type stars , some of them are associated with cregions . since cregions are at a early stage of massive star formation , their impact is limited to their local environment . thus , o6.5v star is the most dominant source of energy responsible for ionization in the region . @xcite showed that the cloud complex is separated mainly into two fragments ecf and wcf , which themselves are embedded in a more diffuse and extended ( @xmath0 90 pc ) molecular cloud complex . most of the mass of wcf ( @xmath0 2.5 @xmath8 10@xmath69 ) is concentrated in a narrow ridge , which borders the western if of sh2 - 252 . this dense , semi - circular shaped shell observed at the eastern boundary of wcf , appears to have been swept up by the pressure from the if generated by the exciting source of sh2 - 252 @xcite . the large size and massive content of wcf suggest that the entire cloud is unlikely to be swept up . we roughly compared the required kinetic energy and the energy produced by the massive o6.5v star to sweep up the entire shell . the combined energy ( radiation plus wind ) liberated by the o6.5v stars is found to be less than the required energy to drive the entire cloud ( wcf+ecf @xmath0 2.7@xmath810@xmath69 ; @xcite ) . thus it is suggested that at least part of the shell in wcf is probably primordial in nature and the entire shell is not swept up by the expansion of the region . however , at the same time it is obvious from fig . [ color - irac ] that the region is in interaction with wcf , with bright semi - circular pdr and compressed neutral material behind it and protruding structures such as ` finger tips ' and a bright rim pointing towards the ionizing source . moreover , we observed recent star formation activity towards this region . the morphology resembles that star formation has occurred in thin shells around the region and star formation seems to be more active in this direction compared to the whole cloud complex . [ color-24 ] shows the colour composite image of the western half of sh2 - 252 made from the 8.0 and 24 @xmath2 m bands . the contours of 1.1 mm dust continuum emission ( taken from bolocam galactic plane survey ; @xcite ) and low resolution radio continuum emission at 1280 mhz ( samal et al . in preparation ) are over plotted in fig . [ color-24 ] . on a large scale we see a ring of molecular material at 1.1 mm aligned parallel to the outer extent of the if of sh2 - 252 seen in optical ( see fig . [ area ] ) , and located just behind the bright pah emission seen in 8 @xmath2 m and also corresponds to the dark region at 24 @xmath2 m . the molecular material in the ring seems to be fragmented at various locations . this shell of molecular material located behind the if provides evidence for the most compressed zone of the wcf fragment . we also observed many class ii and class i ysos in this direction ( fig . [ spatial ] ) , with a majority of them projected behind the pdr , in the vicinity of 1.1 mm emission . apart from the ysos , three cregions ( a , b and c ) are observed in the close vicinity of the 8 @xmath2 m ring , with region a being located slightly behind the pah ring and separated from the ring by a dark lane , consisting of highly dense molecular material . the region a is part of the massive ( m @xmath0 2.5@xmath810@xmath69 @xmath70 ) , large ( @xmath0 16 pc ) wcf cloud ( see fig . [ area ] ) . hence the influence of o6.5v star to region a , if at all exists , must be weak . in general , the configuration ( at least in the western direction ) suggests that the region has collected some of the matter during its expansion at its periphery , like the collect & collapse process of star formation such as in sh2 - 212 ( @xcite ; @xcite ) . collect & collapse process works well in homogeneous and uniform medium , which is definitely not the case here , with openings at various directions and without a clear circular shell structure . however it has been found that collect & collapse process can still be applicable in turbulent medium @xcite . from the extinction map ( fig . [ extmap ] ) we found that the visual extinction in the direction of the pah ring is , @xmath30 @xmath31 5.5 mag , which corresponds to a column density of hydrogen , @xmath71 ) in the region @xmath31 1.03@xmath810@xmath72 atoms @xmath73 , using the relation @xmath30 = 5.34@xmath810@xmath74 , derived from @xcite for @xmath30 = 3.1@xmath8e(b - v ) . the observed column density towards the molecular ring is thus comparable to the column density required ( i.e. , @xmath0 6@xmath810@xmath75 atoms @xmath73 ) in the collected layer for the fragmentation to happen , as predicted by @xcite . assuming the observed star formation near the vicinity of the if might have happened due to collect & collapse process , we compared the observed properties with the predication by the numerical simulation of @xcite . we adopted 1.7@xmath810@xmath76 as ionizing photons s@xmath61 ( for o6.5v star ; @xcite ) , the isothermal sound speed as 0.2 - 0.3 and the density of the neutral material into which the region evolved as 400 @xmath77 @xcite . with these values , the predicted fragmentation time and radius at which fragmentation would take place are found to be 2.3 - 2.9 myr and 10.3 - 11.6 pc , respectively . though there are large uncertainties involved in these calculations , considering the age of the ionizing star is of @xmath0 2 - 4 myr ( paper 1 ) and ages of the class i / ii ysos projected on the shell @xmath0 0.5 myr , as well as the fact that the farthest separation of the shell from the ionizing star is @xmath0 10 pc , the star formation activity could in fact might have taken place in the collected shell . the possible formation of regions b , c and e is in doubt by the same scenario , owing to their large age ( @xmath0 1 - 2 myr ; section [ evolution ] ) . these objects might have formed before the ionization - shock front passed over them , although with the present data it is hard to prove whether the passage of ionization - shock front makes the age of these sources look older . ignoring this fact , we suspect that they are possibly not triggered due to the expanding region , and more likely to be primordial in nature , resulted from the formation and evolution of the molecular cloud itself , which is in agreement with @xcite . the radio source f is an extended one , with no internal heating source . the radio emission is located ahead of the 8 @xmath78 m emission ( see fig . [ color_close ] for a close up view of region f ) and corresponds to the glowing gas in the optical image ( see fig . 2 of paper 1 ) , thus , more likely due to the photo - evaporating gas from the surface of the cloud exposed to radiation . this phenomenon usually occurs when a massive star drives an ionization - shock front into the ism , the external layers of the cloud exposed to radiation starts photo - evaporating , and can be seen as glowing layer of gas . in fig . [ hst ] , the hst - wfpc2 archive image at 0.65 @xmath2 m ( program no : 9091 ; pi : j. hester ) of the extreme edge of the molecular ridge at region f ( i.e. , the brightest portion of the bright - rim feature in fig . 1 of paper 1 ) is shown . apart from the photo - evaporating gas , this better resolution image displays number of ` finger - tips and pillars ' like fragments which are seen silhouetted against the bright background pointing towards the direction of hd 42088 . three ysos are observed at the tip of region f ( see fig . [ spatial ] ) and they are low mass ( @xmath43 2 ) class i / ii sources , thus possibly formed from low mass clouds . assuming that the impact of ionizing radiation to such cloud structures initiates new star formation similar to those in brcs and small globules , we compared the observed incident ionizing flux to the required range of ionizing fluxes that can trigger star formation in fiducial cloud as predicted by @xcite using numerical simulations . @xcite numerically estimated the required ionizing flux range as 10@xmath79 @xmath73 s@xmath61 - 3@xmath810@xmath80 @xmath73 s@xmath61 . they predicted that if the ionizing flux is very low ( @xmath4310@xmath79 @xmath73 s@xmath61 ) then the cloud will be slowly eroded and if the ionizing flux is high ( @xmath3110@xmath80 @xmath73 s@xmath61 ) then the cloud will be rapidly dispersed , with no star formation . we calculated the ionizing photon flux impinging onto these structures from the radio emission from the surface layer of the cloud following @xcite assuming the electron temperature of the photo - evaporating gas as 10@xmath69 k , and it is estimated to be 3@xmath810@xmath79 @xmath73 s@xmath61 . the observed estimation is well within the predicted range to initiate star formation . besides the low - mass ysos at the tip of the pillars , we also observed a clear clustering or concentration of ysos in the dense region of the cloud located behind region f ( see figs . [ color - irac ] and [ spatial ] ) . the age of these ysos is of the order of @xmath0 1 myr , thus indicates the youth of the region . these ysos seem to have resulted from the gravitational fragmentation of the collected material rather than the implosion of pre - existing clumps due to external pressure . the ysos show a chain - like elongated distribution parallel to the if and appear to lie in the collected shell . whereas in the case of radiation driven implosion ( rdi ) , the ysos are expected to lie perpendicular to the direction of progressive if , with younger ysos being located away from the if . the incident photon flux is also on the lower - side of the flux needed to initiate rdi , thus possibly is not enough for a massive cloud to form a cluster of stars . hence , we suspect that the star formation is more likely due to gravitational instability in the collected material , such as seen in the massive condensation of rcw120 , where a chain of ysos is observed parallel to the if @xcite . in fig . [ spatial ] we find the densest concentration of ysos in the close vicinity of the molecular clump located between the cregions a and b ( see fig . 15 of @xcite ) , with many ysos of class i nature ( @xmath0 10@xmath81 yr ) . this cloud clump is associated with water and methanol masers and does not show any radio emission ( down to 0.4 mjy ) . hence , @xcite suggested that it could be a site of high - mass star - forming protocluster in a very early evolutionary stage ( section [ evolution ] ) . we also observed a cluster of ysos , mostly class i ( fig . [ spatial ] ) , thus reflecting it is indeed a site of cluster formation sandwiched between the two relatively evolved cregions a and b. using h@xmath82 observations at 2.12 @xmath2 m , @xcite traced a cometary - arc shaped shocked molecular h@xmath82 emission bounding the ionized gas ( i.e. , ionization bounded side of the cregion a ) in the direction of the molecular clump , possibly reflecting the interaction of the region with the adjacent molecular cloud . similarly , we noticed sharp radio contours for region b , in the direction of the molecular clump ( fig . [ color-24 ] ) . the radio size of region b is larger than regions a , c and e , and is ionized by a similar kind of star ( b0-b0.5 type ; paper 1 ) . thus , if they are evolved in a cloud clump of similar density , the larger size of region b possibly reflects that it is relatively evolved in nature . the region between a and b indeed reflects a site of recent star formation , where much younger sources are found in the matter in between the two relatively evolved cregions a and b. a similar kind of star formation has been observed at the interfaces of two expanding super giant shells lmc4 and lmc5 @xcite and two classical regions s255-s257 @xcite , where it is believed that collision of dense swept up neutral materials associated with individual shells lead to star formation ( @xcite ) . possibly , we are witnessing one such region , where star formation has been enhanced due to a similar kind of process . the overall picture emerges from our analyses is that the star formation activity observed towards wcf is more likely multi - fold and the sources are found to be at different evolutionary stages . though the large scale co map @xcite provides much information about the global structure of the complex , but the region certainly deserves attention for high resolution molecular line observations to explore the detailed properties of the sub - regions . the optically bright , extended region sh2 - 252 is a part of the gemini ob1 association . this region is mainly composed of two small clusters ngc 2175s and teu 136 and four cregions namely a , b , c and e. in this paper , an extensive survey of the star forming complex sh2 - 252 has been undertaken with an aim to explore its hidden young stellar population , their characteristics , spatial distribution , morphology of the region and finally to understand the star formation scenario of the complex for the first time . _ spitzer_-irac , mips photometry ( 3.6 - 24 @xmath2 m ) are combined with 2mass - nir and optical data sets to identify and classify the ysos by their ir excess emission from their circumstellar material . using the ir c - c criteria , we have identified 577 ysos in the complex , of which , 163 are consistent with class i , 400 are consistent with class ii and 14 are consistent with transition disk ysos , suggesting a moderately rich number of ysos in this region . from the cmd and sed based analyses , majority of the ysos are found to have an age distribution between 0.1 - 5 myr and mass in the range of 0.3 - 3.0 m@xmath83 . spatial distribution of the candidate ysos shows that they are mostly clustered around the sub - regions of the complex such as in a , c , e , f and teu 136 . however , majority of the candidate ysos are distributed in the western part of the complex when compared to the east , suggesting enhanced star formation activity towards its west . using the sed and cmd based age analyses , we derived probable evolutionary status of the sub - regions of sh2 - 252 . our analyses suggest that the region a is the youngest ( @xmath0 0.5 myr ) , the regions b , c and e are of similar evolutionary stage ( @xmath0 1 - 2 myr ) and the small clusters ngc 2175s and teu 136 , located towards the east of sh2 - 252 are slightly evolved ( @xmath0 2 - 3 myr ) . morphology of the region in the _ spitzer _ colour images as well as in the 1.1 mm map shows an almost semi - circular ring - like shape towards the western half of the complex . indeed , we find a molecular shell composed of several clumps distributed around the main ionizing source ( i.e. , hd 42088 ) , suggesting that the expansion of the region is collecting the surrounding material , which gives rise to the semi - circular ring shape . we find several candidate ysos distributed over the semi - circular molecular shell which is an evidence for the star formation activity within the shell . finally , by comparing the age of the ionizing source , fragmentation time of the collected molecular shell and age of the ysos , we suggest collect and collapse scenario as one of the the possible mechanisms responsible for the star formation within the shell . we observed the densest concentration of ysos , ( mostly class i , @xmath0 0.5 myr ) at the western outskirts of the complex , within a molecular clump located between the cregions a and b. the correlation between the molecular clump at this location with large number of class i ysos and the associated water and methanol masers suggest that it is indeed a site of cluster formation in a very early evolutionary stage , sandwiched between the two relatively evolved cregions a and b. we conclude that the region is undergoing a complex star formation activity and there is a strong interplay between the radiation from the expanding region to the surrounding molecular material . the region certainly deserves attention for high resolution molecular line observations to further explore its hidden structure . we are thankful to the anonymous referee for useful comments which have improved the contents and presentation of the paper significantly . this publication makes use of data from the two micron all sky survey , which is a joint project of the university of massachusetts and the infrared processing and analysis center / california institute of technology , funded by the national aeronautics and space administration and the national science foundation . this work is based on observations made with the _ spitzer _ space telescope , which is operated by the jet propulsion laboratory , california institute of technology , under contract with nasa . this publication makes extensive use of data products from the hst archive at the canadian astronomy data center ( cadc ) , operated by the national research council of canada with the support of the canadian space agency . jj is thankful for the financial support for part of this study through a stipend from csir , india and also the guidance from the _ spitzer _ support staff for the data analysis is highly acknowledged .

in the color glass condensate approach to the description of high energy heavy ion collisions , one needs to superimpose small random gaussian distributed fluctuations to the classical background field , in order to resum the leading secular terms that result from the weibel instability , that would otherwise lead to pathological results beyond leading order . in practical numerical simulations , one needs to know this spectrum of fluctuations at a proper time shortly after the collision , in the fock - schwinger gauge . in this paper , we derive these fluctuations from first principles , by solving the yang - mills equations linearized around the classical background , with plane wave initial conditions in the remote past . we perform the intermediate steps in light - cone gauge , and we convert the results to the fock - schwinger gauge at the end . we obtain simple and explicit formulas for the fluctuation modes . institut de physique thorique ( ura 2306 du cnrs ) + cea / dsm / saclay , 91191 gif - sur - yvette cedex , france

1307.1765

one of the outstanding theoretical problems in high energy heavy ion collisions is the understanding from first principles of the pressure isotropization and possibly the thermalization of the gluonic matter produced in these collisions . from rhic and lhc data , there is ample evidence that the expansion and cooling of this matter is well described by relativistic hydrodynamics @xcite with a very small viscosity ( characterized by a viscosity to entropy density ratio , @xmath2 , that is fairly close to the value @xmath3 obtained in the strong coupling limit of some qcd - like theories @xcite , and that has been conjectured to be a lower bound ) . this good agreement also suggests that the anisotropy between the transverse and longitudinal ( with respect to the collision axis ) pressures is not too large , because otherwise the viscous corrections could be important and spoil this agreement . however , understanding from first principles why the hydrodynamics models work so well has proven very challenging until now . moreover , there is also a vast amount of data , ranging from deep inelastic scattering to proton - nucleus and nucleus - nucleus collisions , supporting the idea of gluon saturation in high energy collisions involving hadrons or nuclei @xcite . in this regime , the gluon density in the projectiles becomes very large , leading to important non - linear corrections in the evolution of the gluon distribution with energy . these nonlinear effects dynamically generate a dimensionful scale , the saturation momentum @xmath4 , that controls the scattering @xcite . the gluon occupation number is non - perturbatively large , of order @xmath5 , for transverse momenta below @xmath4 , and decreases rapidly above this scale . the saturation momentum increases with the energy of the collision , to reach values of order @xmath6gev for nuclei at lhc energies . since the value of the strong coupling @xmath7 at such scales is around @xmath8 , one may expect to be able to describe these collisions in the color glass condensate effective theory @xcite , that describes the physics of gluon saturation at weak coupling . for this reason , the cgc appears to be a well suited framework in order to try to explain the early isotropization of the system . the state of the system just after such a collision has been calculated at leading order in @xmath7 in the cgc framework @xcite , and one finds that its energy - momentum tensor is very anisotropic , with a negative longitudinal pressure exactly opposite to the energy density ( a trivial consequence of the fact that the chromo - electric and chromo - magnetic fields are parallel to the collision axis just after the collision @xcite ) . at leading order , the subsequent time evolution never leads to the isotropization of the stress tensor @xcite . but it has been noticed long ago that the cgc result at leading order is insufficient , because of the existence of instabilities in the classical solutions of the yang - mills equations @xcite : some of the higher order ( in @xmath7 ) corrections grow exponentially fast with time , and soon become larger than the leading order they are supposed to correct . these instabilities are the manifestation in the cgc framework of the well known weibel instabilities in plasmas with an anisotropic particle distribution @xcite . moreover , a lot of work suggests that these instabilities could play an important role in driving the system towards isotropization and local thermal equilibrium @xcite . in the cgc formalism , these instabilities spoil the naive estimates of the order of magnitude of contributions @xcite , since these estimates usually keep track only of the powers of @xmath7 , implicitly assuming that all the numerical prefactors remain of order unity at all times . in the presence of unstable modes , this is no longer true : some of these coefficients will grow exponentially in time , leading to secular divergences when the time goes to infinity and making the ordinary loop expansion useless after a finite time of order @xmath9 . an improved power counting that tracks these fast growing terms was proposed in ref . @xcite , and it was shown @xcite that one can resum the fastest growing terms by superimposing random gaussian fluctuations to the initial condition of the classical yang - mills equations , and then averaging over these fluctuations . thanks to this resummation , one completely tames the secular terms , and the validity of the resummed result is extended to larger times . as a proof of concept , this resummation was implemented numerically in the case of a @xmath10 scalar field theory . although very different from a yang - mills theory in many respects , this theory has several similar features : it is scale invariant in 3 + 1 dimensions at the classical level , and its classical solutions have instabilities ( here due to parametric resonance ) . it was shown in refs . @xcite that after performing the gaussian average over the fluctuations of the initial classical field , the system evolves towards the equilibrium equation of state , and that its transverse and longitudinal pressures become equal in the case of a system expanding in the longitudinal direction . moreover , the origin of this resummation scheme ( and in particular the fact that it includes the exact nlo result ) completely prescribes the ensemble of these fluctuations : their spectrum can be obtained by computing a 2-point correlator in the presence of a non - trivial background field ( the solution of yang - mills equations at leading order ) . from the analysis of next - to - leading order corrections done in @xcite , this 2-point function can be constructed as follows @xmath11 where @xmath12 is the solution of the yang - mills equations linearized around the classical cgc background , whose initial condition at @xmath13 is a plane wave of momentum @xmath14 , polarization @xmath15 and color @xmath16 . however , this calculation has never been done so far . in ref . @xcite , an alternative way of computing these fluctuations was proposed , based on the existence of an inner product between pairs of these fluctuations ( written here in terms of the proper time @xmath17 , the rapidity @xmath18 and the transverse coordinate @xmath19 ) , @xmath20 @xmath21 denotes the electrical field associated to the gauge potential @xmath22 , defined as : @xmath23 the above inner product is conserved when the fluctuations evolve over the classical background . it is also easy to check that the modes obtained by evolving plane waves from the remote past form an orthonormal ( with respect to the above inner product ) basis of the vector space of fluctuations . it was then suggested that one may avoid solving the linearized equations of motion for the fluctuations from the remote past , and that it could be sufficient to find a complete set of modes that obey the equations of motion locally near a proper time @xmath24 just after the collision has taken place , provided that this set of modes also form an orthonormal basis in the above sense . solving this alternate problem is simpler because one needs only to find _ local _ solutions of the linearized equations of motion , instead of global solutions with prescribed initial conditions at @xmath13 . the reasoning in @xcite was that if one knows a set of orthonormal modes at the time @xmath17 , even if it is not the same set as the one originating from the plane waves , it would generate the same gaussian ensemble of fluctuations provided that the two basis can be related by a unitary transformation . and it is also clear that unitary transformations preserve the inner product defined in eq . ( [ eq : centralps ] ) . it turns out that there is a caveat in this argument : there are also non - unitary transformations that preserve the inner product . such a transformation , when applied to a basis of fluctuation modes , will leave all the inner products unchanged ( and thus transform an orthonormal basis into another orthonormal basis ) but it will lead to a different gaussian ensemble of fluctuations . a very simple example of such a transformation is to multiply all the electrical fields by a constant @xmath15 , while at the same time dividing the gauge potentials by the same constant ) is the complex version of a simplectic product . it is invariant under all the canonical transformations of the fields and their conjugate momenta , that form a superset of the unitary transformations ( where one would apply the same unitary rotation both to the gauge potentials and to the electrical fields ) . ] . obviously this transformation does not change the inner product defined in eq . ( [ eq : centralps ] ) , but it multiplies the variance of the set of gaussian fluctuations by @xmath25 for the electrical fields , and by @xmath26 for the gauge potentials . given the existence of these transformations , one can not be sure that the set of mode functions obtained in @xcite leads to the correct fluctuations . instead , they should be constructed by evolving the plane waves from @xmath13 . in the present paper , we reconsider this question by going back to the original definition of the 2-point function that controls the gaussian spectrum of fluctuations , i.e. the eq . ( [ eq : fluct000 ] ) . using a gauge fixing inspired from ref . @xcite , we explicitly solve the linearized yang - mills equations over the leading order classical background field , with plane waves as the initial condition in the remote past . we obtain rather simple analytical expressions for these solutions , at a proper small positive time @xmath27 ( i.e. just after the collision ) . we provide the results in the fock - schwinger gauge that is commonly employed in the numerical resolution of the yang - mills equations , for a choice of quantum numbers which is appropriate for a numerical implementation on a lattice with a fixed spacing in the rapidity @xmath18 ( as opposed to a discretization with a fixed spacing in the longitudinal coordinate @xmath28 ) . the paper is organized as follows . in the section [ sec : background ] , we recall some well known results for the solution of the classical yang - mills equations in the presence of the two color currents that describe the colliding nuclei . most of the section is devoted to summarizing the derivation of this solution in the @xmath29 gauge , originally performed in ref . @xcite , and on the gauge transformation that one must perform in order to eventually obtain the result in the fock - schwinger gauge . in the section [ sec : fluct ] , we follow a similar strategy in order to solve the linearized yang - mills equations for a small perturbation propagating over this background field . the calculation is subdivided in several stages , corresponding to the successive encounters of the fluctuation with the two nuclei , followed by a final gauge transformation to go from the @xmath29 gauge to the fock - schwinger gauge . the impatient reader may find the final result in eqs . ( [ eq : finalresult ] ) . the section [ sec : concl ] is devoted to concluding remarks , and some more technical material is relegated into several appendices . before going into the details of our calculation , let us state the problem we need to solve , by listing the equations of motion and current conservation constraints that must be satisfied , as well as the boundary conditions that are appropriate in applications to heavy ion collisions . here , we list these equations in a generic form which is valid in any gauge . as we shall see later , specific gauge choices may lead to some simplifications . at leading order in the cgc framework , inclusive observables can be expressed in terms of a gauge field that obeys the classical yang - mills equations , and that vanishes in the remote past ( i.e. before the collision ) : @xmath30=\null & j^{\nu}\ ; , & \left[{{\mathcal{d}}}_\mu , j^{\mu}\right]=\null&0\ ; , \label{eq : ymlo}\\ \lim\limits_{t\to-\infty}{{\mathcal{f}}}^{\mu\nu}=\null&0\ ; , & \lim\limits_{t\to-\infty}j^{\nu}=\null & \delta^{\nu-}\rho_1 + \delta^{\nu+}\rho_2\;.\end{aligned}\ ] ] on the left are the equations obeyed by the gauge potential ( or equivalently the field strength @xmath31 ) , and on the right are the equations satisfied by the external current @xmath32 . in the remote past , it is given simply in terms of the two functions @xmath33 and @xmath34 that represent the color charge distribution in the two nuclei before the collision . however , since its conservation equation involves a covariant derivative , this current can be modified during the collision by the radiated gauge fields . this means that in general , one must view the eqs . ( [ eq : ymlo ] ) as coupled equations . this problem has been solved long ago in @xcite . we just briefly remind the reader of the solution in the rest of this section , and we also discuss an alternate way of solving these equations that has been proposed in @xcite . when extending the cgc to next - to - leading order , one needs to study small perturbations to the gauge field , more specifically those that behave as plane waves before the collision . because the gauge field is entangled with the current via the conservation equation , this in general leads to a small perturbation to the current as well carries a color . a quantum from this wave can be absorbed by one of the charges that contribute to the current @xmath35 , thereby altering its color , and therefore changing the current itself . ] . a linearization of the above equations around the lo solution gives : @xmath36 -\left[{{\mathcal{d}}}^{\nu},a^{\mu}\right]\right ] -ig\left[{{\mathcal{f}}}^{\nu\mu},a_{\mu}\right]=\null&j^{\nu}\ ; , & \lim\limits_{t\to-\infty}a^{\mu}=\null&\epsilon^{\mu } e^{ik\cdot x}\;,\end{aligned}\ ] ] and @xmath37-ig\left[a_\mu , j^{\mu}\right]=\null&0\ ; , & \lim\limits_{t\to -\infty}j^{\nu}=\null&0\;.\end{aligned}\ ] ] note that the change @xmath38 to the current must vanish in the remote past , since this is before the current could possibly have been altered by the plane wave . depending on the gauge choice , the bracket @xmath39 $ ] may vanish and therefore the perturbation of the current is identically @xmath40 . ( but in the collision of two projectiles , the current has both non - vanishing @xmath41 and @xmath42 components , and none of the light - cone gauges can eliminate this term completely ) . in the cgc description of heavy ion collisions , the gauge fields are driven by two color currents @xmath43 and @xmath44 that describe the color carried by the fast partons of the two projectiles . these currents are proportional to delta distributions @xmath45 and @xmath46 , respectively . because of the presence of these singular sources in the classical yang - mills equations , one starts the numerical resolution of the field equations of motion slightly above the forward light - cone , at some small proper time @xmath24 . the evolution of the fields is thus free of these singular sources , but the drawback is that one must know the initial value of the gauge potentials and electrical fields at the starting time @xmath17 . these initial conditions were first obtained in refs . @xcite from the known values of these fields below the light - cone , by a matching procedure that amounts to requesting that all the singularities cancel from the solution . at a proper time @xmath47 immediately after the collision , the initial conditions read-[eq : a1 ] ) in this section , are valid in any representation of the su(n ) algebra . ] @xmath48\ ; , \label{eq : mvfields}\end{aligned}\ ] ] where the fields @xmath49 are the solutions in light - cone gauge of the classical yang - mills equations for a single projectile . for the projectile moving in the @xmath50 direction , we have @xmath51 where @xmath52 ( which can be viewed as the gauge potential of that nucleus in lorenz gauge ) is related to the corresponding color current by @xmath53 a similar set of equations relates the field @xmath54 to the color current @xmath55 of the second nucleus . note that the @xmath56 dependence of @xmath57 is relevant only inside the support of the color currents . outside of these ( infinitesimal ) regions along the light - cones , the wilson lines @xmath58 depend only on the transverse coordinate @xmath19 . this is why in eq . ( [ eq : mvfields ] ) , all the fields have only a transverse dependence . it is sometimes useful in intermediate steps of the calculations to extend the support of these currents to a small but finite range @xmath59 . the limit @xmath60 is always taken at the end of the calculations , and the final answers will all be given for color currents that have an infinitesimal support . in ref . @xcite , the initial conditions ( [ eq : mvfields ] ) have been rederived by doing all the intermediate calculations in light - cone gauge , where crossing the light - cones that support the color currents can be handled more easily . since this is also the gauge choice that we will adopt for the intermediate steps of our calculation of the fluctuations , we reproduce the main steps of @xcite in the rest of this section , in order to outline its key features . in the first derivation of the initial fields of eqs . ( [ eq : mvfields ] ) , different light - cone gauges were used for describing the two projectiles before they collide , by exploiting the fact that there is no causal contact between them until the collision . the main modification introduced in @xcite is to use a unique light - cone gauge , that is employed globally to treat the two projectiles . in this work , we will choose the @xmath29 gauge condition for this purpose . this choice breaks the symmetry between the two nuclei . for the nucleus moving in the @xmath61 direction , the solution of the yang - mills equations in lorenz gauge , @xmath62 fulfills the light - cone gauge condition @xmath29 and therefore does not need to be transformed further . therefore , we just take @xmath63 this is not the case for the nucleus 1 , whose gauge potential in lorenz gauge has a non - zero minus component . we thus need to perform a gauge transformation , @xmath64 the gauge transformation @xmath65 has to be chosen so that it eliminates the minus component , and it turns out that it should be equal to the wilson line @xmath66 introduced earlier in eq . ( [ eq : u1 ] ) . after the transformation , the non - zero components of the field @xmath67 are the transverse ones , and moreover they have the form of a transverse pure gauge has different values at @xmath68 and @xmath69 . ] @xmath70 ( this field is of course identical to the @xmath71 defined in eq . ( [ eq : u1 ] ) . we will denote it @xmath72 here for consistency with the notation used for the field of the nucleus 2 , and to stress the fact that we are now in a different gauge ) . note also that the gauge transformation @xmath65 has no incidence on the field of the first nucleus , since it differs from the identity only at @xmath73 . the figure [ fig : gauget ] summarizes the structure of the gauge potentials before and after the gauge transformation @xmath65 . a legitimate question that arises is what is the advantage in treating the two nuclei in such an asymmetric fashion ? the reason is only technical : many calculations turn out to be simpler in this mixed description . in order to determine the fields just after the collision ( i.e. on the blue surface on the right side of figure ( [ fig : gauget ] ) ) , one can independently study what happens on its left and right branches . indeed , causality prevents the field that travels on the left side of the light - cone from interacting with the field that travels on the right side . ] ( they travel through regions that are separated by space - like intervals ) . the result is the same on the two branches and for infinitesimal @xmath74 or @xmath75 , it is given by ] where all the objects in the right hand side should be in the same representation as the generators @xmath76 . ] @xcite @xmath77 one sees that @xmath78 only depends on @xmath19 on the blue surface of figure ( [ fig : gauget ] ) , in the limit where this surface becomes infinitesimally close to the forward light - cone . note also that this solution is not quite symmetric between the nuclei 1 and 2 . indeed , @xmath79 is non zero , while @xmath80 is identically zero by virtue of the light - cone gauge condition . above the forward light - cone , analytical solutions of the classical yang - mills equations are not known , and one must resort to numerical techniques . in principle , it would be perfectly doable to solve the equations of motion in the gauge @xmath29 , starting with eqs . ( [ eq : fieldslc ] ) as initial conditions . however , above the forward light - cone , the natural coordinates to describe a high energy collision is the @xmath81 system . and consequently , the fock - schwinger gauge condition @xmath82 leads to simpler equations of motion than the light - cone gauge @xmath29 . it is therefore desirable to apply a gauge transformation to the fields of eqs . ( [ eq : fieldslc ] ) in order to satisfy the fock - schwinger gauge condition . this transformation can be done in two stages . first of all , let us apply a gauge transformation @xmath83 , that changes the field of the second nucleus before the collision from @xmath84 into a transverse pure gauge @xmath85 . by doing this , we arrive at a more symmetric description of the collision , where both nuclei produce a transverse pure gauge field prior to the collision . when applied to the fields of eqs . ( [ eq : fieldslc ] ) , this transformation gives the following fields at @xmath47 : @xmath86 the first of eqs . ( [ eq : fieldslc1 ] ) makes an explicit reference to the components of @xmath87 in the adjoint representation . one can therefore also rewrite it as a commutator , @xmath88\ ; .\ ] ] note that after this first stage , we are still in the light - cone gauge @xmath29 , but with a different choice of the residual gauge fixing compared to eqs . ( [ eq : fieldslc ] ) . indeed , since @xmath89 does not depend on @xmath74 , the gauge transformation generated by @xmath89 can not produce a non - zero @xmath90 . as explained in @xcite , the final step to get the fock - schwinger gauge fields is to perform a gauge transform @xmath91 such that @xmath92 where the left hand side is the gauge potential of eqs . ( [ eq : fieldslc1 ] ) in light - cone gauge @xmath29 , and @xmath93 the gauge potential in fock - schwinger gauge . the @xmath94 component of these equations should therefore tell us how to choose @xmath91 in order to achieve the desired transformation . recalling that @xmath95 , and defining also @xmath96 , one then finds @xmath97 note that this formula is only valid for very small values of @xmath24 , since it has been obtained solely from the knowledge of the value of the gauge fields at @xmath47 . applying then this gauge transformation to the other components of the gauge potential , we recover the known results from @xcite , that we have already recalled in eqs . ( [ eq : mvfields ] ) . we now turn to the problem of computing analytically the small fluctuations @xmath22 on top of the background field , with plane wave initial conditions in the remote past . we will perform most of the calculation in the same @xmath29 light - cone gauge that was used in the previous section for the background field , and the gauge transformation to obtain finally the fluctuations in the fock - schwinger gauge will be performed at the very end . the setup of the problem in this gauge is illustrated in the figure [ fig : smallfluc ] , where we indicate the structure of the background field in each relevant region of space - time . in this calculation , we will consider only the propagation of the fluctuations on the right part of the space - time diagram in the figure [ fig : smallfluc ] , i.e. waves that encounter first the nucleus 1 and next the nucleus 2 . naturally , there is a second contribution in which this sequence is reversed , but it is easy to guess it by symmetry at the end of the calculation . note that there is no possibility of cross - talk between these two contributions thanks to causality . this time evolution of a wave starting at @xmath13 can be divided in four steps , illustrated in the figure [ fig:4steps ] : * evolution in the region @xmath98 , before the fluctuation encounters any of the nuclei , * evolution across the trajectory of the first nucleus , * evolution in the region @xmath99 , between the two nuclei , * evolution across the trajectory of the second nucleus . the initial plane wave at @xmath13 is completely characterized by a momentum @xmath100 , a color @xmath16 , and a polarization @xmath15 , and it reads @xmath101 for every momentum @xmath100 , there are two physical polarizations , and we choose their polarization vectors to be mutually orthogonal , @xmath102 . in the rest of this section , we will consistently use the same notation , where the lower indices are the quantum numbers of the initial plane wave at @xmath103 , and the upper indices represent its lorentz and color structure at the current point @xmath104 . from now on we will work in the light cone coordinate system coordinate system only in the section [ sec : tauetacoor ] , when the fluctuation reaches the forward light - cone . ] . the region @xmath98 located below the trajectories of the two nuclei is completely trivial , since none of the nuclei has yet influenced the fluctuation . thus the equation of motion in this region are simply the free linearized yang - mills equations . in the @xmath105 gauge , the plane waves in this region read @xmath106 note that the component @xmath107 of the polarization vector is constrained by gauss s law ( i.e. the one among the four equations of motion that does not contain the derivative @xmath108 , and therefore acts as a constraint at every value of @xmath75 ) , @xmath109 that requires @xmath110 . the two physical polarizations are obtained by choosing the transverse polarization vector @xmath111 , such that @xmath112 . in the rest of this section , we will often omit the subscripts @xmath113 in the notation for the fluctuation , in order to lighten a bit the notations . the first non trivial step of the evolution of the fluctuation is to cross the trajectory of the first nucleus , on the half - line defined by @xmath114 . note that here one can not use the crossing formulas derived in @xcite , since both the structure of the background field and the gauge condition for the fluctuation are different . the first thing to realize is that the fluctuation has a non - zero @xmath115 component ( since we are in the @xmath29 gauge ) , that will induce a precession of the current @xmath116 of this nucleus . therefore , we need to first consider the current conservation equation for the nucleus , @xmath117 for the background field only , the solution reads @xmath118 to compute the change of this current @xmath119 induced by the component @xmath115 of the incoming fluctuation , we need to correct eq . ( [ eq : jcons ] ) to linear order , which gives : @xmath120 recalling the fact that @xmath116 does not depend on @xmath75 , this equation is solved by @xmath121 the operator @xmath122 should be understood as an integration with respect to @xmath75 . we can now write the linearized yang - mills equations that drive the evolution of the fluctuation across the infinitesimal region supporting the sources of the first nucleus , @xmath123 if @xmath124 is the range where the sources of this nucleus are non - zero , then the field strength @xmath125 of the background gauge potential is identically zero for @xmath69 , while inside the strip @xmath124 , its only non - zero component is @xmath126 this allows the following simplifications of eqs . ( [ e : titefluctu ] ) : @xmath127 since we only want to evolve the fluctuation from @xmath128 to @xmath129 , we are interested only in the terms of these equations that can potentially be of order @xmath130 ( i.e. the would behave as @xmath45 in the limit @xmath60 ) and therefore lead to a finite variation of the fluctuation . let us recall that @xmath131 has a finite jump in this strip , and therefore the derivative @xmath132 behaves as @xmath130 . the induced current @xmath133 behaves similarly , since it is proportional to the current @xmath116 . the first of eqs . ( [ eq : fluct1 ] ) has no @xmath108 derivative and can be seen as a constraint at fixed @xmath75 : it is nothing but gauss s law for the small fluctuation in this gauge . more explicitly , it reads @xmath134 which implies that the combination @xmath135 changes by a finite amount when going from @xmath128 to @xmath129 . therefore , the second of eqs . ( [ eq : fluct1 ] ) does not contain any term proportional to @xmath130 , which implies that @xmath136 varies infinitesimally between @xmath128 and @xmath129 . we can now simplify the third equation , by dropping all the terms that are bounded in the limit @xmath137 , which leaves only @xmath138 it is easy to integrate this equation over @xmath74 from @xmath40 to @xmath139 , and since @xmath115 is continuous in the infinitesimal integration domain , it can be taken out of the integral . this leads to @xmath140_{x^+=\epsilon } - \left[a^{j{a}}\right]_{x^+=0 } = -ig\,{{\mathcal{a}}}^{j{ab}}_{1}(x^+=\epsilon,{\bm{x_{\perp}}})\ , \frac{1}{\partial^+}a^{+{b}}_0(x)\;,\label{eq : jumponai}\ ] ] where we have used again the fact that @xmath141 does not depend on @xmath75 . the subscript @xmath40 in @xmath142 in the right hand side is used to indicate that this quantity is the free plane wave described in the previous subsection , in eqs . ( [ eq : freewaves ] ) . in order to be complete , we need to calculate also the variation of the derivative @xmath143 . indeed , although @xmath115 itself varies smoothly , this derivative may have a finite change from @xmath128 to @xmath129 . for that , we integrate gauss s law over @xmath74 from @xmath40 to @xmath139 , @xmath144_{x^+=\epsilon } - \left[\partial^- a^{+{a}}\right]_{x^+=0}= \int_0^\epsilon { \ensuremath{\mathrm{d}}}x^+\,\left[{{\mathcal{d}}}_1^{i{ab}}\partial^-a^{i{b } } + ig(\partial^-{{\mathcal{a}}}^{i{ab}}_{1})a^{i{b}}-j_1^{-{a}}\right]\;.\ ] ] using what we have just derived for @xmath145 , and using the equation of motion of the background field , @xmath146 , we obtain the following result @xmath147_{x^+=\epsilon } - \left[\partial^- a^{+{a}}\right]_{x^+=0 } & = & ig\int_0^\epsilon { \ensuremath{\mathrm{d}}}x^+\,(\partial^-{{\mathcal{a}}}^{i{ab}}_{1})\ , \left(a^{i{b}}_0-\frac{\partial^i}{\partial^+}a^{+{b}}_0\right ) \nonumber\\ & = & ig\,{{\mathcal{a}}}^{i{ab}}_{1}(x^+=\epsilon,{\bm{x_{\perp } } } ) \left(a^{i{b}}_0-\frac{\partial^i}{\partial^+}a^{+{b}}_0\right)\;. \nonumber\\ & & \label{eq : jumponap}\end{aligned}\ ] ] the formulas ( [ eq : jumponai ] ) and ( [ eq : jumponap ] ) , together with the result that @xmath115 varies smoothly while going from @xmath128 to @xmath129 , are the central result of this subsection . one can also check gauss s law at this point , which is a good test of the overall consistency of the solution @xmath148_{x^+=\epsilon } \!\!\!&=&\ ! ig\left[{{\mathcal{d}}}^{i{ac}}_1\partial^-{{\mathcal{a}}}_1^{i{cb}}\right]_\epsilon \frac{1}{\partial^+}a_0^{+{b } } + 2ig\left[\partial^-{{\mathcal{a}}}_1^{i{ab}}\right]_\epsilon a_0^{i{b } } \nonumber\\ & = & 0 \;,\label{eq : gausslaw}\end{aligned}\ ] ] because the background field @xmath149 is independent of @xmath74 for @xmath150 . in this subsection , we consider the evolution of the fluctuation after it has crossed the trajectory of the first nucleus , and before it reaches the second one . the results of the previous subsection provide the initial conditions for this evolution , and the most direct way to perform the next stage is to write the green s formula that relates the value of the fluctuation at any point in the quadrant @xmath99 to this initial data . since in this region the wilson line @xmath151 depends only on @xmath19 , the background field @xmath152 is truly a pure gauge , and the linearized equation of motion ( [ e : titefluctu ] ) for the fluctuation can be written as @xmath153 which means that the gauge rotated fluctuation @xmath154 propagates over the vacuum . one can easily obtain the following green s formula for this free evolution , @xmath155 -\big[\partial^y_{\nu}d^{\mu\nu}_{_r}(x , y)\big]\,\widetilde{a}^{+}(y ) \nonumber\\ & & \qquad\qquad\qquad\qquad + d^{\mu i}_{_r}(x , y)\stackrel{\leftrightarrow}{\partial_y^{+}}\widetilde{a}^{i}(y ) \big\ } \;,\label{eq : greenf}\end{aligned}\ ] ] where @xmath156 is the free retarded propagator in the light - cone gauge @xmath29 , whose expression in momentum space reads : @xmath157 with @xmath158 , @xmath159 . the following formulae will also prove useful later : @xmath160 the green s formula ( [ eq : greenf ] ) is valid everywhere in the region @xmath161 . one can verify that this formula conserves gauss s law , as it should . indeed , since above the @xmath129 line , @xmath66 does not depend on @xmath74 , gauss s law ( [ eq : gausslaudepx ] ) simply becomes @xmath162 which implies that @xmath163 should be independent of @xmath74 . that this is true can easily be checked thanks to eqs . ( [ eq : greenf ] ) and ( [ eq : propretprop ] ) . some technical results that are necessary in order to calculate @xmath164 for @xmath69 are derived in the appendix [ sec : appendixa ] . the results can be written in a more compact form by introducing modified polarization vectors defined by @xmath165 these new polarization vectors satisfy @xmath166 thanks to eq . ( [ eq : greenf ] ) , we obtain : @xmath167\ ; { { \mathcal{\widetilde{u}}}}^{{ac}}_1({\bm{p_{\perp}}}+{\bm{k_{\perp}}})\;\widetilde{\epsilon}^j_{{\bm k } \lambda } \nonumber\\ \widetilde{a}^{+{a}}_{{\bm k } \lambda{c}}(x)&= & 2k^+e^{ik^+x^-}{\int\frac{{\ensuremath{\mathrm{d}}}^2{\bm{p_{\perp}}}}{(2\pi)^2}}\ ; e^{i{\bm{p_{\perp}}}\cdot{\bm{x_{\perp}}}}\ ; \big[e^{i\frac{p^2}{2k^+}x^+}\frac{p^i}{p^2}\nonumber\\ & & \qquad\qquad -\big(\frac{p^i}{p^2}+\frac{k^i}{k^2}\big)\big]\ ; { { \mathcal{\widetilde{u}}}}^{{ac}}_1({\bm{p_{\perp}}}+{\bm{k_{\perp}}})\;\widetilde{\epsilon}^i_{{\bm k } \lambda}\;,\end{aligned}\ ] ] where we denote @xmath168 , @xmath169 , and where @xmath170 is the fourier transform of @xmath171 , @xmath172 undoing the gauge rotation @xmath66 to go back to @xmath22 gives : @xmath173\ ; { { \mathcal{\widetilde{u}}}}^{{bc}}_1({\bm{p_{\perp}}}+{\bm{k_{\perp}}})\;\widetilde{\epsilon}^j_{{\bm k } \lambda}\ ; , \label{eq : aistep3}\\ a^{+{a}}_{{\bm k } \lambda{c}}(x ) & = & 2k^+e^{ik^+x^-}{{\mathcal{u}}}^{{ab}{\dagger}}_1({\bm{x_{\perp}}}){\int\frac{{\ensuremath{\mathrm{d}}}^2{\bm{p_{\perp}}}}{(2\pi)^2}}\;e^{i{\bm{p_{\perp}}}\cdot{\bm{x_{\perp}}}}\ ; \big[e^{i\frac{p^2}{2k^+}x^+}\frac{p^i}{p^2 } \nonumber\\ & & \qquad\qquad -\big(\frac{p^i}{p^2}+\frac{k^i}{k^2}\big)\big]\ ; { { \mathcal{\widetilde{u}}}}^{{bc}}_1({\bm{p_{\perp}}}+{\bm{k_{\perp}}})\;\widetilde{\epsilon}^i_{{\bm k } \lambda}\;. \label{eq : apstep3}\end{aligned}\ ] ] these formulas are valid in the entire quadrant @xmath174 . the last step of the evolution is now to let the fluctuation cross the trajectory of the second nucleus . in the appendix [ app : checks ] , we perform several consistency checks on the formulas ( [ eq : aistep3]-[eq : apstep3 ] ) . the propagation of the fluctuation through the second nucleus is very similar to the situation studied in @xcite . the only difference is that in @xcite , the gauge potential of the nucleus had only an @xmath175 component , proportional to @xmath46 . in the present situation , the nucleus also has a non - zero @xmath131 . despite this important difference , the calculation can be done in a very similar way as in @xcite . in particular , the fact that we are in the @xmath29 light - cone gauge prevents any precession of the color current @xmath55 of the nucleus . the linearized yang - mills equations for the fluctuation therefore take a simpler form , without any source term in the right hand side is the generic notation for the covariant derivative in light - cone gauge , while @xmath176 and @xmath177 are built specifically with the fields @xmath72 and @xmath178 respectively . ] , @xmath179 for the component @xmath180 , using the fact that @xmath181 does not depend on @xmath74 , this gives gauss s law : @xmath182 for @xmath183 , keeping only terms that have a singular behavior in @xmath130 , we get @xmath184 which can be solved by @xmath185 finally , for @xmath186 , the equation of motion reads @xmath187 since @xmath188 in the strip @xmath189 , the previous equation can be simplified into @xmath190 the solution of this equation is known ( see @xcite ) , and agrees with gauss s law , @xmath191 it turns out in the end that the equations ( [ eq : cross2 - 1 ] ) and ( [ eq : cross2 - 2 ] ) are identical to the crossing formulas of @xcite , despite the presence of a non - vanishing @xmath192 . ( [ eq : cross2 - 1 ] ) and ( [ eq : cross2 - 2 ] ) provide the value of the fluctuation in the light - cone gauge @xmath29 on the right branch of the light - cone , just after the collision . our next task will be to convert these expressions into the fock - schwinger gauge . like for the background field itself , the first stage in this process is to go back to the situation where both nuclei are described by transverse pure gauges before the collision . to do so , we first perform a gauge transform @xmath193 which trivially affects the small fluctuations : @xmath194 after this first gauge transformation , the fluctuations on the right branch of the light - cone therefore read @xmath195 in order to check gauss s law at this point , one should recall that after this transformation the transverse background gauge potential is now @xmath196 , and therefore the covariant derivative in eq . ( [ eq : gauss3 ] ) should be modified accordingly . to go to the fock - schwinger gauge , we must perform one last gauge transformation @xmath197 , in analogy with the transformation of eq . ( [ eq : fstrans ] ) for the background field . the crucial point to note is that @xmath197 must differ from @xmath91 ( where @xmath91 is the gauge transformation used to transform the background field into the fock - schwinger gauge ) , since the fluctuation depends on @xmath18 . but since the fluctuation is small compared to the background field , @xmath197 should be close to @xmath91 , @xmath198 , with @xmath199 of order unity . the action of this gauge transformation on the background field and on the small fluctuations can be split as follows : @xmath200 where the light - cone gauge quantities are in the left hand side , and the fock - schwinger gauge quantities carry a fs subscript . when @xmath201 , the gauge rotation @xmath91 goes to @xmath202 , and therefore the transformation of the fluctuation simply becomes @xmath203 where @xmath204 is the covariant derivative constructed with the background field near @xmath47 in the fock - schwinger gauge . like in the case of the background field , we will obtain @xmath199 by requesting that @xmath205 . like for the background field , let us parameterize the @xmath206 components of the fluctuations as follows , @xmath207 from the component @xmath94 of eq . ( [ eq : fstrans - a ] ) , we get @xmath208 and in terms of the coordinates @xmath209 , we obtain @xmath210 injecting this into the @xmath211 equation and using @xmath212 gives @xmath213 where we recall that @xmath214 is given by eqs . ( [ eq : step4 ] ) and ( [ eq : apstep3 ] ) . in terms of this @xmath199 , the gauge transformation formulas for the fluctuation can also be written as @xmath215 ( we recall that all the fields and fluctuations without the fs subscript are in the light - cone gauge @xmath29 . ) it turns out that the terms @xmath216 and @xmath217 will not contribute at lowest non - zero order in @xmath17 . ( [ eq : aistep3]-[eq : apstep3 ] ) , ( [ eq : step4 ] ) , and ( [ eq : omega]-[eq : flucfsint ] ) provide the value in fock - schwinger gauge , just above the forward light - cone , for any fluctuation that started as a free plane wave in the remote past . however , for practical uses of these fluctuations in heavy ion collisions , the numerical implementation will be performed on a lattice that has a fixed extent in the rapidity @xmath18 , which is more appropriate for the description of a system in rapid expansion in the longitudinal direction . anticipating the use of this system of coordinates , it would be desirable to have an ensemble of fluctuations labeled by a quantum number which is the conjugate momentum of rapidity ( that we shall denote @xmath218 in the following ) , instead of the conjugate momentum @xmath219 of the cartesian longitudinal coordinate @xmath28 . obviously , since the two sets of fluctuations both are a basis of the vector space of all fluctuations , there must be a linear transformation to obtain one from the other . the transformation that goes from the basis of fluctuations labeled by @xmath219 to the basis of fluctuations labeled by @xmath218 reads @xmath220 where @xmath221 is the momentum rapidity . indeed , using the fact that the problem is invariant under boosts in the longitudinal direction , the fluctuation @xmath222 must depend on @xmath223 and @xmath18 only via the difference @xmath224 . changing the integration variable @xmath223 in favor of @xmath225 , we readily see that the @xmath18 dependence of the left hand side of eq . ( [ eq : y - to - nu ] ) is of the form @xmath226 . this shows that this transformation indeed leads to fluctuations that have a well defined conjugate momentum @xmath218 to the rapidity @xmath18 . we have now all the ingredients to compute the final form of the fluctuations , @xmath227 ] . after a straightforward but tedious calculation , we obtain the following formulas for the fluctuations and the corresponding electrical fields @xmath228 ( the superscript r indicates that we have only the contribution due to the wave that propagates on the right of the light - cone ) where we denote @xmath229 \ ; . \label{eq : f1}\end{aligned}\ ] ] in order to write the @xmath18 components of the gauge potential and of the electrical field as a covariant derivative acting on the function @xmath230 , we have used the following identity , @xmath231 these formulas ( to be completed by eqs . ( [ eq : centralresult2 ] ) , that give the result when the fluctuations have propagated on the other side of the light - cone ) are the central result of this work . they provide analytical expressions for fluctuations with plane wave initial conditions in the remote past , after they have propagated over the classical background field created in a heavy ion collision , in the fock - schwinger gauge and with a set of quantum numbers appropriate for a discretization on a lattice with a fixed spacing in the rapidity @xmath18 . note that in the first formula , we have written @xmath232 ( with the lorentz index up ) . the corresponding @xmath233 ( with the lorentz index down ) is obtained by multiplying by @xmath234 and therefore becomes very small when @xmath201 . in the limit @xmath201 , the fluctuations of the potentials and electrical fields behave in the same manner as their counterpart in the background field , except for the transverse electrical field @xmath235 . the background field has a transverse electrical field @xmath236 that vanishes like @xmath237 , while its fluctuations @xmath235 go to a non - zero limit when @xmath201 for all the modes @xmath238 . the dependence of these fluctuations on the classical background field is known explicitly , and is entirely contained in the wilson lines that appear in the function @xmath239 , and in the covariant derivatives that appear in some of the eqs . ( [ eq : centralresult ] ) . from this function , it is easy to obtain all the components of the fluctuations and the corresponding electrical fields thanks to eqs . ( [ eq : centralresult ] ) . the numerical evaluation of @xmath230 is rather straightforward , since it only involves a pair of fourier transforms . ( [ eq : centralresult ] ) have been derived by evolving the small fluctuations in the right part of the light cone ( crossing first the nucleus @xmath202 , and then crossing the nucleus @xmath240 ) . to this contribution should be added the contribution obtained by the other ordering of the encounters with the two nuclei , i.e. when the fluctuation propagates on the left side of the light - cone . this extra contribution is completely independent from the one we have just calculated , since by causality they can not talk to each other . this new contribution can be obtained by repeating the same steps as the ones employed so far , but now working in the @xmath241 gauge . this leads to : @xmath242 where the superscript l indicates that this is the partial wave that has propagated on the left part of the light - cone , and where we now denote @xmath243 \ ; . \label{eq : f2}\end{aligned}\ ] ] let us finally add up the results of eqs . ( [ eq : centralresult ] ) and ( [ eq : centralresult2 ] ) , in order to obtain the complete value of the fluctuations just above the forward light cone : @xmath244 with the functions @xmath245 and @xmath246 defined in eqs . ( [ eq : f1 ] ) and ( [ eq : f2 ] ) respectively . these formulas are the analogue for gluons of the eq . ( 14 ) of ref . @xcite , that had been derived for the quark mode functions . analogous formulas are also known for leptons in the qed electromagnetic background created by two colliding electrical charges @xcite . the most obvious check one can perform is that the fluctuations given by eqs . ( [ eq : finalresult ] ) satisfy the equations of motion at lowest order in @xmath17 , i.e. in an infinitesimal domain @xmath247 above the forward light - cone , @xmath248\,\alpha_{_{\rm fs}}^{i { a}}=\null&0\;.\end{aligned}\ ] ] a more stringent test is to check whether gauss s law is still satisfied , because it involves a delicate interplay between the background field and the fluctuation . we indeed find that @xmath249 note that the terms in @xmath250 , that are normally part of the gauss s law for a fluctuation in the fock - schwinger gauge , are of higher order in @xmath17 , and do not play a role here . moreover , the terms in @xmath251 in the solution satisfy independently gauss s law , as they should since they have evolved independently on each side of the light - cone . finally , one can again compute the inner product of two of the fluctuations we have obtained in ( [ eq : finalresult ] ) . the term in @xmath252 does not contribute at lowest order in @xmath17 , and therefore the inner product simply reads @xmath253 using the fact that @xmath254 , we find @xmath255 to prove that this is indeed the correct answer , let us recall what this inner product should be before we made the transformation @xmath256 , @xmath257 using @xmath258 and applying the transformation @xmath259 to the right hand side , we find that the inner product in the basis where we use the quantum number @xmath218 instead of @xmath219 should indeed be given by eq . ( [ eq : pstaueta ] ) . from eqs . ( [ eq : finalresult ] ) , ( [ eq : f1 ] ) and ( [ eq : f2 ] ) it is obvious that the most difficult and time consuming part in evaluating numerically these fields is the computation of the functions @xmath245 and @xmath246 . let us list here the main steps in their computation : * compute the wilson lines @xmath260 that represent the color charge content of the two colliding nuclei in the mclerran - venugopalan model . this is very easy for the su(2 ) gauge group , and a little more involved for su(3 ) . * compute the fourier transform ( over the transverse coordinate @xmath19 ) @xmath261 of these wilson lines . since space is discretized on a lattice , this is a discrete fourier transform , for which there are some very efficient algorithms scale as @xmath262 , while the efficient implementations scale as @xmath263 . ] . * the integration over @xmath264 in eqs . ( [ eq : f1 ] ) and ( [ eq : f2 ] ) is also a discrete fourier transform . this is the work that needs to be done to compute one of the mode functions , with given quantum numbers @xmath265 . when computing a generic perturbation to the gauge potential , one must sum over all these mode functions with random weights . since the @xmath266 dependence of the mode functions is in @xmath226 , the sum over the index @xmath218 can also be viewed as a discrete fourier transform . for computing @xmath267 configurations of these fluctuating fields , on a lattice that has @xmath268 sites in the transverse direction , and @xmath269 sites in the @xmath18 direction , the computational cost scales as @xmath270 this is the estimate for a straightforward implementation . a more careful examination of how the various steps of the calculation depend on each other leads to a better algorithm , whose cost scales as @xmath271 with @xmath272 and @xmath273 two constants . for large @xmath274 and/or @xmath267 , this is significantly better than ( [ eq : cost1 ] ) . the only subtlety arises when discretizing the first order differential operators @xmath275 and the corresponding momenta such as @xmath276 that enter in ( [ eq : finalresult ] ) . they can be discretized either as backward or forward finite differences . the choice between the two is arbitrary , and is completely determined by what kind of discretization is chosen for the derivatives in the linearized gauss law . in this work , we have performed an explicit calculation of the small fluctuations that must be superimposed to the classical cgc field in order to resum the unstable modes of the yang - mills equations . the calculation has been done from first principles , by solving the evolution equation for small fluctuations on top of the classical background field , with plane wave initial conditions in the remote past . although the intermediate steps of the evolution are done in the light - cone gauge @xmath29 , the final results are given in the fock - schwinger gauge @xmath1 . moreover , they are also given in terms of the quantum number @xmath218 , fourier conjugate of the rapidity @xmath18 , which is conserved when the background field in independent of rapidity . fluctuations expressed in terms of @xmath218 are also more suitable for a numerical implementation on a lattice with a fixed spacing in @xmath18 . our final formulas , eqs . ( [ eq : finalresult ] ) , are valid just after the collision , at proper times @xmath247 . by construction , they obey the linearized yang - mills equations with the correct initial condition at @xmath13 , and satisfy gauss s law . they have also been checked to satisfy the expected orthonormality conditions , when we compute the inner product defined in eq . ( [ eq : centralps ] ) . these formulas also turn out to be very compact , and are straightforward to implement numerically . they will be essential in the study of the behavior at early times of the strong color fields produced in high energy heavy ion collisions . indeed , it has been noticed a long time ago that certain modes are subject to the weibel instability and therefore have an exponential growth in time , and that this effect may play a crucial role in the isotropization and thermalization ( the basic idea being that it could be fast thanks to the exponential growth of these modes ) . in the cgc framework , these modes first appear at nlo , where they can give corrections that become sizable ( as large as the lo contribution ) in a short time . it is therefore important to compute these nlo corrections in order to reliably describe the behavior at early times of the fields produced in a collision . and as explained in the introduction , one needs the mode functions derived in this paper in order to perform this calculation . note that with mode functions that differ from the ones derived here , one would still trigger the weibel instabilities since any randomly chosen fluctuations are likely to have a non - zero projection on some of the unstable modes . however , the timescale for the growth of the fields depends on the initial amplitude of the fluctuations , in particular their amplitude relative to that of the background field ( this is were the dependence on the coupling constant @xmath277 comes from , since the background is of order @xmath278 while the fluctuations are of order @xmath202 ) . this timescale also depends on the relative amplitude of the various mode functions , since the growth rate of the weibel instability depends on the quantum numbers @xmath279 and @xmath218 . therefore , in order to assess correctly the early time evolution of the system , it is necessary to use the mode functions calculated here , that have been constructed in order to guarantee an accurate result up to nlo . we thank j. berges , j .- p . blaizot , k. dusling , s. jeon , l. mclerran , s. schlichting , and r. venugopalan for useful discussions related to this work . fg thanks the nuclear theory group at bnl , as well as the university of cape town and nithep , for their hospitality and financial support at various stages of this work . this research is supported by the agence nationale de la recherche project 11-bs04 - 015 - 01 . this appendix provides some formulas that are useful in order to compute the fluctuation @xmath280 via the green s formula ( [ eq : greenf ] ) : @xmath281 @xmath282 @xmath283 @xmath284\ , \widetilde{a}^{+}(y ) = -\widetilde{a}^{+}(x^+=0^+)\ ; , \end{aligned}\ ] ] @xmath285 where @xmath286 denotes the transverse fourier transform of the function @xmath287 , @xmath288 in order to obtain these formulas , one should replace the free retarded propagator by its fourier representation , and perform the integral over the energy in the complex plane thanks to the theorem of residues . first of all , one can check that the gauss s law , given in eq . ( [ eq : gausslaw2 ] ) , is indeed satisfied by eqs . ( [ eq : aistep3]-[eq : apstep3 ] ) . from these formulas , one can readily see that ( [ eq : aistep3]-[eq : apstep3 ] ) @xmath289 using eq . ( [ eq : dudag ] ) , which also implies @xmath290 and @xmath291 , we obtain ( [ eq : gausslaw2 ] ) . another non - trivial check is to compute the inner product on the @xmath292 surface ( i.e. before the fluctuation traverses the trajectory of the second nucleus ) . this inner product reads @xmath293 that is calculated by dividing the integration on @xmath74 in three pieces : @xmath294 , @xmath124 , and @xmath295 . note that the second range gives a finite contribution , despite its infinitesimal size , because @xmath296 behaves as @xmath130 . doing this calculation is tedious but straightforward . using the following identities , @xmath297 a ( somewhat lengthy ) calculation gives @xmath298 the right hand side of this inner product has a somewhat unexpected term , proportional to the integral of the color current of the first nucleus . as we shall see now , this term is correct and is the consequence of the fact that we are in a gauge where the incoming wave induces a change in this current ( because it has a non - zero @xmath115 component that induces a precession of @xmath116 ) . this induced current enters in the equation of motion for the fluctuation itself , and produces this extra term in the inner product . quite generally , in a gauge where such an induced current may appear , the equation of motion of the fluctuation reads @xmath299 because of the induced current in the right hand side , the variation of the inner product between two ( locally space - like ) surfaces @xmath300 and @xmath301 may be non zero . more specifically , one has @xmath302 where @xmath91 is the 4-dimensional domain comprised between the surfaces @xmath300 and @xmath301 . in other words , the inner product is conserved only if there are no induced currents between the two surfaces on which it is calculated . in the situation of interest to us here , the surface @xmath300 is entirely located below the backward light - cone , and the surface @xmath301 is the plane @xmath292 ( just below the trajectory of the second nucleus ) . the first term in the right hand side of eq . ( [ eq : psxpxm ] ) is nothing but @xmath303 . in the right hand side of eq . ( [ eq : consj ] ) , one can perform analytically the integral over @xmath75 , which gives the extra term in the right hand side of eq . ( [ eq : psxpxm ] ) . in this appendix , we derive the transformation from the @xmath29 gauge to the fock - schwinger gauge in the case of fluctuations propagating in the vacuum ( i.e. when the background field is zero ) . in this situation , the fluctuations in light - cone gauge are completely trivial , of the form @xmath304 the transformation to fock - schwinger gauge can be done via eq . ( [ eq : flucfsint ] ) , simplified here thanks to the absence of background field @xmath305 with @xmath306 one obtains easily the following explicit expression for @xmath199 , @xmath307 this leads to @xmath308\ ; , \label{eq : ai - vac } \\ \alpha_{_{\rm fs}}^{\eta}(\tau,\eta,{\bm{x_{\perp } } } ) & = & \frac{k^i\epsilon^i}{k_\perp\tau } \,e^{-i{\bm{k_{\perp}}}\cdot{\bm{x_{\perp}}}}\;\frac{1}{\cosh(y-\eta ) } \big [ e^{ik_\perp\tau\cosh(y-\eta ) } \nonumber\\ & & \qquad\qquad\qquad\qquad + i\frac{e^{ik_\perp\tau\cosh(y-\eta)}-1}{k_\perp\tau\cosh(y-\eta ) } \big]\ ; . \label{eq : aeta - vac}\end{aligned}\ ] ] the final step is to go from the quantum number @xmath219 to the fourier conjugate of rapidity , @xmath218 . this is achieved by a fourier transform of the @xmath223 dependence , @xmath309 after this transformation , eqs . ( [ eq : ai - vac ] ) and ( [ eq : aeta - vac ] ) become respectively @xmath310 \label{eq : ai - vac - nu}\end{aligned}\ ] ] and @xmath311 where @xmath312 is the hankel function defined in terms of the bessel functions as @xmath313 . one can check that the two vacuum solutions ) . ] found in ref . @xcite ( eqs . ( 79 ) and ( 84 ) ) can be rewritten as linear combinations of the present eqs . ( [ eq : ai - vac - nu]-[eq : aeta - vac - nu ] ) . note that if we had started from the vacuum plane wave solutions in the @xmath241 light - cone gauge instead , we would have @xmath314 and the function @xmath315 ( we denote with a prime all the quantities obtained from this alternate starting point ) that defines the transformation to the fock - schwinger gauge would be @xmath316 consequently , the vacuum fluctuations in the fock - schwinger gauge would be replaced by @xmath317\ ; , \label{eq : ai - vac-+ } \\ \alpha_{_{\rm fs}}^{'\eta}(\tau,\eta,{\bm{x_{\perp } } } ) & = & -\frac{k^i\epsilon^{'i}}{k_\perp\tau } \,e^{-i{\bm{k_{\perp}}}\cdot{\bm{x_{\perp}}}}\;\frac{1}{\cosh(y-\eta ) } \big [ e^{ik_\perp\tau\cosh(y-\eta ) } \nonumber\\ & & \qquad\qquad\qquad\qquad + i\frac{e^{ik_\perp\tau\cosh(y-\eta)}-1}{k_\perp\tau\cosh(y-\eta ) } \big]\ ; . \label{eq : aeta - vac-+}\end{aligned}\ ] ] recalling that the polarization vectors @xmath111 in the @xmath29 gauge and @xmath318 in the @xmath241 gauge are related by @xmath319 it is trivial to see that @xmath320 and @xmath321 differ by a gauge transformation that does not depend on @xmath17 . in other words , they correspond to two different ways of fixing the residual gauge freedom in the fock - schwinger gauge . this is why the vacuum limit ( @xmath322 ) of eqs . ( [ eq : finalresult ] ) does not give precisely eqs . ( [ eq : ai - vac - nu]-[eq : aeta - vac - nu ] ) . indeed , eqs . ( [ eq : finalresult ] ) correspond to fixing this residual gauge freedom independently on the two branches of the light - cone .

we discuss a planar variant of the kakeya maximal function in the setting of a vector space over a finite field . using methods from incidence combinatorics , we demonstrate that the operator is bounded from to when and .

math0609337

the kakeya conjecture is a long standing open problem in the field of geometric combinatorics which is concerned with the extent to which a large direction - separated family of thin tubes can be compressed into a small space . there are essentially two formulations of the conjecture : one geometric and the other analytic . in order to state the geometric formulation of the conjecture , we shall need the following fundamental definition : a set @xmath4 is said to be a _ kakeya set _ if for any direction @xmath5 , there exists a unit line segment @xmath6 parallel to @xmath7 such that @xmath8 . the kakeya conjecture is concerned with the dimension of such objects . explicitly , [ kakeyaconjecture : dim ] if @xmath4 is a kakeya set , then @xmath9 . strictly speaking , conjecture [ kakeyaconjecture : dim ] should be interpreted as three separate conjectures , as one can consider the hausdorff dimension ( denoted @xmath10 ) , lower minkowski dimension ( denoted @xmath11 ) or upper minkowski dimension ( denoted @xmath12 ) of such sets ; these different notions of dimension are discussed at great length in most texts on geometric measure theory , for example @xcite and @xcite . by definition , one has @xmath13 for any set @xmath4 . therefore , any progress on the hausdorff version of the kakeya conjecture will immediately imply progress on the other two versions . it is not known , however , whether all three versions of the conjecture are actually equivalent . furthermore , the best known results concerning the conjecture differ for each of the various notions of dimension ( see @xcite and @xcite for examples of this phenomenon ) . the analytic formulation of the conjecture concerns the boundedness of the kakeya maximal function : given a function @xmath14 and a number @xmath15 , we may define the _ kakeya maximal function _ as @xmath16 where @xmath17 denotes the tube of dimensions @xmath18 centered at the point @xmath19 , oriented in the direction @xmath7 . with this definition , we may state the analytic version of the conjecture : for all @xmath20 , and @xmath21 , one has @xmath22 for @xmath23 , where @xmath24 denotes the rotationally invariant probability measure on the unit sphere . it is now known , thanks to an observation of bourgain @xcite , that any progress towards the resolution of the analytic formulation automatically implies progress towards the geometric formulation . in particular , if the kakeya maximal function is bounded on @xmath0 , it follows that kakeya sets have hausdorff ( and , hence , upper and lower minkowski ) dimension at least @xmath25 . modern investigations of this conjecture essentially date back to the 1970 s , when davies @xcite proved that any kakeya set in @xmath26 has hausdorff dimension ( and , hence , upper and lower minkowski dimension ) @xmath27 . later , crdoba @xcite proved the analogous result for the maximal function in @xmath26 . though these two results essentially resolved all questions concerning the kakeya conjecture in @xmath27 dimensions , the conjecture remains open in dimensions @xmath28 and higher . the @xmath29 problem is a variant of the kakeya problem in which one replaces lines with @xmath30-planes . in order to formally describe the @xmath29 problem , we need a few definitions . a set @xmath4 is said to be an @xmath29 _ set _ if for any @xmath30-dimensional subspace @xmath31 , there exists a @xmath30-dimensional unit cube @xmath32 parallel to @xmath33 such that @xmath34 . clearly , an @xmath35 set is simply a kakeya set . the geometric conjecture associated with these sets is essentially the same as the kakeya conjecture : if @xmath4 is an @xmath29 set , then @xmath36 . once again , dimension can be taken as upper minkowski , lower minkowski or hausdorff . one can also define the @xmath29 maximal function : given positive integers @xmath37 , let @xmath38 denote the grassmannian manifold of all @xmath30-dimensional subspaces of @xmath39 . for a function @xmath14 and a real number @xmath40 , we define the @xmath29 _ maximal function _ as @xmath41 where @xmath42 denotes the @xmath43-neighborhood of the @xmath30-dimensional unit cube parallel to @xmath33 and centered at @xmath44 . using this definition , the analytic version of the @xmath29 conjecture is then for all @xmath20 , and @xmath45 , one has @xmath46 for @xmath47 , where @xmath48 denotes the rotationally invariant probability measure on @xmath38 . though a large amount of work has been devoted to the @xmath29 problems mentioned here , as well as various other @xmath30-plane transforms ( for example , see the work of strichartz @xcite , christ @xcite and drury @xcite ) , the previous results most relevant to this paper are those due to oberlin and stein @xcite , themis mitsis @xcite,@xcite , as well as wolff s bound for the kakeya maximal function @xcite we investigate the problem in the setting of vector spaces over finite fields ; this slightly unconventional choice of vector spaces was inspired by a recent paper due to mockenhaupt and tao @xcite . when working in a discrete setting , one is allowed to focus almost exclusively on the geometric and combinatorial aspects of the problem . in particular , problems which arise in the study of kakeya - type problems in euclidean spaces due to multiplicity of scales ( in particular , the troublesome behavior of @xmath43-tubes which intersect at small angles ) are essentially non - existent in the discrete setting . furthermore , working in this setting allows one to import useful tools from other areas of mathematics , such as analytic number theory ( gauss sums , kloosterman sums ) and incidence geometry . there are , of course , a few negative side - effects resulting from the discretization of the problem . for example , taylor approximations of surfaces do nt make sense in @xmath49 , and there is no chance to make use of arguments requiring induction on scales . before stating the main result , a bit of notation is necessary . given a finite field @xmath50 , we let @xmath38 denote the ( grassmannian ) set of all @xmath30-dimensional subspaces of the vector space @xmath49 . we will always assume that @xmath50 is a very large finite field . aside from the fact that the results discussed here are somewhat trivial when @xmath50 is small , working with large fields allows one to develop an intuition regarding the minkowski dimension of the geometric objects we will be studying . when considering lebesgue spaces of functions defined on @xmath51 , we endow this set with a normalized counting measure @xmath52 , so that for any subset @xmath53 , @xmath54 ( where @xmath55 simply denotes cardinality ) . the vector space @xmath49 is endowed with a standard counting measure @xmath56 . with this notation , we may define the @xmath29 maximal function operator as @xmath57 where @xmath58 is some real - valued function on @xmath49 , and @xmath59 . our primary objective here will be to determine a class of lebesgue spaces on which this operator is `` bounded '' . of course , since we are working in vector spaces over finite fields , we wo nt be encountering any divergent integrals . our definition of boundedness will be as follows : [ definition ] let @xmath60 denote the smallest quantity such that @xmath61 holds for all functions @xmath62 . we say that @xmath63 is _ bounded _ from @xmath0 to @xmath1 if @xmath64 as @xmath65 . the notation @xmath66 in the above definition means that for all @xmath20 , there exists some constant @xmath67 such that @xmath68 . similarly , @xmath69 indicates @xmath70 for some constant @xmath71 independent of the field @xmath50 . also , we say @xmath72 ( respectively @xmath73 ) if @xmath74 and @xmath75 ( respectively @xmath76 and @xmath77 ) . one can immediately observe a few necessary conditions on @xmath25 and @xmath78 by examining certain counterexamples . for example , observing the function @xmath79 shows that one must have @xmath80 . also , considering the function @xmath81 for some @xmath30-plane @xmath82 will reveal that one must have @xmath83 . it is conjectured that these necessary conditions are also sufficient , though only partial results are currently available . our main result will be a proof of the following result for the @xmath29 maximal function by means of incidence combinatorial techniques : [ mainthm ] let @xmath84 , @xmath30 be positive integers such that @xmath85 . the operator @xmath86 is bounded when @xmath87 and @xmath88 . as an immediate corollary , we have the following result concerning the geometric version of the @xmath29 problem : let @xmath89 be an @xmath29 set , where @xmath85 are integers . then , @xmath90 it should be noted that theorem [ mainthm ] is in no way the best known result for the @xmath29 problem . in the euclidean case , bourgain @xcite showed that @xmath29 sets have positive @xmath84-dimensional lebesgue measure whenever @xmath91 . furthermore , oberlin ( @xcite , @xcite ) has generalized this result to the corresponding maximal function estimate , and improved the result for certain values of @xmath84 and @xmath30 by making use of recent advances concerning the x - ray transform ( @xcite , @xcite , @xcite ) . the novelty of theorem [ mainthm ] is in its proof ; the proof presented here is entirely combinatorial , whereas the work of bourgain and oberlin incorporates tools from fourier analysis , thus leading to stronger results . this result might be interpreted as an extension of the work of wolff @xcite , who showed that @xmath35 sets have cardinality @xmath92 , and oberlin and stein @xcite , whose work implies that @xmath93 sets have cardinality @xmath94 . a euclidean version of the case @xmath95 was first proven by mitsis @xcite . his proof is an adaptation of the hairbrush " construction used by wolff to demonstrate a bound for the kakeya maximal function . for our purposes , we shall be following the example set by mockenhaupt and tao @xcite who find an alternate proof of wolff s kakeya result by incidence combinatorial methods . furthermore , the two methods are , in a certain sense,equivalent " . for more details , the reader is directed to @xcite . ] . the paper will be organized as follows : in section [ prelim ] , we will introduce the main combinatorial tools which will be used in later computations , and establish a correspondence between bounds on the number of incidences between points and @xmath30-planes and estimates for the @xmath29 maximal function as defined in equation . next , in section [ reductions ] , we will prove a generalized version of wolff s two - ends reduction " . this reduction will allow us to prove non - trivial incidence bounds without being hindered by the existence of certain pathological configurations of points and @xmath30-planes . the remaining sections will be devoted to proving an incidence bound between points and @xmath30-planes . we shall arrive at this bound by estimating the number of @xmath96-simplices arising from a configuration of points @xmath97 and @xmath30-planes @xmath98 which have vertices from @xmath97 and faces from @xmath98 . the lower bound for the number of such simplices is computed by an inductive procedure , and is addressed in sections [ firstproof ] and [ indreg ] . the upper bound for the number of simplices is computed in section [ upperbound ] . i would like to acknowledge terence tao for introducing me to this problem and for his guidance . also , i would like to thank richard oberlin for his careful reading of , and helpful comments concerning this paper . as the proof of the main theorem in this paper will be largely combinatorial , our first task will be to establish some machinery designed to translate incidence combinatorial results into maximal function estimates , and vice versa . an important combinatorial tool which will be used extensively the following set - theoretical version of cauchy - schwarz : [ stcs ] let @xmath99 and @xmath100 be finite sets with some relation @xmath101 between elements @xmath102 and @xmath103 . then , @xmath104 to see how theorem [ stcs ] follows from the traditional cauchy - schwarz inequality , define a function @xmath58 on the set @xmath100 as @xmath105 next , using cauchy - schwarz , we have @xmath106 where the @xmath107 norm is computed with respect to counting measure on the finite set @xmath100 . theorem [ stcs ] then follows by observing that @xmath108 and @xmath109 in practice , the sets @xmath99 and @xmath100 will always denote certain configurations of points , planes and lines , and the relation @xmath110 will denote some sort of geometric incidence . for example , given a configuration of points @xmath97 and @xmath30-planes @xmath98 , we may obtain a lower bound on the number of `` double point - plane incidences '' in terms of the numbers of point - plane incidences and planes : @xmath111 of course , this idea can be generalized into a version of hlder s inequality : @xmath112 the set appearing on the right hand side of both and is known as the _ incidence set _ associated to the points @xmath97 and @xmath30-planes @xmath98 . as this set will be appearing quite often throughout this paper , we ll introduce the following notation : @xmath113 often , the arguments @xmath97 and @xmath98 will be suppressed when they are obvious from the context . the following correlation between maximal function estimates and incidence bounds is used in @xcite to demonstrate a bound for the kakeya maximal function , and we generalize it to suit our purposes as follows : [ max - ic ] given exponents @xmath114 , the bound @xmath64 holds if and only if given any collection of points @xmath115 and any direction separated collection of @xmath30-planes @xmath98 contained in @xmath49 , the following incidence bound holds : @xmath116 before proceeding with the proof , note that the conjectured best possible incidence bound ( corresponding to the necessary conditions @xmath117 and @xmath118 ) is @xmath119 this expression will appear several times throughout the course of the paper . first , we assume @xmath120 . let @xmath97 and @xmath98 be as in the statement of the proposition , and let @xmath121 denote the direction set of @xmath98 ( i.e. each @xmath122 is a parallel translate of exactly one element of @xmath123 ) . then , the result simply follows from hlder s inequality : @xmath124 to prove the converse , it suffices ( by duality ) to show @xmath125 for any pair of functions @xmath58 and @xmath126 ( defined on @xmath49 and @xmath51 , respectively ) , where @xmath127 is some function which translates elements of @xmath51 to affine position . since out notation allows us to lose factors of @xmath128 , we may employ the dyadic pigeonhole principle , and assume that @xmath129 and @xmath130 for sets @xmath115 , and @xmath121 . after making these simplifications , we must now show @xmath131 now , simply define the collection of @xmath30-planes to be @xmath132 . the above equation then follows directly from . in this section , we address an issue which often causes problems when counting incidences between points and ( generally speaking ) algebraic varieties of dimension greater than @xmath133 : it is quite easy to construct a large set of points @xmath97 and a large set of planes @xmath98 such that every point is contained in every plane . the classic example of this behavior can be seen in the following example : a _ type-@xmath134 degenerate configuration _ is a configuration of a set of points @xmath97 and a set of 2-planes @xmath98 such that all of the points in @xmath97 lie on some line @xmath135 , and all of the planes in @xmath98 contain the line @xmath135 . degenerate configuration . ] [ 21degpic ] this configuration is notable because the incidence set associated to it actually attains the worst possible upper bound , @xmath136 . so , it seems that this counterexample should prohibit us from obtaining any sort of nontrivial upper bound on the size of the incidence set which would hold for all possible collections of points and planes . for higher values of @xmath30 , there are even more possible degenerate configurations of this type ; given a positive integer @xmath137 , one can consider a family of @xmath138 points which all lie on some affine @xmath139-plane @xmath140 , and a direction separated collection @xmath141 @xmath30-planes which all contain @xmath140 . we shall call such a configuration _ type-@xmath142 degenerate_. because of the existence of such configurations , it is necessary to place some ( presumably mild ) restrictions on distributions of points and/or planes in question . in recent work concerning incidences between points and surfaces , ( for example , aba and solymosi @xcite , and elekes and tth @xcite ) , point sets were assumed to have certain uniformity properties to prohibit pathological configurations . for our purposes , however , we shall work with arbitrary distributions of points in @xmath49 , and exploit the fact that any family of planes in consideration must be direction separated . the following proposition acts in the same way as the `` two - ends reduction '' first used by wolff @xcite to improve estimates for the kakeya maximal function . we will eliminate the threat posed by configurations in which points tend to cluster along low dimensional subsets of @xmath30-planes from @xmath98 by eliminating the possibility of having sparsely populated @xmath30-planes . the actual proof of this new version of the reduction for the @xmath29 problem is quite different from wolff s ; it is more closely related to drury s work on the x - ray transform @xcite . [ hyperred ] let @xmath115 be a collection of points , and @xmath98 a direction separated collection @xmath30-planes in @xmath49 such that @xmath143 and @xmath144 then , the best possible incidence bound holds : @xmath145 the proof of this proposition will be inductive . we shall first show that holds under the assumption @xmath146 ( hence it is safe to assume @xmath147 ) , and then show that the result holds when @xmath148 for any integer @xmath149 . suppose @xmath150 . then , making use of the trivial estimates @xmath151 and @xmath152 , we have @xmath153 therefore , we may assume @xmath147 . next , assume that @xmath154 for some positive integer @xmath155 . in order to arrive at the desired conclusion , we will estimate the size of the following set @xmath156 using hlder s inequality , we arrive at a lower bound of @xmath157 in order to compute a corresponding upper bound , we break the set @xmath158 into a union of disjoint subsets in the following manner : @xmath159 where @xmath160=j\}\ ] ] now we estimate each of these subsets separately . clearly , we have @xmath161 next , when @xmath162 , we have @xmath163 choices for the @xmath30-plane , @xmath164 choices for a @xmath165-tuple of points which spans @xmath166 $ ] , and @xmath167 choices for the remaining points . therefore , we have @xmath168 finally , in order to estimate @xmath169 , we observe that there are @xmath170 choices for the @xmath171-tuple of points , and ( because the collection of @xmath30-planes is direction separated ) there are at most @xmath172 @xmath30-planes for the collection @xmath98 which can contain the @xmath139-dimensional affine space spanned by the already chosen @xmath171-tuple of points . putting together all of these observations , we have @xmath173 furthermore , since we are assuming @xmath174 , we have @xmath175 whenever @xmath162 . similarly , @xmath176 . therefore , the first term on the right hand side of equation is dominant , and we have @xmath177 a bit of algebraic manipulation then leads us to the following incidence bound : @xmath178 finally , we do a bit more algebra , and observe that @xmath179 since this is true whenever @xmath180 , we are done . it should be noted that in the special case when @xmath181 and @xmath182 , all four of the quantities @xmath183 , @xmath184 , @xmath185 and @xmath186 coincide . these values for @xmath187 and @xmath163 are , of course , the same values that one would see in the construction of a type-@xmath142 degenerate configuration , as was described earlier in this section . these configurations , therefore , are examples of situations in which the worst possible incidence bound coincides with the best possible incidence bound . now that we have eliminated the potential problems caused by the existence of sparsely populated @xmath30-planes , we are in a position to formulate a more flexible version of proposition [ max - ic ] . in order to make use of proposition [ hyperred ] in what follows , we shall need to establish some notation . given a collection of points @xmath97 and a direction separated collection of @xmath30-planes @xmath98 , it is clear that the following equality holds : @xmath188 therefore , by pigeonholing , there exists a subcollection @xmath189 such that @xmath190 the only control we have over the size of the set @xmath191 is the trivial estimate @xmath192 . furthermore , letting @xmath193 denote the incidence set @xmath194 , we have @xmath195 with this notation in place , we may now state and prove the following proposition : [ max - ick ] let @xmath196 be real numbers such that @xmath197 . if the following incidence bound holds @xmath198 then the statement @xmath120 is true with @xmath199 and + @xmath200 . [ bados ] unfortunately , the condition @xmath201 holds only when @xmath202 with the values of @xmath19 , @xmath203 and @xmath204 to be computed in the next section . this technicality prevents us from using the methods described here to demonstrate an entirely geometric proof of the result of oberlin and stein . to start with , if the term @xmath205 dominates , then we arrive at the best possible bound : @xmath206 so , we may also assume @xmath207 . if the term @xmath208 dominates the right - hand side of equation , then ( since our refinements allow us to assume equation holds ) proposition [ hyperred ] shows that we obtain the best possible incidence bound . therefore , we may assume @xmath209 . to finish the proof , assume @xmath210 . since we have refined our sets in such a way that @xmath211 and @xmath192 , we have @xmath212 . taking a convex combination ( making use of the assumption @xmath201 ) of this estimate with the trivial estimate @xmath213 , and applying proposition [ max - ic ] completes the proof . the object of the next three sections will be to prove the following incidence bound : let @xmath115 be a collection of points and @xmath98 a direction separated collection of @xmath30-planes contained in @xmath49 . then , @xmath214 where the sets @xmath193 and @xmath191 are as described in equations and . once this incidence bound has been demonstrated , we may apply proposition [ max - ick ] to obtain the desired bound for the @xmath29 maximal function . the proof presented here is inspired by an argument found in @xcite to prove a similar incidence bound in the case @xmath215 ; for their result , they obtain upper and lower bounds on the number of triangles appearing in the configuration of points and lines ( with sides from the given collection of lines , and vertices from the given collection of points ) . for our purposes here , we will be obtaining upper and lower bounds on the number of @xmath96-simplices appearing in the configuration , each with @xmath216 @xmath30-dimensional faces from the collection @xmath191 , and @xmath216 vertices coming from the collection of points @xmath97 . the proof of the lower bound will be an induction on @xmath30 ( making use of the fact that a @xmath96-simplex is the cone of @xmath30-simplex ) . the upper bound will make use of the direction separatedness of the family of @xmath30-planes . we begin by demonstrating a lower bound for the number of @xmath96-simplices arising from collections of points @xmath97 and @xmath30-planes @xmath191 satisfying certain hypotheses . the actual statement of this lower bound will require quite a bit of terminology and notation , so we begin with a few definitions . let @xmath84 and @xmath30 be integers such that @xmath217 . given a collection of points @xmath115 and @xmath30-planes @xmath191 , we say that this configuration of points and planes satisfies hypothesis h1(@xmath30,@xmath97,@xmath191 ) if @xmath218 let @xmath84 and @xmath30 be integers such that @xmath217 . given a collection of points @xmath115 and @xmath30-planes @xmath191 , we say that this configuration of points and planes satisfies hypothesis h2(@xmath30,@xmath97,@xmath191 ) if @xmath219 observe that these hypotheses are derived from the error terms from the main incidence bound we aim to demonstrate . as the computation of the lower bound the collection of simplices will be inductive , we shall need to investigate the behavior of these hypotheses for varying values of @xmath30 . also , observe that we have not yet mentioned the hypothesis that the set @xmath191 be direction separated ; this hypothesis will not be used until section [ upperbound ] when computing the upper bound on the collection of simplices , and its absence from the lower bound computation greatly simplifies the induction process . in order to carry out the induction , we shall need to construct a few sets from our base sets of @xmath97 , @xmath191 and @xmath193 . the first step towards carrying out these constructions will be observing that the assumptions @xmath220 for each @xmath221 and @xmath222 permit us to make use of proposition [ hyperred ] . this allows us to assume that @xmath223 . furthermore , we may use the simple argument found in the proof of proposition [ max - ick ] to assume that @xmath224 . in other words , the statements h1(@xmath30,@xmath97,@xmath191 ) and h2(@xmath30,@xmath97,@xmath191 ) are true . next , we construct the following set : @xmath225=k-1 \}\ ] ] observe that hlder s inequality , along with proposition [ hyperred ] , gives us a lower bound on @xmath226 of @xmath227 this set also needs to be refined a bit ; we wish to remove elements of the set @xmath228 which are degenerate in the sense that the @xmath229-plane spanned by the @xmath30-tuple @xmath230 carries a small number of points . to make this refinement , we introduce a relation @xmath110 on the set @xmath231 defined as @xmath232$}.\ ] ] since each @xmath221 contains @xmath233 points from @xmath97 , one should expect each @xmath229-dimensional slice of @xmath33 to contain roughly @xmath234 points from @xmath97 . with this intuition , we may refine @xmath228 as follows : @xmath235 to see that this refinement is okay , we need only show that the set @xmath236 is small . a simple estimate will take care of this : we have @xmath237 choices for the @xmath30-plane , each of these @xmath30-planes contains @xmath238 hyperplanes , and each element of the set @xmath239 can have at most @xmath240 @xmath30-tuples of points on any of these hyperplanes . so : @xmath241 this allows us to replace @xmath228 with @xmath242 without doing much harm . next , we will take pairs of elements of the set @xmath242 , and identify them along their @xmath30-tuples : @xmath243 an application of cauchy - schwarz gives us a lower bound on the size of this set : @xmath244 this set also needs to be refined ; we wish to remove elements from @xmath245 which are degenerate in the sense that @xmath246 . such an element of @xmath245 is , in fact , and element of @xmath242 , so this refinement will be okay if we can show that @xmath247 . in order to show this , we make use of hypothesis h1(@xmath30,@xmath97,@xmath191 ) : @xmath248 so , if we define the set @xmath249 as @xmath250 then we have @xmath251 . next , to each element of @xmath249 we wish to add a point from @xmath97 which lives in @xmath252 $ ] . @xmath253)\}\ ] ] since there are @xmath254 points on each plane , and they do not cluster along low dimensional subspaces by proposition [ hyperred ] , we have the following lower bound @xmath255 one last refinement is needed before we can proceed with the construction of simplices . for a given pair @xmath256 such that @xmath257 , we define a function @xmath258 as @xmath259 then , @xmath260 so we may pigeonhole this sum to find a family of plane - point pairs @xmath261 @xmath262 as was the case for the parameter @xmath237 , we have no control over the parameter @xmath263 other than the trivial upper bound @xmath264 . it is somewhat disconcerting that the value of the function @xmath258 may be very small on our set @xmath265 . this potential problem , however , will be ruled out in section [ indreg ] . in summary , we have made the following constructions : @xmath266=k-1\\ \lvert p \cap [ p_1,\dotsc , p_k ] \rvert \geq \frac{\ntili}{10\ntilpi\f } \end{array } \right\}\\ { } & \quad \ntilik \gtrsim \frac{\ntili^{k}}{\ntilpi^{k-1}}\\ v_k & = \left\{((\pi_0,p_1,\dotsc , p_k),(\pi , q_1,\dotsc , q_k))\in \tilik \times \tilik : \begin{array}{l } p_i = q_i \text { for each $ i$}\\ \pi \neq \pi_0 \end{array } \right\}\\ { } & \quad \lvert v_k \rvert \gtrsim \frac{\ntilik^2}{\p^k}\\ v_{k , p } & = \{((\pi_0,\pi , p_1,\dotsc , p_k),x ) \in v_k \times p : x \in p \cap ( \pi \smallsetminus [ p_1,\dotsc , p_k])\}\\ { } & \quad \lvert v_{k , p } \rvert \gtrapprox \vert v_k \rvert \frac{\ntili}{\ntilpi}\\ f(\pi_0,x ) & = \lvert\{(\pi , p_1,\dotsc , p_k ) \in \tilik:((\pi_0,\pi , p_1,\dotsc , p_k),x ) \in v_{k , p}\}\rvert\\ { } & \quad \sum f(\pi_0,x ) = \nvkp \gtrapprox \nvk\frac{\ntili}{\ntilpi}\\ d(p,\tilpi ) & = \{(\pi_0,x ) \in \tilpi \times p : x \notin \pi_0 , f(\pi_0,x ) \gtrapprox \nvk\frac{\ntili}{\ntilpi \lvert d(p,\tilpi ) \rvert}\}\\ { } & \quad \lvert d(p,\tilpi ) \rvert \leq \p\ntilpi\end{aligned}\ ] ] with this notation established , we may finally begin to compute a lower bound on the number of simplices . [ lowerlemma ] given an arrangement of points @xmath97 and @xmath30-planes @xmath191 satisfying hypotheses h1(@xmath30,@xmath97,@xmath191 ) and h2(@xmath30,@xmath97,@xmath191 ) , let @xmath267 denote the set of @xmath96-simplices in @xmath49 with faces from @xmath191 and vertices from @xmath97 . then , @xmath268 where the set @xmath249 is as defined in equation . [ logloss ] heuristically , one can easily compute this lower bound on @xmath269 . since a point from @xmath97 and a @xmath30-plane from @xmath98 are incident with probability @xmath270 , and a simplex consists of @xmath271 incidences amongst @xmath272 points and @xmath272 @xmath30-planes , it follows that @xmath273 under the unrealistic assumption that all events of point - plane incidence are independent . to validate this heuristic rigorously , we shall make several refinements in order to ensure a certain amount of geometric regularity regarding the distribution of points within each @xmath30-plane , hence creating a logarithmic loss in the end result . if a new procedure for demonstrating the heuristically obvious lower bound can be established , then the incidence bound might be slightly improved ( i.e. we can replace the symbol @xmath274 with @xmath275 ) . the proof of this lemma will be a somewhat complicated induction on @xmath30 . for the sake of clarity , it will be useful to briefly outline the proof before proceeding . given a collection of points @xmath97 and @xmath30-planes @xmath191 , let c1(@xmath30,@xmath97,@xmath191 ) denote the statement that the conclusion of lemma [ lowerlemma ] holds . given a collection of points @xmath97 and @xmath30-planes @xmath191 , let c2(@xmath30,@xmath97,@xmath191 ) denote the statement that @xmath276 with this terminology established , we may describe the inductive procedure . first , we will prove that c2(@xmath277,@xmath278,@xmath279 ) is true for any collection of points @xmath278 and `` affine @xmath277-planes '' ( i.e. points ) in an ambient space of any dimension . though the case @xmath280 is a bit absurd , it will work ( formally ) as the base of the induction . the reader who is unsatisfied with this may choose to refer to the arguments found in @xcite , and begin the induction at @xmath281 . next , we make the following essentially trivial observation : let @xmath30 be an integer , @xmath282 . given a collection of points @xmath278 , and a collection of @xmath30-planes @xmath279 ( living in an ambient space of any dimension strictly larger than @xmath30 ) , we have the following implication : @xmath283 the proof of this sub - lemma is simply a computation making use of bounds we have already demonstrated . use hypotheses h1(@xmath30,@xmath278,@xmath279 ) and h2(@xmath30,@xmath278,@xmath279 ) to assume all of the regularity conditions needed for the estimates @xmath284 where the sets @xmath285 and @xmath286 are defined as in equations and ( respectively ) . then , simply insert these bounds into equation . the final ingredient needed to prove lemma [ lowerlemma ] is a statement the c2(@xmath287,@xmath288,@xmath289 ) @xmath290 c1(@xmath30,@xmath97,@xmath191 ) ( under suitable hypotheses ) , where the sets @xmath288 and @xmath289 are collections of points and @xmath229-planes ( respectively ) to be defined later . with this argument in place , the proof will then , schematically , look like the following : @xmath291 unfortunately , there are two technicalities we must dispense with before beginning with the above program . the method described will not work unless we can ensure that simplices formed at every level of the induction are non - degenerate . in order to avoid potential non - degeneracies , we must show that the hypotheses h1(@xmath30,@xmath97,@xmath191 ) and h2(@xmath30,@xmath97,@xmath191 ) imply a new pair of hypotheses h1(@xmath287,@xmath292,@xmath279 ) and h2(@xmath287,@xmath292,@xmath279 ) for suitably defined sets of points @xmath292 , and @xmath229-planes @xmath279 ( and so on for @xmath293-planes , @xmath294-planes , etc ... ) . in order to simplify the exposition , we will assume , for the moment , that the regularity hypotheses hold at every level of the induction . then , in section [ indreg ] , we will verify that this is indeed the case . to begin the proof , we let @xmath278 denote a collection of points , and @xmath279 a collection of affine @xmath277-planes in some ambient space ( of any finite dimension @xmath295 ) . we wish to show that the conclusion c2(@xmath277,@xmath278,@xmath279 ) holds ; this essentially means that we must show @xmath296 this estimate , however , is trivial after observing that @xmath297 , and @xmath298 denotes the set of all line segments with distinct endpoints in the set @xmath299 . in order to begin the inductive part of the proof , we will need to introduce a family of @xmath229-planes . given a set of points @xmath97 , and @xmath30-planes @xmath191 satisfying h1(@xmath30,@xmath97,@xmath191 ) and h2(@xmath30,@xmath97,@xmath191 ) , fix an element @xmath300 ( as defined in equation ) . for this particular pair , we define a family of @xmath229-planes @xmath289 defined as : @xmath301 , ( ( \pi_0,\pi , p_1,\dotsc , p_k),x ) \in v_{k , p } \end{array}\right\}\ ] ] also , we need to identify a class of points @xmath302 which will interact with the set @xmath289 . in section [ indreg ] we will verify that there exists a set @xmath303 such that @xmath304 and the hypotheses h1(@xmath287,@xmath288,@xmath289 ) and h2(@xmath287,@xmath288,@xmath289 ) are true for each pair @xmath305 . therefore , we may assume that the configurations of points and planes @xmath306 induced from this large subclass enjoy the same regularity properties as @xmath307 . next , we observe that the set of @xmath96-simplices in @xmath308 which contain the @xmath30-plane @xmath309 as a face and the point @xmath310 as a vertex are in one to one correspondence with the set of @xmath30-simplices contained in @xmath309 whose faces belong to @xmath289 and vertices belong to @xmath288 . in other words , each such @xmath96-simplex is seen as the cone of some @xmath30-simplex contained in @xmath309 . so , once we have a lower bound on the number of @xmath30-simplices contained in @xmath309 , we will obtain a lower bound on the number of @xmath96-simplices containing @xmath309 and @xmath310 . we formalize this as follows : let @xmath311 denote the incidence set arising from the collection of points @xmath312 and the collection of @xmath229-planes @xmath289 . also , define the set @xmath313=k-1 \end{array } \right\}\ ] ] next , we observe that ( by the construction of the sets @xmath289 and @xmath288 ) there exists a one - to - one correspondence between the set @xmath314 and the set @xmath315 . so , we have the following : @xmath316 now we invoke the inductive hypothesis : letting @xmath317 denote the set of @xmath30-simplices contained in @xmath309 with faces from @xmath289 and vertices from @xmath288 , we assume the statement c2(@xmath287,@xmath288,@xmath289 ) is true . hence , @xmath318 next , we use the previous computation to reinterpret this lower bound as : @xmath319 now let @xmath320 denote the set of @xmath96-simplices which contain the @xmath30-plane @xmath309 as a face and the point @xmath310 as a vertex . recall that every simplex in @xmath320 is the cone of a simplex from @xmath317 . so , @xmath321 next observe that for any @xmath309 , @xmath322 . so , we have @xmath323 to finish the computation , simply observe that this bound holds for any pair @xmath324 . so , simply multiply the previous bound by @xmath263 , and then insert the trivial bound @xmath325 . this finishes the proof of the lemma . the material discussed in this section should essentially be viewed as an appendix to the previous section . in particular , we wish to verify that the simplices constructed in the previous section are not degenerate in the sense that their faces ( edges , vertices , hyper - edges , etc . ) do not collapse onto each other . this nondegeneracy can be demonstrated by showing that the hypotheses h1(@xmath139,@xmath292,@xmath279 ) and h2(@xmath139,@xmath292,@xmath279 ) hold for most of the arrangements of points and @xmath139-planes @xmath326 appearing throughout the induction . before beginning the process of verifying these hypotheses , it will be useful to formally describe the construction of the collections of points and @xmath139-planes we shall be working with . the procedure is virtually identical to the one seen in the previous section for constructing the collections @xmath288 and @xmath289 . let @xmath292 be a collection of points and @xmath279 a collection of @xmath139-planes embedded in some @xmath171-dimensional space , and assume that the hypotheses h1(@xmath139,@xmath292,@xmath279 ) and h2(@xmath139,@xmath292,@xmath279 ) have been verified . working with these sets , we may ( as we have already done with @xmath30-planes ) construct the set @xmath327 : @xmath328 \subset \sigma_0 \cap \sigma\\ \dim[p_1,\dotsc , p_r ] = r-1\\ x \in \sigma \smallsetminus \sigma_0 , \sigma \neq \sigma_0\\ \lvert[p_1,\dotsc , p_r ] \cap q\rvert \geq \frac{\lvert i(q,\sigma ) \rvert}{10\lvert \sigma \rvert\f } \end{array } \right\}.\ ] ] since we re assuming the hypotheses h1(@xmath139,@xmath292,@xmath279 ) and h2(@xmath139,@xmath292,@xmath279 ) hold , we may compute a lower bound for the size of this set as @xmath329 next , we define a function @xmath330 on the collection of all pairs @xmath331 such that @xmath332 as @xmath333 clearly , we have @xmath334 so , we may once again use a dyadic pigeonholing argument to find a nice collection @xmath335 such that @xmath336 now , we are in a position to define the relevant collections of @xmath337-planes necessary for the construction . given a pair @xmath338 , we define a collection of @xmath337-planes embedded in @xmath339 as @xmath340$},\\ ( ( \sigma_0,\sigma , p_1,\dotsc , p_r),x ) \in v_{r , p}(q,\sigma ) \end{array}\right\}.\ ] ] also , we may simply define the associated collection of points as @xmath341 as it turns out , the verification of h2 at each level of the induction is far more elementary than the verification of h1 ; in fact , h2 essentially holds by definition . we state this formally as follows : let @xmath307 be an arrangement of points and @xmath30-planes embedded in some ambient space as described in equations and . given a pair @xmath342 , define collections of points @xmath288 and @xmath229-planes @xmath289 embedded in @xmath309 as defined in equations and respectively . then , we have the following : @xmath343 for each of the pairs @xmath344 . furthermore , this regularity holds at all lower levels of the induction in the sense that ( using the notation already established in this section ) @xmath345 for each of the pairs @xmath346 . first , let us recall the definitions of the sets @xmath288 and @xmath289 : @xmath347 , ( ( \pi_0,\pi , p_1,\dotsc , p_k),x ) \in v_{k , p } \end{array}\right\}.\end{aligned}\ ] ] next , let us recall the definition of @xmath242 : @xmath348 = k-1,\\ \lvert [ p_1,\dotsc , p_k ] \cap p \rvert \geq \frac{\ntili}{10\ntilpi\f } \end{array } \right\}.\ ] ] suppose that h2(@xmath287,@xmath288,@xmath289 ) failed for some pair @xmath300 ; this would imply @xmath349 this , however , is a contradiction to the assumption that h2(@xmath30,@xmath97,@xmath98 ) holds , as the above definitions imply the following string of inequalities : @xmath350 or , @xmath351 we omit the proof for the remaining levels of the induction , as it is virtually identical to what we have already shown . in order to show that property h1 is inherited at all stages of the induction , one needs to do a bit more work . the reason for this is that property h1 does not simply pass freely from level @xmath30 to level @xmath287 by definition ( as was essentially the case for property h2 ) . loosely speaking , the statement h1(@xmath30,@xmath97,@xmath191 ) states that most maximal - rank @xmath30-tuples of points in the set @xmath97 are incident to many @xmath30-planes from the set @xmath191 ( in this case @xmath281 , this general property is usually referred to as `` bilinearity '' ) . to illustrate the difficulty , consider the case @xmath352 , @xmath353 . choose a pair @xmath354 , and construct the sets @xmath288 and @xmath289 in the usual fashion . we would like to say that the hypothesis h1(@xmath133,@xmath288,@xmath289 ) is satisfied , or ( more informally ) that most of the points in @xmath288 are incident to many lines from @xmath289 . a natural attempt to prove such a statement is to observe that , by property h1(@xmath27,@xmath97,@xmath191 ) , most pairs of points in @xmath97 are incident to many planes from @xmath98 . in particular , given a point @xmath355 , there are many planes incident to the pair @xmath356 . unfortunately , since @xmath353 , these @xmath27-planes are not obligated to intersect @xmath309 in a line , hence not contributing to the set @xmath289 . therefore , one must come up with a different means by which to verify h1(@xmath133,@xmath288,@xmath289 ) . let @xmath307 be an arrangement of points and @xmath30-planes embedded in some ambient space as described in equations and . given a pair @xmath342 , define collections of points @xmath288 and @xmath229-planes @xmath289 embedded in @xmath309 as defined in equations and respectively . then , we have the following : @xmath357 for at least half of the pairs @xmath261 . furthermore , this regularity holds at all lower levels of the induction in the sense that ( using the notation already established in this section ) @xmath358 for at least half of the pairs @xmath346 . given @xmath324 , define sets @xmath288 and @xmath289 as above . assuming that h1(@xmath287,@xmath288,@xmath289 ) fails when @xmath359 , where @xmath360 and @xmath361 , we will obtain upper and lower bounds for the size of the set @xmath362 as defined in equation . for convenience , let us recall this definition : @xmath363)\}\ ] ] assuming h1(@xmath30,@xmath97,@xmath191 ) and h2(@xmath30,@xmath97,@xmath191 ) hold , we may use the computations from the previous section , and obtain a lower bound of @xmath364 before proceeding , we will refine the set @xmath365 to the following slightly smaller subset @xmath366 since there is a uniform ( throughout the set @xmath265 ) lower bound on the number of objects of the form @xmath367 such that @xmath368 , and @xmath369 , we have @xmath370 . next , we compute an upper bound for the size of @xmath371 in two slightly different ways . first of all , there are @xmath372 choices for the pair @xmath261 . once this pair has been fixed , we make use of the estimate @xmath373 , and observe that there are @xmath374 choices for the @xmath229-tuple @xmath375 . next , assuming that h1(@xmath287,@xmath288,@xmath289 ) fails , it follows that there are at most @xmath376 @xmath30-planes from @xmath191 which contain the @xmath229-tuple @xmath375 and contribute a @xmath229-plane to the set @xmath289 . this means that there are at most @xmath377 choices for the remaining point @xmath378 . putting this all together , we have an estimate of @xmath379 also , we will need to make another , seemingly cruder estimate of the same set . this estimate is obtained in the same manner as equation , except we use the trivial estimate @xmath380 in place of the estimate @xmath381 . this yields an upper bound of @xmath382 combining the upper bounds and with the lower bound , we obtain the incidence bounds @xmath383 and @xmath384 respectively . next , for each integer @xmath30 , we define a quantity @xmath385 as @xmath386 while this quantity may seem a bit odd , the reader will notice that it satisfies the inequalities @xmath387 for every @xmath30 , and ( after a great deal of tedious algebra ) , the following miracle occurs : @xmath388 as this is precisely the incidence bound that we are trying to demonstrate , it follows that we may assume h1(@xmath287,@xmath288,@xmath289 ) holds for most pairs @xmath300 . next , given an integer @xmath389 , we assume that collections of points @xmath292 and @xmath390-planes @xmath279 ( as constructed at the start of this section ) satisfy hypothesis h1(@xmath391,@xmath292,@xmath279 ) : @xmath392 suppose that h1(@xmath393,@xmath394,@xmath395 ) fails when @xmath396 for some set @xmath397 where@xmath398 . then , we have the following : @xmath399 in order to show that one may assume that this situation does nt occur , we will proceed in a manner similar to the way we addressed the @xmath400 case , however we will be counting objects which are a bit more complicated that elements of the set @xmath365 . given integers @xmath30 and @xmath135 , we define a @xmath401-_chain _ to be a @xmath272-tuple of points @xmath402 , and an @xmath135 tuple of @xmath30-planes @xmath403 such that @xmath404 spans a @xmath96-dimensional space , and for any positive integer @xmath405 , and any choice of @xmath406 @xmath30-planes from the @xmath135-tuple @xmath407 , the space @xmath408 is @xmath409-dimensional , and spanned by some @xmath410-tuple of points from @xmath272-tuple @xmath404 . the set of all @xmath401-chains constructed from the sets @xmath97 and @xmath191 will be denoted @xmath411 . as the definition of a @xmath401-chain is somewhat complicated , one might wish to regard them as @xmath96-simplices ( with faces from @xmath191 and vertices from @xmath97 ) which are missing @xmath412 of their faces , but none of their vertices . this construction is motivated by the fact that when @xmath413 , the intersection of the @xmath414 @xmath30-planes in any @xmath415 chain in the set @xmath416 will be a @xmath417-plane from the set @xmath395 . therefore , if we assume that equation holds , we will be able to demonstrate an upper bound for @xmath418 . then , after computing a corresponding lower bound ( mostly by means of cauchy - schwarz and pigeonholing techniques ) , we will be able to show that the incidence bound holds . our first task will be to compute a lower bound for a certain large subclass of @xmath416 . unfortunately , this computation is much more complicated than it was in the case @xmath400 ; the reason for this is that cauchy - schwarz alone does not seem to be powerful enough to count @xmath401-chains when @xmath419 . therefore , we shall use a `` coning procedure '' , much like was seen in the previous section for counting simplices , in order to estimate @xmath420 from below . the basic scheme for this construction will be to realize any @xmath415-chain as the cone of a @xmath421-chain , which in turn is the cone of a @xmath422-chain , and so on ( see figure [ chainconepic ] ) . eventually , elements of the set @xmath416 will be described as hypercones of @xmath423-chains . since cauchy - schwarz _ is _ useful for estimating chains of length @xmath27 , we will arrive at the desired lower bound for @xmath418 , by estimating the size of a certain class of @xmath423-chains , and then making a few observations . -chain as the cone of a @xmath424-chain . the three @xmath28-planes in the @xmath425-chain on the right are @xmath426 $ ] , @xmath427 $ ] and @xmath428 $ ] . ] [ chainconepic ] this computation will require a bit of extra notation . as was shown in the previous section , it is possible to find a collection of disjoint pairs of points and @xmath30-planes @xmath429 such that for each @xmath430 one has @xmath431 where the function @xmath58 is defined ( for pairs @xmath432 ) as @xmath433 so , for each @xmath430 , one can define families @xmath434 and @xmath435 of points and @xmath229-planes contained in @xmath309 . by definition , these collections satisfy the following estimates : @xmath436 next , we observe that any element of the set @xmath416 which contains the point @xmath437 as a vertex and the @xmath30-plane @xmath309 as a face must be the cone of a unique element of the set @xmath438 . this observation allows us to conclude that @xmath439 so , by a simple application of the pigeonhole principle , we may choose a single value @xmath430 such that @xmath440 therefore , we have reduced our problem to that of estimating the size of the sets @xmath438 . at this point , the procedure repeats itself , this time with the collections @xmath434 and @xmath435 of points and @xmath229-planes . since these estimates are hold uniformly for a specifically chosen ( via the pigeonhole principle ) pair @xmath430 , we will no longer use these parameters as subscripts when describing induced sets of points or @xmath229-planes . for example , the set @xmath441 will be used to denote an induced family of @xmath229-planes relative to some generic @xmath442 . this convention is purely in the interest of simplifying notation , and will be used for analogous constructions to appear later in the proof . furthermore , the sets @xmath395 referred to in equation will henceforth be denoted @xmath443 . in general , once the sets @xmath444 and @xmath445 have been constructed ( for some integer @xmath446 ) , one can choose a collection @xmath447 of disjoint pairs of points and @xmath448-planes , and use this collection to define ( as was described at the beginning of this section ) to create families @xmath449 and @xmath450 of points and @xmath451-planes which satisfy the estimates @xmath452 furthermore , by identifying elements of the sets @xmath453 and @xmath454 , we have the estimate @xmath455 since these observations hold for any integer @xmath446 , we arrive at the following conclusion : @xmath456 at this point , since we have reduced to problem to estimating the size of a family of @xmath423-chains , we may use cauchy - schwarz . the construction of elements of the set @xmath457 is virtually identical to the construction of the sets @xmath458 , except that we add an extra point to each of the @xmath390-planes , rather than just one of them . therefore , cauchy - schwarz yields an estimate of @xmath459 furthermore , since we are assuming ( by induction ) that the hypothesis h1(@xmath391,@xmath460,@xmath461 ) holds , it follows that these @xmath423-chains are non - degenerate ( i.e. the two @xmath390-planes contained in each are distinct ) . explicitly , the degenerate @xmath390-chains which we would like to discard are simply elements of the set @xmath462 equipped with an additional pair of points , so it has size @xmath463 . applying the hypothesis h1(@xmath391,@xmath460,@xmath461 ) yields @xmath464 therefore , we have @xmath465 before preceding , we must refine the set @xmath466 to a slightly smaller set @xmath467 defined as @xmath468 making use of the uniform lower bound estimate from equation , and the fact that @xmath469 , we have @xmath470 we shall denote the elements of the set @xmath416 which are cones of @xmath423-chains from @xmath466 as @xmath471 . next , it is necessary to simplify the expression on the right hand side this expression . making use of equations and , we have @xmath472 holds when @xmath473 . furthermore , making use of the trivial estimate @xmath474 we arrive at an estimate of @xmath475 after repeating this procedure @xmath476 times , and making use of the trivial estimate @xmath325 , we finally arrive at the desired conclusion @xmath477 next , we shall compute a complementary upper bound for @xmath478 in two different ways ( as was done for the case @xmath400 ) . first , choose a pair @xmath432 such that @xmath479 ; there are clearly at most @xmath480 choices for this pair . by definition , the remaining @xmath481 points must lie inside @xmath309 , and they will uniquely define the remaining @xmath139 @xmath30-planes once chosen . so , once the number of ways to choose the remaining points has been computed , we will have an upper bound for @xmath478 . since @xmath482 , and these points can not cluster on low dimensional subspaces of @xmath309 , there are @xmath483 choices for the first @xmath393 points of the @xmath415-chain . next , assuming equation holds , there are at most @xmath484 choices for the @xmath417-plane found at the intersection of the @xmath171-fold intersection of the @xmath30-planes in the @xmath415 chain . this means that we have at most @xmath485 choices for the next point . finally , there are @xmath486 choices for the remaining @xmath139 points . these observations yield the bound @xmath487 next , we make the same computation , only use the cruder estimate @xmath488 instead of @xmath489 . this method yields the bound @xmath490 in order to finish the proof , we simply combine the upper and lower bounds we have now computed for @xmath491 . making use of the result from the previous computation , and the estimates and , we have @xmath492 and @xmath493 respectively . as it turns out , our main estimate is simply a convex combination of these two estimates . to be more specific , if we define the quantity @xmath494 as @xmath495 then , one can compute that @xmath496 for appropriate values of @xmath30 and @xmath139 , and @xmath497 now that we have established a lower bound on the number of simplices appearing in the configuration of points and @xmath30-planes , we must find an upper bound . this task requires making a few observations about the set @xmath249 . first of all , by construction , the pair of @xmath30-planes in any element of @xmath498 spans some affine @xmath96-dimensional space @xmath499 . since our original family of @xmath30-planes ( and hence every refinement of that family ) is direction separated , we have @xmath500 for all @xmath498 . so , when constructing a simplex from some @xmath501 by choosing the remaining @xmath30 faces from @xmath499 , we have @xmath502 choices for each face . before going ahead with this construction , however , one must observe that an element @xmath498 is not merely a pair of @xmath30-planes ; each such @xmath503 comes equipped with a @xmath30-tuple of points on its `` spine '' ( the intersection of the two @xmath30-planes ) . in order to manage this technicality , recall that ( from the refined definition of @xmath242 ) , the spine of any @xmath498 must carry at least @xmath504 points from @xmath97 . so , if we delete the points from each element of @xmath249 to create the following set @xmath505 we have the following bound : @xmath506 now that we have dealt with this possible over - counting , we may construct simplices from each element @xmath507 by choosing the @xmath30 remaining faces from @xmath499 . this yields an upper bound in the number of simplices of @xmath508

the pseudoscalar - vector - vector ( ) correlator is constructed using two meson multiplets in the vector and two in the pseudoscalar channel . the parameters are constrained by the operator product expansion at leading order where two or all three momenta are considered as large . demanding in addition the brodsky lepage limit one obtains ( in the chiral limit ) a pion - vector - vector ( ) correlator with only one free parameter . the singly virtual pion transition form factor and the decay width of are independent of this parameter and can serve as cross - checks of the results . the free parameter is determined from a fit of the- transition form factor . the resulting correlator is used to calculate the decay widths and and finally the widths of the rare decay and of the dalitz decay . incorporating radiative qed corrections the calculations of decays are compared to the ktev results . we find a deviation of 2 or less for the rare pion decay . 13.20.cz decays of mesons , 13.40.gp electromagnetic form factors , 11.30.rd chiral symmetry , 12.40.vv vector - meson dominance

1507.00478

two major challenges of contemporary particle physics are the search for beyond - standard model physics and a better understanding of the non - perturbative low - energy sector of the strong interaction . typically both aspects intermix when it comes to high - precision determinations of low - energy quantities and the corresponding standard model predictions . if a low - energy observable is potentially influenced by quantum effects from new particles , then it is also influenced by hadronic loop effects . the latter often if not always constitute the main uncertainty of the standard model prediction . on the other hand , new physics can only be revealed if a significant deviation between experiment and standard model calculation is observed . of course , this requires small uncertainties for both the experimental and the standard model result . our poor understanding of non - perturbative qcd could provide a serious hurdle for a reliable standard model calculation and/or for a reliable uncertainty estimate of such a calculation . two quantities of current interest which might indicate some deviation between experiment and the standard model are the gyromagnetic ratio of the muon @xcite and the rare decay of the neutral pion into electron and positron @xcite . an important quantity that enters both observables is the pion transition form factor , i.e. the three - point correlator between a neutral pion and two electromagnetic currents . two tasks are carried out in the present work : first , the pion - vector - vector ( @xmath1 ) correlator is determined by combining high - energy quark - based information with low(er)-energy hadronic information . we follow the general approach proposed in @xcite with some refinements . second , we explore some phenomenological consequences of our correlator and focus in particular on the rare pion decay @xmath13 . here we include also qed radiative corrections along the lines of @xcite . two limits of qcd are of particular interest for low - energy hadron physics : the chiral limit where the masses of the two or three lightest quarks are neglected @xcite and the limit where the number of quark colors , @xmath14 , is sent to infinity @xcite . concerning the @xmath1 correlator the chiral anomaly fixes the low - energy strength unambiguously in the chiral limit . for a large number of colors there are infinitely many , infinitely narrow , i.e. stable , quark - antiquark states for every combination of quantum numbers . they show up as poles in the @xmath15-point correlators of quark currents . in the real world of three colors the hadrons generically turn to unstable resonances because the hadronic interactions do not vanish any more . the poles in the correlators turn to cuts ( and poles in other riemann sheets ) . the cuts start at the corresponding many - body thresholds . thus the relevance of the large-@xmath14 limit for the real world is highest , if one considers hadrons which are narrow and/or kinematical regions where there are no ( significant ) cuts . this is our guiding principle when exploring phenomenological consequences . concerning form factors there are no cuts in the space - like region . in the standard model the rare decay of the pion into electron and positron is caused by a loop where the pion first turns into a pair of ( real or virtual ) photons ; see fig . [ fig : piee ] below . at this vertex the pion transition form factor sneaks in . if the form factor was replaced by a constant , the loop would diverge @xcite . in qcd the pion transition form factor is suppressed for large virtualities @xcite . this leads to a finite result for the @xmath11-@xmath16-@xmath17 amplitude at the one - loop level of the standard model calculation . thus for a quantitative determination of the branching ratio of the considered rare pion decay it is necessary to know where and how fast the pion transition form factor reaches its asymptotic form which in turn depends on the various combinations of virtualities . these considerations show that one needs information from various qcd regimes : the threshold regime governed by the chiral anomaly , the regime of hadronic resonances , and , finally , the regime of asymptotically high energies dictated by quarks and asymptotic freedom . these regimes are connected in the approach of @xcite where the operator product expansion ( ope ) for various three - point correlators of quark currents is matched to a hadronic ansatz that satisfies the chiral constraints for the low - energy limit . this ansatz is furthermore based on a truncation of the infinite tower of stable hadronic states that appears in the limit of a large number of colors . in principle one can work out arbitrary many orders in the ope and match to the parameters that emerge with the tower of hadron states . however , the higher orders in the ope contain unknown quark and gluon condensates of high dimensionality . thus in practice the model dependence emerges from a selection of the to be matched ope constraints and from the choice where to truncate the tower of hadron states . using one hadron multiplet per channel and leading - order ope constraints has been studied in detail in @xcite . this truncation is called `` lowest - meson dominance '' ( lmd ) . in the present work we will explore the consequences of having two hadron multiplets per channel . in the language of @xcite our approach would be called `` lmd+v+p '' . to avoid this clumsy name we decided to introduce the name `` two - hadron saturation '' ( ths ) . concerning our quantity of interest , the @xmath1 correlator and the corresponding pion transition form factor , the starting point on the level of quark currents is the pseudoscalar - vector - vector ( @xmath0 ) correlator . the consequences of lmd for this quantity have been studied in @xcite . an application to the rare pion decay to electron and positron was presented in @xcite . two vector multiplets have also been considered in @xcite ; see also @xcite where this has been used for the rare pion decay . what makes our approach different from previous works is that we explore in detail the consequence of two multiplets in _ each _ channel and that we fit and/or compare to data on the @xmath18 correlator . in fact , including a second multiplet in the vector channel involves the energy region of about 1.4 gev @xcite . in this region there is also a pseudoscalar multiplet . thus the extension from lmd to two multiplets for a channel suggests to use two multiplets for _ every _ channel . concerning the second aspect , the interrelation to the @xmath18 correlator , we will come back to this issue below , after discussing in more detail the pertinent high - energy constraints . there is yet one more short - distance limit to be considered . instead of studying the high - energy limit of correlators of quark currents ( ope ) , one can also study the high - energy limit of correlators that involve specific asymptotic states like hadrons or photons together with one or several quark currents . in particular the high - energy behavior of the pion - photon - vector correlator has recently gained much attention since the babar data @xcite seem to contradict the brodsky lepage ( b - l ) scaling limit @xcite while the belle data @xcite seem to support it . using ths we are able to satisfy all leading - order ope constraints for the @xmath0 correlator and in addition the b - l constraint for the pion - photon - vector correlator . while lmd satisfies the same ope constraints , it violates the b - l constraint as can be easily deduced from the explicit form given in @xcite ; see also the discussion in @xcite . as we will show below , the constraints from the leading order of the ope , from b - l and from the chiral anomaly together fix the ths approach to the @xmath1 correlator up to one single parameter , which we call @xmath19 . if the invariant mass of one of the vector currents in this correlator is set to zero , i.e. for the pion - photon - vector correlator , then @xmath19 drops out . in other words we have full predictive power for this correlator . aiming at the rare pion decay into electron and positron one needs the full information on the @xmath1 correlator for arbitrary invariant masses of the two vector currents . in principle , the parameter @xmath19 would be best determined from data on the @xmath1 correlator with both invariant masses being different from real photons . unfortunately such data do not exist . in this situation we turn to the second - best choice . projecting the @xmath1 correlator on one of the vector mesons that we include in ths yields a 3-point correlator for the pion , the vector meson and a quark current with vector quantum numbers . given that our approach is based on the large-@xmath14 limit where mesons are approximated by infinitely narrow states , it is suggestive to use a narrow vector meson . since phenomenologically and in the large-@xmath14 limit the pion decouples from the @xmath20 meson @xcite , we are left with the @xmath4 meson as the best choice for a vector meson . consequently we will use data on the @xmath4-@xmath5 transition form factor @xcite to fix our remaining parameter @xmath19 . the @xmath1 correlator obtained in this way from ths shows an awesome behavior : if the virtuality of one vector current becomes large while the other is set to zero ( photon case ) , the asymptotic b - l limit is reached rather fast , resembling essentially strict vector - meson dominance ( vmd ) @xcite . the scale that defines where the approach to the asymptotic limit sets in is basically given by the mass of the @xmath21/@xmath4 meson . on the other hand , if both vector currents have the same large virtuality , the corresponding asymptotic limit is reached very late for the @xmath1 correlator as obtained from ths . this finding points to the relevance of details of hadronic physics above 1 gev . for this case the @xmath1 correlator from vmd falls off much faster than demanded by qcd , while for lmd the asymptotic limit is reached much earlier than for ths . since the rare decay @xmath13 is sensitive to both high- and low - energy physics , it is interesting to study how this intriguing behavior of the @xmath1 correlator obtained from ths influences the branching ratio of this rare process . before going into the details of our findings we shall compare our approach to related ones from the literature . the @xmath4-@xmath5 transition form factor and related quantities have also been addressed in @xcite where the ground - state vector mesons are treated as light degrees of freedom . by construction the approach is restricted to low energies , i.e. high - energy constraints were not considered . nonetheless , it turns out that the ths result for the @xmath4-@xmath5 transition form factor is numerically very close to the one from @xcite . conceptually close in spirit to ths is the lagrangian approach utilized , e.g. , in @xcite ( earlier references can be traced back from these works ) . also here hadron resonances in the large-@xmath14 limit are used to interpolate between the low - energy region governed by chiral perturbation theory and the high - energy region governed by the ope or quark scaling considerations . in @xcite a @xmath0 and a corresponding @xmath1 correlator are constructed with one multiplet in the vector channel and two multiplets in the pseudoscalar channel ( i.e. the goldstone bosons and one resonance multiplet ) . for an extension to two multiplets in the vector channel see @xcite . the @xmath1 correlator from @xcite satisfies the b - l constraint , but the @xmath0 correlator satisfies only one leading - order ope constraint , not all of them . as shown in @xcite the lagrangian utilized in @xcite is not capable to provide correlators that satisfy all ope constraints ; see also the discussion in @xcite . instead of the other ope constraints on the @xmath0 correlator a high - energy constraint based on quark counting rules is imposed on the @xmath22 correlator in @xcite . in our work we impose high - energy constraints on the @xmath0 correlator ( ths satisfies _ all _ leading - order ope constraints ) and on the @xmath1 correlator ( the b - l limit ) , but not on the @xmath22 or @xmath23 correlator . our philosophy is that we take the first multiplets to resemble the corresponding ground - state physical particles but the second multiplets to mimic the effect of the tower of infinitely many excited states . since we study with @xmath0 an order parameter of chiral symmetry breaking @xcite one expects that the second multiplets are close to the physical states that are the first excitations on top of the ground states . however , the weighted average of the whole tower of states might shift the effective mass higher up . in the present work we explore the uncertainty of the ths approximation by changing the masses of the second multiplets from the first to the second physical excitations . concerning the @xmath1 correlator we expect to obtain reasonable results because the pion might not resolve too many details of the intermediate - energy region . thus replacing the tower of excited resonances by the respective lowest excitation and demanding high - energy constraints for the @xmath1 correlator might be good enough . in contrast , a @xmath22 or @xmath23 correlator resolves more from the intermediate - energy region because the vector meson induces a larger mass scale . ( the same would apply to a @xmath24 correlator . ) therefore to satisfy high - energy constraints in this case we expect that one would need a more detailed modeling than just having ths , i.e. one excitation ( plus the ground state ) in each channel . it might be worth to explore a three - hadron saturation scenario , but this is beyond the scope of the present work . therefore we demand constraints on the @xmath0 and @xmath1 correlators , but disregard constraints for three - point correlators of one pseudoscalar meson , one vector meson and one vector quark current . for the same reason we only consider leading , but not subleading high - energy / ope constraints . from a formal point of view our approach is close to the successive pad approximations as utilized , e.g. , in @xcite . our correlators are also approximated by rational functions . in our approach , however , the poles of the correlators are related to physical states ( in the large-@xmath14 limit ) . in contrast , in the pad framework of @xcite one determines the rational functions by fits to data . it is not the purpose of this pad approach to look for the poles of the obtained rational functions . in fact , there are no physical restrictions from outside that would make sure that these poles correspond in any way to physical hadrons . but this is what s - matrix theory suggests ( in the large-@xmath14 limit ) : anything beyond polynomials should be caused by unitarity , analyticity , crossing symmetry and physical states . our work is complementary to the dispersion theoretical approach of @xcite ; see also @xcite . while we can not expect to reach the accuracy of a dispersive approach concerning low - energy quantities , our framework has the advantage to provide a smooth and physical connection between the low- and high - energy region and between the quark based and hadron based correlators . in practice , dispersive calculations are based on an excellent account of the low - energy region ( up to about 1 gev ) and a high - energy completion , i.e. a matching to the high - energy behavior deduced from s - matrix theory , qcd or qcd related approaches ; see the discussion in @xcite concerning the @xmath1 correlator . as already discussed , for our doubly virtual @xmath1 correlator there are regions in the virtualities where the asymptotic regime is reached only at very high energies . in practice this might imply that a naive matching of a dispersive calculation to the asymptotic regime might miss part of the physics present at intermediate energies . clearly it is worth to explore the interplay of a dispersive calculation with ths in the future . the rest of the paper is structured in the following way : in section [ sec : ff ] we construct the @xmath0 and @xmath1 correlators subject to high- and low - energy constraints . the results for the @xmath1 correlator are compared to data and to other approaches ( lmd and vmd ) . it is studied how the shape of the correlator changes when varying the remaining free parameter @xmath19 in a reasonable range . in addition , a model uncertainty is estimated by varying the mass of the second vector - meson multiplet between the _ physical _ masses of the first and second excitation . in section [ sec : pheno ] the @xmath23 correlator is constructed . the parameter @xmath19 is determined from a fit to the @xmath4-@xmath5 transition form factor . the widths or branching ratios , respectively , for the corresponding decays @xmath25 , @xmath26 , and @xmath27 are determined for ths ( and also for lmd and vmd ) . actually the first decay does not depend on @xmath19 . we find good agreement between ths and the experimental results . we address the rare pion decay to electron and positron in section [ sec : piee ] . including radiative corrections the branching ratio of this process is calculated and we compare again the ths result to other approaches . for the ths case this branching ratio is sensitive to @xmath19 . direct comparison to the experimental value from ktev @xcite seems to suggest that a discrepancy persists on the level of 2@xmath12 . in section [ sec : pieegam ] we study the properties of the singly virtual pion transition form factor in the low - energy region , such as slope and curvature . note that these quantities do not depend on the parameter @xmath19 . for the dalitz decay @xmath28 we calculate the decay width taking into account next - to - leading order ( nlo ) radiative corrections . these are evaluated along the lines of @xcite . again we compare ths to other approaches . with the full set of radiative corrections at hand , we take a fresh look on the ktev result . considering some radiative corrections that were not accounted for in the analysis suggests that the previously stated discrepancy might be even reduced to 1.5@xmath12 . the main message here , however , is that the radiative corrections are now theoretically under control @xcite and can be used in future data analyses . in section [ sec : outlook ] we provide an outlook how ths can be further utilized and how the scheme can be extended . finally , an appendix is added to provide the pseudoscalar form factors in terms of the loop integrals required for the rare pion decay . comparing in detail various approaches throughout the present work reveals that the vmd form factor proves to be phenomenologically very successful in most applications , in spite of the facts that vmd is so simple and partially possesses an improper high - energy behavior . however , the ths form factor , which satisfies all the considered constraints , works very well , too . still , the mean value of the ktev result remains a challenge for all approaches , even though the discrepancy has been reduced by a considerable level . we provide a brief discussion on this point in the corresponding sections . following @xcite we introduce the @xmath0 correlator @xmath29 by @xmath30|0\rangle \end{split}\ ] ] with @xmath31 . the vector and pseudoscalar current , respectively , are defined by.]@xmath32 in the above formulae we have used @xmath33 = \frac12\delta^{ab}\,,\quad d^{abc } \equiv2\,\text{tr}\big[\{t^a , t^b\}t^c\big ] \ , . \label{eq : normdef}\ ] ] for @xmath34 we have @xmath35 where @xmath36 denote the gell - mann matrices in flavor space . since we utilize the large-@xmath14 limit we have to deal with flavor nonets , i.e. with u(3 ) instead of su(3 ) . the formulae ( [ eq : normdef ] ) provide a natural extension to @xmath37 . working in the chiral limit and considering two meson multiplets in the vector and two in the pseudoscalar channel , we propose a correlator of the form @xmath38 here @xmath39 is the most general polynomial in its arguments that is symmetric in the second and the third argument . for our purpose , a polynomial of order four is sufficient inasmuch as higher powers of momenta in the numerator are not allowed due to the desired high - energy behavior that we will specify below . thus we start with a term containing 22 monomials and associated free parameters . schematically this looks as follows : @xmath40\\ & + c_2[(p^2)^3q^2+(q^2)^3p^2]+c_3(r^2)^2p^2q^2+\,\dots \ , . \end{split } \label{eq : poly}\ ] ] the quantities @xmath41 and @xmath42 in ( [ eq : corr ] ) denote the usual low - energy constants of chiral perturbation theory @xcite . we will specify them further when needed . @xmath43 denotes the mass of the second multiplet , i.e. first excitation , in the pseudoscalar channel . the first multiplet ( ground state ) is , of course , the massless multiplet of goldstone bosons . the masses of the lowest two vector - meson multiplets are denoted by @xmath44 and @xmath45 . as already spelled out in section [ sec : intro ] we use for @xmath44 the mass of the ground - state vector - meson multiplet ( in the chiral limit ) . we assume that this mass is approximately given by the mass of the @xmath4 or @xmath21 meson . for @xmath45 it is suggestive to use the physical mass of the first excitation in the vector channel . however , we do not use a fixed mass here , but study the impact of a variation of @xmath45 on our results . in this way we explore the uncertainty caused by the higher - lying excitations that are neglected in ths . in practice we vary @xmath45 in the range @xmath46\,\text{mev , } \label{eq : rangev2}\ ] ] which is the interval between the masses of the first and the second physical excitation @xcite . finally we note that the same logic applies to the mass @xmath43 of the second pseudoscalar multiplet . for the @xmath0 correlator one should study the impact of a variation of @xmath43 . as we will see below , however , this mass does not show up in the final expression for the @xmath1 correlator . we now demand that the ansatz ( [ eq : corr ] ) satisfies all the relevant high- as well as low - energy constraints in order to minimize the number of free parameters introduced in ( [ eq : poly ] ) . starting with the general correlator ( [ eq : corr ] ) we apply the following leading - order ope constraints @xcite : @xmath47 @xmath48 it turns out that the third ope constraint @xcite @xmath49 is automatically fulfilled by our ansatz . here @xmath50 denotes a function ( actually a two - point correlator @xcite ) that depends only on @xmath51 and not on the other kinematic variables @xmath52 and @xmath53 . next , we define the @xmath1 correlator : @xmath54 where @xmath55 denotes the overlap between the pion field and the pseudoscalar quark current . with the usual conventions from chiral perturbation theory one obtains @xmath56 @xcite . for large @xmath57 one then finds @xmath58 which means that in agreement with @xcite we have @xmath59 hence , this condition is satisfied automatically on account of the ope constraints . actually , for the quantity @xmath60 also the subleading order in the high - energy expansion is known in terms of a quark - gluon condensate @xcite ; see also the corresponding discussion in @xcite . however , we refrain from incorporating this as an additional constraint . the reason is the same as to why we do not use constraints from transition form factors of hadronic resonance states . one might become too sensitive to the details of the intermediate mass region where we use one hadronic state to describe effectively the whole infinite tower of large-@xmath14 excited states . if one used three instead of two hadronic states per channel , the subleading high - energy constraints might provide a viable input to pin down the growing number of resonance parameters . but this is clearly beyond the scope of the present work . instead of involving subleading orders in the high - energy expansion we apply the b - l constraint @xcite @xmath61 we define the pion transition form factor as @xmath62 and match at the photon point to the chiral anomaly , i.e. to the wess zumino witten term @xcite , @xmath63 together the constraints ( [ eq : ope1 ] ) , ( [ eq : ope2 ] ) , ( [ eq : bl ] ) and ( [ eq : wzw ] ) provide us with a @xmath1 correlator that appears to have only one free dimensionless parameter @xmath19 : @xmath64\bigg\}\ , . \end{split } \label{eq : fffinal2}\ ] ] note that our result is independent of the mass @xmath43 of the first pseudoscalar excitation . this happens due to the fact that at the end of the day we could conveniently rescale the only free parameter left . from the structure of the result ( [ eq : fffinal2 ] ) it can be read off that @xmath19 emerges from @xmath65 in ( [ eq : poly ] ) on account of @xmath66 we note in passing that @xmath19 scales with @xmath67 . a comparison to the work of @xcite is in order here concerning the number of free parameters . we shall compare our ths approach to lmd+v of @xcite , i.e. to the case of one pseudoscalar and two vector multiplets . we recall that ths = lmd+v+p in the language of @xcite . after applying the ope constraints to ths we are left with 12 free parameters for the @xmath0 correlator . this compares to 7 free parameters for the case of lmd+v . the low - energy constraint ( [ eq : wzw ] ) fixes always one more parameter . once we focus on the @xmath1 correlator we are left with 4 parameters for ths . for lmd+v one demanding that the @xmath1 correlator of ( [ eq : bl ] ) drops like @xmath68 limits the free parameters to 2 for both cases ths and lmd+v . demanding ( [ eq : bl ] ) _ quantitatively _ and not just a scaling with @xmath68 yields one free parameter for ths as well as for lmd+v . to summarize , after applying all constraints ths has more parameters for the @xmath0 correlator than lmd+v : 8 in ths versus 4 in lmd+v . but concerning the @xmath1 correlator one ends up with one free parameter in both cases . according to @xcite the @xmath1 correlator should satisfy the inequality @xmath69 it turns out that the ths expression ( [ eq : fffinal2 ] ) satisfies ( [ eq : dc ] ) for @xmath70 . we will see below that the values for @xmath19 that we obtain from fitting to experimental data lie well within this range . before further constraining @xmath19 from data it is illuminating to study the qualitative shape of the @xmath1 correlator when @xmath19 is varied . in addition , we compare our results to similar approaches from the literature , namely to the vmd correlator @xcite @xmath71 and to the lmd expression @xcite @xmath72\ , . \label{eq : fflmd } \end{split}\ ] ] see also @xcite for a comparison of various correlators . first of all , we note that the vmd result grossly violates ( [ eq : f / q2 ] ) while it is not far off from the b - l constraint ( [ eq : bl ] ) . on the other hand , lmd satisfies ( [ eq : f / q2 ] ) , but does not satisfy ( [ eq : bl ] ) . by construction the ths correlator satisfies all the mentioned constraints , but it is interesting to see how fast or slow the asymptotic limits are reached . this is shown in fig . [ fig : ff ] . to facilitate a comparison with data we show the pion transition form factor ( times the virtuality @xmath51 ) instead of the @xmath1 correlator . note that the relation ( [ eq : defpitff ] ) between the pion transition form factor and the @xmath1 correlator only amounts to a rescaling . in the first panel of fig . [ fig : ff ] we display the symmetric doubly virtual pion transition form factor . this plot shows three different _ types _ of lines and a gray band . we shall first discuss the lines : the dash - dotted lmd line approaches very quickly the asymptotic qcd result given by ( [ eq : f / q2 ] ) . the dashed vmd line falls stronger than what is required by qcd . the full lines labeled by values for @xmath19 show the ths result for a mass of @xmath45 taken in the middle of the interval ( [ eq : rangev2 ] ) . all the full lines will approach the lmd line at very large momenta . one sees , however , that typically the ths lines reach this limit rather late . in particular , for @xmath73 the ths form factor peaks at rather high momentum values and with large magnitude in comparison to vmd and lmd . it reaches the asymptotic limit only very slowly . for the case @xmath74 we see that the associated line is very close to lmd , but peaks later and approaches the asymptotic limit from above . for @xmath75 we get the limiting case where ths is monotonically growing . going further down with @xmath19 a wiggle appears , which for @xmath76 and low virtualities resembles very much the vmd behavior . for negative @xmath19 the ths results start to undershoot vmd at low virtualities . moving to even lower @xmath19 values there is always a region where the ths result becomes negative . intuitively we find it hard to believe that the transition form factor of a ground - state hadron ( the pion ) would display so many wiggles as suggested by the full lines for small values of @xmath19 . indeed , the determination of @xmath19 , which will be carried out in the next section , reveals a @xmath19 value of about 21 . thus we find the qualitative situation that the ths result for the pion transition form factor ( times the virtuality @xmath51 ) overshoots the asymptotic limit , peaks at rather large momenta and approaches the asymptotic limit rather slowly from above . the gray bands in both panels of fig . [ fig : ff ] constitute our final ths predictions for the pion transition form factor . the results from the next section are anticipated where @xmath19 is further constrained . to obtain the gray bands all input for our ths pion transition form factor ( @xmath41 , @xmath44 ) is varied within the respective experimentally allowed regions , @xmath77 as described in the next section , and @xmath45 is varied within region ( [ eq : rangev2 ] ) . the whole gray band in the first panel shows the qualitative behavior described previously . it also points to a significant quantitative uncertainty of our prediction for the symmetric doubly virtual pion transition form factor . in other words , data on this form factor or in general on any doubly virtual pion transition form factor would be highly welcome to better constrain our approach . in the second panel of fig . [ fig : ff ] we display the singly virtual pion transition form factor . this plot shows two lines , a very narrow gray band and data . we first discuss the gray band , which nearly resembles a full line since its width is so small . obviously the narrow gray band of the second panel contrasts with the rather broad band of the first panel . however , the singly virtual pion transition form factor is independent of @xmath19 as can be easily deduced from ( [ eq : fffinal2 ] ) when putting @xmath52 to zero . the largest uncertainty comes from a variation of @xmath45 according to ( [ eq : rangev2 ] ) . even this variation does not lead to very different results , which causes the curves to nearly collapse to one single line . thus ths has rather high predictive ( or rather _ postdictive _ ) power for the singly virtual pion transition form factor . another qualitative difference between the gray bands of the first and the second panel is the fact that the ths result for the singly virtual pion transition form factor reaches the asymptotic ( b - l ) limit rather early . we regard this intriguing behavior of our correlator as one of the highlights of our work : the ths pion transition form factor shows an early ( late ) onset of the asymptotic behavior for the singly ( symmetric doubly ) virtual pion transition form factor . it would be extremely interesting to see if this is supported by data in the future . finally , we compare the ths result for the singly virtual pion transition form factor to vmd and lmd and to data . the lmd result for @xmath78 behaves as a constant for large @xmath79 . thus , it can not even qualitatively explain the data since , of course , it diverges after being multiplied by @xmath51 . this behavior can be seen in the second panel of fig . [ fig : ff ] where the lmd result is given by the dash - dotted line . the results for vmd ( dashed line ) and ths are fairly close . cleo data @xcite are only available for low virtualities ( @xmath80gev@xmath81 ) and prefer vmd to some extent though ths appears to be acceptable as well . the situation turns around for higher virtualities . ths describes the belle data @xcite better than vmd . the early onset of the asymptotic behavior of the ths result for the singly virtual pion transition form factor makes ths essentially identical to the b - l limit . not surprisingly then , ths agrees well with the belle data and is at odds with the babar data @xcite that are not fully compatible with the belle data . we do not attempt to contribute to a clarification of the differences between belle and babar . in view of these complications we have decided to stick to the b - l constraint ( [ eq : bl ] ) right from the start . our strategy now is to obtain the parameter @xmath19 of the newly proposed @xmath1 correlator ( [ eq : fffinal2 ] ) from the @xmath23 correlator . thus we turn to @xmath4 data for the reasons specified in section [ sec : intro ] , namely the lack of doubly virtual data for the pion transition form factor and the fact that the @xmath4 meson is fairly long lived to resemble the situation of the large-@xmath14 approximation . we introduce the overlap between an @xmath4 meson and the vector current @xmath82 , i.e. @xmath83 where we assume that the @xmath4 meson does not contain hidden strangeness , which is a fairly good approximation to the real world @xcite . for later convenience we also define the @xmath84-@xmath4 coupling strength @xmath85 the modulus of which agrees with @xmath86 as introduced in @xcite . with this at hand we can obtain @xmath87 from the @xmath88 decay process . a direct calculation from the lorentz invariant matrix element @xmath89 yields after averaging and summing over polarizations @xmath90^{-\frac12}\,,\ ] ] where @xmath91 is the speed of the electron in the rest frame of the decaying @xmath4 meson . taking into account that @xmath92 , we can write @xmath93 using the values @xmath94 and @xmath95mev @xcite we find @xmath96kev . together with @xmath97mev we obtain from ( [ eq : fv1 ] ) the following value for the coupling strength : @xmath98 it is convenient to introduce the @xmath18 transition form factor by @xmath99 this turns out to be the central quantity to determine @xmath19 and to address the phenomenologically interesting decays @xmath3 and @xmath100 , where @xmath101 denotes a lepton . for the special case when @xmath102 we find ( see also @xcite for a similar expression for the case of one vector multiplet ) @xmath103\,.\ ] ] hence the decay width for @xmath3 is independent of @xmath19 , which provides us with full predictive power for this particular decay and can therefore serve as a cross - check of our formalism . we will come back to this decay after having determined @xmath19 from the measured @xmath4-@xmath5 transition form factor . in fact , the parameter @xmath19 , which constitutes the only unknown parameter of our proposed @xmath1 correlator , can be obtained by a fit to the na60 data @xcite for the normalized transition form factor @xmath104 in the low - energy time - like region of @xmath51 with the result @xmath105 the result is displayed in fig . [ fig : ffomega ] as a gray band . the uncertainty comes mainly from the fitting procedure ( and so from the error bars of the data ) and from the variation of the second - multiplet mass @xmath106 inside the considered region ( [ eq : rangev2 ] ) . our fit to the na60 data has a @xmath107 per degree of freedom of 1.5 . apparently we obtain a rather satisfying fit to the na60 data except for the last two or three data points at the largest values of the dimuon mass . at present , none of the hadron - theory approaches to this @xmath4 transition form factor @xcite is able to understand these last data points . recently it has been suggested in @xcite using a dispersive calculation and high - energy constraints that these data points might be incompatible with qcd . clearly it would be highly desirable to obtain additional data for this @xmath4 transition form factor , in particular from experiments where the complete final state @xmath108 can be reconstructed and not only the dilepton ; see @xcite for more details . in the following we will stick to our result for @xmath19 as given in ( [ eq : kappa ] ) and obtained from a fit to the full range of na60 data since it is the best that one can do in the present situation . however , we briefly discuss two alternatives : if we performed an alternative fit to the na60 data rejecting the last three data points , we would get @xmath109 , which is fairly compatible with our full fit . the @xmath107 per degree of freedom would reduce to 0.8 . if we rejected the na60 data altogether and regarded the dispersive calculation of @xcite as the `` truth '' , we would find @xmath110 . the result of this last fit is also shown in fig . [ fig : ffomega ] as the dark gray band . if we return to the first panel of fig . [ fig : ff ] we observe that one obtains the same qualitative features with such values of @xmath19 . nevertheless , we will not use these results further on and will take into account only the result ( [ eq : kappa ] ) of the all - data fit . as for the pion transition form factor we shall compare our result to similar approaches from the literature . the form factor from @xcite , @xmath111 labeled by `` tl '' in fig . [ fig : ffomega ] , lies within our uncertainty band in spite of the fact that the derivation is based on a very different approach . the results for vmd and lmd can be obtained from ( [ eq : ffvmd ] ) and ( [ eq : fflmd ] ) , respectively , using ( [ eq : ffq2 ] ) and ( [ eq : ompinorm ] ) . for instance , the vmd expression is @xmath112 as can be seen in fig . [ fig : ffomega ] the results for both vmd and lmd deviate significantly from the data and therefore from our approach . they also deviate from the results of the dispersive calculation of @xcite , which is shown as the black band in fig . [ fig : ffomega ] . as already noted we have performed an alternative fit of our ths expression to the dispersive result the dark gray band in fig . [ fig : ffomega ] . it should be noted that our rational function can not fit the cusp structure in the low - energy region @xmath113 $ ] that emerges in the dispersive calculation from a cross - channel inelasticity ; see @xcite for a detailed discussion . we repeat our statement that additional data on this transition form factor would be highly welcome . 0.45 ( 208,152 ) ( 180,-230 ) ( 208,-152)(288,-152 ) ( 288,-152)(352,-88 ) ( 192,-152)[lb ] ( 368,-88)[lb ] ( 368,-224)[lb ] ( 352,-216)(288,-152)46.5 ( 288,-152)(14,14)(0)0.882 0.5 ( 220,165 ) ( 165,-230 ) ( 208,-152)(288,-152 ) ( 288,-152)(352,-88 ) ( 192,-152)[lb ] ( 368,-88)[lb ] ( 296,-192)[lb ] ( 336,-184)(288,-152)44.5 ( 336,-184)(368,-152 ) ( 368,-216)(336,-184 ) ( 380,-160)[lb ] ( 380,-224)[lb ] ( 288,-152)(14,14)(0)0.882 to further explore the validity of the ths scheme we study selected decay channels of the @xmath4 meson . it appears to be convenient to introduce the matrix element @xmath114 using previous definitions ( [ eq : ffp2q2 ] ) and ( [ eq : ffq2 ] ) this can be reduced to @xmath115 in ( [ eq : mpq1 ] ) the factors 2/3 and 1/3 emerge from the respective overlap of the @xmath4 meson with the singlet and the 8th component of the octet current , i.e. the factors come from ( [ eq : normdef ] ) . these factors actually sum to unity , which is not surprising from a flavor su(2 ) point of view where the @xmath4 is a pure singlet . we start with the prediction of the decay width of the @xmath3 process , which is depicted in fig . [ fig : omega]a . the matrix element @xmath116 does not depend on @xmath19 as already mentioned . this provides us with a pure prediction . after simple manipulations we find @xmath117 where we have used ( [ eq : fv1 ] ) to obtain the last expression . taking @xmath118 , @xmath119mev @xcite , @xmath120mev , @xmath121mev @xcite and @xmath106 in the range ( [ eq : rangev2 ] ) we obtain the values listed in tab . [ tab : gwpi0 g ] . . decay width of the process @xmath3 as obtained from different approaches and from data @xcite . we estimate an uncertainty only for our ths case . [ cols="^,^,^,^,^",options="header " , ] in the first row the nlo correction is presented . naively one might get from the definitions ( [ eq : dalitznlosl ] ) and ( [ eq : dalitzxlo ] ) the impression that the pion transition form factor does not enter @xmath122 . this is , however , not the case . the nlo correction involves , of course , the four - body phase - space integrations and the pion transition form factor can not be scaled out . this is due to the integration over the @xmath123 variable , which is tacitly performed also on the left - hand side of ( [ eq : dalitznlosl ] ) . on the other hand , taking into account the correction to the differential decay width , we would get an expression independent of the form factor . this statement then holds only when leaving out the 1@xmath84ir part , where the form - factor - dependent loop integration is performed . nonetheless , concerning model uncertainties , it is encouraging that numerically the nlo correction is to a very large extent independent of the used pion transition form factor . using the respective value of the nlo correction , the second row in tab . [ tab : dalitz ] shows the corresponding branching ratio if no cut on additional photons is applied , i.e. the branching ratio of the dalitz decay for the whole kinematic region @xmath124 $ ] . the results of this row should be compared to the experimental value @xmath125 , which has been calculated using inputs from @xcite . the agreement with the values of the second row in tab . [ tab : dalitz ] is very satisfying . in the previous section we have stated the value ( [ eq : bktev ] ) as the outcome of the ktev analysis . however , what ktev has determined at first place is from @xcite in contrast to 0.232 as stated in @xcite . this value for @xmath126 corresponds to ( 65mev/@xmath127)@xmath81 . ] @xmath128 we note in passing that for the quantity displayed in the last row of tab . [ tab : dalitz ] the corresponding value used by ktev to obtain finally the result ( [ eq : bktev ] ) is 3.19% @xcite . this value is close but not equal to any of the results in the last row of tab . [ tab : dalitz ] . having the radiative corrections determined from theory suggests to calculate directly the ratio ( [ eq : ktev232 ] ) instead of the derived quantity ( [ eq : bktev ] ) and to compare the calculation to the experimental result . however , the result ( [ eq : ktev232 ] ) is not a direct measurement either , but is , of course , based on an interplay of measurements , monte carlo simulations , acceptance corrections , etc . @xcite . by the time the ktev measurements were analyzed not all qed radiative corrections were available . to evaluate the decay width of the dalitz decay the radiative corrections from @xcite have been used . in particular the 1@xmath84ir contributions have not been included in @xcite . in addition , in the monte carlo generator of ktev the radiative corrections for the @xmath129 decay have been adopted from @xcite ; see @xcite . however , in @xcite some approximations have been used that have been proven to be misleading after the exact calculation has been performed @xcite . therefore it would be somewhat misleading to compare a theory result with all radiative corrections included to an experimental result where only part of these corrections have been taken into account . on the other hand , it might be illuminating to provide a rough estimate of the impact of these differences . before presenting such an estimate we would like to repeat the take - home message from the first section : all nlo qed radiative corrections are now available @xcite and can be taken into account in future analyses of data on the pion decays @xmath13 and @xmath130 . in spite of the fact that not all radiative corrections have been taken into account in the analysis that led to the result ( [ eq : ktev232 ] ) we do not feel legitimated to modify this experimental result . instead we will calculate the ratio of ( [ eq : ktev232 ] ) within ths the results from the other approaches are comparable . ] , but assign an uncertainty to it , which is related to the neglect of the 1@xmath84ir contributions.ir contributions . ] it should be stressed , however , that this is not an entirely valid approach either . after all , as soon as one can calculate the nlo qed radiative corrections , they do not constitute an uncertainty of a result but rather a well - defined shift of the result . nonetheless , we proceed in the described way to obtain a rough estimate of the uncertainty caused by neglecting part of the radiative corrections . the quantity of our interest can be generally expressed and expanded as @xmath131\,.\ ] ] in our case we have @xmath132 and @xmath133 , which we calculate up to nlo using the complete set of corrections . the uncertainties are then given by the 1@xmath84ir contributions to the nlo virtual radiative corrections , i.e. by the interference of the diagrams shown in fig . [ fig : d1gir ] . hence @xmath134}\frac{\gamma^\text{lo}(\pi^0\to e^+e^-\gamma\,,\;x>0.95)}{\gamma^\text{lo}(\pi^0\to e^+e^- ) } \ , , \label{eq : sa } \end{split}\ ] ] which numerically yields @xmath135 . for the second term we get @xmath136}\ , , \label{eq : sb } \end{split}\ ] ] for which we find @xmath137 . we can thus conclude that the total uncertainty would be at the level of 1% . for completeness we list the values that we have used to get the above stated numbers . the input for ( [ eq : sa ] ) is @xmath138 for the 1@xmath84ir contribution to the nlo virtual radiative correction for the decay width of the dalitz decay . for the ratio of the lo decay widths of the dalitz decay with @xmath139 and the @xmath9 process we have @xmath140 . in ( [ eq : sb ] ) we have utilized for the overall nlo correction to the dalitz decay @xmath141 and finally @xmath142 . in general , we can see that the radiative corrections are getting more important with higher values of @xmath126 . taking values from tab . [ tab : ktev ] and tab . [ tab : dalitz ] and the previously estimated relative uncertainty of 1% ( we state this separately ) we find @xmath143 this result is to be compared with the ktev value ( [ eq : ktev232 ] ) , yielding a discrepancy of approximately 1.5@xmath12 . we stress one more time that the result ( [ eq : finallabel ] ) should be read with caution . several directions are conceivable how the ths scheme might be extended or combined with other approaches . most straightforward would be to extend ths to other correlators ; see also @xcite . eventually one might get cross correlations and in that way more information , e.g. , by fitting to additional sets of data . in the present work we have merely used the @xmath0 correlator as an intermediate step to obtain the @xmath1 correlator . of course , the @xmath0 correlator is also interesting in its own right . for instance , as it has been worked out in @xcite , the low - energy expansion of this correlator determines some of the low - energy constants of the chiral lagrangian at order @xmath144 in the anomalous sector @xcite , i.e. the corrections to the wess zumino witten lagrangian . having determined some of the parameters of the @xmath0 correlator within ths by demanding high - energy constraints one could provide estimates and/or cross - correlations for these low - energy constants following the procedure outlined in @xcite . as already spelled out in section [ sec : intro ] we refrained from using high - energy constraints for quantities like the @xmath23 correlator since the vector meson might probe already too many details of the intermediate - energy region ( above 1 gev ) where we have approximated the tower of infinitely many states by one effective state per channel . extending the scheme from ths to three hadrons per channel , albeit introducing more parameters in the first place , might allow to use quark - scaling relations along the lines of @xcite for @xmath145 and similar correlators which in turn could help to keep the number of free parameters manageable . in the same spirit one can use subleading orders in the high - energy expansion in terms of qcd condensates @xcite . in the present work we have used the chiral limit to construct the @xmath0 correlator . in line with this approximation we have focused on the @xmath1 and @xmath23 correlator where no strange quarks are involved . for a reasonable extension from the pion to the whole pseudoscalar multiplet ( a nonet in the large-@xmath14 limit ) one has to go beyond the chiral limit . the prospect of such an endeavor would be the possibility to tackle the transition form factors of the @xmath146 and @xmath147 mesons and their rare decays into electron and positron or muon and anti - muon . the simplest way to go beyond the chiral limit would be to utilize a pertinent lagrangian . in extensions of chiral perturbation theory @xcite one frequently uses a lagrangian where the vector mesons are represented by antisymmetric tensor fields @xcite . as discussed in @xcite a given resonance lagrangian of formal chiral order @xmath144 might not be capable of satisfying all desired ope constraints . indeed , as shown in @xcite this problem appears at least if only one vector - meson multiplet is used for the @xmath0 correlator . concerning the vector instead of the antisymmetric tensor representation the same negative result has been reported in @xcite . it might appear that the first - order approach suggested in @xcite offers more flexibility here . alternatively or in addition one might consider the inclusion of resonance terms of chiral order @xmath148 . the following consideration might demonstrate that one can expect to create additional terms that would not spoil the ope right away : if one normalizes the @xmath0 correlator to the corresponding chiral - anomaly expression , a @xmath144 resonance lagrangian would produce `` corrections '' that schematically are given by @xmath149 we have replaced all resonance masses by @xmath150 and all momenta by @xmath151 . this is sufficient to discuss the high- and low - energy limits . the factor of @xmath152 in the numerator of emerges from the fact that the resonance lagrangian is assumed to be of chiral order @xmath144 while the chiral anomaly is of order @xmath153 . the limitation @xmath154 comes from the fact that a three - point correlator is considered . at high energies a single term with @xmath155 definitely spoils the ope . coupling constants between various contributions must be adjusted such that cancellations appear . of course , this is the reason why the ope is capable to tell something about the resonance parameters . a resonance lagrangian of chiral order @xmath148 produces terms with the schematic structure @xmath156 here the terms with @xmath157 spoil the ope and must cancel . the terms with @xmath158 , however , produce structures that could be compatible with the ope but are structurally different from the ones in . thus the inclusion of resonance terms of chiral order @xmath148 might help to obtain an agreement between a lagrangian approach and the ope . needless to say that the systematic construction of a @xmath148 lagrangian would be extremely tedious ; see @xcite and references therein . in practice , according to @xcite , leading - order ope constraints together with a singly virtual pion transition form factor that vanishes at large momenta can be achieved by a resonance lagrangian with two vector - meson multiplets in the antisymmetric tensor representation . note , however , that the concrete b - l limit ( [ eq : bl ] ) has not been used in @xcite . nonetheless , the results of @xcite suggest that ths can be reproduced from a resonance lagrangian of order @xmath144 in the chiral counting . while leaving the chiral limit might be systematized by a lagrangian approach , our second basic assumption , the large-@xmath14 limit is much harder to remedy . on purpose we stuck to narrow states ( @xmath5 , @xmath4 ) and to the space - like or low - energy time - like region where the effects of inelasticities are minor . nonetheless this induces uncertainties , which are hard to assess on a quantitative level . at low energies an excellent way to deal with inelasticities is to use them in ones favor by a dispersive setup @xcite . a powerful framework emerges when combined with chiral low - energy constraints , qcd high - energy constraints , and precise data on the scattering amplitudes that come into play via the inelasticities . in practice the scheme becomes intractable at intermediate energies ( roughly above 1 gev ) where more and more many - particle channels open up . even if one studies low - energy quantities one might need some information from the intermediate - energy regime if the qcd high - energy limit is reached very late . this is exactly what we saw for the ths result of the @xmath1 correlator . this suggests to study the impact of the ths results on a dispersive calculation by using ths in the intermediate- and high - energy space - like regime where its results should be most trustworthy . in the present work we have determined the @xmath1 correlator with a focus on the rare pion decay into electron and positron . in turn this decay might offer some window to observe low - energy traces of physics beyond the standard model . however , the @xmath1 correlator is also interesting because it contains information about the intrinsic structure of the pion . hence it could be illuminating to figure out what the ths result implies for the pion distribution function @xcite . t.h . would like to thank for the hospitality of uppsala university and to all his colleagues there who made a very friendly atmosphere . we also acknowledge valuable discussions with our colleagues k. kampf and j. novotn from prague . we are indebted to m. knecht for bringing @xcite to our attention . the present work was supported by the charles university grants gauk 700214 and svv 260219/2015 , by the czech science foundation grant gar 15 - 18080s and by the project `` study of strongly interacting matter '' ( hadronphysics3 , grant agreement no . 283286 ) under the 7th framework program of the eu . in section [ sec : piee ] we have defined the invariant amplitude for the process @xmath9 using the expression @xmath159 . in this appendix we summarize its explicit form for various approaches that lead to a rational function for the pion transition form factor . we make use of the standard passarino veltman @xcite scalar one - loop integrals @xmath42 and @xmath160 . the only uv divergent function is then @xmath42 . its explicit form will be given here as a reference point for the used notation : @xmath161 ^ 2}\\ & = i\pi^2\bigl[\frac 1\varepsilon-\gamma_e+\log4\pi+\log\left(\frac{\mu^2}{m^2}\right)\bigr]\ , , \end{split}\ ] ] where we have introduced @xmath162 . note that differences of @xmath42 functions are finite . it will then be manifest from the form in which the results are presented , that the divergent parts cancel yielding a finite amplitude as desired . it is also convenient to introduce the following combination of the three - point scalar one - loop function @xmath160 and the klln triangle function @xmath57 : @xmath163 for instance , we present two @xmath164 functions that are used further in this appendix : @xmath165 @xmath166 now we can finally list the results for @xmath167 as obtained from the various approaches to the pion transition form factor : @xmath168\\ & + c^\prime_0(m_e^2,m_e^2,m_\pi^2,0,m_e^2,0)\\ & -2\,c^\prime_0(m_e^2,m_e^2,m_\pi^2,0,m_e^2 , m_{v_1}^2)\\ & + c^\prime_0(m_e^2,m_e^2,m_\pi^2 , m_{v_1}^2,m_e^2 , m_{v_1}^2 ) \bigg\}\ , , \end{split}\ ] ] @xmath169\\ & -\bigg(\frac { m_{v_1}^2}{m_\pi^2}-1\bigg)\big[b_0(m_e^2,m_e^2 , m_{v_1}^2)-b_0(0,m_e^2,m_e^2)-2\big]\\ & + c^\prime_0(m_e^2,m_e^2,m_\pi^2,0,m_e^2 , m_{v_1}^2)\\ & -c^\prime_0(m_e^2,m_e^2,m_\pi^2 , m_{v_1}^2,m_e^2 , m_{v_1}^2 ) \bigg\}\ , , \end{split}\ ] ] @xmath170\\ & + \frac2{m_\pi^2}\big[7\,b_0(m_e^2,m_e^2 , m_{v_2}^2)-b_0(m_e^2,m_e^2 , m_{v_1}^2)\\ & -6\,b_0(0,m_e^2,m_e^2)\big]\bigg)\bigg]+\bigg [ m_{v_1}\leftrightarrow m_{v_2}\bigg]\\ & -\frac{2 } { m_{v_2}^2- m_{v_1}^2}\bigg(1+\frac{2\kappa m_{v_1}^4 m_{v_2}^4}{(4\pi f)^6m_\pi^2}\bigg)\\ & \times\big[b_0(m_e^2,m_e^2 , m_{v_2}^2)-b_0(m_e^2,m_e^2 , m_{v_1}^2)\big]\bigg\}\\ % & + c^\prime_0(m_e^2,m_e^2,m_\pi^2,0,m_e^2,0)\\ & -\bigg\{\frac { m_{v_2}^2 } { m_{v_2}^2- m_{v_1}^2}\bigg(2-\frac{3}{n_c}\frac{(4\pi f)^2 } { m_{v_2}^2}\bigg)\\ & \times c^\prime_0(m_e^2,m_e^2,m_\pi^2,0,m_e^2 , m_{v_1}^2)-\frac { m_{v_2}^4}{2 ( m_{v_2}^2- m_{v_1}^2)^2}\\ & \times\bigg[2+\frac{\kappa m_{v_1}^4}{n_c(4\pi f)^4}-\frac{(4\pi f)^2}{n_c m_{v_2}^2}\bigg(6+\frac { m_{v_1}^2 } { m_{v_2}^2}\bigg)\bigg]\\ & \times c^\prime_0(m_e^2,m_e^2,m_\pi^2 , m_{v_1}^2,m_e^2 , m_{v_1}^2)\bigg\}-\bigg\ { m_{v_1}\leftrightarrow m_{v_2}\bigg\}\\ & -\frac { m_{v_1}^2 m_{v_2}^2 } { ( m_{v_2}^2- m_{v_1}^2)^2}\\ & \times\bigg[2+\frac{\kappa m_{v_1}^2 m_{v_2}^2}{n_c(4\pi f)^4}-\frac{14}{n_c}(2\pi f)^2\bigg(\frac1 { m_{v_1}^2}+\frac1 { m_{v_2}^2}\bigg)\bigg]\\ & \times c^\prime_0(m_e^2,m_e^2,m_\pi^2 , m_{v_1}^2,m_e^2 , m_{v_2}^2 ) \bigg\}\ , . \tag{\stepcounter{equation}\theequation}\end{aligned}\ ] ]

in this work we find the capacity of a compound finite - state channel with time - invariant deterministic feedback . the model we consider involves the use of fixed length block codes . our achievability result includes a proof of the existence of a universal decoder for the family of finite - state channels with feedback . as a consequence of our capacity result , we show that feedback does not increase the capacity of the compound gilbert - elliot channel . additionally , we show that for a stationary and uniformly ergodic markovian channel , if the compound channel capacity is zero without feedback then it is zero with feedback . finally , we use our result on the finite - state channel to show that the feedback capacity of the memoryless compound channel is given by . compound channel , feedback capacity , finite state channel , directed information , causal conditioning probability , gilbert - elliot channel , universal decoder , code - trees , types of code - trees , sanov s theorem , pinsker s inequality

0711.0705

the compound channel consists of a set of channels indexed by @xmath1 with the same input and output alphabets but different conditional probabilities . in the setting of the compound channel only one actual channel @xmath2 is used in all transmissions . the transmitter and the receiver know the family of channels but they have no prior knowledge of which channel is actually used . there is no distribution law on the family of channels and the communication has to be reliable for all channels in the family . blackwell et al . @xcite and independently wolfowitz @xcite showed that the capacity of a compound channel consisting of memoryless channels only , and without feedback , is given by @xmath3 where @xmath4 denotes the input distribution to the channel , @xmath5 denotes the conditional probability of a memoryless channel indexed by @xmath2 , and the notation @xmath6 denotes the mutual information of channel @xmath7 for the input distribution @xmath8 , i.e. , @xmath9 the capacity in ( [ eqn : capacitymemorylesscompoundnofb ] ) is in general less than the capacity of every channel in the family . wolfowitz , who coined the term `` compound channel , '' showed that if the transmitter knows the channel @xmath2 in use , then the capacity is given by @xcite @xmath10 where @xmath11 is the capacity of the channel indexed by @xmath2 . this shows that knowledge at the transmitter of the channel @xmath2 in use helps in that the infimum of the capacities of the channels in the family can now be achieved . in the case that @xmath12 is a finite set , then it follows from wolfowitz s result that @xmath13 is the feedback capacity of the memoryless compound channel , since the transmitter can use a training sequence together with the feedback to estimate @xmath2 with high probability . in this paper we show that when @xmath12 is not limited to finite cardinality , the feedback capacity of the memoryless compound channel is given by @xmath14 . one might be tempted to think that for a compound channel with memory , feedback provides a means to achieve the infimum of the capacities of the channels in the family . however this is not necessarily true , as we show in example [ example : compoundge ] , which is taken from @xcite and applied to the compound gilbert - elliot channel with feedback . that example is found in section [ section : compoundge ] . a comprehensive review of the compound channel and its role in communication is given by lapidoth and narayan @xcite . of specific interest in this paper are compound channels with memory which are often used to model wireless communication in the presence of fading @xcite . lapidoth and telatar @xcite derived the following formula for the compound channel capacity of the class of finite state channels ( fsc ) when there is no feedback available at the transmitter . @xmath15 where @xmath16 denotes the initial state of the fsc , and @xmath17 and @xmath18 denote the input distribution and channel conditional probability for block length @xmath19 . lapidoth and telatar s achievability result makes use of a universal decoder for the family of finite - state channels . the existence of the universal decoder is proved by feder and lapidoth in @xcite by merging a finite number of maximum - likelihood decoders , each tuned to a channel in the family @xmath12 . throughout this paper we use the concepts of causal conditioning and directed information which were introduced by massey in @xcite . kramer extended those concepts and used them in @xcite to characterize the capacity of discrete memoryless networks . subsequently , three different proofs tatikonda and mitter @xcite , permuter , weissman and goldsmith @xcite and kim @xcite have shown that directed information and causal conditioning are useful in characterizing the feedback capacity of a point - to - point channel with memory . in particular , this work uses results from @xcite that show that gallager s 4,5 ) upper and lower bound on capacity of a fsc can be generalized to the case that there is a time - invariant deterministic feedback , @xmath20 , available at the encoder at time @xmath21 . in this paper we extend lapidoth and telatar s work for the case that there is deterministic time - invariant feedback available at the encoder by replacing the regular conditioning with the causal conditioning . then we use the feedback capacity theorem to study the compound gilbert - elliot channel and the memoryless compound channel and to specify a class of compound channels for which the capacity is zero if and only if the feedback capacity is zero . the proof of the feedback capacity of the fsc is found in section [ section : converse ] , which describes the converse result , and section [ section : achievability ] , where we prove achievability . as a consequence of the capacity result , we show in section [ section : compoundge ] that feedback does not increase the capacity of the compound gilbert - elliot channel . we next show in section [ section : iff ] that for a family of stationary and uniformly ergodic markovian channels , the capacity of the compound channel is positive if and only if the feedback capacity of the compound channel is positive . finally , we return to the memoryless compound channel in section [ section : memorylesscompound ] and make use of our capacity result to provide a proof of the feedback capacity . the notation we use throughout is as follows . a capital letter @xmath22 denotes a random variable and a lower - case letter , @xmath23 , denotes a realization of the random variable . vectors are denoted using subscripts and superscripts , @xmath24 and @xmath25 . we deal with discrete random variables where a probability mass function on the channel input is denoted @xmath26 and @xmath27 denotes a mass function on the channel output . when no confusion can result , we will omit subscripts from the probability functions , i.e. , @xmath28 will denote @xmath29 . the problem we consider is depicted in figure [ fig:1 ] . a message @xmath30 from the set @xmath31 is to be transmitted over a compound finite state channel with time - invariant deterministic feedback . the family @xmath12 of finite state channels has a common state space @xmath32 and common finite input and output alphabets given by @xmath33 and @xmath34 . for a given channel @xmath35 the channel output at time @xmath21 is characterized by the conditional probability @xmath36 which satisfies the condition @xmath37 . the channel @xmath2 is in use over the sequence of @xmath19 channel inputs . the family @xmath12 of channels is known to both the encoder and decoder , however , they do not have knowledge of the channel @xmath2 in use before transmission begins . [ ] [ ] [ 0.8]@xmath38[][][0.8]message [ ] [ ] [ 0.8]encoder [ ] [ ] [ 0.7]@xmath39 [ ] [ ] [ 0.8]@xmath40 [ ] [ ] [ 0.8]@xmath41 [ ] [ ] [ 0.8 ] [ ] [ ] [ 0.7]@xmath42 [ ] [ ] [ 0.8]@xmath43 [ ] [ ] [ 0.8]@xmath41[][][0.8]@xmath44 [ ] [ ] [ 0.8]decoder [ ] [ ] [ 0.8]@xmath45 [ ] [ ] [ 0.8]feedback generator [ ] [ ] [ 0.8]@xmath46 [ ] [ ] [ 0.8]@xmath47 [ ] [ ] [ 0.8]@xmath41 [ ] [ ] [ 0.8]unit delay [ ] [ ] [ 0.8]@xmath48 [ ] [ ] [ 0.8]@xmath41 [ ] [ ] [ 0.8]@xmath49[][][0.8]estimated [ ] [ ] [ 0.8]message the message @xmath30 is encoded such that at time @xmath21 the codeword symbol @xmath50 is a function of @xmath30 and the feedback sequence @xmath51 . for notational convenience , we will refer to the input sequence @xmath52 as simply @xmath53 . the feedback sequence is a time - invariant deterministic function of the output @xmath54 and is available at the encoder with a single time unit delay . the function performed on the channel output @xmath54 to form the feedback @xmath55 is known to both the transmitter and receiver before communication begins . the decoder operates over the sequence of channel outputs @xmath56 to form the message estimate @xmath57 . for a given initial state @xmath58 and channel @xmath35 , the channel causal conditioning distribution is given by @xmath59 additionally we will make use of massey s directed information @xcite . when conditioned on the initial state and channel , the directed information is given by @xmath60 our capacity result will involve a maximization of the directed information over the input distribution @xmath61 which is defined as @xmath62 we make use of some of the properties provided in @xcite in our work , including the following three which we restate for our problem setting . 1 . @xmath63 ( * ? ? ? * eq . ( 3 ) ) * lemma 1 ) [ haimchainrulecausalcond ] 2 . @xmath64 , where random variable @xmath65 denotes the state of the finite - state channel ( * ? ? ? * lemma 5 ) 3 . from ( * ? ? ? * lemma 6 ) [ haimdirectedinfoequivi ] , @xmath66 note that properties [ haimchainrulecausalcond ] ) and [ haimdirectedinfoequivi ] ) hold since @xmath67 for our feedback setting , where it is assumed that the state @xmath16 is not available at the encoder . for a given initial state @xmath16 and channel @xmath2 the average probability of error in decoding message @xmath44 is given by @xmath68 where @xmath69 is a function of the message @xmath44 and of the feedback @xmath70 . the average ( over messages ) error probability is denoted @xmath71 , where @xmath72 . we say that a rate @xmath73 is achievable for the compound channel with feedback as shown in figure [ fig:1 ] , if for any @xmath74 there exists a code of fixed blocklength @xmath19 and rate @xmath73 , i.e. @xmath75 , such that @xmath76 for all @xmath77 and @xmath58 . equivalently , rate @xmath73 is achievable if there exists a sequence of rate-@xmath73 codes such that @xmath78 this definition of achievable rate is identical to that given in previous work on the compound channel without feedback . a different definition for the compound channel with feedback could also be considered ; for instance , in @xcite , the authors consider codes of variable blocklength and define achievability accordingly . the capacity is defined as the supremum over all achievable rates and is given in the following theorem . [ thrm : fbcapcompoundfsc ] the feedback capacity of the compound finite state channel is given by @xmath79 theorem [ thrm : fbcapcompoundfsc ] is proved in section [ section : converse ] , which shows the existence of @xmath80 and proves the converse , and section [ section : achievability ] , where achievability is established . we first state the following proposition , which shows that the capacity @xmath80 as defined in theorem [ thrm : fbcapcompoundfsc ] exists . the proof is found in appendix [ app : existenceproof ] . [ prop : existencec ] let @xmath81 then @xmath82 is well defined and converges for @xmath83 . in addition , let @xmath84 then @xmath85 to prove the converse in theorem [ thrm : fbcapcompoundfsc ] , we assume a uniform distribution on the message set , for which @xmath86 . since the message is independent of the channel parameters @xmath87 and we apply fano s inequality as follows . @xmath88 for any code we have @xmath89 and therefore @xmath90 by combining the above statement with proposition [ prop : existencec ] we have @xmath91 then for a sequence of codes of rate @xmath73 with @xmath92 , this implies @xmath93 . before proving achievability , we mention a simple case which follows from previous results . if the set @xmath12 has finite cardinality , then achievability follows immediately from the results in ( * ? ? ? * theorem 14 ) , which are true for any finite state channel with feedback . hence , we can construct a finite state channel where the augmented state is @xmath94 and by assuming that the initial distribution is positive for all @xmath95 then we get that for any latexmath:[$\theta \in \theta , @xmath73 is achievable if @xmath97 more work is needed in the achievability proof when the set @xmath12 is not restricted to finite cardinality . this is outlined in the following subsections in three steps . in the first step , we assume that the decoder knows the channel @xmath2 in use and we show in theorem [ thrm : achievabilitythetaknown ] that if @xmath98 and if the decoder consists of a maximum - likelihood decoder , then there exist codes for which the error probability decays uniformly over the family @xmath12 and exponentially in the blocklength . the codes used in showing this result are codes of blocklength @xmath99 where each sub - block of length @xmath38 is generated i.i.d . according to some distribution . in the second step , we show in lemma [ lemma : typeoncodetree ] that if instead the codes are chosen uniformly and independently from a set of possible blocklength-@xmath99 codes , then the error probability still decays uniformly over @xmath12 and exponentially in the blocklength . in the third and final step , we show in theorem [ thrm : univdecoderforseparable ] and lemma [ lemma : compoundfscisstronglysep ] that for codes chosen uniformly and independently from a set of blocklength-@xmath99 codes , there exists a decoder that for every channel @xmath77 achieves the same error exponent as the maximum - likelihood decoder tuned to @xmath2 . in the sections that follow , @xmath100 denotes the set of probability distributions on @xmath101 causally conditioned on @xmath102 . we begin by proving that if the decoder is tuned to the channel @xmath35 in use , i.e. , if the decoder knows the channel @xmath2 in use , and if @xmath103 then the average error probability approaches zero . this is proved through the use of random coding and maximum likelihood ( ml ) decoding . the encoding scheme consists of randomly generating a _ code - tree _ for each message @xmath44 , as shown in figure [ f_codetree](b ) for the case of binary feedback . a code - tree has depth @xmath19 corresponding to the blocklength and level @xmath21 designates a set of @xmath104 possible codeword symbols . one of the @xmath104 symbols is chosen as the input @xmath50 according to the feedback sequence @xmath105 . the first codeword symbol is generated as @xmath106 . the second codeword symbol is generated by conditioning on the previous codeword symbol and on the feedback , @xmath107 for all possible values of @xmath108 . for instance , in the binary case , @xmath109 , two possible values ( branches ) of @xmath110 will be generated and the transmitted codeword symbol will be selected from among these two values according to the value of the feedback @xmath111 . subsequent codeword symbols are generated similarly , @xmath112 for all possible @xmath105 . for a given feedback sequence @xmath70 , the input distribution , corresponding to the distribution on a path through the tree of depth @xmath19 , is @xmath113 [ ] [ ] [ 0.8]@xmath114 [ ] [ ] [ 0.8]@xmath41 [ ] [ ] [ 0.8]@xmath115 [ ] [ ] [ 0.8]@xmath116 [ ] [ ] [ 0.8]@xmath117 [ ] [ ] [ 0.8]@xmath118 [ ] [ ] [ 0.8]@xmath119 [ ] [ ] [ 0.8]@xmath120 [ ] [ ] [ 0.8]@xmath114 [ ] [ ] [ 0.8]@xmath41 [ ] [ ] [ 0.8]@xmath115 [ ] [ ] [ 0.8]@xmath41 [ ] [ ] [ 0.8]@xmath115 [ ] [ ] [ 0.8]@xmath41 [ ] [ ] [ 0.8]@xmath121 [ ] [ ] [ 0.8]@xmath41 [ ] [ ] [ 0.8]@xmath117 [ ] [ ] [ 0.8]@xmath41 [ ] [ ] [ 0.8]@xmath117 [ ] [ ] [ 0.8]@xmath41 [ ] [ ] [ 0.8]@xmath117 [ ] [ ] [ 0.8]@xmath41 [ ] [ ] [ 0.8]@xmath119 [ ] [ ] [ 0.8]@xmath122 [ ] [ ] [ 0.8]@xmath41 [ ] [ ] [ 0.8]@xmath123 [ ] [ ] [ 0.8]@xmath124 [ ] [ ] [ 0.8]@xmath125 [ ] [ ] [ 0.8]@xmath126 ( no feedback ) [ ] [ ] [ 0.8](a ) codeword ( no feedback ) [ ] [ ] [ 1](b ) code - tree [ ] [ ] [ 1](c ) concatenated code - tree a code - tree of depth @xmath19 is a vector of @xmath127 symbols , where @xmath128 and each element in the vector takes value from the alphabet @xmath129 . we denote a random code - tree by @xmath130 and a realization of the random code - tree by @xmath131 . the probability of a tree @xmath132 is uniquely determined by @xmath133 . for instance , consider the case of binary feedback , @xmath134 , and a tree of depth @xmath135 , for which @xmath136 . a code - tree is a vector @xmath137 where @xmath138 is the symbol sent at time @xmath116 , @xmath139 is the symbol sent at time @xmath118 for feedback @xmath140 , and @xmath141 is the symbol sent at time @xmath118 for feedback @xmath142 . then @xmath143 which is uniquely determined by @xmath144 . in general , for a code - tree of depth @xmath19 , the following holds . @xmath145 a code - tree for each message @xmath44 is randomly generated , and for each message @xmath44 and feedback sequence @xmath70 the codeword @xmath146 is unique . the decoder is made aware of the code - trees for all messages . assuming that the ml decoder knows the channel @xmath2 in use , it estimates the message as follows . @xmath147 as shown in @xcite , since @xmath148 is uniquely determined by @xmath44 and @xmath105 and since @xmath149 is a deterministic function of @xmath150 , we have the equivalence @xmath151 so the ml decoder can be described as @xmath152 let @xmath153 denote the average ( over messages ) error probability incurred when a code of blocklength @xmath19 is used over channel @xmath2 with initial state @xmath16 . the following theorem bounds the error probability uniformly in @xmath95 when the decoder knows the channel @xmath77 in use . the theorem is proved in appendix [ app : achievabilitythetaknown ] . [ thrm : achievabilitythetaknown ] for a compound fsc with initial state @xmath58 , input alphabet @xmath33 , and output alphabet @xmath34 , assuming that the decoder knows the channel @xmath2 in use , then there exists a code of rate @xmath73 and blocklength @xmath99 , where @xmath154 and @xmath38 is chosen such that @xmath155 , for which the error probability @xmath156 of the ml decoder satisfies @xmath157 for any @xmath77 , where @xmath158 the result in theorem [ thrm : achievabilitythetaknown ] is shown by the use of a randomly - generated code - tree of depth @xmath99 for each message @xmath44 . for every feedback sequence @xmath159 , the corresponding path in the code - tree is generated by the input distribution @xmath160 given by @xmath161 where @xmath162 is the distribution that achieves the supremum in @xmath163 . the random codebook @xmath164 used in proving theorem [ thrm : achievabilitythetaknown ] consists of @xmath165 code - trees . each code - tree in the codebook is a concatenated code - tree with depth @xmath99 consisting of @xmath166 code - trees , each of depth @xmath38 . for a given feedback sequence @xmath159 ( corresponding to a certain path in the concatenated code - tree ) the codeword is generated by @xmath167 . an example of a concatenated code - tree is found in figure [ f_codetree](c ) . in this subsection we show that the result in theorem [ thrm : achievabilitythetaknown ] implies that the error probability can be similarly bounded when codewords are chosen uniformly over a set . in other words , we convert the random coding exponent given in theorem [ thrm : achievabilitythetaknown ] , where it is assumed that the codebook consists of concatenated code - trees of depth @xmath99 in which each sub - tree of depth @xmath38 is generated i.i.d . according to @xmath162 , to a new random coding exponent for which the concatenated code - trees in the codebook are chosen uniformly from a set of concatenated code - trees . this alternate type of random coding , where the concatenated code - trees are chosen uniformly from a set , is the coding approach subsequently used to prove the existence of a universal decoder . we first introduce the notion of types on code - trees . let @xmath168 denote the concatenation of @xmath166 depth-@xmath38 code - trees @xmath169 , where @xmath170 is defined in ( [ eqn : codetreenumsymbols ] ) and @xmath171 . the type ( or empirical probability distribution ) of a concatenated code - tree @xmath168 is the relative proportion of occurrences of each code - tree @xmath172 . equivalently , @xmath166 multiplied by the type of @xmath168 indicates the number of times each depth-@xmath38 code - tree from the set @xmath173 occurs in the concatenated code - tree @xmath168 . let @xmath174 denote the set of types of concatenated code - trees of depth @xmath99 . let @xmath175 denote the average probability of error incurred when a code - tree of depth @xmath19 and rate @xmath73 drawn according to a distribution @xmath176 is used over the channel @xmath177 . we now prove the following result . [ lemma : typeoncodetree ] given @xmath178 , let @xmath179 denote the distribution given by the n - fold product of @xmath180 , i.e. , @xmath181 for a given type @xmath182 , let @xmath183 denote the distribution that is uniform over the set of concatenated code - trees of type @xmath184 . for every distribution @xmath185 there exists a type @xmath182 whose choice depends on @xmath180 and @xmath166 but not on @xmath177 such that @xmath186 for all @xmath177 , where @xmath187 tends to 0 as @xmath188 . the proof follows the approach of ( * ? ? ? * lemma 3 ) except that our codebook consists of code - trees rather than codewords ; we include this proof for completeness in describing the notion of types on code - trees . given a codebook @xmath164 of rate @xmath189 chosen according to @xmath190 , we can construct a sub - code @xmath191 of rate @xmath73 in the following way . let @xmath192 denote the type with the highest occurrence in @xmath164 . the number of types in @xmath164 is upper bounded by @xmath193 , so the number of concatenated code - trees of type @xmath192 is lower bounded by @xmath194 . we construct the code @xmath191 by picking the first @xmath165 concatenated code - trees of type @xmath192 . since @xmath191 is a sub - code of @xmath164 , its average probability of error is upper bounded by the average probability of error of @xmath164 times @xmath195 . conditioned on @xmath192 , the codewords in @xmath191 are mutually independent and uniformly distributed over a set of concatenated code - trees of type @xmath192 . since @xmath164 is a random code , the type @xmath192 is also random , and let @xmath196 denote the distribution of @xmath192 . pick a realization of the type @xmath192 , denoted @xmath184 , that satisfies @xmath197 . ( this is possible since the number of types is upper bounded by @xmath198 . ) then @xmath199 and @xmath200 combining this result with theorem [ thrm : achievabilitythetaknown ] , we have that there exists a type @xmath182 such that when the codewords are chosen uniformly from the type class of @xmath184 , given by the distribution @xmath201 , the average probability of error is bounded as @xmath202 \right\}\end{aligned}\ ] ] it is then possible to choose @xmath203 such that for all @xmath204 , @xmath205 and @xmath206 which implies that the probability of error is bounded as @xmath207 we next show that when a codebook is constructed by choosing code - trees uniformly from a set , there exists a universal decoder for the family of finite - state channels with feedback . this result is shown in the following four steps . * we define the notion of a strongly separable family @xmath12 of channels given by the causal conditioning distribution . the notion of strong separability means that the family is well - approximated by a finite subset of the channels in @xmath12 . * we prove that for strongly separable @xmath12 and code - trees chosen uniformly from a set , there exists a universal decoder . * we describe the universal decoder which `` merges '' the ml decoders tuned to a finite subset of the channels in @xmath12 . * we show that the family of finite - state channels given by the causal conditioning distribution is a strongly separable family . our approach follows precisely the approach of feder and lapidoth @xcite except that our codebook consists of concatenated code - trees ( rather than codewords ) and our channel is given by the causal conditioning distribution . let @xmath168 denote a concatenated code - tree of depth @xmath99 , @xmath208 where @xmath209 , and let @xmath210 denote a set of such code - trees , @xmath211 . as described in lemma [ lemma : typeoncodetree ] , @xmath210 will be the set of code - trees of type @xmath182 and the code - tree for each message will be chosen uniformly from this set , i.e. @xmath212 for any @xmath213 . as described below , for a given output sequence @xmath214 , ml decoding will correspond to comparing the functions @xmath215 , @xmath213 . note that comparing the functions @xmath215 is equivalent to comparing the channel causal conditioning distributions since @xmath216 as shown below . @xmath217 in the above , @xmath218 holds since @xmath105 is a known , deterministic function of @xmath219 and @xmath220 holds since the code - tree @xmath168 together with the feedback sequence @xmath105 uniquely determines the channel input @xmath148 . for notational convenience , the results below on the universal decoder are stated for blocklength @xmath19 , where @xmath130 denotes a code - tree of depth @xmath19 and @xmath221 denotes a set of such code - trees . these results extend to the set of concatenated code - trees @xmath210 and any exceptions are described in the text . furthermore , we introduce the following notation : @xmath222 denotes the ml decoder tuned to channel @xmath2 ; @xmath223 denotes the average ( over messages and codebooks chosen uniformly from a set ) error probability when decoder @xmath224 is used over channel @xmath2 ; and @xmath225 denotes the average ( over messages ) error probability when codebook @xmath164 and decoder @xmath224 is used over channel @xmath2 . a family of channels @xmath226 defined over common input and output alphabets @xmath227 is said to be _ strongly separable _ for the input code - tree sets @xmath228 , @xmath229 , if there exists some @xmath230 that upper bounds the error exponents in the family , i.e. , that satisfies @xmath231 such that for every @xmath74 and blocklength @xmath19 , there exists a subexponential number @xmath232 ( that may depend on @xmath233 and on @xmath234 ) of channels @xmath235 @xmath236 that well approximate any @xmath77 in the following sense : for any @xmath77 there exists @xmath237 , @xmath238 , so that @xmath239 and @xmath240 [ def : strongsep ] the notion of strong separability means that the family @xmath12 is well - approximated by a finite subset @xmath241 of the channels in the family . in order to prove that the family of finite - state channels with feedback is separable , we will need a value @xmath233 that satisfies ( [ eqn : defstrsepmu ] ) . the error probability @xmath242 is lower bounded by the probability that the output sequence @xmath243 corresponding to two different messages is the same for a given realization of the channel and code - tree . for a random code - tree this is lower bounded by a uniform memoryless distribution on the channel output . then @xmath244 and a suitable candidate for @xmath233 is @xmath245 . the following theorem shows the existence of a universal decoder for a strongly separable family and input code - tree sets @xmath221 . the proof follows from the proof of theorem 2 in @xcite except that we replace the channel conditional distribution @xmath246 with the causal conditioning distribution @xmath247 . [ thrm : univdecoderforseparable ] if a family of channels defined over common finite input and output alphabets @xmath248 is strongly separable for the input code - tree sets @xmath228 , then there exists a sequence of rate-@xmath73 blocklength-@xmath19 codes @xmath249 and a sequence of decoders @xmath250 such that @xmath251 the universal decoder @xmath252 in theorem [ thrm : univdecoderforseparable ] is given by `` merging '' the ml decoders tuned to channels @xmath253 , @xmath254 , that are used to approximate the family @xmath12 . in order to describe the merging of the ml decoders , we first present the ranking function @xmath255 . a ml decoder tuned to the channel @xmath2 can be described by a ranking function @xmath255 defined as the mapping latexmath:[\[m_{\theta } : b_{nm } \times { \cal y}^{nm } \rightarrow \{1,2,\hdots , @xmath168 that is most likely given output @xmath214 , rank 2 denotes the second most likely code - tree , and so on . for a given received sequence @xmath214 , every code - tree in the set @xmath210 is assigned a rank . for code - trees @xmath257 , @xmath258 by ( [ eqn : pyconda = pycauscondx ] ) , comparing the function @xmath215 is equivalent to comparing the channel causal conditioning distribution @xmath259 . letting @xmath222 denote the ml decoder tuned to @xmath2 , we can describe the decoder as @xmath260 where @xmath261 represents the code - tree chosen for message @xmath44 , @xmath262 . in the case that multiple code - trees maximize the likelihood @xmath215 for a given @xmath214 , the ranking function @xmath255 determines which code - tree ( and correspondingly message ) is chosen by the decoder . in the case that the same code - tree from @xmath210 is chosen for more than one message , the ranks will be identical and a decoding error will occur . note that for a given output sequence @xmath214 , the decoder @xmath263 will not always return the code - tree @xmath213 for which @xmath264 , since the code - tree @xmath168 may or may not be in the codebook . now consider a set of @xmath265 channels from the family @xmath12 , given by @xmath266 . the codebooks for these @xmath265 channels will be drawn randomly from the set @xmath210 . ( note that the same set @xmath210 is used for all channels @xmath253 since , as shown in lemma [ lemma : typeoncodetree ] , the type @xmath182 is chosen independent of the channel @xmath177 . ) the @xmath265 ml decoders matched to these channels , denoted @xmath267 , can be merged as shown in @xcite . the merged decoder @xmath268 is described by its ranking function @xmath269 which is a mapping latexmath:[\[m_{u_k } : b_{nm } \times { \cal y}^{nm } \rightarrow \{1,2,\hdots , for each output sequence @xmath214 . the ranking @xmath269 is established for a given @xmath214 by assigning rank 1 to the code - tree for which @xmath271 , rank 2 to the code - tree for which @xmath272 , rank 3 to the code - tree for which @xmath273 , and so on . after considering the code - trees with rank 1 for all @xmath274 , the code - trees with rank 2 in @xmath274 , @xmath275 are considered in order and added into the ranking @xmath269 . the process continues until the code - trees with rank @xmath276 for all @xmath274 have been assigned a rank in @xmath269 . throughout this process , if a code - tree has already been ranked , it is simply skipped over , and its original ( higher ) ranking is maintained . the rank of a code - tree in @xmath269 can be upper bounded according to its rank in @xmath274 as shown in @xcite and stated as follows . @xmath277 this bound on the rank in @xmath269 implies another ( looser ) upper bound . @xmath278 equation ( [ eqn : upperboundrankmerged ] ) can be used to upper bound the error probability when sequences output from the channel @xmath279 are decoded by the merged decoder @xmath268 . this is a key element of the proof of theorem [ thrm : univdecoderforseparable ] . finally , we state the lemma below , which shows that the family of finite - state channels defined by the causal conditioning distribution is strongly separable . together with theorem [ thrm : univdecoderforseparable ] , this establishes existence of a universal decoder for the problem we consider , and completes our proof of achievability . the family of all causal - conditioning finite - state channels defined over common finite input , output , and state alphabets @xmath280 is strongly separable in the sense of definition [ def : strongsep ] for any input code - tree sets @xmath228 . [ lemma : compoundfscisstronglysep ] see appendix [ app : compoundfscisstronglysep ] . the gilbert - elliot channel is a widely used example of a finite state channel . it has a state space consisting of ` good ' and ` bad ' states , @xmath281 and in either of these two states , the channel is a binary symmetric channel ( bsc ) . the gilbert - elliot channel is a stationary and ergodic markovian channel , i.e. , @xmath282 is satisfied and the markov process described by @xmath283 is a stationary and ergodic process . for a given channel @xmath2 , the bsc crossover probability is given by @xmath284 for @xmath285 and @xmath286 for @xmath287 . the channel state @xmath288 forms a stationary markov process with transition probabilities @xmath289 for a given @xmath2 , the gilbert - elliot channel is equivalent to the following additive noise channel @xmath290 where @xmath291 denotes modulo-2 addition and @xmath292 . conditioned on the state process @xmath293 , the noise @xmath294 forms a bernoulli process given by @xmath295 for a given channel @xmath2 , the capacity of the gilbert - elliot channel is found in @xcite and is achieved by a uniform bernoulli input distribution . the following example illustrates that the feedback capacity of a channel with memory is in general _ not _ given by @xmath296 as in the memoryless case . @xcite [ example : compoundge ] consider the example of a gilbert - elliot channel where @xmath297 for @xmath298 with feedback . the compound feedback capacity of this channel is zero because assuming that we start in the bad state , for any blocklength @xmath19 , the channel that corresponds to @xmath299 , will remain in the bad state for the duration of the transmission with probability @xmath300 . while the channel is in the bad state the probability of error for decoding the message is positive with or without feedback , hence no reliable communication is possible . however if we fix @xmath2 , then the capacity @xmath301 is at least @xmath302 , because we can use a deep enough interleaver to make the channel look like memoryless bsc with crossover probability @xmath303 . a gilbert - elliot channel is described by the four parameters @xmath304 and @xmath284 that lie between 0 and 1 and for any fixed @xmath19 , @xmath305 is continuous in those parameters . the continuity of @xmath305 follows from the fact that @xmath306 is continuous in the four parameters for any @xmath307 , and also because ( as shown in appendix [ app : compoundfscisstronglysep ] in eqns . ( [ eqn : equivfederlapidoth63 ] ) and ( [ eqn : equivfederlapidoth64 ] ) ) we can express @xmath305 as @xmath308 let us denote by @xmath309 the closure of the family of channels . hence instead of @xmath310 we can write @xmath311 since @xmath309 is compact and since @xmath312 is continuous in @xmath177 . now , let @xmath313 denote the uniform distribution over @xmath314 . we have @xmath315 where @xmath218 follows from the fact that @xmath316 and @xmath220 follows from the fact that for any channel a uniform distribution maximizes its capacity . therefore we can restrict the maximization to the uniform distribution @xmath317 instead of @xmath318 . hence feedback does not increase the capacity of the compound gilbert - elliot channel . this result holds for any family of fscs for which the uniform input distribution achieves the capacity of each channel in the family and is closely related to alajaji s result @xcite that feedback does not increase the capacity of discrete additive noise channels . in this section we show that the capacity of a compound channel that consists of stationary and uniformly ergodic markovian channels is positive if and only if it is positive for the case that feedback is allowed . the intuition of this result comes mainly from lemma [ lemma : directed0iff ] that states that @xmath319 the reason our proof is restricted to the family of channels that are stationary and uniformly ergodic markovian is because for this family of channels we can show that the capacity is zero only if for every finite @xmath19 , @xmath320 a stationary and ergodic markovian channel is a fsc where the state of the channel is a stationary and ergodic markov process that is not influenced by the channel input and output . in other words , the conditional probability of the channel output and state given the input and previous state is given by @xmath321 where the markov process , described by the transition probability @xmath283 , is stationary and ergodic . we say that the family of channels is _ uniformly ergodic _ if all channels in the family are ergodic and for all @xmath322 there exists an @xmath323 such that for all @xmath324 @xmath325 where @xmath326 is the stationary ( equilibrium ) distribution of the state for channel @xmath2 . we define the sequence @xmath327 as @xmath328 [ thrm : zerofbciffzeronfbc ] the channel capacity of a family of stationary and uniformly ergodic markovian channels is positive if and only if the feedback capacity of the same family is positive . since a memoryless channel is a fsc with only one state , the theorem implies that the feedback capacity of a memoryless compound channel is positive if and only if it is positive without feedback . the theorem also implies that for a stationary and ergodic point - to - point channel ( not compound ) , feedback does not increase the capacity for cases that the capacity without feedback is zero . the stationarity of the channels in theorem [ thrm : zerofbciffzeronfbc ] is not necessary since according to our achievability definition , if a rate is less than the capacity , it is achievable regardless of the initial state . we assume stationarity here in order to simplify the proofs . the uniform ergodicity is essential to the proof that is provided here but there are also other family of channels that have this property . for instance , for the regular point - to - point gaussian channel this result can be concluded from factor two result that claims that feedback at most doubles capacity ( c.f . , @xcite ) . the proof of theorem [ thrm : zerofbciffzeronfbc ] is based on the following lemmas . we refer the reader to appendix [ app : lemmasforiff ] for the proofs of these lemmas . [ lemma : qkqm_inequality ] for any channel with feedback , if the input to the channel is distributed according to @xmath329 then @xmath330 [ lemma : capacity_stationary_ergodic_markovian ] the feedback capacity of a family of stationary and uniformly ergodic markovian channels is @xmath331 the limit of @xmath327 exists and is equal to @xmath332 . [ lemma : directed0iff ] let the input distribution to an arbitrary channel be uniform over the input @xmath314 , i.e. , @xmath333 . if under this input distribution @xmath334 , then the channel has the property that @xmath335 for all @xmath336 and this implies that @xmath337 _ proof of theorem [ thrm : zerofbciffzeronfbc ] _ : let @xmath338 denote the capacity without feedback and @xmath339 denote the capacity with feedback . @xmath340 is trivial . to show that @xmath341 , we use lemma [ lemma : capacity_stationary_ergodic_markovian ] to conclude that since @xmath342 then @xmath343 and therefore for any @xmath344 , @xmath345 in order to conclude the proof , we show that if ( [ eqn : compound_no_feedback2 ] ) holds , then it also holds when we replace @xmath346 by @xmath347 . since @xmath348 is continuous in @xmath349 and since the set @xmath12 is a subset of the unit simplex which is bounded , then the infimum over the set @xmath12 can be replaced by the minimum over the closure of the set @xmath12 . since ( [ eqn : compound_no_feedback2 ] ) holds also for the case that @xmath346 is restricted to be the uniform distribution , then lemma [ lemma : directed0iff ] implies that the channel that satisfies @xmath335 for all @xmath350 is in the closure of @xmath12 and therefore @xmath351 recall that the capacity of the memoryless compound channel ( without feedback ) is @xcite @xmath352 wolfowitz also showed @xcite that when @xmath2 is known to the encoder , the capacity of the memoryless compound channel is given by switching the @xmath353 and the @xmath354 , i.e. , @xmath355 in this section we make use of theorem [ thrm : fbcapcompoundfsc ] to show that ( [ eqn : capacityencoderknowstheta ] ) is equal to the feedback capacity of the memoryless compound channel . based on wolfowitz s result it is straightforward to show that if the family of memoryless channels is finite , @xmath356 , then the feedback capacity of the compound channel is given by switching the @xmath354 and the @xmath357 , @xmath358 this result can be achieved in two steps . given a probability of error @xmath359 , first , the encoder will use @xmath360 uses of the channels in order to estimate the channel with probability of error less than @xmath361 . since the number of channels is finite such an @xmath360 exists . in the second step the encoder will use a coding scheme with blocklength @xmath166 adapted for the estimated channel to obtain an error probability that is smaller than @xmath361 . hence we get that the total error of the code of length @xmath362 is smaller than @xmath363 . for the case that the number of channels is infinite , the argument above does not hold , since there is no guarantee that for any @xmath359 there exists a blocklength @xmath364 such that a @xmath365 code achieves an error less than @xmath363 for all channels in the family . however , we are able to establish the feedback capacity using our capacity theorem for the compound fsc , and the result is stated in the following theorem . [ thrm : capacitymemorylesscompound ] the feedback capacity of the memoryless compound channel is @xmath366 theorem [ thrm : capacitymemorylesscompound ] is a direct result of theorem [ thrm : fbcapcompoundfsc ] and the following lemma . [ l_memoryless_eq ] for a family @xmath12 of memoryless channels we have @xmath367 the proof of lemma [ l_memoryless_eq ] requires two lemmas , which we state below . the proofs of lemmas [ l_donotloosemuch ] and [ l_estimate_channel ] are found in appendix [ app : lemmasformemorylesscompound ] . [ l_donotloosemuch ] let @xmath368 and @xmath369 . for two conditional distributions @xmath370 and @xmath371 with @xmath372 where @xmath373 as @xmath374 . [ l_estimate_channel ] for any @xmath375 , any @xmath322 and any channel @xmath376 , there exists an @xmath360 such that we can choose a channel @xmath377 as a function of @xmath360 inputs and outputs such that @xmath378 where @xmath379 denotes the @xmath380 distance between the estimated channel @xmath381 and the actual channel @xmath376 , i.e. , @xmath382 _ proof of lemma [ l_memoryless_eq ] : _ we prove the equality by showing the following two inequalities hold : @xmath383 where @xmath384 as @xmath385 . inequality ( [ e_memoryless1 ] ) is proved by the fact that @xmath386 is less than or equal to @xmath387 and by the fact that for a memoryless channel an i.i.d input maximizes the directed information . @xmath388 in order to prove inequality ( [ e_memoryless2 ] ) we consider the following input distribution . the first @xmath360 inputs are used to estimate the channel and we denote the estimated channel as @xmath389 . after the first @xmath360 inputs , the input distribution is the i.i.d distribution that maximizes the mutual information between the input and the output for the channel @xmath390 . according to lemma [ l_estimate_channel ] , we can estimate the channel to within an @xmath380 distance smaller than @xmath322 with probability greater than @xmath391 , where @xmath375 . according to lemma [ l_donotloosemuch ] , by adjusting the input distribution to a channel that is at @xmath380 distance less than @xmath234 from the actual channel in use , we lose an amount that goes to zero as @xmath392 . under the input distribution described above we have the following sequence of inequalities . @xmath393 * and ( f ) follow from a change of notation . * follows the fact that we sum fewer elements . the parameter @xmath360 is a function of @xmath322 and @xmath375 and is determined according to lemma [ l_estimate_channel ] . for brevity of notation we denote @xmath394 simply as @xmath360 . * follows from the fact that @xmath395 . * follows from the fact that the estimated channel is a random variable denoted as @xmath396 and it is a deterministic function of @xmath397 as described in lemma [ l_estimate_channel ] . * follows by restricting the input distribution @xmath347 to one that uses first @xmath360 uses of the channel to estimate as described in lemma [ l_estimate_channel ] , and then uses an i.i.d distribution , i.e. , for @xmath398 , @xmath399 . * follows from the fact that with probability @xmath391 we have that the @xmath380 distance @xmath400 and by applying lemma [ l_donotloosemuch ] , which states that for this case we lose @xmath401 where @xmath402 as @xmath392 . * follows from the fact that @xmath403 is identical to @xmath404 . finally , since @xmath360 is fixed for any @xmath322 , @xmath375 then we can achieve any value below @xmath405 for large @xmath19 . therefore inequality ( [ e_memoryless2 ] ) holds . the compound channel is a simple model for communication under channel uncertainty . the original work on the memoryless compound channel without feedback characterizes the capacity @xcite , which is less than the capacity of each channel in the family , but the reliability function remains unknown . an adaptive approach to using feedback on an unknown memoryless channel is proposed in @xcite , where coding schemes that universally achieve the reliability function ( the burnashev error exponent ) for certain families of channels ( e.g. , for a family of binary symmetric channels ) are provided . by using the variable - length coding approach in @xcite , the capacity of the channel in use can be achieved . in our work , we consider the use of fixed length block codes and aim to ensure reliability for every channel in the family ; as a result , our capacity is limited by the infimum of the capacities of the channels in the family . for the compound channel with memory that we consider , we have characterized an achievable random coding exponent , but the reliability function remains unknown . the encoding and decoding schemes used in proving our results have a number of practical limitations , including the memory requirements for storing codebooks consisting of concatenated code - trees at both the transmitter and receiver as well as the complexity involved in merging the maximum - likelihood decoders tuned to a number of channels that is polynomial in the blocklength . as such , our work motivates a search for more practical schemes for feedback communication over the compound channel with memory . the proposition is nearly identical to ( * ? ? ? * proposition 1 ) except that we replace @xmath406 by @xmath407 and @xmath408 by @xmath61 using results from @xcite on directed mutual information and causal conditioning . we first prove the following lemma , which is needed in the proof of proposition [ prop : existencec ] . the lemma shows that directed information is uniformly continuous in @xmath347 . for our time - invariant deterministic feedback model , @xmath409 , and the lemma holds for any such feedback . [ lemma : continuitydirectedinfo ] _ ( uniform continuity of directed information ) _ if @xmath410 and @xmath411 are two causal conditioning distributions such that @xmath412 then for a fixed @xmath413 @xmath414 directed information can be expressed as a difference between two terms @xmath415 . let us consider the total variation of @xmath416 , @xmath417 by invoking the continuity lemma of entropy ( * ? ? ? * theorem 2.7 , p33 ) we get , @xmath418 and @xmath419 are the entropies induced by @xmath420 and @xmath421 , respectively . now let us consider the difference @xmath422 . @xmath423 by combining inequalities ( [ eq_hdiff1 ] ) and ( [ eq_hdiff2 ] ) we conclude the proof of the lemma . by lemma [ lemma : continuitydirectedinfo ] , @xmath424 is uniformly continuous in @xmath425 . since @xmath425 is a member of a compact set , the maximum over @xmath425 is attained and @xmath82 is well - defined . next , we invoke a result similar to ( * ? ? ? * lemma 5 ) . given integers @xmath426 and @xmath38 such that @xmath427 , input sequences @xmath428 and @xmath429 with corresponding output sequences @xmath430 and @xmath431 , let @xmath425 be defined as @xmath432 then @xmath433 this result follows from ( * ? ? ? * lemma 5 ) and ( * ? ? ? * lemma 5 ) . finally , if we let @xmath434 and @xmath435 achieve the maximizations in @xmath436 and @xmath437 , respectively , then we have @xmath438 or equivalently , @xmath439 clearly @xmath440 , and by the convergence of a super - additive sequence , @xmath441 . the theorem is proved through a collection of results in @xcite and @xcite . let @xmath442 denote the error probability of the ml decoder when a random code - tree of blocklength @xmath19 is used at the encoder . @xmath443 the following corollary to ( * theorem 8) bounds the expected value @xmath444 $ ] , where the expectation is with respect to the randomness in the code . the result holds for any initial state @xmath16 . [ cor : boundeproberror ] suppose that an arbitrary message @xmath44 , @xmath445 , enters the encoder with feedback and that ml decoding tuned to @xmath2 is employed . then the average probability of decoding error over the ensemble of codes is bounded , for any choice of @xmath446 , @xmath447 , by @xmath448 \leq ( e^{nr}-1)^{\rho } \sum_{y^n } \left [ \sum_{x^n } q(x^n||z^{n-1 } ) p(y^n||x^n,\theta)^{\frac{1}{1+\rho } } \right]^{1+\rho}.\ ] ] identical to ( * ? ? ? * proof of theorem 8) except that @xmath349 is replaced by @xmath247 . next , we let @xmath153 denote the average ( over messages ) error probability incurred when a code - tree of blocklength @xmath19 is used over channel @xmath2 with initial state @xmath16 . using corollary [ cor : boundeproberror ] , we can bound @xmath153 as in the following corollary to ( * ? ? ? * theorem 9 ) [ cor : boundproberrorfe ] for a compound fsc with @xmath449 states where the codewords are drawn independently according to a given distribution @xmath450 and ml decoding tuned to @xmath2 is employed , the average probability of error @xmath153 for any initial state @xmath58 , channel @xmath77 , and @xmath446 , @xmath451 is bounded as @xmath452 where @xmath453^{1+\rho}\end{gathered}\ ] ] identical to ( * ? ? ? * proof of theorem 9 ) except for : ( i ) we replace @xmath305 by @xmath454 , ( ii ) we consider the error averaged over all messages ( rather than the error for an arbitrary message @xmath44 ) , and ( iii ) we assume a fixed input distribution @xmath425 rather than minimizing the error probability over all @xmath425 . the two results stated above provide us with a bound on the error probability , however , the bound depends on the channel @xmath2 in use . instead , we would like to bound the error probability uniformly over the class @xmath12 . to do so we cite the following two lemmas from previous work . [ lemma : fsuperadditive ] given @xmath455 and @xmath185 , let @xmath456 and define @xmath457 then @xmath458 as defined in corollary [ cor : boundproberrorfe ] satisfies @xmath459 identical to ( * ? ? ? * proof of lemma 11 ) except that we replace @xmath305 by @xmath454 . [ lemma : bounde ] @xmath460 the lemma follows from ( * ? ? ? * lemma 2 ) , which holds for a channel @xmath177 and input distribution @xmath461 satisfying @xmath462 and @xmath463 . we now follow the technique in @xcite by using lemmas [ lemma : fsuperadditive ] and [ lemma : bounde ] to bound the error probability independent of both @xmath16 and @xmath2 . for a given rate @xmath98 , let @xmath464 and pick @xmath38 in such a way that @xmath155 . then @xmath465 let @xmath466 be the input distribution that achieves the supremum in @xmath163 , i.e. , @xmath467 next , we use @xmath162 to define a distribution @xmath468 for a sequence of length @xmath99 , @xmath154 , as follows . @xmath469 for this new input distribution and sequence of length @xmath99 , we can bound the error exponent @xmath470 as shown below . @xmath471 where @xmath218 is due to lemma [ lemma : fsuperadditive ] , @xmath220 follows from lemma [ lemma : bounde ] , and @xmath472 follows from ( [ eqn : qm * ] ) . as in @xcite , we can maximize the lower bound on the error exponent by setting @xmath473 . with this choice of @xmath446 we have @xmath474 theorem [ thrm : achievabilitythetaknown ] follows by combining ( [ eqn : lowerboundfnm ] ) with the result in corollary [ cor : boundproberrorfe ] ( for blocklength @xmath99 ) . to prove the lemma , we must first establish two equalities relating the channel causal conditioning distribution @xmath454 to the channel probability law @xmath475 . the following set of equalities hold . @xmath476 where @xmath218 is due to ( * ? ? ? * lemma 2 ) and @xmath220 follows from our assumption that the input distribution @xmath69 does not depend on the state sequence @xmath477 . by the chain rule for causal conditioning ( * lemma 1 ) , ( [ eqn : chainrulecc ] ) implies that @xmath478 also , @xmath479 where @xmath472 follows from the definition of the compound finite - state channel . having established equations ( [ eqn : equivfederlapidoth63 ] ) and ( [ eqn : equivfederlapidoth64 ] ) , lemma [ lemma : compoundfscisstronglysep ] follows immediately from ( * ? ? ? * lemma 12 ) , where the conditional probability @xmath475 is quantized and the quantization cells are represented by channels @xmath480 . the proof of our result differs only in that the upper bound on the error exponents in the family is given by @xmath481 . the proof of lemma [ lemma : qkqm_inequality ] is based on an identity that is given by kim in ( * ? ? ? * eq . ( 9 ) ) : @xmath482 _ proof of lemma [ lemma : qkqm_inequality ] : _ using kim s identity we have @xmath483 now we bound the sum in the last equality , @xmath484 where @xmath218 follows from the assumption that @xmath485 . _ proof of lemma [ lemma : capacity_stationary_ergodic_markovian ] _ : the proof consists of two parts . in the first part we show that @xmath486 is sup - additive and therefore @xmath487 . in the second part we prove the capacity of the family of stationary and uniformly ergodic markovian channels by showing that @xmath488 where @xmath82 is defined in ( [ eqn : c_ndefinition ] ) . _ first part : _ we show that the sequence @xmath327 is sup - additive and therefore the limit exists . let integers @xmath426 and @xmath38 be such that @xmath427 and denote input distributions @xmath489 , and @xmath435 in shortened forms as @xmath490 , and @xmath180 . we have , @xmath491\nonumber \\ & \stackrel{}{\geq}&\max_{q_kq_m } \left[\inf_{\theta } i(x^k\to y^k |\theta)+\inf_{\theta } i(x_{k+1}^n\to y_{k+1}^n |\theta)\right]\nonumber \\ & \stackrel{=}{}&\max_{q_k } \inf_{\theta } i(x^k\to y^k |\theta)+\max_{q_m}\inf_{\theta } i(x_{k+1}^n\to y_{k+1}^n |\theta)\nonumber \\ & \stackrel{(c)}=&\max_{q_k } \inf_{\theta } i(x^k\to y^k |\theta)+\max_{q(x^m||z^{m-1})}\inf_{\theta } i(x^m\to y^m |\theta)\nonumber \\ & \stackrel{=}{}&kc_k^{markovian}+mc_m^{markovian},\end{aligned}\ ] ] where @xmath218 follows by restricting the maximization to causal conditioning probabilities of the product form @xmath492 , @xmath220 follows from lemma [ lemma : qkqm_inequality ] , and @xmath472 follows from stationarity of the channel . _ second part : _ we show that @xmath493 . due to lemma 5 in @xcite , @xmath494 , therefore it is enough to prove that latexmath:[\[\label{eqn : diff_s0_s0 } \lim_{n \rightarrow \infty}\frac{1}{n } \left [ \max_{q_{x^n||z^{n-1 } } } \inf _ { \theta } i(x^n\to y^n|s_0,\theta ) - \max_{q_{x^n||z^{n-1 } } } \inf _ { \theta , s_0 } i(x^n\to y^n , always positive , hence it is enough to upper bound it by an expression that goes to zero as @xmath385 . again by lemma 5 in @xcite we can bound the second term in ( [ eqn : diff_s0_s0 ] ) , @xmath496 where ( a ) holds for every @xmath497 and is due to lemma [ lemma : qkqm_inequality ] and ( b ) holds by the stationarity of the channel . hence , ( [ e_bound_directed ] ) implies that we can bound the difference , @xmath498 inequality ( a ) is due to the fact that @xmath499 and due to ( [ e_bound_directed ] ) . inequality ( b ) holds since for a uniformly ergodic family of channels , @xmath500 for all @xmath501 implies that for any input distribution @xmath502 and any channel @xmath2 , @xmath503 ) by @xmath19 , and since @xmath234 can be arbitrarily small and @xmath426 is fixed for a given @xmath234 , then ( [ eqn : diff_s0_s0 ] ) holds . _ proof of lemma [ lemma : directed0iff ] _ : from the assumption of the lemma we have @xmath504 by assuming a uniform input distribution , @xmath505 and by using the fact that if the kullback leibler divergence @xmath506 is zero , then @xmath507 for all @xmath508 , we get that ( [ eqn : directed_as_divergence ] ) implies that @xmath335 for all @xmath509 . it follows that @xmath510 \\ & = & \max_{q_{x^n||y^{n-1 } } } e[0]=0.\end{aligned}\ ] ] _ proof of lemma [ l_donotloosemuch ] : _ the proof is based on the fact that @xmath511 is uniformly continuous in @xmath376 , namely for any @xmath8 , @xmath512 where @xmath513 as @xmath374 ( the uniform continuity of mutual information is a straightforward result of the uniform continuity of entropy ( * ? ? ? * theorem 2.7 ) ) . we have , @xmath514 where the last inequality is due to ( [ e_uniform_cont_mutual ] ) . we conclude the proof by bounding the last term in ( [ e_bounding_diff ] ) by @xmath515 , which implies that if we let @xmath516 then ( [ e_l_diff ] ) holds . @xmath517 similarly , we have @xmath518 , and therefore @xmath519 _ proof of lemma [ l_estimate_channel ] : _ the channel @xmath520 is chosen by finding the conditional empirical distribution induced by an input sequence consisting of @xmath521 copies of each symbol of the alphabet @xmath129 . we estimate the conditional distribution @xmath522 separately for each @xmath523 . we insert @xmath524 for @xmath525 uses of the channel and we estimate the channel distribution when the input is @xmath524 as the type of the output which is denoted as @xmath526 . from sanov s theorem ( cf . * theorem 12.4.1 ) ) we have that the probability that type @xmath526 will be at @xmath380-distance larger than @xmath527 from @xmath522 is upper bounded by @xmath528 where @xmath529 denotes the divergence between the two distributions . using pinsker s inequality ( * lemma 12.6.1 ) we have that @xmath530 and therefore , @xmath531 the term @xmath532 goes to zero as @xmath38 goes to infinity for @xmath533 and therefore , for any @xmath534 we can find an @xmath38 such that @xmath535 . finally we have , @xmath536 where the inequality on the right is due to the union bound . the authors would like to thank their advisors - anthony ephremides and tsachy weissman - as well as prakash narayan for useful discussions on this topic and andrea goldsmith for organizing the roundtable research discussion at isit06 which led to the conception of this work . h. h. permuter , t. weissman , and a. j. goldsmith , `` finite state channels with time - invariant deterministic feedback , '' sep 2006 , submitted to ieee trans . inform . theory . availble at http://arxiv.org/abs/cs/0608070v1 .

the kpno international spectroscopic survey ( kiss ) is an objective - prism survey designed to detect extragalactic emission - line objects . it combines many of the features of previous slitless spectroscopic surveys with the advantages of modern ccd detectors , and is the first purely digital objective - prism survey for emission - line galaxies ( elgs ) . here we present the third list of elg candidates selected from our red spectral data , which cover the wavelength range 6400 to 7200 . in most cases , the detected emission line is h . the current survey list covers the region of the noao deep wide - field survey ( ndwfs ) . this survey covers two fields ; the first is degrees square and located at ra = 14 30 , = 34 30 ( b1950 ) , the second is degrees and centered at ra = 2 7.5 , = -4 44 . a total area of 19.65 deg is covered by the kiss data . a total of 261 candidate emission - line objects have been selected for inclusion in the survey list ( 13.3 per deg ) . we tabulate accurate coordinates and photometry for each source , as well as estimates of the redshift , emission - line flux and line equivalent width based on measurements of the digital objective - prism spectra . the properties of the kiss elgs are examined using the available observational data . when combined with the wealth of multi - wavelength data already available for the ndwfs fields , the current list of kiss elgs should provide a valuable tool for studying star - formation and nuclear activity in galaxies in the local universe .

astro-ph0508334

surveys for galaxies containing active galactic nuclei ( agns ) or strong star - formation activity have been an important area of extragalactic astronomy for decades . many fruitful surveys have been carried out with wide - field schmidt telescopes used in conjunction with objective prisms . an overview of previous surveys is given in salzer _ etal . _ ( 2000 ) , along with a sampling of the types of applications that such surveys have for the study of the extragalactic universe . we have been carrying out a modern objective - prism survey for the past several years . called the kpno international spectroscopic survey ( kiss ) , it combines many of the advantages of older surveys with the use of state - of - the - art ccd detectors , providing superior depth and data quality . the digital nature of kiss has many advantages over the older photographic surveys of this type ( e.g. , markarian 1967 ; smith , aguirre & zemelman 1976 ; macalpine , smith & lewis 1977 ; pesch & sanduleak 1983 ; wasilewski 1983 ; markarian , lipovetskii , & stepanian 1983 ; zamorano _ etal . _ 1994 ; popescu _ etal . _ 1996 ; surace & comte 1998 ; ugryumov _ etal . _ 1999 ) . besides the obvious factors of higher sensitivity and speed , we stress the importance of being able to measure the completeness limits and selection function of the survey directly from the data used to derive the catalogs of kiss emission - line galaxies ( elgs ) . this is not possible with photographic survey material , and makes kiss particularly useful for statistical studies of galaxian activity in the nearby universe . the current survey lists cover the area of the sky included in the noao deep wide - field survey ( ndwfs ; jannuzi & dey 1999 , jannuzi _ etal . _ in preparation , dey _ etal . _ in preparation ) . ndwfs is a deep optical and nir imaging survey carried out in two well separated fields . all optical data were taken on the noao 4-m telescopes in the bri bandpasses , while jhk imaging was carried out on the kpno 2.1-m telescope . the fields were both covered to a uniform depth of b @xmath9 26.6 ( and correspondingly deep in the other five bands ) . we chose to observe these fields as part of kiss because of the expectation that they would become well observed at many wavelengths as various groups studied the properties of the ndwfs galaxies . while the primary science goals of the ndwfs focus on galaxies at redshifts well beyond the filter - imposed redshift limit of kiss ( z @xmath10 0.095 ) , the volume covered by our survey is sufficiently large to provide a good - sized sample of star - forming galaxies and agns . bolstered by the large amount of data becoming available for the galaxies in the ndwfs area at radio , fir , nir , uv , x - ray , and optical wavelengths , the kiss elgs should allow for a number of detailed statistical studies of activity in galaxies in the local universe . while the kiss data are completely independent of the ndwfs data , they can be used to complement and extend the usefulness of the latter . this is the fifth paper in the kiss series . the first presents a complete description of the survey method , including a discussion of the survey data and its associated uncertainties ( salzer _ etal . _ 2000 ; hereafter paper i ) . the first and second survey lists of h@xmath0-selected elgs , informally referred to as the red survey , are given in salzer _ etal . _ ( 2001 ; hereafter kr1 ) and gronwall _ etal . _ ( 2004 ; hereafter kr2 ) , while the first list of [ ] -selected galaxies ( the blue survey ) is found in salzer _ etal . _ ( 2002 ; hereafter kb1 ) . the current paper follows a format similar to kr1 and kr2 ; for the sake of brevity , the reader is referred to kr1 , kr2 , and paper i for many details . the observational data and image processing are described in section 2 , while the new list of elg candidates is presented in section 3 . the properties of the new list of h@xmath0-selected elgs are described in section 4 , while our results are summarized in section 5 . all survey data were acquired using the 0.61-meter burrell schmidt telescope . the detector used for all data reported here was a 2048 @xmath11 4096 pixel site ccd . the ccd is identical to the one used for the kr2 list , however this is not the same ccd that was used for kr1 or kb1 , giving a different image scale and field - of - view . the ccd has 15-@xmath12 pixels , yielding an image scale of 1.43 arcsec / pixel at the newtonian focus of the telescope . the overall field - of - view was 50 @xmath11 100 arcmin , and each image covered 1.37 square degrees . the long dimension of the ccd was oriented north - south during our survey observations . the red survey spectral data were obtained with a 4@xmath5 prism , which provided a reciprocal dispersion of 17 / pixel at h@xmath0 . the spectral data were obtained through a special filter designed for the survey , which covered the spectral range 6400 7200 ( see figure 1 of paper i for the filter transmission curve ) . the two ndwfs fields each cover an area of 9 sq . the spring ( botes ) field is centered at ra = 14@xmath2 30@xmath3 , dec = 34@xmath5 30@xmath6 ( b1950 ) . it consists of a 3 @xmath11 3 degree square field . the fall ( cetus ) field is a 2.3 @xmath11 4.0 degree area centered at ra = 2@xmath2 7.5@xmath3 , dec = -4@xmath5 44@xmath6 ( b1950 ) . the layout of the botes field allowed us to cover the ndwfs area with two rows of kiss fields , with four fields per row . there is essentially zero overlap in declination between the two rows of fields . in addition , due to larger than normal pointing offsets between the direct and spectral fields ( see below ) , there are modest gaps between some of the fields within a given row . the net result is that the kiss data for the botes field only cover 8.08 sq . deg . , rather than the full 9 sq . deg . for the cetus fields , we again utilized two rows of kiss fields . however , in this case there is substantial overlap between the upper and lower rows , due to the fact that the declination extent of this ndwfs region is smaller . furthermore , we needed 6 kiss fields per row to cover the full 4.0 degrees of ra . despite the declination overlap , the cetus kiss fields cover a total area of 12.57 sq . , substantially larger than the area of the ndwfs fields . the total area covered by the kiss observations is 19.65 sq . deg . as with our previous survey strips , we obtained images of each survey field both with and without the objective prism on the telescope . the images taken without the prism ( referred to as direct images ) were obtained through standard b and v filters . the direct images were photometrically calibrated , and provide accurate astrometry and photometry for all sources in the survey fields . we used uniform exposure times for all survey fields : 4 @xmath11 720 s for the objective - prism ( spectral ) data , and 2 @xmath11 300 s for v and 1 @xmath11 600 s in b for the direct images . the telescope was dithered by a small amount ( @xmath1310 arcsec ) between exposures . table [ table : tab1 ] lists the observing runs during which the current set of survey fields were observed . the first column gives the ut dates of the run , while the second column indicates the number of nights on which observations were obtained . at least some data were obtained on 12 of 14 scheduled nights ( 86% ) . the last two columns indicate the number of direct and spectral images , respectively , obtained during each run . it was common practice to observe in both direct and spectral modes during parts of each run , although it was not always the case that the direct and spectral images of a given field were obtained during the same run . all data reduction took place using the image reduction and analysis facility ( iraf ) software . a special package of iraf - based routines that were written by members of the kiss team was used for most of the data analysis . full details of the observing procedures and data reduction methods are given in paper i and kr1 . the selection of the third red ( h@xmath0 ) list of elg candidates was carried out in precisely the same fashion as the first and second red lists ( kr1 and kr2 ) . full details are presented in paper i and kr1 . to briefly summarize , we use our automated kiss software to evaluate the extracted objective - prism spectrum of each object located within a survey field . all objects with spectral features that rise more than five times the local noise above the continuum level are flagged as potential elgs . this 5@xmath14 threshold is the primary selection criterion of the survey , and was arrived at after substantial testing during the early phases of the kiss project . following the initial automated selection , all candidates are visually examined , and spurious sources are removed from the sample . finally , the objective - prism images are scanned visually for sources that might have been missed by the software . these tend to be objects where the emission line is redshifted to the red end of the objective - prism spectrum , so that the software can not detect continuum on both sides of the line . the combination of our automated selection process and our careful visual checking helps to ensure a high degree of reliability that the kiss elg candidates are real , and that the sample is largely complete for all objects with 5@xmath14 emission lines . as described in kr1 , we also flag objects that have emission lines between 4@xmath14 and 5@xmath14 during our selection process . these 4@xmath14 detections represent objects with somewhat weaker emission lines than the main kiss sample , but that are nonetheless valid elg candidates . however , these objects do not constitute a statistically complete sample in the same sense as the main ( @xmath15 5@xmath14 ) list . we report the 45@xmath14 sources in a secondary list of elg candidates ( see appendix ) , which should be thought of as a supplement to the main kiss catalog . this list of extra " ( or kissrx ) objects likely includes a number of interesting sources . the list of elg candidates selected in the third red survey is presented in table [ table : tab2 ] . because the survey data includes both spectral images and photometrically - calibrated direct images , we are able to include a great deal of useful information about each source , such as accurate photometry and astrometry and estimates of the redshift , emission - line flux and equivalent width . only the first page of the table is printed here ; the complete table is available in the electronic version of this paper . the contents of the survey table are as follows . column 1 gives a running number for each object in the survey with the designation kissr @xmath16 , where kissr stands for `` kiss red '' survey . this is to distinguish it from the blue kiss survey ( kb1 ) . the kr1 and kr2 survey lists included kissr objects 12157 , and here we present kissr objects 21582418 . columns 2 and 3 give the object identification from the kiss database tables , where the first number indicates the survey field ( f@xmath16 ) , and the second number is the identification number within the field table for that galaxy . this identifier is necessary for locating the kiss elgs within the survey database tables . columns 4 and 5 list the right ascension and declination of each object ( j2000 ) . the formal uncertainties in the coordinates are 0.25 arcsec in ra and 0.20 arcsec in declination . column 6 gives the b magnitude , while column 7 lists the b@xmath17v color . for brighter objects , the magnitude estimates typically have uncertainties of 0.05 mag , increasing to @xmath130.10 mag at b = 20 . paper i includes a complete discussion of the precision of both the astrometry and photometry of the kiss objects . an estimate of the redshift of each galaxy , based on its objective - prism spectrum , in given in column 8 . this estimate assumes that the emission line seen in the objective - prism spectrum is h@xmath0 . follow - up spectra for @xmath151600 elg candidates from the two red survey lists ( kr1 , kr2 and the current list ) show that this assumption is correct in the vast majority of cases . only four elgs in the current list that possess follow - up spectra ( 3% ) are high redshift objects where a different line ( typically [ ] and/or h@xmath18 ) appears in the objective - prism spectrum . the formal uncertainty in the redshift estimates is @xmath19 = 0.0028 ( see section 4.1.3 ) . columns 9 and 10 list the emission - line flux ( in units of 10@xmath20 erg / s / cm@xmath8 ) and equivalent width ( in ) measured from the objective - prism spectra . the calibration of the fluxes is discussed in section 4.1.2 . these quantities should be taken as being representative estimates only . a simple estimate of the reliability of each source , the quality flag ( qflag ) , is given in column 11 . this quantity , assigned during the line measurement step of the data processing , is given the value of 1 for high quality sources , 2 for lower quality but still reliable objects , and 3 for somewhat less reliable sources . column 12 gives alternate identifications for kiss elgs which have been cataloged previously . this is not an exhaustive cross - referencing , but focuses on previous objective - prism surveys which overlap part or all of the current survey area : markarian ( 1967 ) and case ( pesch & sanduleak 1983 ) . the markarian survey overlaps both the botes and cetus fields , while the case survey only overlaps the botes field . also included are objects in common with the _ uppsala general catalogue of galaxies _ ( ugc , nilson 1973 ) . a total of 261 elg candidates are included in this third list of h@xmath0-selected kiss galaxies . the total area covered by the third red survey strip is 19.65 deg@xmath8 , meaning that there are 13.3 kiss elgs per square degree . for the first , second , and third red lists combined , the surface density is 16.4 galaxies deg@xmath22 , and if the lower significance kissrx objects are included the density is 20.7 elgs deg@xmath22 . this compares to the surface density of 0.1 galaxies deg@xmath22 from the markarian ( 1967 ) survey , and 0.56 galaxies per deg@xmath8 from the h@xmath0-selected ucm survey ( zamorano _ etal . _ 1994 ) ; the present survey is much deeper despite the redshift limit inherent in our detection method . it is interesting to note that the fraction of 4@xmath14 5@xmath14 kissrx elg candidates is substantially higher for the sample presented here than for the first and second red survey areas . as discussed below , we believe that this difference is caused by the somewhat different noise characteristics of the ccd used for the current survey . for example , elgs that would have been 5.0@xmath14 objects when observed with the previous ccd might be detected as 4.8@xmath14 sources in the current data . the net effect would be to lower the number of objects in the main survey list and to shift some of them into the kissrx list . if both the lower - significance kissrx objects and the _ bona fide _ kissr objects are combined , the surface density of elg candidates is essentially constant for all three red survey lists . -0.5 in = 6.5 in 0.5 in -0.5 in = 6.5 in 0.5 in -0.1 in = 5.0 in 1.0 in of the 261 objects cataloged , 167 were assigned quality values of qflag = 1 ( 64% ) , 74 have qflag = 2 ( 28% ) , and 20 have qflag = 3 ( 8% ) . based on our follow - up spectra to date , 99% ( 87 of 88 ) of the sources with qflag = 1 are _ bona fide _ emission - line galaxies , compared to 83% ( 24 of 29 ) with qflag = 2 and 78% ( 7 of 9 ) with qflag = 3 . overall , 94% of the objects with follow - up spectra are _ bona fide _ elgs . the properties of the kiss galaxy sample are described in the next section . figure [ fig : find1 ] shows an example of the finder charts for the kiss elgs . these are generated from the direct images obtained as part of the survey , and represent a composite of the b- and v - band images . figure [ fig : spec1 ] displays the extracted spectra derived from the objective - prism images for the first 24 elgs in table 2 . finder charts and spectral plots for all 261 objects in the current survey list , along with finder charts for the kissrx objects , are available in the electronic version of this paper . a supplementary table containing an additional 158 elg candidates with 4@xmath14 to 5@xmath14 emission lines is included in the appendix of this paper ( table [ table : tab3 ] ) . these additional galaxies do not constitute a statistically complete sample , and should therefore be used with caution . however , there are likely many interesting objects contained in this supplementary list . hence , following the precedent established in kr1 and kr2 , we list these objects in order to give a full accounting of the elgs in the area surveyed . due to the manner in which the survey is carried out , a great deal of observational data are available for all of the kiss elg candidates cataloged in the current paper . this includes accurate astrometry and b and v photometry for each source , as well as estimates of the redshift , h@xmath0 + [ ] line fluxes and equivalent widths . the combination of these data allow us to acquire a fairly complete picture of the make - up of the kiss sample . however , the quantities derived from the objective - prism spectra are inadequate for detailed analyses . first , the low resolution of the spectra limits the accuracy of the redshifts measured ( see below ) . further , the combination of low resolution and limited spectral coverage prevent us from using the survey data to ascertain the activity type of the elgs ( e.g. , agn vs. star - forming ) . hence follow - up spectra obtained with a higher dispersion spectrograph are required for a complete understanding of the kiss elgs . nonetheless , much can be gleamed about the survey constituents with the data currently available . we present an overview of the properties of our new sample of kiss elgs below . the b - band apparent magnitude distribution for the 261 kiss elgs in the current survey list is shown in figure [ fig : appmag]a . the median apparent b magnitude is 17.91 . this value is somewhat brighter than those of the kr1 and kr2 survey lists , which have median apparent magnitudes of b = 18.08 and 18.13 , respectively . however , it is clear that kiss still probes substantially deeper than previous objective - prism surveys : the median apparent magnitude for the h@xmath0-selected ucm survey ( prez - gonzlez _ etal . _ 2000 ) is b @xmath9 16.1 , and the [ ] -selected michigan ( um ) survey ( salzer _ etal . _ 1989 ) has a median apparent magnitude of b = 16.9 . indicated in the figure is the completeness limit of the markarian survey , b = 15.2 ( mazzarella & balzano 1989 ) . -0.1 in = 5.0 in 1.0 in the distribution of the b@xmath17v colors of the third red survey list is shown in figure [ fig : appmag]b . the median color is 0.67 , which is identical to that of the first red survey list and very close to that of the kr2 survey list ( b@xmath17v = 0.69 ) . this color is representative of an sb galaxy ( roberts & haynes 1994 ) . the ucm survey shows a similar color distribution , and has a median color of b@xmath17r = 0.71 ( prez - gonzlez _ etal . _ 2000 ) . in contrast , the [ ] -selected kb1 and um surveys have color distributions that are significantly shifted to the blue , with median b@xmath17v colors of 0.50 and 0.55 , respectively ( kb1 ; salzer _ etal . _ 1989 ) . this is a selection effect caused by the use of different emission lines for detection in the different surveys . the h@xmath0-selected samples include a broader spectrum of elgs , including many more luminous star - forming galaxies and liners which tend to be dominated by older , redder stellar populations . in addition , they are able to detect galaxies with higher levels of intrinsic reddening . the [ ] -selected samples are dominated by lower luminosity , lower metallicity galaxies which are dominated by younger stellar populations and have lower levels of internal absorption and reddening . while the h@xmath0-selected surveys tend to include both types of elgs , they are dominated by the more luminous galaxies . in contrast , the blue - selected surveys tend to not select the redder galaxies at all . the distribution of equivalent widths ( ews ) for the third red survey list is shown in figure [ fig : ew ] . we assume that the line we measure in the survey spectra is h@xmath0 blended with the [ ] @xmath236584,6548 lines . based on follow - up observations obtained to date , we know this assumption to be true for the vast majority of red survey objects . the three lines are blended at the resolution of the objective - prism spectra . the [ ] @xmath236731,6717 doublet is well resolved from the blended h@xmath0 + [ ] lines , and is often seen in survey spectra from strong - lined objects . the ew distribution peaks in the 40 50 bin , which indicates that kiss is fairly complete for objects with equivalent widths greater than @xmath24 . the median equivalent width of h@xmath0 + [ ] is 50.5 , with the majority of elgs having equivalent widths of less than 100 . this median ew is approximately 25% higher for this sample than for the two previous red survey lists . the noise level in the third red survey list data is slightly higher than for the first and second red survey list . we attribute this to the use of a different ccd for the newer sample , which had somewhat worse noise characteristics than the previous ccd . this shift in noise characteristics results in a selection of 5@xmath14 elg candidates that have relatively stronger lines . as we mention above the fraction of 4@xmath14 5@xmath14 objects is higher for the survey list presented here than for the two previous red survey lists , which is what we expect due to a higher noise level . -0.1 in = 5.0 in 1.0 in the calibration of the flux scale is a two - step process . the objective - prism spectra for each field are first corrected for throughput variations and atmospheric extinction . this places all line fluxes on the same _ relative _ flux scale . the fluxes are then calibrated on an absolute scale , using information obtained from the follow - up spectra . from a sample of 126 follow - up observations , we selected 49 objects that were classified as star - forming galaxies , had spectral quality @xmath25 or 2 , and had been observed with a long - slit spectrograph under photometric conditions . all spectra were obtained during the same observing run , and are of galaxies located in the botes field . since the fluxes measured from the objective - prism spectra are a combination of the h@xmath0 and [ ] lines , we use the fluxes from our slit spectra for the sum of these three lines . figure [ fig : lfluxcal ] shows a plot of the ratio of objective - prism flux ( in counts ) to spectroscopic flux _ versus _ the equivalent width measured from the follow - up spectra . the majority of the emission - line flux from a point source was included in the follow - up spectra , since the slit width used was 2.0 . some galaxies , however , have a larger angular extent and the emission lines originate in an extended region . since our long - slit measurements do not include all the h@xmath0 emission from these sources , they tend to have large flux ratios . we restricted the calibration sample to those galaxies with an objective - prism - to - spectroscopic flux ratio of less than 20 and an equivalent width greater than 40 , which left 25 galaxies . the emission regions of these galaxies are essentially point sources . the median flux ratio of the calibration sample is 12.61 ; the mean is 12.92 with a standard deviation of 4.19 and an error in the mean of 0.84 . we adopted the reciprocal of the median value as our calibration value , or @xmath26 ergs / s / cm@xmath8 per count . -0.1 in = 5.0 in 1.0 in the calibration value is applied to the measured objective - prism line fluxes to convert their instrumental fluxes ( in counts ) to calibrated fluxes ( in ergs / s / cm@xmath8 ) . in figure [ fig : lflux ] we show the distribution of observed h@xmath0 + [ ] line flux values for the 261 kiss elgs . the median value is @xmath27 ergs / s / cm@xmath8 , which is @xmath28 higher than the values found for the the first two red survey lists , suggesting that the data used for the current survey is less sensitive that those used for the first and second red survey lists . as we mention above , this difference is likely due to the slightly higher noise level in the kr3 data . however , the median line flux of the third red survey list is substantially fainter than that of the ucm sample , which is @xmath29 ergs / s / cm@xmath8 ( based on follow - up spectra of gallego _ etal . _ 1996 ) . -0.0 in = 5.0 in 1.0 in as mentioned earlier , one of the strengths of kiss is that the selection function and completeness limit can be derived using the survey data directly , rather than relying on secondary information ( e.g. , line strengths measured from follow - up spectra ) . the calibrated objective - prism line fluxes are used to determine the completeness limit of the survey , following the procedure described in gronwall _ etal . _ ( 2005 , in preparation ) . briefly , we convert the line fluxes into pseudo - magnitudes the line magnitude m@xmath30 , and then apply a v / v@xmath31 analysis ( e.g. , schmidt 1968 , huchra & sargent 1973 ) to the complete sample of 261 galaxies . the results are presented in table [ table : comptab ] . column ( 1 ) lists m@xmath32 , the value of m@xmath30 for which @xmath33 is being computed . column ( 2 ) lists the total number of elgs brighter than that m@xmath30 level , while columns ( 3 ) and ( 4 ) give the numbers of objects in the volume - limited and flux - limited subsamples , respectively ( see below ) . note that some objects may start out in the flux - limited sample at brighter values of m@xmath32 , then move into the volume - limited sample at fainter values of m@xmath32 . column ( 5 ) lists the mean v / v@xmath31 for the flux - limited subsample . column ( 6 ) shows the number of galaxies that need to be added to the sample at each m@xmath32 level to maintain @xmath34 , and column ( 7 ) lists the cumulative number of galaxies added at all magnitudes brighter than the given magnitude level to maintain @xmath33 . column ( 8) shows the percentage of objects that are in the flux - limited subsample , which decreases continuously as m@xmath30 becomes fainter . column ( 9 ) lists the completeness percentage of the flux - limited subsample as a function of m@xmath30 . these latter two quantities are plotted in figure [ fig : complete ] . -0.1 in = 5.0 in 1.0 in the interpretation of the results of the v / v@xmath31 test follows exactly the discussion found in gronwall _ etal . _ ( 2005 , in preparation ) for the kr1 sample . rather than repeating that discussion here , we will simply summarize the main results . it is important to realize that because of the redshift limit imposed by the survey filter , objects in the sample can be either line - flux - limited or volume - limited objects , depending on the strength of their h@xmath0 + [ ] emission and their redshift . objects with sufficiently strong lines will have values of v@xmath31 that exceed the effective volume of the survey set by the redshift limit . such objects are volume limited . as the limiting line flux ( parameterized by m@xmath30 ) decreases , a given object may actually switch from being in the flux - limited category to the volume - limited category . as seen in the table , for faint limiting line fluxes ( fainter m@xmath30 ) the majority of the kiss elgs are in the volume - limited subsample . this is illustrated by the dashed line in figure [ fig : complete ] . we see that the kiss sample is 100% complete to m@xmath30 = 15.3 , which is very similar to the results for the kr1 and kr2 samples ( gronwall _ etal . _ 2004 , 2005 ) . this completeness limit includes 184 kiss elgs , or 70 .. 5% of the full sample . as is often done , one can construct a correctably complete " sample by extending the line - flux limit down to even lower values . for example , at m@xmath30 = 15.9 , the sample is still 69.2% complete , but now includes 91.2% of the sample . we also derive redshifts from the objective - prism spectra . for objects with follow - up observations , we can compare the survey redshifts to the redshifts derived from the long - slit spectra ( figure [ fig : zcompare ] ) . in general the agreement between @xmath35 ( objective - prism redshift ) and @xmath36 ( follow - up redshift ) is excellent . only four objects deviate substantially from the equality line . two of these are active galaxies at @xmath37 which are not shown in the diagram . they were detected due to their [ ] @xmath385007 and h@xmath18 lines , respectively . the remaining two objects are kissr 2336 and kissr 2320 ; two relatively large , well - resolved disk galaxies that both have emission regions which are offset from the center of the galaxy . because the dispersion of the objective - prism spectra is in the north - south direction , a spectrum of an object that has an emission region spatially offset north or south of the center will yield an incorrect estimate of the redshift of the emission line . only a small minority of kiss objects are affected by this . for the first red survey list ( kr1 ) the survey redshifts above @xmath39 showed a systematic offset from the redshifts determined from follow - up spectroscopy . the reason for the offset is that as the h@xmath0 + [ ] line in the objective - prism spectrum begins to shift out of the survey bandpass , only the lower - redshift portion of the line is detected . a correction was applied to the kr1 survey redshifts , as described in paper i. the objects plotted in figure [ fig : zcompare ] do not display this offset , and no correction was applied to the survey redshifts . the second red survey list ( kr2 ) also did not show any systematic offset . the reason for the difference is probably the better pixel scale of the ccd used for the second and third red survey lists . when we calculate the rms scatter of @xmath35 about the equality line we use only objects with @xmath40 , as was done for kr1 . we exclude the four most deviant objects that we described in the previous paragraph . the resulting rms scatter is 0.0032 ( 950 km / s ) , which is marginally higher than the value found for the first two red survey lists ( 0.0028 , or 840 km / s ) . -0.1 in = 5.0 in 1.0 in the distribution of the objective - prism redshifts is shown in figure [ fig : zhist ] . the botes and cetus fields are shown in separate panels . the middle panel shows the redshift distribution for a comparison sample of galaxies from zwicky _ etal . _ ( 1961 1968 ; hereafter cgcg ) . the redshifts for the 97 cgcg galaxies are taken from falco _ etal . _ ( 1999 ) . since the surface density of the cgcg catalog is fairly low , we included objects located in a region of @xmath41 degrees , centered at the location of the botes field . the cgcg catalog does not extend far enough south to overlap the cetus field , hence the comparison sample applies only to the upper ( botes ) redshift sample . the botes void ( kirshner _ etal . _ 1981 ) is clearly visible between @xmath42 and @xmath43 in figure [ fig : zhist]a . even thought the ndwfs field is located far south of the nominal void center , the impact of the void is unmistakable in the redshift distribution . there is a significant density enhancement seen at redshifts between 0.0275 and 0.0375 , just in front of the void . this is most likely associated with the hercules supercluster . a modest peak in the redshift distribution of the cgcg sample is also seen in this redshift range . the latter becomes quite sparse beyond this redshift . the cgcg sample also shows a strong peak at z = 0.0125 which is less prominent in the kiss distribution . of the 41 cgcg galaxies that constitute this peak , only four are found inside the botes field survey area . the galaxies in this peak appear to fall along a large - scale structure feature that falls mostly outside the ndwfs area . beyond the botes void the kiss sample displays a fairly flat redshift distribution out to z @xmath13 0.085 , after which it begins to drop off . the distribution drops to zero at z = 0.095 , because kiss can not detect galaxies via the h@xmath0 line beyond this distance since it redshifts completely out of the survey filter at this point . the flat distribution and drop - off between 0.085 and 0.095 are characteristic of what is seen with the kr1 and kr2 samples as well . -0.1 in = 5.0 in 1.0 in the redshift distribution for the cetus field ( figure [ fig : zhist]c ) is dominated by a low density region at low redshifts that reaches out to z = 0.035 . this is part of the large void that dominates the foreground in the south galactic cap . there is no hint of the pisces - perseus supercluster at z @xmath9 0.020 ; the cetus field is far enough south to miss the supercluster . there is a strong density enhancement at z = 0.040 . beyond this point , the kiss redshift distribution is again fairly flat out the the point where the survey filter starts to exclude the h@xmath0 line . note that the location of the cetus field is such that only very sparse redshift information of the normal " galaxies exists . hence , we do not have a suitable comparison sample as we do with the botes field . the median redshifts of the two ndwfs kiss samples are quite similar to the values found for kr1 ( median z = 0.063 ) and kr2 ( median z = 0.061 ) . the availability of both an accurate apparent magnitude and a redshift estimate allows us to compute the absolute magnitude for each source . using the values listed in table 2 for the b magnitude and objective - prism redshift , we compute m@xmath44 for all elg candidates in the current list . h@xmath45 = 75 km / s / mpc is adopted , and a correction for galactic absorption is applied using the values for a@xmath44 compiled by schlegel _ etal . _ ( 1998 ) . in both fields the galactic absorption is small , with typical values of 0.04 0.06 mags in the botes field , and 0.09 0.11 mags in the cetus region . an explicit assumption made is that the line seen in the objective - prism spectrum is in fact h@xmath0 . previous observations of kr1 and kr2 elgs suggest that for roughly 2.5% of the kiss elg candidates the line seen in the objective - prism spectrum is some other line ( usually [ ] @xmath385007 ) . hence we might expect 6 - 7 of the elg candidates in table 2 to fall in this category . a histogram showing the distribution of m@xmath44 for the kiss elgs in the current survey list is shown in figure 10a . for comparison , in figure 10b we plot the absolute magnitude distribution for the 97 cgcg galaxies used in the redshift distribution comparison in the previous subsection . note that for this presentation we have combined the two portions of the kiss ndwfs sample into a single histogram . the median absolute magnitudes of the kiss and cgcg galaxies differ significantly : while the cgcg has a median m@xmath44 = @xmath1719.78 ( i.e. , close to m@xmath46 ) , the kiss elgs have a median absolute magnitude fully 2/3rds of a magnitude fainter . this is consistent with our previous survey lists . that is , kiss is especially sensitive to intermediate and low luminosity galaxies when compared to a magnitude - limited sample like the cgcg . despite the fact that the current sample of kiss elgs have systematically lower luminosities than the cgcg galaxies located in the same area of the sky , they are , on average , _ higher _ in luminosity than either of the previous two h@xmath0-selected kiss lists . the median m@xmath44 values for kr1 and kr2 are @xmath1718.96 and @xmath1718.64 , respectively . the apparent reason for the differences between the three lists is the relative paucity of lower redshift galaxies in the current survey . as mentioned above , the cetus and botes fields of the ndwfs exhibit very low numbers of galaxies at redshifts below 0.035 and 0.025 , respectively this is precisely where kiss is most sensitive to dwarf elgs . to be sure , there are still plenty of low luminosity galaxies present in the current survey list . however , the fraction of the survey with m@xmath44 @xmath15 @xmath1718 is less than in both kr1 and kr2 . at the high luminosity end , the kiss elgs appear to be deficient in galaxies with luminosities above m@xmath46 ( m@xmath44 @xmath47 @xmath1720 ) . while the kiss elgs cover the same absolute magnitude range as the magnitude - limited cgcg galaxies , the proportion of higher luminosity galaxies is much lower among the kiss galaxies . this most likely occurs for two reasons . first , the kiss sample is redshift limited at high luminosities , meaning that it does not probe arbitrarily large volumes of space for the highest luminosity objects as magnitude - limited samples do . second , the detection of any galaxy by kiss requires that the emission line observed be bright enough to stand out against the stellar continuum of the galaxy in the objective - prism spectrum . that is , there must be a minimum contrast between the continuum and the line . this is effectively an equivalent width limit . for luminous galaxies , a larger star - forming event or a stronger level of agn activity is needed for the emission lines to exceed this implicit equivalent width threshold . since these intense activity levels are fairly rare , there are fewer detected elgs among the higher luminosity host galaxies . the combination of these two effects means that there are fewer kiss elgs at luminosities above m@xmath46 . table [ table : tab2 ] lists cross - references for kiss elgs which are also cataloged in previous photographic surveys for active and star - forming galaxies , and we note the ugc numbers for the objects that are listed in the _ uppsala general catalogue of galaxies _ ( nilson , 1973 ) . the first red survey area overlapped with four major active galaxy surveys : markarian ( 1967 ) , case ( pesch & sanduleak 1983 ) , wasilewski ( 1983 ) , and ucm ( zamorano _ etal . _ the third red survey area , like kr2 , overlaps with only the markarian and case surveys . the markarian survey overlaps both the botes and cetus fields . there are , however , no markarian galaxies in either field . this is not too surprising , since the surface density of markarian galaxies is small ( 0.1 per sq . the case survey overlaps only with the botes field , and there are 18 case objects in this area . however , two of them lie just outside the area covered by the kiss objective - prism images . they are both included in the kiss direct images , but as we mention in 2 , the spectral and direct images do not always cover the exact same area . two additional case objects ( 459 and 460 ) are emission regions within the same galaxy , and we choose to count them as one object for the purpose of the comparison with kiss . of the resulting 15 case objects , 13 ( 87% ) are recovered by kiss in the main survey ( i.e. , table 2 ) . this fraction of recovered objects is somewhat higher than what was found for the first two red surveys ( 73% and 72% respectively ) . both of the two case galaxies that kiss does not recover are listed as color - selected in the case survey papers . neither one is listed in the secondary kiss survey list with 4@xmath14 to 5@xmath14 elg candidates . a large fraction of case galaxies have h@xmath0 lines with equivalent widths less than 30 ( salzer _ etal . _ 1995 ) , and kiss is not as sensitive to this type of object . the ugc catalog overlaps with the botes field , and there are eight ugc galaxies in this area . four of them are also kiss galaxies . weak emission lines appear to be present in the objective - prism spectra of the remaining four galaxies , but only at the @xmath133@xmath14 level . we present the third list of h@xmath0-selected emission - line galaxy candidates ( and fourth list overall ) from the kpno international spectroscopic survey ( kiss ) . all data presented here were obtained with the 0.61-m burrell schmidt telescope . kiss is an objective - prism survey , but differs from older such surveys by virtue of the fact that it utilizes a ccd as the detector . while we sacrifice areal coverage relative to classical photographic surveys , we benefit from the enormous gain in sensitivity that ccds provide over plates . we readily detect strong - lined elgs as faint as b = 21 . in addition , the pan - chromatic nature of ccds allows us greater wavelength agility compared to photographic surveys . even with the use of our survey filter , which restricts the detection of elgs to z @xmath47 0.095 , we are sensitive to a broader range of galaxian redshifts than the older photographic objective - prism surveys ( paper i ) . the combination of higher sensitivity , lower noise , and larger volumes surveyed yield huge improvements in the depth of the resulting survey . with the kissrx objects included , kiss finds @xmath15200 times more agn and starburst galaxy candidates per unit area than did the markarian ( 1967 ) survey , and @xmath1337 times more than the ucm survey ( zamorano _ etal . _ 1994 ) . the current installment of kiss includes 261 elg candidates selected from 20 red survey fields covering a total of 19.65 deg@xmath8 . this yields a surface density of 13.3 galaxies per deg@xmath8 . we are sensitive to the h@xmath0 emission line with redshifts up to @xmath130.10 . the survey fields presented here are located at ra = 14@xmath2 30@xmath3 , @xmath4 = 34@xmath5 30@xmath6 ( b1950 ) , and at ra = 2@xmath2 7.5@xmath3 , @xmath4 = -4@xmath5 44@xmath6 . these fields were chosen to coincide with the location of the noao deep wide - field survey ( jannuzi & dey 1999 ) . for each object in the catalog we tabulate accurate equatorial coordinates , b & v photometry , and estimates of the redshift and line strength measured from the objective - prism spectra . we also provide finder charts and extracted spectral plots for all objects . in addition to the main survey list , we include a supplementary list of 158 elg candidates with weaker ( lower significance ) emission lines . this newest list brings the total of h@xmath0-selected kissr elgs to 2418 objects present in three survey regions . in addition , we have cataloged another 638 extra " kissrx candidates that are detected in the survey with a lower significance level , the total number of cataloged elgs is 3056 , contained in a survey area of just 147.6 sq . the overall surface density of kiss elgs is thus 20.7 per sq . one of the advantages of our survey method is the large amount of basic data that we acquire for each object . this in turn allows us to parameterize the constituents of the survey and to develop a fairly complete picture of the overall sample without the need for extensive follow - up observations . we present an overview of the survey properties for the current list of elg candidates . the median apparent magnitude of the current sample is b = 17.91 . this is somewhat brighter than the values found for kr1 and kr2 ( b = 18.08 and 18.13 , respectively ) , but it is still substantially fainter than previous elg surveys . objects fainter than b = 20 are routinely cataloged . line strengths measured from the objective - prism spectra show that kiss is sensitive to objects with h@xmath0 + [ ] equivalent widths of less than 20 , and that most objects with ew @xmath15 40 are detected . the median emission - line flux of the kiss sample is nearly three times lower than that of the ucm survey ( gallego _ etal . _ 1996 ) . the luminosity distribution of the kiss elgs is heavily weighted toward intermediate- and low - luminosity galaxies , although we are still sensitive to luminous agn and starbursting galaxies . the median absolute magnitude of m@xmath44 = @xmath1719.11 underscores the fact that strong - lined galaxies of the type cataloged by kiss tend to be less luminous than the types of objects found in more traditional magnitude - limited samples . despite the fact that one can learn a great deal about each kiss elg from the survey data alone , it is still necessary to obtain higher dispersion follow - up spectra in order to arrive at a more complete understanding of each object . for example , due to the low - dispersion nature of the objective - prism spectra it is not possible to distinguish between agn and star - formation activity in the kiss galaxies . further , the redshifts derived from the kiss spectral data are too coarse to be used in detailed spatial distribution studies . we have obtained follow - up spectra for 100% of kiss elgs in the kr1 and kb1 survey lists , and we are in the process of obtaining spectra for the objects in the remaining survey lists ( kr2 and the current list . these follow - up spectra will allow us to better assess the nature of the individual galaxies , which in turn enables the sample to be used for a wide variety of science applications , many of which are outlined in paper i. examples include a series of multi - wavelength studies of the properties of kiss elgs in the radio ( van duyne _ etal . _ 2004 ) and x - rays ( stevenson _ etal . _ 2002 , datta _ etal . _ 2006 , in preparation ) , plus studies currently underway in the mid- and far - ir ( iras and spitzer ) and near - ir ( 2mass ) , as well as ongoing studies of the metal abundances in kiss star - forming galaxies ( e.g. , lee _ etal . _ 2004 , salzer _ etal . _ 2005 ) . we gratefully acknowledge financial support for the kiss project through nsf presidential faculty award to jjs ( nsf - ast-9553020 ) , which was instrumental in initiating the original international collaboration , as well as continued support via nsf - ast-0071114 and nsf - ast-0307766 . cg also acknowledges support from nsf - ast-0137927 . we thank the numerous colleagues with whom we have discussed the kiss project over the past several years , including jess gallego , rafael guzmn , rob kennicutt , david koo , trinh thuan , alexei kniazev , yuri izotov , janice lee , jason melbourne and jose herrero . finally , we wish to thank the astronomy department of case western reserve university for maintaining the burrell schmidt during the period of time when the survey observations reported here were obtained . as we explained in section 3 , the main survey objects are selected based on the presence of a 5@xmath14 emission feature in their spectra . because of the high sensitivity of the survey data , many objects were detected that have apparent emission lines with strengths that are only slightly weaker than the 5@xmath14 limit . we decided to exclude such objects from the main survey list since one of the primary goals of the kiss project is to construct a deep but statistically complete sample of elgs . early tests involving follow - up spectroscopy carried out on fields where objects were selected to lower thresholds showed that 5@xmath14 detections were nearly always real sources , while objects between 4@xmath14 and 5@xmath14 tended to be real but also included a fair number ( @xmath1325% ) of spurious sources . however , these objects are nonetheless valid elg candidates , and this list of objects likely includes a number of interesting objects . therefore , rather than ignore these weaker - lined elg candidates entirely , we are publishing them in a supplementary table . listed in table [ table : tab3 ] are 158 elg candidates that have emission lines detected at between the 4@xmath14 and 5@xmath14 level . the format of table [ table : tab3 ] is the same as for table [ table : tab2 ] , except that the objects are now labeled with kissrx numbers ( ` x ' for extra ) . the kissrx numbers start at 481 since we presented 480 kissrx objects in kr1 and kr2 . the full version of the table , as well as finder charts for all 158 kissrx galaxies , are available in the electronic version of the paper . the supplementary elg sample has characteristics similar to those of the main survey elgs , although with some notable differences . the median h@xmath0 equivalent width is 40.9 , roughly 20% lower than the value for the main sample . the kissrx galaxies are somewhat fainter ( median b magnitude of 18.65 ) and significantly redder ( median b@xmath17v = 0.81 ) . their median redshift is slightly higher than that of either the botes or cetus field ( 0.067 ) , and their median luminosity is nearly a magnitude fainter ( @xmath1718.23 ) . hence , the supplementary elg list appears to be dominated by intermediate luminosity galaxies with somewhat lower rates of star - formation activity ( lower equivalent widths , redder colors ) than the elgs in the main sample . the differences between the kissr and kissrx objects in the current paper are similar to those seen between the two samples in kr1 and kr2 . jannuzi , b. t. , & dey , a. , 1999 , in _ photometric redshifts and the detection of high redshift galaxies _ , asp conference series , vol . r. weymann , l. storrie - lombardi , m. sawicki , and r. brunner , p. 111 rrrllcccrrcl 2158 & h0202 & 4755 & 2 01 59.3 & -4 41 03.5 & 17.88 & 0.59 & 0.0286 & 53 & 26 & 2 & + 2159 & g0202 & 3879 & 2 02 18.2 & -3 42 09.9 & 18.97 & 1.37 & 0.0871 & 34 & 21 & 3 & + 2160 & g0202 & 3858 & 2 02 18.8 & -3 36 50.7 & 18.43 & 0.50 & 0.0415 & 72 & 64 & 1 & + 2161 & g0202 & 3653 & 2 02 26.0 & -4 11 05.9 & 15.25 & 0.47 & 0.0159 & 353 & 50 & 1 & + 2162 & g0202 & 3213 & 2 02 39.4 & -4 09 01.7 & 15.91 & 0.59 & 0.0465 & 316 & 28 & 2 & + 2163 & g0202 & 3037 & 2 02 44.0 & -3 52 35.2 & 18.49 & 0.38 & 0.0143 & 46 & 36 & 2 & + 2164 & g0202 & 3055 & 2 02 44.1 & -4 11 17.8 & 19.05 & 0.51 & 0.0430 & 39 & 60 & 2 & + 2165 & g0202 & 2729 & 2 02 53.9 & -4 07 19.7 & 16.45 & 0.84 & 0.0435 & 138 & 14 & 1 & + 2166 & g0202 & 2315 & 2 03 05.8 & -3 50 24.8 & 16.57 & 0.90 & 0.0427 & 38 & 4 & 3 & + 2167 & g0202 & 2317 & 2 03 06.4 & -4 15 05.9 & 19.48 & 0.92 & 0.0811 & 39 & 47 & 2 & + + 2168 & g0202 & 2329 & 2 03 06.5 & -4 27 14.1 & 17.11 & 0.41 & 0.0155 & 245 & 72 & 1 & + 2169 & g0202 & 2135 & 2 03 13.2 & -4 18 47.3 & 19.01 & 0.83 & 0.0909 & 50 & 94 & 3 & + 2170 & g0202 & 1836 & 2 03 21.4 & -4 03 31.0 & 18.05 & 0.87 & 0.0650 & 81 & 33 & 1 & + 2171 & h0202 & 2122 & 2 03 25.5 & -5 04 24.4 & 20.01 & 0.51 & 0.0547 & 69 & 3054 & 3 & + 2172 & g0202 & 1224 & 2 03 42.3 & -4 45 18.3 & 17.19 & 0.81 & 0.0370 & 54 & 9 & 3 & + 2173 & g0202 & 1148 & 2 03 42.9 & -3 49 15.4 & 16.78 & 0.71 & 0.0469 & 410 & 57 & 1 & + 2174 & g0202 & 786 & 2 03 54.0 & -3 53 00.1 & 15.76 & 0.63 & 0.0425 & 863 & 65 & 1 & + 2175 & g0202 & 646 & 2 03 58.9 & -4 01 15.2 & 17.96 & 0.96 & 0.0617 & 68 & 24 & 1 & + 2176 & g0202 & 550 & 2 04 00.9 & -3 16 08.8 & 21.00 & 1.68 & 0.0902 & 33 & 342 & 3 & + 2177 & h0202 & 807 & 2 04 08.2 & -4 50 47.1 & 17.78 & 0.53 & 0.0771 & 117 & 97 & 1 & + + 2178 & g0202 & 420 & 2 04 08.5 & -4 49 15.4 & 18.13 & 1.17 & 0.0829 & 45 & 25 & 2 & + 2179 & g0202 & 372 & 2 04 09.5 & -4 28 20.2 & 18.33 & 0.79 & 0.0793 & 89 & 54 & 1 & + 2180 & h0202 & 549 & 2 04 16.9 & -4 48 41.0 & 17.29 & 0.70 & 0.0869 & 77 & 24 & 1 & + 2181 & g0202 & 76 & 2 04 17.0 & -3 39 15.2 & 19.17 & 0.90 & 0.0777 & 66 & 101 & 2 & + 2182 & h0202 & 434 & 2 04 20.3 & -4 54 58.1 & 14.99 & 0.28 & 0.0178 & 151 & 32 & 1 & + 2183 & g0205 & 5741 & 2 04 21.7 & -3 42 40.0 & 17.71 & 0.83 & 0.0586 & 54 & 22 & 1 & + 2184 & g0205 & 5425 & 2 04 30.4 & -3 36 09.1 & 17.71 & 0.76 & 0.0558 & 72 & 23 & 1 & + 2185 & g0205 & 5416 & 2 04 31.4 & -4 03 34.6 & 19.56 & 0.67 & 0.0387 & 59 & 168 & 1 & + 2186 & g0205 & 5375 & 2 04 33.4 & -4 29 39.6 & 17.83 & 0.45 & 0.0483 & 97 & 65 & 1 & + 2187 & h0205 & 6171 & 2 04 35.9 & -5 30 00.0 & 18.12 & 0.78 & 0.0764 & 151 & 92 & 1 & + ccccccccc 13.0 & 4 & 4 & 0 & 0.5922 & 0 & 0 & 100.00 & 100.00 + 13.1 & 5 & 5 & 0 & 0.5900 & 0 & 0 & 100.00 & 100.00 + 13.2 & 5 & 5 & 0 & 0.5139 & 0 & 0 & 100.00 & 100.00 + 13.3 & 7 & 7 & 0 & 0.5889 & 0 & 0 & 100.00 & 100.00 + 13.4 & 8 & 8 & 0 & 0.5688 & 0 & 0 & 100.00 & 100.00 + 13.5 & 9 & 9 & 0 & 0.5468 & 0 & 0 & 100.00 & 100.00 + 13.6 & 12 & 12 & 0 & 0.5928 & 0 & 0 & 100.00 & 100.00 + 13.7 & 16 & 16 & 0 & 0.6271 & 0 & 0 & 100.00 & 100.00 + 13.8 & 20 & 20 & 0 & 0.6256 & 0 & 0 & 100.00 & 100.00 + 13.9 & 22 & 20 & 2 & 0.5666 & 0 & 0 & 90.91 & 100.00 + 14.0 & 28 & 24 & 4 & 0.6046 & 0 & 0 & 85.71 & 100.00 + 14.1 & 35 & 28 & 7 & 0.6352 & 0 & 0 & 80.00 & 100.00 + 14.2 & 38 & 31 & 7 & 0.5891 & 0 & 0 & 81.58 & 100.00 + 14.3 & 44 & 36 & 8 & 0.5833 & 0 & 0 & 81.82 & 100.00 + 14.4 & 54 & 42 & 12 & 0.5818 & 0 & 0 & 77.78 & 100.00 + 14.5 & 67 & 50 & 17 & 0.5946 & 0 & 0 & 74.63 & 100.00 + 14.6 & 81 & 56 & 25 & 0.6012 & 0 & 0 & 69.14 & 100.00 + 14.7 & 98 & 67 & 31 & 0.6173 & 0 & 0 & 68.37 & 100.00 + 14.8 & 108 & 69 & 39 & 0.5764 & 0 & 0 & 63.89 & 100.00 + 14.9 & 120 & 76 & 44 & 0.5646 & 0 & 0 & 63.33 & 100.00 + 15.0 & 139 & 87 & 52 & 0.5642 & 0 & 0 & 62.59 & 100.00 + 15.1 & 154 & 93 & 61 & 0.5480 & 0 & 0 & 60.39 & 100.00 + 15.2 & 166 & 99 & 67 & 0.5251 & 0 & 0 & 59.64 & 100.00 + 15.3 & 184 & 106 & 78 & 0.5122 & 0 & 0 & 57.61 & 100.00 + 15.4 & 195 & 107 & 88 & 0.4822 & 4 & 4 & 54.87 & 96.40 + 15.5 & 206 & 114 & 92 & 0.4612 & 6 & 10 & 55.34 & 91.94 + 15.6 & 216 & 117 & 99 & 0.4330 & 9 & 19 & 54.17 & 86.03 + 15.7 & 225 & 120 & 105 & 0.4104 & 10 & 29 & 53.33 & 80.54 + 15.8 & 231 & 119 & 112 & 0.3821 & 13 & 42 & 51.52 & 73.91 + 15.9 & 238 & 117 & 121 & 0.3761 & 10 & 52 & 49.16 & 69.23 + 16.0 & 244 & 110 & 134 & 0.3417 & 17 & 69 & 45.08 & 61.45 + 16.1 & 253 & 104 & 149 & 0.3333 & 13 & 82 & 41.11 & 55.91 + 16.2 & 257 & 94 & 163 & 0.3116 & 16 & 98 & 36.58 & 48.96 + 16.3 & 259 & 88 & 171 & 0.2757 & 21 & 119 & 33.98 & 42.51 + 16.4 & 260 & 83 & 177 & 0.2521 & 22 & 141 & 31.92 & 37.05 + 16.5 & 261 & 76 & 185 & 0.2258 & 24 & 165 & 29.12 & 31.54 + rrrllcccrrcl 481 & g0202 & 5949 & 2 01 17.5 & -3 54 59.0 & 18.12 & 0.57 & 0.0396 & 43 & 19 & 2 & + 482 & g0202 & 5213 & 2 01 41.7 & -4 16 42.6 & 21.07 & 1.88 & 0.0816 & 79 & 116 & 3 & + 483 & g0202 & 4890 & 2 01 49.8 & -3 46 50.7 & 18.98 & 0.77 & 0.0729 & 92 & 71 & 2 & + 484 & g0202 & 4563 & 2 01 57.6 & -3 17 19.8 & 19.27 & 0.76 & 0.0407 & 29 & 43 & 3 & + 485 & g0202 & 4294 & 2 02 07.1 & -4 05 04.2 & 19.13 & 1.05 & 0.0597 & 59 & 32 & 2 & + 486 & g0202 & 4158 & 2 02 11.2 & -4 07 33.9 & 17.76 & 1.03 & 0.0849 & 50 & 19 & 3 & + 487 & g0202 & 2393 & 2 03 04.4 & -4 28 15.8 & 17.89 & 1.03 & 0.0370 & 51 & 18 & 2 & + 488 & g0202 & 2057 & 2 03 13.9 & -3 19 42.6 & 20.58 & 0.57 & 0.0767 & 49 & 228 & 3 & + 489 & g0202 & 1608 & 2 03 27.2 & -3 17 14.9 & 17.74 & 0.15 & 0.0375 & 46 & 32 & 2 & + 490 & g0202 & 1499 & 2 03 31.6 & -3 46 56.4 & 19.07 & 0.70 & 0.0873 & 39 & 33 & 3 & + + 491 & h0202 & 1892 & 2 03 32.0 & -4 21 50.5 & 20.42 & 1.16 & 0.0515 & 64 & 105 & 2 & + 492 & h0202 & 1443 & 2 03 46.9 & -4 35 15.3 & 19.10 & 0.82 & 0.0546 & 43 & 6 & 3 & + 493 & h0202 & 749 & 2 04 09.5 & -4 28 19.8 & 18.34 & 0.75 & 0.0733 & 121 & 69 & 2 & + 494 & g0205 & 4694 & 2 04 53.7 & -4 00 57.6 & 19.44 & 0.24 & 0.0135 & 31 & 111 & 3 & + 495 & g0205 & 4052 & 2 05 12.1 & -3 28 55.0 & 19.35 & 0.74 & 0.0674 & 93 & 50 & 2 & + 496 & g0205 & 3412 & 2 05 33.2 & -3 57 25.3 & 18.78 & 0.65 & 0.0926 & 44 & 54 & 2 & + 497 & g0205 & 3002 & 2 05 45.8 & -3 15 37.4 & 23.42 & 3.92 & 0.0713 & 67 & 84 & 3 & + 498 & g0205 & 2790 & 2 05 52.7 & -3 23 46.0 & 18.30 & 0.82 & 0.0827 & 20 & 11 & 3 & + 499 & h0205 & 3297 & 2 06 00.2 & -4 40 31.8 & 23.69 & 3.69 & 0.0673 & 40 & 322 & 3 & + 500 & h0205 & 3071 & 2 06 08.5 & -5 20 22.8 & 19.37 & 0.73 & 0.0744 & 32 & 35 & 2 & + + 501 & g0205 & 2173 & 2 06 14.5 & -4 28 14.9 & 18.16 & 0.13 & 0.0341 & 53 & 45 & 2 & + 502 & g0205 & 1969 & 2 06 21.7 & -4 13 34.2 & 16.73 & 0.67 & 0.0346 & 75 & 11 & 2 & + 503 & g0205 & 1809 & 2 06 25.4 & -3 29 24.8 & 21.11 & 1.16 & 0.0684 & 28 & 57 & 2 & + 504 & g0205 & 1512 & 2 06 35.9 & -3 38 17.3 & 16.35 & 0.27 & 0.0284 & 115 & 12 & 3 & + 505 & g0205 & 1496 & 2 06 37.6 & -4 19 43.9 & 19.82 & 1.48 & 0.0388 & 30 & 19 & 3 & + 506 & h0205 & 1656 & 2 06 50.8 & -5 32 50.7 & 18.52 & 0.84 & 0.0245 & 32 & 19 & 3 & + 507 & g0205 & 943 & 2 06 54.0 & -3 35 21.2 & 19.68 & 1.20 & 0.0682 & 40 & 28 & 2 & + 508 & h0205 & 1333 & 2 07 01.1 & -5 47 59.0 & 18.66 & 0.66 & 0.0345 & 44 & 32 & 3 & + 509 & g0205 & 417 & 2 07 10.9 & -3 30 36.5 & 19.50 & 0.89 & 0.0763 & 84 & 362 & 2 & + 510 & g0205 & 52 & 2 07 21.7 & -3 15 56.5 & 19.89 & 0.81 & 0.0091 & 52 & 64 & 3 & +

we present results from quantitative modeling and spectral analysis of the high mass x - ray binary system vela x-1 obtained with the _ chandra _ high energy transmission grating spectrometer . the observations cover three orbital phase ranges within a single binary orbit . the spectra exhibit emission lines from h - like and he - like ions driven by photoionization , as well as fluorescent emission lines from several elements in lower charge states . the properties of these x - ray lines are measured with the highest accuracy to date . in order to interpret and make full use of the high - quality data , we have developed a simulator , which calculates the ionization and thermal structure of a stellar wind photoionized by an x - ray source , and performs monte carlo simulations of x - ray photons propagating through the wind . the emergent spectra are then computed as a function of the viewing angle accurately accounting for photon transport in three dimensions including dynamics . from comparisons of the observed spectra with results from the simulator , we are able to find the ionization structure and the geometrical distribution of material in the stellar wind of vela x-1 that can reproduce the observed spectral line intensities and continuum shapes at different orbital phases remarkably well . we find that the stellar wind profile can be represented by a cak - model with a star mass loss rate of ( 1.52.0 ) , assuming a terminal velocity of 1100 km s . it is found that a large fraction of x - ray emission lines from highly ionized ions are formed in the region between the neutron star and the companion star . we also find that the fluorescent x - ray lines must be produced in at least three distinct regions ( 1 ) the extended stellar wind , ( 2 ) reflection off the stellar photosphere , and ( 3 ) in a distribution of dense material partially covering and possibly trailing the neutron star , which may be associated with an accretion wake . finally , from detailed analysis of the emission line profiles , we demonstrate that the stellar wind dynamics is affected by x - ray photoionization .

astro-ph0607025

in a high mass x - ray binary ( hmxb ) , a compact object , either a neutron star or a black hole , sweeps up material as it orbits through the stellar wind of a massive o- or b - type companion star . when the material is accreted onto the compact object , a fraction of its gravitational potential energy is converted into x - rays , which in turn ionizes and heats the surrounding stellar wind . x - ray photons from the compact object are reprocessed by the ionized material , resulting in discrete spectral features , which carry a wealth of information about the geometry and physical state of the material in the system . the compact object can , therefore , be used as a radiation source to probe the structure of the stellar wind to derive the physical parameters that characterize its nature . vela x-1 is the archetype hmxb consisting of a neutron star and a massive companion star , and as such it has been the most well - studied object in this class since its discovery . it is an eclipsing hmxb system containing a pulsar with a pulse period of 283 s @xcite and an orbital period of 8.964 days @xcite . the optical companion star , hd 77581 , is a b0.5ib supergiant @xcite , which drives a stellar wind with a mass - loss rate in the range ( 1 7 ) @xmath2 10@xmath3 yr@xmath1 @xcite . the terminal velocity of the stellar wind was determined to be 1100 km s@xmath1 from the p - cygni profile of uv resonance lines @xcite . its average intrinsic x - ray luminosity is @xmath4 10@xmath5 erg s@xmath1 , which is consistent with accretion of the stellar wind by the gravitational field of a neutron star . the first attempt to model the global x - ray emission line spectrum of vela x-1 was presented by @xcite using the ccd spectrum obtained with . they characterized the wind using a standard velocity profile of ob stars proposed by @xcite ( cak model ) . by adjusting the mass loss rate , they found a statistically acceptable fit to the spectrum . however , since the energy resolution of is not sufficiently high to resolve any of the emission lines predicted by the ionization model , the intensity measurement of each line is limited due to blending . in addition , since the expected x - ray line shifts and widths are much smaller than the width of the instrument response , they were not able to extract any information about the dynamics of the x - ray emitting gas . the spectral resolution ( @xmath6 @xmath4 1001000 ) of the transmission grating spectrometers on board _ chandra _ have provided a detailed view of the emission line spectrum , including accurate measurements of line intensities and velocity profiles in a number of hmxbs ( vela x-1 : @xcite , @xcite ; gx 301@xmath72 : @xcite ; cyg x-3 : @xcite ; 4u 1700@xmath737 : @xcite ) . @xcite present the first high - resolution x - ray spectrum of vela x-1 observed during eclipse . they detect a narrow rrc from with an electron temperature of @xmath4 10 ev , which implies that the stellar wind is indeed driven by photoionization . @xcite present the spectra of vela x-1 observed during the three orbital phase ranges eclipse , @xmath8 = 0.25 , and @xmath8 = 0.50 . they report the first detection of multiple emission lines at orbital phase @xmath8 = 0.5 and analyzed the spectral changes between the three orbital phases . in this paper , we present a quantitative analysis of the x - ray spectra of vela x-1 first presented by @xcite using the data obtained with _ chandra _ high energy transmission grating spectrometer ( hetgs ) during three orbital phase ranges . we construct a simulator to compute the physical conditions of the surrounding gas for a given binary configuration , stellar wind parameters , and x - ray luminosity . continuum photons that originate from the compact object interact with the atoms to produce emission lines and continua , which are propagated through the wind accounting for all relevant radiative transfer effects including line scattering / absorption , continuum absorption , and compton scattering . the emerging spectra are calculated and compared with the observed orbital - phase - resolved spectra , which provide constraints on the global physical properties of the ionized stellar wind . this paper is organized as follows . in 2 , we describe the _ chandra _ observations and summarize the data reduction procedures . the reduced spectra are fitted using simple empirical models to extract the general properties of the stellar wind . in 3 , we provide a full description of the simulator , which is used to model the binary system and to calculate the resulting spectra as a function of the orbital phase . in 4 , we describe a self - consistent picture of the vela x-1 system derived from the simulator in comparison to the data . we finally summarize our results in 5 . vela x-1 was observed with the _ chandra _ hetgs / acis - s during three different orbital phase ranges within a single binary orbit in february 2001 ( 1 ) @xmath8 = 0.2370.278 , ( 2 ) @xmath8 = 0.4810.522 , and ( 3 ) @xmath8 = 0.9800.093 , hereafter referred to as phase 0.25 , phase 0.50 , and eclipse , respectively . the observation dates and exposure times are summarized in table [ tbl : velax1obslist ] . a schematic drawing of the binary orbit and the phase ranges are shown in fig . [ fig : velax1orbit ] . all of the data are processed using ciao v2.3 , and spectral analyses are performed using xspec . since the zeroth order image was severely piled - up during phase 0.25 and phase 0.50 , the locations of the zeroth order image were determined by finding the intersection of the dispersed events as well as the streak events , which are produced during ccd frame transfer . this gives us a zeroth order location that is accurate to within 0.1 pixels , which is much smaller than the instrument line response . the positive and negative spectral orders are inspected separately to check that systematic offsets do not exist . we then apply spatial filters for both the medium energy grating ( meg ) and the high energy grating ( heg ) data , and subsequently use an order sorting mask by using the energy information obtained with acis - s . only the first order events are used in our spectral analyses . the background events are estimated from the adjacent region to the dispersed event region and are found to contribute at most 5 % for the eclipse data and 3% for the phase 0.25 and phase 0.50 data . the background is subtracted in the spectral analyses . figure [ fig : velacont1 ] shows the heg spectra for the three orbital phases . x - rays emitted from the neutron star are directly observed during phase 0.25 and phase 0.50 as continuum spectra . as expected from the geometry of the binary system , the spectrum taken in eclipse is dominated by line emission and scattered components . in order to parameterize the properties of the continuum , we fit the spectra with a photo - absorbed power - law function . since the spectrum of phase 0.25 is affected less by absorption , we leave both the hydrogen column density and the photon index as free parameters in the fit . on the other hand , for phase 0.5 , we fix the photon index to the value obtained from phase 0.25 and calculate the absorption . we assume that the chemical abundance of metals in the absorbing material is 0.75 times the cosmic abundance @xcite , which is known to be a representative value for typical ob - stars @xcite . the derived parameters from the spectral fits are listed in table [ tbl : cont_fit ] . the best - fit models are superimposed on the spectra in fig . [ fig : velacont1 ] . the absorption - corrected luminosity is determined to be identical for observations in phase 0.25 and in phase 0.50 and corresponds to 1.6 @xmath2 10@xmath5 erg s@xmath1 in the 0.510 kev range , assuming a distance of 1.9 kpc @xcite . a large number of emission lines are clearly detected in the spectra of phase 0.50 and eclipse as shown in fig . [ fig:050_line_spec ] and fig . [ fig:000_line_spec ] . emission lines from h - like and he - like s , si , mg and ne can be seen , in addition to fluorescent lines from fe , ca , s , and si ions in lower charge states . we note that emission lines from the same ions are detected in the spectra of both phase 0.50 and eclipse . in fig . [ fig : si_line_spec ] , we show the three spectra in a narrow energy range of @xmath9 kev , which contains the k@xmath10 transitions of si . an intense ly@xmath10 line from h - like si and fully - resolved he - like triplet lines are clearly seen in the phase 0.50 and eclipse spectra . at slightly lower energy , a spectacular complex of emission lines are also detected , which we identify as k@xmath10 fluorescent lines from si in lower charge states . in the phase 0.5 spectrum , the near - neutral lines ( ) are blended at @xmath11 kev , and a series of lines above this energy are resolved into those from . this implies that the stellar wind in this system contains an extremely wide range in ionization parameter . the eclipse spectrum also shows a similar structure , but with much lower intensities . the centroid of energies , the widths and the intensities of each line are determined by representing them with single gaussian components . in the spectral fittings , we use the poisson likelihood statistics , instead of the @xmath12 statistics , because the numbers of photons in some of the bins in the spectra are small . the derived parameters for phase 0.50 and eclipse are listed in table [ tbl:050_line_fit ] and table [ tbl:000_line_fit ] , respectively . we also show list in table [ tab : comparison_lines ] the line intensity ratios between phase 0.50 and eclipse for the h- and he - like lines that are detected with a statistical significance of more than 5 @xmath13 . the h - like lines are brighter during phase 0.5 by a factor of 810 compared to during eclipse , and similarly the he - like lines are 47 times stronger . the resolving power of the hetgs allows us to measure doppler shifts to an accuracy of @xmath4 100 km s@xmath1 . figure [ fig : comparison_doppler ] compares the line profiles of si ly@xmath10 and mg ly@xmath10 between phase 0.50 and eclipse . the offset in the line center energies are clearly seen . we measure the velocity shifts of all of the k@xmath10 lines of h - like and he - like ions , which are plotted in fig . [ fig : dopplershift ] . though some fluctuations are seen , there is a general trend that the lines are blue - shifted during phase 0.50 , and they are red - shifted during eclipse as expected from an outflowing wind irradiated by an x - ray source . the velocity shifts have values in the range in @xmath14 300600 km s@xmath1 ( table [ tab : comparison_lines ] and widths of @xmath15 km s@xmath1 . radiative recombination continuum ( rrc ) is detected clearly from h - like ne . we fitted the rrc spectra using the `` redge '' model in xspec . the electron temperatures are determined to be @xmath16 ev and @xmath17 ev during phase 0.50 and eclipse , respectively . iron k@xmath10 fluorescent lines are detected in all three orbital phases . the profiles of these lines are shown in fig . [ fig : vela_fek ] . the parameters derived from the spectral fits with single gaussian models are listed in table [ tab : fekalpha ] . the equivalent width of the iron k@xmath10 line is measured to be 116 ev and 51 ev for phases 0.50 and 0.25 , respectively . during eclipse , a high equivalent width of 844 ev is observed . as shown in fig . [ fig : vela_fek ] , there is an indication of a compton shoulder in the iron k@xmath10 line at phase 0.50 . some general global properties of the circumsource medium can be deduced from the results of the simple fitting procedures described in [ sec : continuum ] and [ sec : line ] . the continuum from the neutron star is highly absorbed during phase 0.5 whereas the absorption column density is reduced during phase 0.25 . this implies that there is an absorber located behind the neutron star as viewed from the companion star . the average continuum intensities are nearly identical during the two phase ranges . the presence of strong soft x - ray lines during phase 0.5 , however , suggests that this absorber is localized and does not cover the x - ray emission line regions , which most likely extend beyond the size of the absorber . the emission lines during phase 0.50 is approximately an order of magnitude brighter than those during eclipse . this indicates that a significant fraction of the x - ray line emission is produced in a region between the neutron star and the companion star , which is occulted during eclipse . the velocity shifts and widths provide additional constraints on the properties of the emission region . the measured doppler shifts of @xmath18 km s@xmath1 are much smaller than the terminal velocity of vela x-1 ( @xmath19 km s@xmath1 ; @xcite ) . for a @xcite stellar wind velocity profile of isolated ob stars , this velocity is achieved at only 0.25 stellar radii from the photosphere . the doppler broadening of emission lines is also small ( @xmath20 km s@xmath1 ) . these observational results shows that the emission site of lines from highly ionized ions is distributed on a plane between the neutron star and the companion star , with most of the emission coming from near the line of centers . finally , the detection of a narrow radiative recombination continuum from h - like ne during phase 0.5 and in eclipse with electron temperatures of 68 ev provides direct evidence that the emission lines from highly ionized gas in vela x-1 are driven through photoionization . other possible sources of mechanical heating are , therefore , not important in characterizing the ionization structure of at least the highly - ionized regions of the stellar wind . the x - ray emission lines detected from vela x-1 can be interpreted as due to emission from a gas photoionized by the continuum radiation from the neutron star . when line photons are created through recombination cascades following photoionization , they interact with the surrounding gas as they propagate out of the binary system , and this is controlled by the ionization structure and the density distribution of the gas in the stellar wind . therefore , by investigating the properties of the emission lines , we can infer the physical conditions , geometrical distribution , and the velocity structure of the wind . in order to compute the ionization structure and to reproduce the wealth of information provided by the high - resolution hetgs spectra , it is necessary to model the emission using realistic physical models of the stellar wind in a binary system . we have developed three - dimensional monte carlo radiative transfer code to simulate the interactions of x - rays with the particles in the stellar wind . our simulation code consists of two main components : ( 1 ) : : calculation of the ionization structure . ( 2 ) : : monte carlo simulation for radiative transfer . in part ( 1 ) , we construct a map of the ion abundance and temperature in the stellar wind . in part ( 2 ) , we use the results from part ( 1 ) and propagate x - ray photons from the neutron star out of the binary system . the simulation is performed on a three - dimensional grid of cells , which is capable of treating complex geometries . radiative transfer effects that are important for highly ionized media are included . the medium is allowed to have arbitrary velocities , and lorentz transformations are performed between the cells , which is necessary to calculate the correct output line profiles . our detailed procedures are described in the following subsections . the ionization balance and electron temperature are determined primarily by the particle density and the local flux and spectral shape of the ionizing x - rays . we compute the three - dimensional ionization structure by first dividing the binary system into square cells of equal size . since the system is assumed to be cylindrically symmetric along the line of centers , we compute the ion fractions and temperatures on a two - dimensional grid as shown in fig . [ fig : geom_grid ] . each cell is @xmath21 cm along the side and the system consists of @xmath22 cells . the density structure around the companion star is specified from the velocity profile determined from the uv observation and an assumed mass - loss rate , which is treated as a free parameter . this assigns a particle density to each of the cells . the intrinsic luminosity and spectrum shape of the x - ray source as determined from the observed spectrum are then placed in the cell where the neutron star is located . the ion fractions and the electron temperatures of the cells immediately surrounding the neutron star are computed using this initial continuum shape as the input ionizing radiation field . the computation moves outwards to the next layer of cells , this time , using a different continuum shape , which accounts for absorption by the inner cells . this procedure continues until the outermost cells are reached . we have use the xstar program to calculate the ion fractions h , he , c , n , o , ne , mg , si , s , ar , ca , and fe , and the electron temperature in each cell . a flowchart of the procedures is shown in fig . [ fig : sim_scheme1 ] . this procedure ignores the diffuse radiation field as the source of additional ionization , which is a small contributor to the total photoionization rate at least for the ions observed in our spectra . a more accurate treatment of this effect might involve an iterative scheme , which is not necessary for our present analysis , but may be important for extremely optically thick media . the map of the ion fractions calculated by the method described in the previous subsection assures equilibrium between photoionization and recombination . we then use this ionization structure to propagate again the continuum photons originating from the neutron star to compute the emitted x - ray spectra as a function of the orbital phase . this is done by accounting for radiative transfer effects in detail and using monte carlo methods as described below . the cells are set up somewhat differently from those used in the calculation of the ionization structure . the cells sizes are chosen so that each of them are not optically thick to the strongest resonance lines . smaller cells are required near the companion star where both the density and density gradient are high , and we choose cubic cell sizes of @xmath21 cm on each side . the cells are increased in size outwards and the outermost cells of the simulation space are larger by a factor of 7.5 compared to the inner ones as shown in fig . [ fig : geom_grid ] . we start the monte carlo simulation with a photon at the position of the central neutron star . the photon energy is drawn from the observed power - law spectral energy distribution and is initially given a random direction . this photon is then traced through the cells and interacts with the ions in the stellar wind , which then generates a secondary photon through radiative recombination , radiative transitions , and fluorescent emission . all of the photons including those from the central continuum and subsequent line photons produced by interactions are propagated until they either completely escape the simulation space or are destroyed by one of the physical processes described below . the emergent photons are then selected depending on the viewing angle of the system , and are finally histogrammed to produce a spectrum . when a photon interacts with a particle , one of three physical processes is allowed to occur : ( 1 ) photoionization , ( 2 ) photoexcitation , and ( 3 ) compton scattering . for a given photon energy , the absorption and scattering cross sections and hence the optical depths are calculated for each ion within a given cell . interaction probabilities are then assigned to each possible process and are drawn via monte carlo . although process ( 1 ) produces a free electron and process ( 3 ) changes the energy of a free electron , we do not trace them in our simulations , but instead simply assume that they are thermalized before any further interaction takes place . as mentioned earlier , we assume that photoionization equilibrium is established locally everywhere in the plasma . therefore , photoionization of h- or he - like ions is always followed by radiative recombination and radiative cascades to the ground level . for ions of lower charge states , k - shell photoionization is followed by either fluorescent emission or auger decay . after photoexcitation , the ion in an excited state produces one or more photons in the downward radiative transitions that eventually lead to the ground level . by the combination of the photoexcitation and the downward radiative transitions , the resonance scattering of the lines is incorporated in the monte carlo simulation . for the h - like and he - like ions , we compute all of the atomic quantities with the flexible atomic code ( fac ) . this includes energy levels , oscillator strengths , radiative decay rates , photoionization cross sections from all levels between the ground state up to @xmath23 . radiative recombination rates onto each of levels as a function of the electron temperature are also computed from the photoionization cross sections using the milne relation . collisional transfers and uv photoexcitations in the he - like ions are not included in the current version of the code . photoionization of ions with three or more electrons de - excites by either an emission of fluorescent photon or ejection of one or more auger electrons . the k - shell photoionization cross sections are computed using the fitting formulae provided by @xcite , which roughly accounts for the change in the k - edge energy and cross section as functions of the charge state . the fluorescence yields and k@xmath10 energies , however , are assumed to be fixed for a given element and their values are adopted from @xcite and @xcite , respectively , for neutral atoms . we also do not account for k - shell ionization of li - like ions , which can lead to the production of the he - like forbidden line . we use l - shell cross sections from epdl97 , which is also valid for neutral atoms . the l@xmath24 , l@xmath25 and l@xmath26-shell cross sections are taken into account for the ions with three , five , and six electrons , respectively . finally , we do not include l - shell fluorescent emission , since the yields are typically much smaller compared to the k - shell fluorescence yields . we consider compton scattering only by free electrons , which is an excellent approximation for highly ionized media . the differential cross section is given by the the klein - nishina formula and the electrons velocities are drawn from a maxwellian distribution with the local electron temperature . doppler shifts due to the motion of the stellar wind are accounted for in all of the processes . when a photon enters a cell , the cross sections are calculated for the photon energy in the co - moving frame and lorentz transformed back into the rest frame . the doppler shifts could affect the line strength through the resonance line scattering . a velocity gradient can change the ratio of the line escape . in this monte carlo simulation , such doppler effects on the lines are included , though the reproducibility of the velocity gradient is limited by the geometry cell size . we first search for the value of the mass loss rate that reproduces the observed line intensities as a function of the orbital phase . we assume that the velocity structure of the stellar wind is represented by a generalized cak @xcite profile : @xmath27 where @xmath28 is the distance from the center of the star , @xmath29 is the stellar radius , and @xmath30 is the terminal velocity . we use a terminal velocity of 1100 km s@xmath1 as determined from uv observations of vela x-1 @xcite and fix the value of @xmath31 to 0.80 @xcite . for a given velocity profile and a mass loss rate @xmath32 , the wind density can be calculated by applying the equation of mass continuity assuming spherical symmetry , @xmath33 where @xmath34 is the gas mass per hydrogen atom , which is @xmath35 for cosmic chemical abundances . this specifies the density at each point in the stellar wind , and the observed x - ray luminosity can be used to obtain a map of the ionization parameter . the binary system parameters used in the simulations are listed in table [ tbl : simparams ] . the x - ray radiation field from the neutron star is modeled based on the parameter determined from the fits to the observed spectra at phase 0.25 and phase 0.5 . since no change in the average intrinsic luminosity is observed between these phases , we assume that the luminosity remains unchanged during the eclipse phase as well . the power - law spectrum with a photon index @xmath36 is assumed to extend up to 20 kev , which is the cut - off energy detected by previous hard x - ray observations with _ ginga _ @xcite and @xcite . the corresponding 0.520 kev luminosity is 3.5 @xmath2 10@xmath5 erg s@xmath1 . in the following subsection , we discuss the results of our simulation for mass loss rates of 5.0 @xmath2 10@xmath37 , 1.0 @xmath2 10@xmath38 , 1.5 @xmath2 10@xmath38 , and 2.0 @xmath2 10@xmath38 @xmath39 yr@xmath1 . figure [ fig : map_abund ] shows maps of the h - like ion fraction of si @xmath40 at three different values for the mass loss rate . at a mass loss rate of 5.0 @xmath2 10@xmath37 @xmath39 yr@xmath1 , the concentration of h - like si is highest near the mid - plane of the binary system as shown on the left panel of fig . [ fig : map_abund ] . as the mass loss rate is increased , the density everywhere in the wind is also increased . for a fixed ionization parameter , the distance to the neutron star has to be reduced , so the region moves closer towards the neutron star as shown in the middle and right panels of fig . [ fig : map_abund ] . multiplying the maps in fig . [ fig : map_abund ] by the local density of the stellar winds gives the maps in fig . [ fig : map_density ] . these show the absolute abundances of h - like si @xmath41 and the region where the x - ray photons of the emission line are produced . according to these simulations , a large fraction of emission lines from highly ionized ions such as h - like si are produced in the region between the companion star and the neutron star . this is the region that is obscured by the companion star during eclipse . it turns out that the mass loss rate is sensitive to the ratio of line fluxes between phase 0.5 and eclipse , because the size of the region becomes larger when we increase the mass loss rate . if a different value for the terminal velocity is assumed , the mass loss rate will differ such that the quantity @xmath42 remains fixed . the simulations are run for the various values of the mass loss rate . in order to calculate the observed spectra at different orbital phases , we designate the direction of the outgoing photon by the angle @xmath43 that the momentum vector makes relative to the line of centers ( see , fig . [ fig : direction ] ) . the phase 0.5 spectrum is constructed from the outgoing photons with @xmath44 and the eclipse spectrum is made from photons with @xmath45 . in this calculation , photons directly coming from the neutron star are neglected since only the emission lines and scattered components are of interest . although many emission lines are resolved in the chandra spectra , we use only the ly@xmath10 line to estimate the mass loss rate . this line is one of the brightest lines observed in the data . since the energy of this line is sufficiently high , the uncertainties associated with foreground absorption is lower than the low energy lines . at this energy , an uncertainty of 1 @xmath2 10@xmath46 @xmath47 in @xmath48 , corresponds to only a @xmath49% uncertainty in the ly@xmath10 intensity . the simulated intensity ratio of the ly@xmath10 lin between phase 0.5 and eclipse as a function of the mass loss rate is shown in fig . [ fig : sih_05_ec ] . the errors associated with these points are due to the finite number of photons generated in the simulation . the horizontal lines indicate the @xmath50 range in the observed value . as mentioned earlier in [ sec : velaion ] , the ratio becomes larger as the mass loss rate is increased . the intensities of the si ly@xmath10 in the phase 0.5 and in the eclipse are shown in fig . [ fig : sih_line_intensity ] . from these three plots in fig . [ fig : sih_05_ec ] and fig . [ fig : sih_line_intensity ] , the mass loss rate is estimated to be ( 1.52.0 ) @xmath0 . the estimated mass loss rate of the stellar wind is consistent with the observed x - ray continuum luminosity . the mass accreting rate onto the neutron star @xmath51 is given by @xmath52 where @xmath32 is the mass loss rate of the stellar wind , and @xmath53 is the distance of the neutron star from the center of the ob star . @xmath54 is the accretion radius express as : @xmath55 where @xmath56 is the relative velocity of the neutron star and the stellar wind . the gravitational energy of the accreting material is converted into x - rays . the x - ray luminosity resulting from this accretion will simply be the rate at which gravitational energy is released : @xmath57 where it is assumed that most of this energy is liberated near the neutron star surface ( of radius @xmath58 ) . if @xmath59 @xcite , @xmath60 km , @xmath61 km s@xmath1 ( @xmath62 km s@xmath1 , @xmath63 km s@xmath1 ) , @xmath64 and @xmath65 are applied , an x - ray luminosity of @xmath66 erg s@xmath1 is obtained , which agrees with the observed luminosity of 3.5 @xmath2 10@xmath5 erg s@xmath1 within a factor of 2 . from the eclipse spectrum obtained by , @xcite estimated that the mass loss rate associated with the highly ionized component of the stellar wind is @xmath67 , which is a factor of 58 lower than the current estimate . the lower mass loss rate was derived most likely because only the eclipse data was used . as the mass loss rate is increased , the intensities of the emission lines from highly ionized ions become higher . when the mass loss rate is increased further , the photoionized region is occulted by the companion star and the intensities of the emission lines during the eclipse phase become lower . this tendency for the is shown in fig . [ fig : sih_line_intensity ] ( left ) . there were two solutions for the eclipse spectrum , and , @xcite found the solution of the lower mass rate . however , we note that the intensity of emission lines in the phase 0.50 obtained with _ chandra _ can not be explained with such a lower mass loss rate . we , therefore , have concluded the higher mass loss rate of 1.52.0 @xmath68 . these measurements were probably not possible with the moderate spectral resolution of the instruments on board . adopting a mass loss rate of 1.5 @xmath2 10@xmath38 @xmath39 yr@xmath1 estimated from the line , we use the results of the monte carlo simulation to generate broadband spectra including both the continuum and emission lines from other ionic species for the three orbital phases observed with the hetgs . photons with outgoing momenta in the @xmath43 ranges of 010 , 8595 , and 170180 are used for the phase 0.50 , phase 0.25 , and eclipse model spectra , respectively . figure [ fig : all_data_model ] shows both the observed spectra above @xmath69 kev and the simulated model spectra convolved through the response of the hetgs . the simulated spectra are multiplied by the effects of inter - stellar absorption ( @xmath70 @xmath47 ) in addition to intrinsic absorption by the stellar wind , which depends on the observed binary phase . this matches the observed continuum shape remarkably well at both phases 0.25 and eclipse . at phase 0.5 , however , the observed column density of @xmath71 @xmath47 is in excess of the galactic plus stellar wind by an amount @xmath72 @xmath47 . we apply this additional absorption component to the continuum emission to match the data . the simulated spectra at the three orbital phases are shown in fig . [ fig : sim_spec ] . the observed emission line spectra in comparison with the simulations are shown in fig . [ fig : simobs_spec ] . the lack of obvious discrete spectral features at phase 0.25 is consistent with the model predictions , which show p - cygni lines with comparable emission and absorption equivalent widths and are therefore washed out after convolution with the spectral response . we note that the additional absorption component observed at phase 0.50 does not appear to be affecting the emission lines , which implies that this absorber is localized to a small region along the line of centers and in front of the neutron star at phase 0.5 . the simulated spectra match fairly well the emission lines from h - like and he - like ions , as well as the radiative recombination continuum emission . the ratios of simulated line intensities to the observed ones are plotted for those lines that are detected at greater than 5 @xmath13 in fig . [ fig : compare_intensity ] . all of the simulated line intensities are within the observed values to within a factor of @xmath73 . one of the uncertainties of the line intensities is the absorption , and it is difficult to derive the line intensities precisely by unfolding the absorption effect . in the simulation , the effects of inter - stellar absorption ( @xmath70 @xmath47 ) and intrinsic absorption by the stellar wind ( @xmath74 @xmath47 ) are assumed . if there is an additional absorption of 10% of the assumed value , the intensities of ne lines are reduced to @xmath4 50% . additionally , the absorption cross sections are very sensitive to the ionization degree of the absorption matter . fluorescence emission lines , in particular the bright iron line complex , provide information about the distribution of cold material around the x - ray source . figure [ fig : vela_fek ] shows the iron k@xmath10 line spectrum at the three orbital phases , together with the models obtained from the simulations described in the previous subsection . during eclipse , the iron k@xmath10 line intensity calculated from the simulation is consistent with the observed one . therefore , it seems likely that the iron k@xmath10 line observed in eclipse is produced in the cool regions of the extended stellar wind . on the other hand , we find that this same model can not explain the observed intensity of the k@xmath10 line in the other two phases . the equivalent widths from the simulations are only 4 ev and 11 ev for phases 0.25 and 0.50 , whereas the observed values are 51 ev and 116 ev , respectively . as an obvious source of additional line production regions , we first consider the surface of the companion star . using monte carlo simulations and adopting the geometrical parameters listed in table [ tbl : simparams ] , we obtain reflected iron k@xmath10 equivalent widths of 29 ev and 65 ev during phase 0.25 and phase 0.50 , respectively , assuming that the metal abundance of the companion star is also 0.75 times the cosmic value . therefore , the stellar wind and the companion surface combined yield equivalent widths of 33 ev ( phase 0.25 ) and 76 ev ( phase 0.50 ) . the observed values are still in excess these values by 18 ev and 40 ev at phase 0.25 and 0.5 , respectively . we note the presence of excess absorption of @xmath75 = 1.7 @xmath2 10@xmath76 @xmath47 observed during phase 0.5 relative to that at phase 0.25 . this suggests that a distribution of cold localized material lies along the line of centers somewhere behind the neutron star as viewed from the companion . we therefore expect that at least some amount of fe - k line photons are produced in this cloud . this has already been pointed out by @xcite using the _ tenma _ data . in order to estimate the contribution of the cloud to the iron line flux , we have performed a monte carlo simulation of a partially covered cloud irradiated by an x - ray source . the geometrical configuration is shown schematically in fig . [ fig : partial_cloud ] . the thickness of the cloud is assumed to be @xmath77 @xmath47 and we assume a uniform density . the cloud is assumed to be neutral and its metal abundance is set to 0.75 cosmic . the resulting emission spectra as functions of the orbital phase with the solid angle as the free parameter are computed through the same procedures as in the previous simulations , and the simulated fe k@xmath10 line equivalent widths are measured at the three orbital phases . figure [ fig : eqw_cover ] shows the relation between the solid angle subtended by the cloud and the iron line equivalent width . as can be seen from this figure , the missing equivalent widths outside of eclipse ( 18 ev at phase 0.25 and 40 ev at phase 0.50 ) can be explained if the cloud covers 2540 % of the sky as viewed from the neutron star . although this geometrical configuration may be oversimplified , its general characteristics are consistent with the fact that the spectrum observed at phase 0.25 does not exhibit heavy absorption and the fact that soft x - ray lines are still observed during phase 0.5 . @xcite reported a gradual increase of average absorption column density between orbital phase @xmath78 and @xmath79 from @xmath75 @xmath80 1 @xmath2 10@xmath81 @xmath47 to @xmath75 @xmath80 3 @xmath2 10@xmath76 @xmath47 observed in the _ tenma _ data of vela x-1 . the enhanced absorption at orbital phase 0.50 , and the localized cloud responsible for the iron line production can be attributed to an `` accretion wake '' , which is produced by a combination of several physical effects , including gravitational , rotational , and radiation pressure forces , and x - ray heating . numerical simulations by @xcite , @xcite , @xcite and @xcite exhibit such structures trailing the accreting neutron star . when the observed line profiles of a few of the brightest lines are compared with the output of the monte carlo simulations , we find that there is a mismatch in the average velocity shift at both phase 0.5 and eclipse . figure [ fig : comparison_doppler ] shows comparison of the observed and ly@xmath10 lines with the simulations convolved with the instrument response . although the sign of the shifts are correct , the simulations seem to predict a larger velocity shift than seen in in the data . figure [ fig : dopplershift ] shows the velocity shifts for all of the bright highly ionized emission lines detected in the spectrum . the average velocity shifts ( @xmath82 ) observed between phase 0.5 and eclipse lie range in @xmath4 300600 km s@xmath1 whereas the simulations predict @xmath83 1000 km s@xmath1 , which is larger approximately by a factor of two . one of the most obvious candidate mechanisms that is capable of reducing the wind velocity is x - ray photoionization . in the cak - model of isolated ob stars , a stellar wind is driven by the radiation force on the large number of resonance lines that lie in the uv . in particular , the dominant ions that are responsible for this are the l - shell ions of c , n and o , which have strong valence shell transitions in the uv range and have large abundances compared to the other metals . in hmxbs , there are regions in the stellar wind where x - ray photoionization destroys the l - shell ions of c , n , and o into k - shell ions and , therefore , reduce the amount of radiative force . this effect is maximized near the neutron star where the x - ray flux is high and which also happens to coincide with the region where the wind would otherwise be accelerated in the absence of the x - ray source . as a result , the stellar wind velocity within this region should be lower than that predicted by the cak - model . the effect of x - ray irradiation on line - driven stellar winds in hmxb systems has been studied theoretically by @xcite and @xcite , and they have also been seen in uv spectral observations @xcite . in order to investigate how the stellar wind velocity is affected due to x - ray ionization , we have performed a simple one - dimensional calculation to estimate the velocity profile along the line of centers . the extension of this calculation to three - dimensions is required to self - consistently incorporate into the monte carlo simulator and predict the emission line spectra , which is beyond the scope of this paper . nevertheless , as we show below , this estimate in 1-d is still useful for understanding semi - quantitatively the mismatch between the observed and predicted velocity shifts . in this calculation , we make two simplifying assumptions . we first assume that the wind is driven radiatively by uv resonance absorption and that there are no other relevant forces that provide additional acceleration . the other assumption is that the force is proportional to the local number densities of c , n and o ions with more than one electron in the l - shell ( i.e. , li - like and lower charge states ) . figure [ fig : cakmodel_v_f_r ] shows the velocity as a function of distance along the line of centers in the cak - model and the radiative force , which is numerically derived from the cak - model velocity structure . as can been seen from these figures , the radiation force peaks in the region between the companion star surface and the location of the neutron star . this is , however , the region where most of the x - ray lines are produced in this system . therefore , the acceleration of the stellar wind in this region must be reduced since the c , n , and o ions inside this region are primarily h - like , he - like , or fully - stripped , and they alone can not provide the force required to maintain the acceleration . we estimate the magnitude of this effect as follows . we start with the cakvelocity profile , which has an implicit assumption about the force as a function of distance , and can calculate the density everywhere in the wind . this can then be used to derive the ionization structure and , in particular , the distribution of c , n , and o ions , which provides an estimate of the new force and hence the velocity profile . this process can be repeated and iterated to find an ionization structure that is self - consistent with the velocity profile through the following steps , 1 . start from the velocity structure ( @xmath84 ( cak - model ) ) , and , calculate a force structure ( @xmath85 ) , then , obtain @xmath86 . 2 . start from @xmath87 , and , calculate @xmath88 , then obtain @xmath89 . 3 . start from @xmath90 , and , calculate @xmath91 , then obtain @xmath92 . 4 . start from @xmath93 , and , calculate @xmath94 , then obtain @xmath95 . 5 . start from @xmath96 , and , calculate @xmath97 , then obtain @xmath98 . 6 . start from @xmath99 , and , calculate @xmath100 , then obtain @xmath101 . the computed velocity structures ( @xmath102 , @xmath86 , @xmath89 , @xmath92 , @xmath95 , @xmath98 and @xmath101 ) are plotted in fig . [ fig : velocity_calc ] . as shown , the solution has converged after the 5th iteration . in this simple model , the stellar wind velocity at the point of the neutron star is only @xmath4 180 km s@xmath1 compared to @xmath4 570 km s@xmath1 predicted by the cak - model . this reduction in velocity of @xmath4 390 km s@xmath1 is remarkably similar to the observed velocity offset between the model and data observed during phase 0.5 , where most of the emission is believed to originate from near the neutron star . our one dimensional calculations confirm that the photoionization by the x - ray radiation can make the stellar wind near the neutron star slower by a factor of 23 compared to an undisturbed cak wind . this results are in agreement with the observed doppler shifts in the emission lines . the x - ray radiation of the neutron star is disturbing the flow of the cak - model stellar wind in the vela x-1 system , and then , the velocity of the stellar wind in the highly photoionized region becomes slower than that estimated from cak - model . in order to understand the flow of stellar wind precisely and obtain the x - ray spectrum from the photoionized stellar wind , further considerations should be needed . a three dimensional calculation of the electromagnetic hydrodynamics would be required for a fully quantitative analysis . the lower velocity of the stellar wind due to the x - ray photoionization leads to a density increase and changes of emission line intensities . in that case , an adjustment of the mass loss rate could be needed to reproduce the observed spectra . though the mass loss rate is proportional to the density and the velocity , how large the adjustment is required is not simple . the neutron star goes around the companion star , and , the photoionized region moves in association with the orbital motion of the neutron star . it takes about 0.5 days for the stellar wind to reach the position of the neutron star from the surface of the companion star assuming the cak - model . it is not negligible in comparison to the orbital period of the neutron star ( 8.96 days ) , and , calculations including the time evolution due to the orbital motion are required . additionally , gas pressure forces not only in the radial directions but along the circumferential directions are important since the stellar wind flow is not spherically symmetry any more . vela x-1 was observed by the _ chandra _ hetgs at three distinct orbital phase ranges . in order to understand the geometry and the dynamics of the x - ray irradiated stellar wind in this system , we have developed a monte carlo spectral simulator to predict the emission line spectrum for a given set of stellar wind parameters . the model spectra were then compared to the observed spectra to construct a self - consistent picture that explains the global properties of the x - ray spectra . the qualitative aspects of this model and its various components are illustrated in fig [ fig : hmxb_illust ] . the x - ray emission line intensities of highly ionized h- and he - like ions are well - reproduced by a cak - model of a stellar wind with a mass loss rate of ( 1.52.0 ) @xmath103 and a terminal velocity of 1100 km s@xmath1 irradiated by the x - ray continuum radiation from the neutron star . through comparisons of the spectra observed at eclipse and at phase 0.50 , we find that most of the x - ray recombination emission lines are produced in the region between the neutron star and the companion star ( see fig . [ fig : hmxb_illust ] ) where both the gas density and the local x - ray flux are high . this region is occulted by the companion star during the eclipse phase , which explains the relative strengths of the emission line intensities during phase 0.50 and eclipse . the same set of model parameters also nicely explains the lack of obvious discrete spectral features observed at phase 0.25 . the properties of the intense iron fluorescent lines , which are observed throughout the binary orbit , are sensitive to the distribution of cold material relative to the location of the neutron star . although the iron fluorescent line intensity observed during eclipse can be explained entirely by fluorescence in the extended stellar wind , the line intensities during the other two phases are far brighter than predicted by the same model . we find that , in addition to the stellar wind and the x - ray irradiated surface of the companion star , a cold cloud partially covering the neutron star is required to explain the data . from the quantitative analysis , we find that a cloud with a thickness of @xmath77 @xmath47 and covers 2540 % of the solid angle viewed from the neutron star , explains both the iron k@xmath10 line intensities as well as the excess column density observed during phase 0.5 . the one observed feature that this model fails to reproduce is the velocity shift of the highly - ionized recombination emission lines , which are observed to be smaller in magnitude by several hundred @xmath104 compared to the predictions . using a simple 1-d model , we find that destruction of uv - absorbing ions by x - ray photoionization can explain this reduction in velocity . however , the fully self - consistent 3-d model including the x - ray photoionization effect still remains an issue to be resolved . a full 3-d electromagnetic hydrodynamics calculation including time dependent effects would be required for the further study . in addition , there are much more things to be done with this data set . these include study of the compton shoulder in the iron k line and the ratio of he - like triplet lines , which constrain the physical conditions in the emission region . there is also much more to be improved in the monte carlo code . for example , the code for ions from li - like to near neutral is needed for a detailed study of the si k line complex shown in fig . [ fig : si_line_spec ] . such topics should be subjects of the next paper . 37 band , i.m . , trzhaskovskaya , m.b . , verner , d.a . , & yakovlev , d.g . 1990 , , 237 , 267 barziv , o. , kaper , l. , van kerkwijk , m.h . , telting , j.h . , & van paradijs , j. 2001 , , 377 , 925 blondin , j.m . , kallman , t.r . , fryxell , b.a . , & taam , r.e . 1990 , , 356 , 591 blondin , j.m . , stevens , i.r , & kallman , t.r . 1991 , , 371 , 684 blondin , j.m 1994 , , 435 , 756 bord , d.j . , mook , d.e . , petro , l. , & hiltner , w.a . 1976 , , 203 , 689 boroson , b. , vrtilek , s.d . , kallman , t. & corcoran , m. , 2003 , , 592 , 516 brucato , r.j . , & kristian , j. 1972 , , 173 , 105 castor , j.i . , abbott , d.c . , & klein , r.i . 1975 , , 195 , 157 dupree , a.k . 1980 , , 238 , 969 feldman , u. 1992 , physica scripta . 46 , 202 forman , w. , jones , c. , tananbaum , h. , gursky , h. , kellogg , e. , & giacconi , r. 1973 , , 182 , 103 goldstein , g. , huenemoerder , d.p . & blank , d. 2004 , , 127 , 2310 hiltner , w.a . , werner , j. , & osmer , p. 1972 , , 175 , l19 hutchings , j.b . 1976 , , 203 , 438 inoue , h. 1985 , , 40 , 317 kallman , t.r . , & white , n.e . 1982 , , 261 , 35 kaper , l. , hammerschlag - hensberge , g. , & van loon , j.th . 1993 , , 279 , 485 krause , m.o . 1979 , j. phys . chem . ref . data , 8 , 307 kreykenbohm , i. , kretschmar , p. , wilms , j. , staubert , r. , kendziorra , e. , gruber , d.e . , heindl , w.a . , & rothschild , r.e . 1999 , , 341 , 141 larkins , f.b . 1977 , at . data and nucl . data tables , 20 , 313 macgregor , k.b . , & vitello , p.a.j . 1982 , , 259 , 267 makishima , k. , mihara , t. , nagase , f. , & tanaka , y. 1999 , , 525 , 978 masai , k. 1984 , ap&ss , 106 , 391 mcclintock , j.e . 1976 , , 206 , 99 paerels , f. , cottam , j. , sako , m. , liedahl , d.a . , brinkman , a.c . , van der meer , r.l.j . , kaastra , j.s . , & predehl , p. 2000 , , 533 , l135 pauldrach , a. , puls , j. , & kudritzki , r.p . 1986 , 164 , 86 prinja , r.k . , barlow , m.j . , & howarth , i.d . 1990 , , 361 , 607 sadakane , k. , hirata , r. , jugaku , j. , kondo , y. , matsuoka , m. , tanaka , y. , & hammerschlag - hensberge , g. 1985 , , 288 , 284 sako , m. , liedahl , d.a . , kahn , s.m . , & paerels , f. 1999 , , 525 , 921 salem , s.i . , panossian , s.l . , & krause , r.a . 1974 , at . data and nucl . data tables , 14 , 91 sato , n. , hayakawa , s. , nagase , f. , masai , k. , dotani , t. , inoue , h. , makino , f. , makishima , k. , & ohashi , t. 1986 , , 38 , 731 sawada , k. , matsuda , t. & hachisu , i. 1986 , , 221 , 679 schulz , n.s . , canizares , c.r . , lee , j.c . , & sako , m. 2002 , , 564 , l21 van kerkwijk , m.h . , van paradijs , j. , hammerschlag - hensberge , g. , kaper , l. , & sterken , c. 1995 , , 303 , 483 van pradijs , j. , zuiderwijk , e.j . , takens , r.j . , & hammerschlag - hensberge , g. 1977 , , 30 , 195 watanabe , s. , et al . 2003 , , 597 , l37 cccccc 0.25 & 1.45 @xmath105 0.03 & 1.01 @xmath105 0.01 & 3.0 @xmath2 10@xmath106 & 1.6 @xmath2 10@xmath5 & 2872./ 3318 + 0.50 & 18.5 @xmath105 0.3 & 1.01 ( fixed ) & 1.5 @xmath2 10@xmath106 & 1.6 @xmath2 10@xmath5 & 1000./ 1051 + ccccc @xmath107 & @xmath108 & [email protected] & ca ii xii k@xmath10 & + [email protected] & @xmath109 & @xmath110 & s xvi ly@xmath10 & + @xmath111 & @xmath112 & @xmath113 & s xv r & + @xmath114 & @xmath115 & @xmath116 & s iv viii k@xmath10 & + [email protected] & @xmath117 & @xmath118 & si xiv ly@xmath10 & + 127@xmath10524 + [email protected] & [email protected] & @xmath119 & si xiii r & @xmath120 + @xmath121 & @xmath122 & @xmath123 & si xiii i & @xmath124 + @xmath125 & [email protected] & [email protected] & si xiii f & @xmath126 + [email protected] & [email protected] & @xmath127 & si ii vi k@xmath10 & + @xmath128 & @xmath129 & [email protected] & al xiii ly@xmath10 ? & + @xmath130 & @xmath131 & @xmath132 & al xii r ? & + [email protected] & [email protected] & [email protected] & mg xi & + @xmath133 & @xmath134 & @xmath135 & fe xxiv ? & + [email protected] & [email protected] & [email protected] & mg xii ly@xmath10 & @xmath136 + @xmath137 & [email protected] & [email protected] & mg xi r & @xmath138 + @xmath139 & [email protected] & @xmath140 & mg xi i & @xmath141 + [email protected] & [email protected] & @xmath142 & mg xi f & @xmath143 + @xmath144 & @xmath145 & @xmath146 & fe xxi ? & + @xmath147 & [email protected] & @xmath148 & ne x ly@xmath149 & + [email protected] & @xmath150 & @xmath151 & ne x ly@xmath31 & + @xmath152 & @xmath153 & @xmath154 & ne ix & + @xmath155 & @xmath156 & @xmath157 & ne ix & + @xmath158 & @xmath159 & @xmath160 & ne x ly@xmath10 & @xmath161 + [email protected] & [email protected] & @xmath162 & ne ix r & @xmath163 + @xmath164 & @xmath165 & @xmath166 & ne ix i & @xmath167 + @xmath168 & @xmath169 & @xmath170 & ne ix f & @xmath171 + ccccc @xmath172 & @xmath173 & @xmath174 & ca ii xii k@xmath10 & + @xmath175 & @xmath176 & @xmath177 & ar vi ix k@xmath10 & + @xmath178 & @xmath179 & @xmath180 & s xvi ly@xmath10 & + [email protected] & [email protected] & @xmath181 & s iv viii k@xmath10 & + [email protected] & [email protected] & @xmath182 & si xiv ly@xmath10 & @xmath7314@xmath10551 + @xmath183 & [email protected] & @xmath184 & si xiii r & @xmath185 + @xmath186 & @xmath187 & @xmath188 & si xiii i & @xmath189 + @xmath190 & @xmath191 & @xmath192 & si xiii f & @xmath193 + @xmath194 & [email protected] & @xmath195 & si ii vi k@xmath10 & + @xmath196 & @xmath197 & @xmath198 & mg xi or fe xxiii & + @xmath199 & @xmath200 & @xmath201 & mg xi & + @xmath202 & @xmath203 & @xmath204 & mg xii ly@xmath10 & @xmath205 + [email protected] & @xmath206 & @xmath207 & mg xi r & @xmath7373@xmath10569 + @xmath208 & [email protected] & @xmath209 & mg xi i & @xmath210 + @xmath211 & @xmath212 & @xmath213 & mg xi f & @xmath214 + @xmath215 & @xmath216 & @xmath217 & fe xxi ? & + @xmath218 & @xmath219 & @xmath220 & ne x ly@xmath149 & + [email protected] & @xmath221 & @xmath222 & fe xx ? & + @xmath223 & @xmath224 & @xmath225 & ne x ly@xmath31 & + @xmath226 & [email protected] & @xmath227 & ne ix & + @xmath228 & [email protected] & @xmath229 & ne x ly@xmath10 & @xmath230 + @xmath231 & @xmath232 & @xmath233 & ne ix r & @xmath234 + [email protected] & @xmath235 & @xmath236 & ne ix i & @xmath237 + @xmath238 & [email protected] & @xmath239 & ne ix f & @xmath240 + rcc si xiv ly@xmath10 & 10.3 @xmath105 1.3 & 441 @xmath105 56 + si xiii resonance & 5.81 @xmath105 0.83 & 506 @xmath105 62 + si xiii forbidden & 6.68 @xmath105 0.95 & 345 @xmath105 86 + mg xii ly@xmath10 & 8.94 @xmath105 1.37 & 436 @xmath105 58 + mg xi resonance & 5.70 @xmath105 1.00 & 493 @xmath105 78 + mg xi forbidden & 4.50 @xmath105 1.01 & 205 @xmath105 123 + ne x ly@xmath10 & 5.18 @xmath105 1.04 & 490 @xmath105 60 + ne ix resonance & 5.27 @xmath105 2.14 & 549 @xmath105 153 + ne ix forbidden & 4.13 @xmath105 1.49 & 498 @xmath105 198 + lrl geometry & & + binary separation @xmath53 & 53.4 @xmath248 & @xcite + companion star radius @xmath29 & 30.0 @xmath248 & @xcite + stellar wind & & + velocity structure @xmath249 & @xmath250 & @xcite + terminal velocity @xmath30 & 1100 km s@xmath1 & @xcite + @xmath31 & 0.80 & @xcite + metal abundance ( @xmath251 ) & 0.75 cosmic & for typical ob - stars + & & @xcite + x - ray radiation & & + luminosity(0.520 kev ) & 3.5 @xmath2 10@xmath5 erg s@xmath1 & this observation + & & ( 1.6 @xmath2 10@xmath5 erg s@xmath1 ( 0.510 kev ) ) + spectrum shape & power - law ( @xmath252 = 1.0 ) & this observation + energy range & 13.6 ev 20.0 kev & +

the aim of this paper ( inspired from a problem of habegger ) is to describe the set of cubical decompositions of compact manifolds mod out by a set of combinatorial moves analogous to the bistellar moves considered by pachner , which we call bubble moves . one constructs a surjection from this set onto the the bordism group of codimension one immersions in the manifold . the connected sums of manifolds and immersions induce multiplicative structures which are respected by this surjection . we prove that those cubulations which map combinatorially into the standard decomposition of for large enough ( called mappable ) , are equivalent . finally we classify the cubulations of the 2-sphere . + _ keywords and phrases _ : cubulation , bubble / np - bubble move ,-vector , immersion , bordism , mappable , embeddable , simple , standard , connected sum . .

math9804046

stellar moves were first considered by alexander ( @xcite ) who proved that they can relate any two triangulations of a polyhedron . alexander s moves were refined to a finite set of local ( bistellar ) moves which still act transitively on the triangulations of manifolds , according to pachner ( @xcite ) . using pachner s result turaev and viro proved that certain state - sums associated to a triangulation yield topological invariants of 3-manifolds ( see @xcite ) . recall that a bistellar move ( in dimension @xmath1 ) excises @xmath3 and replaces it with @xmath4 , where @xmath3 and @xmath4 are complementary balls , subcomplexes in the boundary of the standard @xmath5-simplex . for a nice exposition of pachner s result and various extensions , see @xcite . the turaev - viro invariants carry less information than the reshetikhin - turaev invariants , which are defined using dehn surgery presentations instead of triangulations . in fact the latter have a strong 4-dimensional flavor , as explained by the theory of shadows developed by turaev ( see @xcite ) . this motivates the study of state - sums based on cubulations , as an alternative way to get intrinsic invariants possibly containing more information ( e.g. the phase factor ) . cubical complex _ is a complex @xmath6 consisting of euclidean cubes , such that the intersection of two cubes is a finite union of cubes from @xmath6 , once a cube is in @xmath6 all its faces are still in @xmath6 , and no identifications of faces of the same cube are allowed . a _ cubulation _ of a manifold is specified by a cubical complex pl homeomorphic to the manifold . in order to apply the state - sum machinery to these decompositions we need an analogue of pachner s theorem . specifically , n.habegger asked ( see problem 5.13 from r.kirbys list ( @xcite ) ) the following : suppose @xmath7 and @xmath8 are pl - homeomorphic cubulated @xmath1-manifolds . are they related by the following set of moves : excise @xmath3 and replace it by @xmath4 , where @xmath3 and @xmath4 are complementary balls ( union of @xmath1-cubes ) in the boundary of the standard @xmath5-cube ? these moves will be called _ bubble _ moves in the sequel . among them , those for which @xmath3 or @xmath4 does not contain parallel ( when viewed in the @xmath9-cube ) faces are called _ np - bubble _ moves . there are @xmath9 distinct np - bubble moves @xmath10 , @xmath11 and their inverses , where the support @xmath3 of @xmath10 is the union of exactly @xmath12 cubes . for @xmath13 there is one bubble move which is not a np - bubble ( see picture 1 ) . set @xmath14 for the set of cubulations of a closed manifold @xmath7 , @xmath15 for the equivalence classes of cubulations mod np - bubble moves and @xmath16 for the equivalence classes of cubulations mod bubble moves . the answer to habegger s question , as it states , is negative because the triangle and the square are not bubble equivalent . in fact , for @xmath17 the move @xmath18 divides an edge into three edges and so @xmath19 . therefore a complete answer would rather consist of a description of @xmath16 . another way is to avoid the difficulties of a direct approach by looking for a sufficiently large class of cubulations having an intrinsic characterization and within which the cubulations are equivalent . the aim of this paper is to formulate some partial solutions along these lines . for instance , one associates to each cubulation @xmath20 of @xmath7 , a codimension one normal crossings immersion @xmath21 in @xmath7 . in this way one obtains a surjective map from the set of ( marked ) cubulations mod bubble moves to the bordism set of immersions . the latter has a homotopical description via the pontryagin - thom construction . we conjecture that this surjection is a bijection . on the other hand let us restrict to cubulations which can be combinatorially mapped into the standard cubulation of some euclidean space ( called mappable ) . one can approximate an ambient isotopy between two cubulations by some cubical sub - complexes of the standard cubulation . the path of cubical approximations is locally constant except for a finite number of critical values of the parameter , when a jump described by a bubble move occurs . as a consequence two mappable cubulations are bubble equivalent . we prove that the connected sum of cubulations mod bubble moves is well - defined , and this is compatible with the composition map for immersions . finally we consider the case of @xmath22 and show by a direct combinatorial proof that @xmath23 . * acknowledgements * : part of this work was done during the author s visit at university of palermo and university of columbia , whose support and hospitality are gratefully acknowledged . i m thankful to e.babson , j.birman , c.blanchet , r.casali , c.chan , l.guillou , n.habegger , t.kashiwabara , a.marin , d.matei , s.matveev , v.ponaru , r.popescu , v.sergiescu and the referee for helpful discussions , suggestions and improvements . we outline the combinatorial approach in higher dimensions from @xcite . for a cubulation @xmath24 of the @xmath1-manifold @xmath7 the component @xmath25 of the @xmath2-vector @xmath26 counts the number of @xmath27-dimensional cubes in @xmath28 . the orbit of the @xmath2-vector @xmath29 under bubble moves has the form @xmath30 , where @xmath31 is a lattice . therefore we have an induced map @xmath32 taking values in a finite abelian group . [ section ] there exist nonzero even numbers @xmath33 such that the projection @xmath34 is surjective . the greatest such numbers @xmath35 verify @xmath36 , @xmath37 , @xmath38 , @xmath39 , @xmath40 . see @xcite for the proof . let @xmath41 be the class of @xmath29 in @xmath42 and @xmath43 be the reduced elements modulo @xmath44 . notice that @xmath31 is not a product lattice in general . for instance , when @xmath45 there is an additional invariant @xmath46 . a natural problem is to compute the image @xmath47 for given @xmath7 . some partial results for the mod 2 reductions @xmath48 are known . this is equivalent to characterize those @xmath2-vectors mod 2 which can be realized by cubulations of the manifold @xmath7 . there are constraints for the existence of a simplicial polyhedron with a given @xmath2-vector and fixed topological type . for convex simplicial polytopes one has mcmullen s conditions ( see @xcite ) . the complete characterization of the @xmath2-vectors of simplicial polytopes ( and pl - spheres ) was obtained in @xcite . the analogous problem of the realization of @xmath2-vectors by cubical polytopes has also been addressed in some recent papers , for example @xcite and references therein . the new feature is that , unlike for the simplicial case , there are parity restrictions on the @xmath2-vectors ( see @xcite ) . the relationship between cubical pl @xmath1-spheres and the immersions was described in the following result of babson and chan ( see @xcite ) : [ alba1]proposition there exists a cubical @xmath1-sphere @xmath6 with given @xmath49 if and only if there exists a codimension 1 normal crossings immersion @xmath50 such that @xmath51 , where @xmath52 denotes the euler characteristics , and @xmath53 is the set of @xmath27-tuple points . there is a wide literature on immersions , and especially on the function @xmath54 counting the number of multiple @xmath1-points mod 2 , which was considered first by freedman ( @xcite ) . earlier banchoff @xcite has proved that the number of triple points of a closed surface @xmath55 immersed in @xmath56 is @xmath57 . there is an induced homomorphism @xmath58 on the abelian group @xmath59 of bordism classes of immersions of @xmath60-manifolds in @xmath61 . now @xmath54 is surjective ( i.e. nontrivial ) if and only if @xmath62 . from the results concerning the function @xmath54 obtained in @xcite we deduce that the @xmath2-vectors of a @xmath1-sphere have the following properties ( see also @xcite ) : 1 . for @xmath13 we have @xmath63 and @xmath64 and thus @xmath65 . 2 . for @xmath45 , @xmath66 , @xmath67 . the boy immersion @xmath68 has a single triple point and so there exists a pl 3-sphere with an odd number of facets . therefore @xmath69 . the characterization of @xmath70 is reduced to a homotopy problem : @xmath62 if and only if 1 . either @xmath1 is @xmath71 or @xmath72 . 2 . or else @xmath73 , with @xmath74 , and there exists a framed @xmath1-manifold with kervaire invariant 1 . the latter is known to be true for @xmath75 . if we consider only edge - orientable cubulations ( see @xcite ) then @xmath70 is known . the edge - orientability is equivalent to the orientability of the manifold immersed in @xmath76 and the restriction of the map @xmath54 to the subgroup of oriented bordism classes was computed in @xcite . in particular @xmath77 if @xmath78 . to a surface cubulation we can associate a set of immersed circles @xmath79 obtained from the union of arcs joining the opposite sides in each square . the cubulation is _ simple _ if the circles @xmath79 are individually embedded in the respective surface . simple is equivalent to mappable for the cubulations of @xmath80 ( see below ) . a cubulation is called _ semi - simple _ if each image circle @xmath81 has an even number of double points , which form cancelling pairs . two double points form a _ cancelling pair _ if they are connected by two distinct and disjoint arcs . [ alba1]theorem the np - bubble moves act transitively on the set of simple cubulations of @xmath80 . the orbit of the standard cubulation is the set of semi - simple cubulations . the map @xmath82 is an isomorphism between @xmath22 and @xmath83 . let us consider the set @xmath84 ( respectively @xmath85 in the orientable case ) of bordisms of codimension 1 nc ( i.e. normal crossings ) immersions in the manifold @xmath7 . two nc - immersions @xmath86 of the @xmath60-manifolds @xmath87 are _ bordant _ if there exists a proper nc - immersion @xmath88 $ ] of some cobordism @xmath8 between @xmath89 and @xmath90 , such that the restriction of @xmath2 to @xmath87 is isotopic to @xmath91 . using general position arguments one may get rid of the nc - assumption . a _ marked _ cubulation is a cubulation @xmath20 of the manifold @xmath7 , endowed with a pl - homeomorphism @xmath92 of its subjacent space @xmath93 , considered up to isotopy . if a bubble move is performed on @xmath20 , then there is a natural marking induced for the bubbled cubulation . thus it makes sense to consider the set @xmath94 of marked cubulations mod bubble moves . we associate to each marked cubulation @xmath20 a codimension 1 nc - immersion @xmath95 ( the cubical complex @xmath96 was called the derivative complex in @xcite ) . each cube is divided into @xmath97 equal cubes by @xmath1 hyperplanes which we call sections . when gluing together cubes in a cubical complex the sections are glued accordingly . the union of the hyperplane sections form the image of a codimension 1 nc - immersion . in the differentiable case one uses a suitable smoothing when gluing the faces . if the cubulation @xmath20 is edge - orientable ( see @xcite ) , and @xmath7 is oriented then @xmath96 is an oriented manifold . [ alba1]theorem the map @xmath98 induces a surjection @xmath99 . the theorem 1.3 says that the map @xmath100 is injective for @xmath101 . we conjecture that @xmath100 is bijective . in particular @xmath94 would depend only on the homotopy type of @xmath7 and the functor @xmath102 which associates to @xmath7 the set @xmath94 would be ( homotopically ) representable . notice that @xmath103 , where @xmath104 is the mapping class group of @xmath7 , i.e. the group of homeomorphisms of @xmath7 up to isotopy . using the classical pontryagin - thom construction ( see e.g. @xcite ) it follows that @xmath105 $ ] , and @xmath106 $ ] , where @xmath107 is the one point compactification , @xmath108 denotes the loop space , @xmath55 the reduced suspension and the brackets denote the set of the homotopy classes of maps . moreover @xmath109 is the first cohomotopy group @xmath110 . the cohomotopy groups of spheres can be computed : @xmath111 , where @xmath112 is the @xmath1-th stable homotopy group , and @xmath113 . it is known that @xmath114 , @xmath115 , and a few values of the stable stems are tabulated below : [ cols="^,^,^,^,^,^,^,^,^,^",options="header " , ] let us introduce now the set @xmath116 of bordisms of cubulations of the manifold @xmath7 . the cubulations @xmath117 and @xmath118 are _ bordant _ if there exists a cubulation @xmath20 of @xmath119 $ ] whose restrictions on the boundaries are the @xmath120 . the identity induces a map @xmath121 . the question on the existence of an inverse arrow is similar to wall s theorem about the existence of formal deformations between simple homotopy equivalent @xmath1-complexes through @xmath5-complexes ( for @xmath122 ) . remark that any two cubulations become bordant when suitably subdivided . consider some cubulation of the sphere @xmath76 which is bubble equivalent to the standard one . we can view the bubble moves as the result of gluing and deleting @xmath5-cubes ( after some thickening ) to the given cubulation . it follows that any such cubulation bounds , i.e. it is the boundary of a cubulation of the @xmath5-ball . for instance a polygon bounds iff it has an even number of edges . for @xmath123 it might be true that the boundary of a ball cubulation is bubble equivalent to the standard one , but the result can not be extended to @xmath124 . this is analogous to the existence of non - shellable triangulations of the ball for @xmath125 . we define a _ shuffling _ to be a sequence of moves where shellings ( i.e. adding iteratively cells , each intersecting the union of the previous ones along a ball ) alternate with inverse shellings . an equivalent statement of pachner s theorem is that all triangulations can be shuffled . in the case of cubulations the first obstruction for shuffling is that the cubulation bounds . however there exist cubulations which bound but can not be shuffled for @xmath126 . consider for example the connected sum @xmath127 , where @xmath128 is the generator of the third stable stem . we will prove below that the connected sum of cubulations makes @xmath129 a monoid . but @xmath127 bounds and if it can be shuffled then @xmath28 should have order 2 . this is impossible because @xmath100 is a monoid homomorphism and the bordism group is @xmath130 . cubical complexes , as objects of study from a topological point of view , were also considered by novikov ( @xcite , p.42 ) which asked whether a cubical complex of dimension @xmath1 embeds in ( or can be mapped to ) the @xmath1-skeleton of the standard cubic lattice of some dimension @xmath8 . these are called _ embeddable _ and respectively _ mappable _ cubulations . by the standard cubic lattice ( or the standard cubical decomposition ) is meant the usual partition of @xmath131 into cubes with vertices in @xmath132 . several results were obtained in @xcite . [ alba1]theorem the mappable cubulations of a pl manifold @xmath7 are bubble equivalent . let us say that a cubulation is _ simple _ if no path in which consecutive points correspond to edges which are opposite sides of some square of the cubulation , contains two orthogonal edges from the same cube . the cubulation is _ standard _ if any two of its cubes are either disjoint or have exactly one common face . an immediate observation is that embeddable cubulations are standard and simple and mappable cubulations are simple . on the other hand the simplicity is very close to the mappability , at least for manifolds with small fundamental group . we have for instance the following results of karalashvili ( @xcite ) and dolbilin , shtanko and shtogrin ( @xcite ) : 1 . the double ( i.e. the result of dividing each @xmath12-dimensional cube in @xmath133 equal cubes ) of a simple cubulation is mappable . a simple cubulation of a manifold @xmath7 satisfying @xmath134 is mappable . 3 . from a simplicial decomposition @xmath55 one constructs a cubulation @xmath135 by dividing each @xmath1-simplex into @xmath9 cubes . then the cubical decomposition @xmath135 is embeddable . in particular cubulations coming from triangulations are bubble equivalent . also the simple cubulations of the sphere are equivalent . notice that the set of simple ( or mappable ) cubulations is not closed to arbitrary bubble moves . in general the simplicity is not preserved by the move @xmath136 . the connected sum of the manifolds @xmath7 and @xmath8 is denoted by @xmath137 . when appropriately extended to cubulations @xmath138 depends on various choices , but after passing to bubble equivalence classes these ambiguities disappear . [ alba1]theorem there exists a map @xmath139 induced by the connected sum of cubulations . as a consequence @xmath138 induces also a composition map @xmath140 . on the other hand there is a natural composition map on the sets of bordisms of immersions , by using the connected sum away from the immersions . we prove that the map @xmath100 is functorial : [ alba1]theorem we have a commutative diagram @xmath141 we believe that the monoid @xmath129 is actually a group . notice that @xmath142 has a group structure for any @xmath7 ( induced from the cohomotopy group structure ) , but we do nt know whether this can be lifted to @xmath94 . let @xmath6 be the derivative complex ( having the connected components @xmath79 ) associated to a cubulation @xmath20 of the manifold @xmath7 and let @xmath143 be the associated immersion . in order to rule out some pathologies we restrict here to the _ combinatorial _ cubulations , meaning that the star of each vertex ( or the link ) is a pl ball ( respectively a pl sphere ) of the right dimension . let us first show that we have an induced map @xmath144 . consider the local picture of a bubble move , viewed in the boundary of the @xmath5-cube . the set of sections on the boundary are intersections of the hyperplane sections of the @xmath5-cube with the faces . let @xmath3 and @xmath4 be the @xmath1-balls interchanged by a bubble move . then the union of hyperplane sections in the @xmath5-cube yields a bordism between the immersions @xmath145 and @xmath146 in the @xmath5-ball . thus immersions associated to equivalent cubulations are cobordant . the immersion @xmath147 is _ pseudo - spine _ if the closures of the connected components of the complementary @xmath148 are balls , and the image of @xmath147 is connected . the immersion is called _ admissible _ if each connected component @xmath149 of the set @xmath150 of @xmath12-tuple points is a pl ball and we have @xmath151 ( for all @xmath152 , where @xmath153 denotes the closure ) . observe that any nc - immersion is cobordant to an admissible pseudo - spine immersion . in fact consider a set of small ( bounding ) spheres embedded in @xmath7 , transverse to the immersion @xmath147 . if they are sufficiently small they cut the connected components of the complementary @xmath148 into balls . we add sufficiently many so that all connected components of the strata @xmath150 are divided into balls by the additional spheres . the immersion @xmath154 whose image consists of the union of @xmath155 with the spheres is cobordant to @xmath147 . in fact let us choose some balls in @xmath156 $ ] bounded by the spheres in @xmath157 . the required cobordism is obtained by putting the balls in standard position with respect to @xmath158 $ ] . furthermore we want to associate a cubulation @xmath20 to the immersion @xmath147 such that @xmath159 . for an admissible pseudo - spine immersion one takes the corresponding cell decomposition of the manifold , then the dual decomposition is just the cubulation we are looking for . the pseudo - spine condition implies that the dual decomposition has the structure of a cubical complex and the admissibility is required for the cubulation comes endowed with a natural marking @xmath160 . the proof of theorem 1.5 goes as follows : we show first that embeddable cubulations are equivalent and then that a mappable cubulation is equivalent to an embeddable one . for the first step consider @xmath8 sufficiently large so that both cubulations @xmath161 and @xmath162 embed into the standard cubulation @xmath163 of the euclidean space and such that there exists an ambient isotopy carrying @xmath161 into @xmath162 . the image of @xmath161 during the isotopy is denoted by @xmath164 . we define an approximation @xmath165 of the manifold @xmath164 , which is a cubulated sub - manifold of @xmath163 , it is sufficiently close to @xmath164 , and when @xmath166 varies the family @xmath165 is either locally constant or changes around a `` critical value '' by a bubble move . we realize an arbitrarily fine approximation by taking the cubical structure be @xmath167 $ ] , based on cubes whose edges are of length @xmath168 , for small @xmath168 . a way to do that is to divide each cube of the lattice into @xmath169 equal cubes . then the initial cubulations @xmath161 and @xmath162 are replaced by some iterated doublings , say @xmath170 and @xmath171 . it remains to prove that @xmath170 is equivalent to @xmath161 . [ section ] [ gigi ] let @xmath161 and @xmath162 be two cubulations of a pl manifold which are embedded in the standard cubical lattice @xmath163 . then there exists @xmath172 arbitrarily large , such that @xmath170 and @xmath171 are bubble equivalent . [ alb]proposition if @xmath161 is embeddable , then for big enough @xmath172 the iterated doubling @xmath170 is bubble equivalent to @xmath161 . notice that the analogous statement for np - bubble moves is false in general . [ alb]proposition let @xmath161 be a mappable cubulation . then @xmath161 is np - bubble equivalent to an embeddable cubulation . consider a cubulation @xmath173 $ ] , of codimension at least 1 . let us denote by @xmath174 $ ] the union of all hyperplanes defining the cubulation @xmath167 $ ] , which can be written as a disjoint union of the strata @xmath174^{(m)}$ ] consisting of all open codimension @xmath172 cubes . set @xmath20 for the cube given by the equations @xmath175 . let @xmath176 be the @xmath177-complex obtained from @xmath178 by adding the hyperplanes @xmath179 . denote by @xmath180 the star of the origin in @xmath176 i.e. the union of cells having the form @xmath181 , where @xmath182 . [ alb]definition the disk @xmath183 is a standard model in @xmath20 if @xmath183 is properly embedded in @xmath20 ( and transverse to @xmath178 ) , @xmath183 is contained in @xmath180 and the origin lies in @xmath184 . [ alb]definition let @xmath20 be a n - cube of the cubulation @xmath167 $ ] . we say that @xmath185 is standard with respect to c if the following conditions are fulfilled : 1 . @xmath185 is transversal to @xmath174\cap c$ ] . 2 . there exists an isotopy supported on @xmath186^{(0)}\cap c)$ ] , if the codimension of @xmath185 is at least 2 , and respectively @xmath186^{(0)}\cup \lambda[\varepsilon]^{(1)}\cap c)$ ] , if the codimension is precisely 1 , which transforms @xmath187 in a standard model . finally @xmath185 is standard ( or in standard position ) with respect to @xmath167 $ ] ( or @xmath174 $ ] ) if @xmath185 is standard w.r.t . all cubes . observe first that a plane @xmath149 transverse to the boundary of @xmath20 is in standard position w.r.t . @xmath20 . in fact using a recurrence argument the intersection of @xmath149 with any face @xmath188 can be put in standard position by means of an isotopy . the union of standard models for @xmath189 over the faces is a pl sphere and the cone centered at the origin on it is isotopic to @xmath190 and hence it is a standard model . consider a submanifold @xmath185 which is standard w.r.t . the lattice @xmath191 . then there is an isotopy transforming @xmath185 into @xmath192 , where @xmath192 intersects each cube along a standard model . in fact this can be done in each cube that @xmath185 intersects and the standard models for different cubes have disjoint interiors . it suffices to check the compatibility of the boundary gluings : if @xmath185 cuts two adjacent cubes @xmath20 and @xmath193 then the standard models of @xmath187 and @xmath194 can be glued together . first the neighborhood of the common face is determined by the standard model of @xmath185 inside the face . hence we have to prove the uniqueness of the standard model for @xmath195 , which follows by a recurrence argument on the dimension . therefore the cubical complex @xmath192 is uniquely determined by @xmath185 and @xmath191 , and it will be called _ the standard model _ of @xmath185 w.r.t . @xmath191 . let consider the cubulations @xmath161 and @xmath162 of a @xmath1-manifold @xmath7 , embedded in @xmath163 . there exists an isotopy with compact support @xmath196\longrightarrow { \bf r}^n$ ] , with @xmath197 being the identity and @xmath198 . for big enough @xmath8 one can choose the isotopy @xmath199}$ ] to be an embedding . we assume that the cubulations @xmath161 and @xmath162 are embedded in the standard cubulation given by the affine lattice @xmath200 , which has the origin translated into @xmath201 . notice that a cubulation @xmath202 is automatically in standard position w.r.t . moreover the intersection of @xmath161 with each cube of @xmath191 is a standard model and so @xmath203 . we obtained a submanifold @xmath204)$ ] of @xmath131 , whose boundary @xmath205 is in standard position w.r.t . the lattice @xmath191 . we claim that for big enough @xmath172 there exists an isotopy carrying @xmath206 into @xmath185 , such that @xmath207 and @xmath205 are isometric and @xmath185 is in standard position w.r.t . the lattice @xmath208 $ ] . there exists a subdivision @xmath208 $ ] ( for large @xmath172 ) , such that @xmath206 becomes standard w.r.t . @xmath208 $ ] , after a small isotopy which is identity near the boundary . we translate @xmath206 into the lattice whose origin is at @xmath209 . then @xmath161 and @xmath162 transform into @xmath170 and @xmath210 , and they are in standard position w.r.t . @xmath208 $ ] . the last condition is an open condition , so we can keep fixed @xmath206 near the boundary during the isotopy . let us denote by @xmath211 the tube describing an isotopy between @xmath170 and @xmath171 , which is in standard position w.r.t . @xmath208 $ ] . there exist topologically trivial tubes @xmath212 between @xmath161 and @xmath170 , and respectively @xmath213 between @xmath162 and @xmath171 , which are in a standard positions w.r.t . @xmath208 $ ] . then set @xmath214 . therefore we derived a pl cylinder @xmath215 $ ] which interpolates between @xmath161 and @xmath162 . notice that the cubical structure of @xmath161 in @xmath208 $ ] is that of @xmath170 in @xmath191 . in general one can not shell the boundary from @xmath161 to @xmath162 . the tube @xmath185 carries a pl foliation by submanifolds @xmath216 , where @xmath147 states for the isotopy carried by @xmath185 . the leaf @xmath164 does not contain _ flat directions _ in the cube @xmath217 if @xmath164 does not contain any segment parallel to some vector in @xmath218 . using a small isotopy one can get rid of flat directions in all leaves furthermore there exists some @xmath172 such that either @xmath164 is standard w.r.t . @xmath219 $ ] , or else @xmath164 contains vertices of the lattice @xmath208 $ ] . the first alternative would hold if we are allowed to move slightly @xmath164 , using an arbitrary small isotopy ( in order to achieve the transversality ) . on the other hand the leaf is not transverse iff it contains vertices from @xmath208 $ ] , because there are not flat directions . further `` to be in standard position '' is an open condition and so one can choose the constant @xmath172 such that for each @xmath166 either the leaf @xmath164 is in standard position w.r.t . the cube @xmath220 $ ] , or else @xmath164 contains vertices from @xmath220 $ ] . the set of those exceptional @xmath166 for which the second alternative holds is finite because each critical leaf contains at least one vertex from @xmath220 $ ] and the different leaves are disjoint . observe that we can change the isotopy @xmath164 such that no exceptional leaf contains more than one vertex from @xmath208 $ ] , while keeping all the other properties we obtained upon now . furthermore there is such a @xmath172 which is convenient for all cubes that @xmath185 intersects . if one replaces a leaf @xmath164 by the standard model @xmath221)$ ] then we get a family of cubulations embedded in @xmath222 $ ] . this family should be locally constant , until @xmath166 reaches an exceptional value @xmath223 ( where the standard model can not be defined ) . set @xmath217 for the cube of size @xmath224 centered at the ( exceptional ) vertex . the intermediary set @xmath225}\cap u$ ] , for @xmath226 , is a trivial cobordism properly embedded in @xmath217 . using a small isotopy one can change @xmath227 into a union of planes passing through the vertex . if @xmath228 is the set of the @xmath229 directions of the coordinate axes around the vertex lying in @xmath185 we put @xmath230 , @xmath231 . each such direction is dual to a face of a cube @xmath232 in the dual lattice where the standard models @xmath233)$ ] live . we have @xmath234 , and @xmath235 since @xmath236 separates the directions which are cut by @xmath237 from the directions cut by @xmath238 . let @xmath239 and @xmath240 be respectively the union of faces of @xmath232 duals to the directions in @xmath241 and @xmath242 respectively . then @xmath239 and @xmath240 are pl balls because @xmath243 is a ball , as well as @xmath244 . we need only to see that both are non - void . if @xmath239 is empty then @xmath237 would be contained in a half cube @xmath245 of the lattice @xmath208 $ ] . then @xmath246 will be a cylinder , and thus it can not be standard , contradicting our hypothesis . therefore the standard model @xmath247)$ ] is obtained from @xmath248)$ ] by means of the bubble move @xmath249 having the support on @xmath232 . this proves proposition 3.1 . actually a stronger statement concerning sub - complexes of the standard lattice @xmath163 is true . we will ask also the bubble moves which pass from one cubulation to the other to be _ embedded _ in @xmath163 . this means that each bubble move which exchanges the balls @xmath3 and @xmath4 has the property that the cube bounded by @xmath250 is contained in the skeleton of @xmath163 . two isotopic lattice knots ( or graphs ) in @xmath56 are bubble equivalent , by means of bubble moves which can be realized on the lattice of @xmath56 and which avoid self - crossings . consider now a lattice @xmath251-manifold ( or complex ) @xmath252 , and a preferred coordinate axis defining a height function @xmath253 . the preimage of the open interval @xmath254 , for @xmath255 is an open pl cylinder because an open @xmath1-cell @xmath256 must be vertical with respect to @xmath257 . this means that @xmath258 , where @xmath2 is a @xmath60-cell whose projection @xmath259 is a single point . a horizontal cell is one whose image under @xmath257 is a point . therefore we derive the sub - complexes @xmath260 , @xmath261 , with @xmath262 , such that @xmath263 and @xmath264 let @xmath265 be the union of interiors of the horizontal cells in @xmath266 . then one can decompose the sets @xmath267 as follows : @xmath268 , @xmath269 , where @xmath270 , and @xmath271 . using a recurrence argument we assume that @xmath272 and @xmath273 are bubble equivalent for some @xmath12 , by means of the sequence of embedded bubble moves @xmath274 . then the same sequences of bubble moves transforms @xmath275 into @xmath276 . since @xmath277 , there exist some cone constructions over the bubble moves @xmath274 , which are realized in the @xmath278-dimensional lattice , and relate @xmath279 $ ] to @xmath280 $ ] , as follows . if the bubble move @xmath274 touches only @xmath281 , then consider the usual cone of @xmath274 , which is also a bubble move in one more dimension . if the bubble move @xmath274 touches @xmath282 ( or @xmath283 , on the other side ) then construct an extension with one more dimension for @xmath274 , by using the horizontal flat . the slices @xmath279 $ ] , and @xmath254 can be glued now back , and we obtain a bubble equivalence between @xmath7 and a @xmath284-dilatation , meaning that the dilatation acts trivially in the direction of the chosen axis . the same procedure works for any other coordinate axis . then the product of all such dilatations is a homothety of factor a power of @xmath285 , hence the claim follows . a mappable cubulation is not embeddable for two reasons : either it is non - standard or else the map to the lattice is not injective . both accidents can be resolved using np - bubble moves . the cubulation @xmath20 is @xmath12-_standard _ if its @xmath12-skeleton is standard , i.e. two cubes of dimensions at most @xmath12 are either disjoint or else they have exactly one common face . assume for the moment that the @xmath1-dimensional cubulation @xmath20 is @xmath60-standard . let @xmath286 be a combinatorial map , locally an isometry on each cube . the singularities of the map @xmath2 are therefore either _ foldings _ or _ double points_. a _ double point singularity _ is when two disjoint cubes @xmath28 and @xmath287 have the same image @xmath288 . the codimension of the cubes is called _ the defect _ of the double points . if the defect is positive then @xmath289 , @xmath290 , where @xmath291 and @xmath292 ( respectively @xmath293 and @xmath294 ) are top dimensional cubes having a common face . a _ folding _ corresponds to a pair of cubes having a common face and the same image under @xmath2 . in order to get rid of double points of positive defect one performs a @xmath18 followed by a @xmath136 move on @xmath291 and @xmath292 respectively , in an additional dimension . if the defect of the double point is zero then we can settle using a cubical whitney trick , as follows . consider @xmath193 be obtained by a @xmath18 move on @xmath287 . perform a @xmath18 move on @xmath295 in an additional dimension . then @xmath296 embeds into @xmath297 , since the interiors of the new cubes do not intersect the cubes of @xmath298 . there exists an extension @xmath299 of @xmath300 sending @xmath301 onto @xmath302 , which is combinatorial and has less singularities than @xmath2 . iterate this procedure until an embeddable cubulation is obtained . in order to solve a folding of two cubes one uses a @xmath18 move over one folding cube , and on its image . the folding is replaced then by a double point singularity . it remains to see how we can use np - bubble moves in order to assume the @xmath60-standardness . using @xmath60-dimensional np - bubble moves @xmath303 one can transform @xmath304 into a standard complex . it suffices to observe that the action of the @xmath1-dimensional np - bubble move @xmath305 in one more dimension agrees with that of @xmath303 on the @xmath60-skeleton . this ends the proof of the theorem 1.5 . [ alb]corollary the simple cubulations of a manifold @xmath7 satisfying @xmath306 are bubble equivalent . we want to find out whether a cubulation which is bubble equivalent to a mappable one is mappable itself . in general the answer is negative , and we have to restrict to np - bubble moves . the set of embeddable cubulation is not stable under np - bubbles either . in fact an embeddable cubulation may become non - standard , after performing some @xmath307 moves . we will show that this is the only accident which can happen . more precisely , we say that @xmath7 and @xmath8 are _ standard np - bubble _ equivalent if there exists a chain of np - bubble moves joining @xmath7 and @xmath8 among standard cubulations . also the simplicity is preserved by all np - bubble moves but @xmath136 , hence the mappability can not be np - bubble invariant . two cubulations are _ simply np - bubble _ equivalent if they are np - bubble equivalent among simple cubulations . let @xmath7 be a mappable cubulation and @xmath308 be the support of a @xmath136 move . the move is _ rigid _ with respect to @xmath7 if there is a combinatorial map @xmath309 for which @xmath310 is the union of two orthogonal @xmath1-cubes . otherwise , either @xmath310 consists of a single cube ( @xmath2 is a _ folding _ ) or @xmath310 is the union of two cubes lying in the same @xmath1-plane ( @xmath2 is _ flexible _ ) . the following result is a coarse converse of the theorem 1.5 : [ alb]proposition [ albi ] 1 . consider @xmath7 be mappable and @xmath8 be a cubulation which is np - bubble equivalent to @xmath7 using only rigid @xmath136 moves . then @xmath8 is mappable . if @xmath7 is embeddable then @xmath311 is mappable . 3 . if @xmath7 is embeddable and @xmath8 is standard np - bubble equivalent to @xmath7 then @xmath8 is embeddable . 4 . the class of mappable cubulations is closed to simple np - bubble equivalences . let @xmath309 be a combinatorial map into the standard cubulation and @xmath312 be the support of a @xmath10 move ( i.e. the union of @xmath12-cubes ) . since @xmath2 is non - degenerate on each cube either @xmath2 is an embedding on @xmath313 or else @xmath2 is a folding and @xmath314 . also @xmath2 is always an embedding on the support @xmath315 of @xmath307 . if the move is rigid ( for @xmath314 ) or @xmath2 is an embedding on the support then there exists a cube @xmath316 , such that @xmath317 . in fact there is an unique embedding of @xmath318 ( @xmath319 ) and respectively @xmath315 into @xmath297 , up to isometry . the map @xmath2 extends then to a map @xmath320 over @xmath321 using the cube @xmath20 . this ends the proof of the first part of the proposition . let us introduce some notations and definitions from @xcite , for the sake of completeness . two edges @xmath322 and @xmath323 of a cubic complex @xmath162 are said to be _ equivalent _ if there exists a sequence of edges joining them , in which any two successive edges @xmath324 are opposite sides of some square in @xmath162 . an edge equivalence class is called _ simple _ if all the edges in it belonging to a single cube in @xmath162 are parallel . an equivalence class is called _ orientable _ if all the edges in it can be oriented such that whenever two equivalent edges are parallel to each other , their orientations are parallel . we consider a partition @xmath325 of the edge equivalence classes into certain families of classes @xmath326 such that each class has a fixed orientation and two perpendicular edges are members of equivalence classes that belong to different families . let @xmath327 be an oriented edge path in the cubulation @xmath162 . let @xmath328 be + 1 if the direction of travel of @xmath329 on the edge @xmath330 coincides with the orientation of @xmath330 , and -1 otherwise . consider then the following formal sum , taking values in the free @xmath331-module generated by the symbols @xmath332 : @xmath333 where @xmath334 is the family to which the edge @xmath322 belongs . we can state now ( see @xcite , p. 305 - 306 ) : [ alb]proposition 1 . a simple cubulation @xmath162 maps into the skeleton of @xmath163 , where @xmath8 is the affine dimension of the image , if and only if there exist orientations of the equivalence classes and a partition @xmath325 of these classes into @xmath8 families such that for any closed edge path @xmath329 in @xmath162 , we have @xmath335 . a simple and standard cubulation @xmath162 embeds into the skeleton of @xmath163 , where @xmath8 is the affine dimension of the image , if and only if there exist orientations of the equivalence classes and a partition @xmath325 of these classes into @xmath8 families such that , for an edge path @xmath329 in @xmath162 , we have @xmath335 if and only if the path @xmath329 is closed . we are ready to prove now the second part : according to proposition 3.8 the cubulation @xmath7 ( which is mappable , hence simple ) admits a partition @xmath325 of the equivalence classes of edges such that the development map @xmath336 vanishes on all closed curves . the orientations of the edge equivalence classes of @xmath7 naturally induce orientations for @xmath311 . further any loop in @xmath311 can be deformed to a loop in @xmath7 hence @xmath311 is mappable from the previous criterion , provided that @xmath311 is simple . the simplicity is not preserved by @xmath136 , in general , but we asked the cubulation be an embeddable . if @xmath311 wo nt be simple , then the support of the @xmath136 move should consist of two _ twin _ cells . this means that we have two cells @xmath322 and @xmath323 which have a common face @xmath2 , and the two layers parallel to @xmath2 containing @xmath322 and @xmath323 coincide . but the layers of an embeddable cubulation can not be self - tangent ( see @xcite ) , hence @xmath311 is simple . the third and fourth statements in proposition [ albi ] follow the same way . remark that the collection of non - trivial homotopy class of immersions @xmath81 ( of the connected components @xmath79 of the derivative complex @xmath6 ) is invariant to np - bubble moves . in particular @xmath15 is infinite if the manifold @xmath7 has non - trivial topology . if we consider two homothetic cubulations @xmath185 and @xmath337 , then in general they can not be np - bubble equivalent , because any non - trivial homotopy class appears @xmath338 times more in the latter . thus the np - bubble equivalence may be interesting only for pl - spheres . we believe that any two simple cubulations of the sphere are np - bubble equivalent , as it happens for @xmath80 . assume the manifolds we consider in the sequel are connected . we wish to prove that the connected sum of cubulations induces a monoid structure on @xmath129 . we believe that @xmath339 inherits also a monoid structure . consider the cubulations @xmath340 and @xmath341 of the manifolds @xmath7 and @xmath8 , and choose two cells @xmath342 , @xmath343 . a cell is always a top dimensional cube in this section . let @xmath166 be a length @xmath344 cubical cylinder made up from @xmath345 $ ] by removing the interiors of @xmath346 and @xmath347 . we define therefore the map `` connected sum of cubulations '' @xmath348 , by @xmath349 with the obvious identifications of the boundaries . we make some remarks before we proceed to prove the theorem 1.6 . we will freely use during the proof the fact that the tube @xmath166 can be changed into another tube while it remains mappable . the gluing of the tube @xmath166 requires some self - identification of the boundary @xmath350 . we fix an arbitrary identification @xmath351 . then the other boundary of the tube can be glued to @xmath352 in @xmath97 ways corresponding to the elements of the symmetry group @xmath353 of the @xmath1-cube . the relative twist @xmath354 measures the difference between two gluings on @xmath323 . notice that there is no there is no canonical choice in gluing @xmath323 . all tubes of length @xmath344 are bubble equivalent rel boundary so that their length has not to be specified . then the connected sum cubulation depends on the choice of @xmath322 , @xmath323 and of the relative twist . one can assume that @xmath20 and @xmath183 are standard cubulations . we want to define a _ developing map _ @xmath355paths in @xmath356cells in @xmath357 . the map @xmath55 depends on the particular data @xmath358 we chose . a _ path _ in @xmath183 is a sequence of cells starting at @xmath323 , consecutive cells having a common face . it suffices to define the map @xmath55 for the trivial path then use a recurrence on the length of the path . if the path is trivial , consisting of the cell @xmath323 , we define @xmath359 . further choose a cell @xmath360 having a common face @xmath293 with @xmath323 . let @xmath291 be the face of @xmath322 which is opposite to @xmath293 using the tube @xmath166 . then the face @xmath2 neighboring @xmath322 and intersecting it along @xmath291 is by definition @xmath361 , where @xmath329 is the path @xmath362 . let @xmath363 be a tube isometric to @xmath166 which is glued on @xmath20 and @xmath183 along @xmath2 and @xmath360 respectively , such that @xmath364 is a cylinder on the union of two neighboring cells . here @xmath365 $ ] is the common face of the tubes . this condition determines uniquely the gluing twist @xmath366 of @xmath363 . let @xmath206 denotes the cubulation obtained by gluing @xmath367 to @xmath20 and @xmath183 , along @xmath368 and @xmath369 respectively . then @xmath206 is obtained from @xmath370 by a sequence of @xmath136 moves ( along the tube @xmath363 ) and a final @xmath371 move . the same procedure transforms @xmath372 into @xmath206 , hence @xmath370 and @xmath372 are equivalent . using a recurrence one proves that @xmath370 and @xmath373 are equivalent for any path @xmath329 , where @xmath374 is a tube isometric to @xmath166 whose gluing twist is that induced by @xmath329 . set @xmath375 . we have to prove that , possibly using bubble moves on the initial cubulations , we have @xmath376 . for any loop @xmath329 based at @xmath322 one knows that @xmath377 . consider a loop @xmath329 of length @xmath172 and assume that the last segment of the curve @xmath378}$ ] , for @xmath379 is _ straight_. a curve is _ straight _ if it consists of a sequence of cells @xmath330 , each cell @xmath330 having a common face @xmath91 with the preceding cell @xmath380 , so that the faces @xmath91 and @xmath381 are opposite faces in @xmath330 . then a straight curve @xmath378}$ ] defines a strip @xmath382 consisting of the cells of the maximal straight extension of @xmath378}$ ] . set @xmath383 . the image of the strip @xmath382 under the developing map is also a strip @xmath384 . when a basepoint cell @xmath322 , and a preferred direction are fixed a discrete flow is defined on the strip @xmath382 . the action of @xmath385 on the cell @xmath292 is the cell @xmath386 which is @xmath12 steps forward in the given direction , starting from @xmath292 . notice that there is also a flow defined on the other strip @xmath387 . we claim that @xmath388 ( after using bubble moves ) . let @xmath389 , @xmath390 $ ] , be a cell of @xmath391 , located between @xmath322 and @xmath392 . one performs a @xmath18-move on @xmath291 and set @xmath393 for the natural extension of @xmath329 to @xmath394 . we can express @xmath395 in terms of the flow on the strip @xmath387 as @xmath396 because @xmath393 is also straight and its length was increased by @xmath285 . thus @xmath397 holds , but the strip is finite hence the @xmath331-action has cyclic orbits , implying that @xmath398 . a _ bicoloring _ of @xmath20 associates a color @xmath399 to each cell @xmath322 such that adjacent cells have different colors . for any two cells @xmath2 and @xmath360 there exists a system of strips allowing to pass from @xmath2 to @xmath360 using only the action of @xmath400 on these strips , except for the case when @xmath20 admits bicolorings and @xmath401 . however any bicolored cubulation can be transformed into one having no bicolorings using one @xmath18-move . this shows that @xmath376 . recall that given a path @xmath329 between @xmath322 and @xmath2 one associates a twist for the corresponding tube over @xmath2 . set @xmath402 for the twist over @xmath2 associated to a straight path @xmath403 . let us perform a @xmath18 move over @xmath2 . the curve @xmath329 has several lifts @xmath404 ( not necessarily straight ) relating @xmath322 and @xmath405 through the cells of @xmath406 . it is simple to check that the set of twists over @xmath2 induced by these paths is all of @xmath407 . therefore the connected sum does not depend on the choice of the twist . this proves the theorem 1.6 . the composition law for immersions is the disjoint union of immersions inside the connected sum of manifolds , where the latter is made away from the immersions . the immersion associated to the connected sum of cubulations is obtained from the initial immersions by some surgery which involves only local data . we will prove that this surgery can be also realized by a local relative cobordism of the associated immersions . consider the manifolds @xmath7 and @xmath8 with their respective cubulations @xmath185 and @xmath206 . the connected sum cubulation @xmath408 is obtained by removing the cells @xmath322 from @xmath185 and @xmath2 from @xmath206 . one uses @xmath18 moves on @xmath322 and @xmath2 in order to reduce the cubulated piping tube to a boundary identification @xmath409 . the bordism classes of @xmath410 and @xmath411 are preserved under these transformations . we consider that the connected sum of manifolds is done by means of a piping tube which is close to the images @xmath412 and @xmath413 . both @xmath412 and @xmath413 have as local models the set of coordinate hyperplanes around the origin in @xmath0 . the surgery which changes @xmath414 into @xmath415 excises @xmath412 and @xmath413 and replaces them by a cylinder with the same boundary . one can realize the surgery on the piping tube , after pushing the local models thru it . a small isotopy moves them outside of a longitude and then the configurations embed in the ball obtained from the tube by cutting it along the longitude . it suffices therefore to show that the surgery can be realized by a cobordism in the ball which is a product on the boundary . for @xmath13 the different slices of the cobordism are given in the picture 4 . assume that the local models living in a ball have the corresponding hyperplane sections parallel to each other . let us consider two parallel hyperplane sections @xmath291 and @xmath293 of the respective local models @xmath412 and @xmath413 . then one constructs the neighborhood of @xmath416 around the piping tube by removing the interiors of @xmath291 and @xmath293 and gluing back the cylinder of boundary @xmath417 . this transformation can be realized also by a cobordism of immersions since it is represented by the local picture around a critical point of index 1 , where the images of the immersions are the non - critical levels before and respectively after passing through the critical value . the local picture can be made transversal with respect to the other coordinate hyperplanes of the immersions which were left untouched . one composes the cobordisms associated to all @xmath1 pairs of parallel hyperplane sections . the restrictions on the boundary of these cobordism are products and so we can glue the composition with the trivial cobordism outside the local pictures . thus we derived a cobordism between @xmath416 and @xmath418 . notice that , in general , the map @xmath419 should depend on the length of the tube @xmath166 . the case when @xmath7 and @xmath8 are spheres could be different however . observe first that the set @xmath420 is infinite . in fact the number of components with odd self - intersections of the associated immersion is an invariant , because the only move creating new components is @xmath18 , and each new created component is an embedded circle . set @xmath421 for the sum over the various connected components @xmath79 of the number of self - intersections . the similarity with the reidemeister moves in the plane suggests that @xmath420 is the set of framed circles in the plane . unlike the case of reidemeister moves , the image of the immersion remains connected , so its components can not be separated using bubble moves . on the other hand the move @xmath136 can create / annihilate a pair of self - intersections . it is not clear that the singularities can always be paired such that suitable np - bubble moves destroy all pairs of singularities and so each transformed circle has @xmath422 singularities . if this is true it will remain to prove that all configurations of circles among which there are exactly @xmath172 singular circles are np - bubble equivalent . this would establish an isomorphism between @xmath420 and @xmath423 . we are able to prove this statement for the case @xmath424 ( and @xmath425 ) : [ section ] call a disk bounded by two disjoint arcs a _ biangle_. a biangle is _ tight _ if no other arc intersects its interior . observe that using np - bubble moves one can transform all minimal biangles into tight biangles . we use isotopies of the boundary of the biangle ( @xmath136-moves ) reducing the number of squares contained in the disk and @xmath371 moves . the possible configurations of tight biangles can be rather complicated , even if the cubulation is simple . let us show that the tight triangles can travel along the edges ( sliding ) . this means that a regular homotopy moving a tight biangle along the arcs can be realized by np - bubble moves . the proof of this claim is contained in the picture below . we marked in a little rectangle the area on which the bubble move acts : set @xmath185 for the union of two transversal arcs . an arc which intersects three times one branch of @xmath185 and then the other arc can be simplified by a composition of bubble moves which we call a s move : now it is easy to obtain that the set of semi - simple cubulations mod np - bubble moves is equal to the set of simple cubulations mod np - bubble moves . in fact any tight biangle coming from the same component can be transformed into tight biangles on different components : assume that the cubulation is simple . the union of the circles @xmath79 for @xmath426 , divides the sphere into polygonal faces , each of them having at least two vertices . we claim that , either there are no biangles involving an arc from @xmath427 or else @xmath427 is a small circle contained in the union of two faces which intersects minimally ( i.e. twice ) the common edge . indeed , suppose that there exists a minimal tight biangle . consider the face which is adjacent to the biangle and not containing it . if @xmath427 does not satisfy the claim then we can simplify the biangle using slidings and s - moves , as it is shown below : thus @xmath427 can be transformed into a circle which does not support any biangle . further use of the moves @xmath136 and @xmath371 allows us to isotope @xmath427 into the union of two faces as claimed . one continues the simplification procedure with the other components @xmath79 and the proposition 5.1 follows . a similar proof works when the circle @xmath427 is allowed to have one self - crossing . now @xmath427 is transformed into a a figure eight contained in the union of two faces which intersects minimally their common edge and forms one biangle . in order to prove that @xmath23 one has to get rid of those self - intersections which are not cancelling pairs the following figures describe the simplifications obtained with the additional move @xmath428 for two adjacent self - intersections : further , if the two self - intersections are separated by additional arcs then use @xmath136-moves and slide across these arcs . it follows that any dual graph can be transformed using bubble moves into one satisfying @xmath429 , for each component @xmath79 . the next step is to show that two components `` can be added '' . let @xmath427 and @xmath430 be two components having non - void intersection . there exists an equivalent configuration in which @xmath427 and @xmath430 are replaced by the circle @xmath431 verifying @xmath432 , the other @xmath79 s ( @xmath433 ) are left unchanged and the additional components have @xmath434 . it suffices to do that in the case when @xmath435 . then the kinks can be added and transformed in consecutive self - intersections which have already been solved . thus , if one adds those components which are not embedded , we obtain a planar graph whose components are embedded , except possibly for one which has @xmath436 . we have therefore @xmath437 . if @xmath438 then the result of the previous section shows that all these configurations are np - bubble equivalent . if @xmath439 it means that @xmath20 is a figure eight in the plane . the remark after the proof of proposition 5.1 ends the proof of the theorem 1.3 . dolbilin , m.a . shtanko and m.i.shtogrin . a topology for the family of parameterizations of cycles modulo 2 in a three dimensional cubic lattice . , 295(1987 ) , 19 - 23 , translated in _ soviet math.dokl._ , 36(1988 ) . dolbilin , m.a . shtanko and m.i.shtogrin . the problem of parameterizations of cycles modulo 2 in a three dimensional cubic lattice . , 52(1988 ) , 355 - 377 , translated in _ russian acad.sci.izvestya math . _ , 32(1989 ) .

we establish a correspondence between the su(2 ) witten reshetikhin turaev invariant for the seifert manifold and ramanujan s mock theta functions .

math-ph0506073

according to jacobi , theory of elliptic functions was given birth in december 23 , 1751 , when euler was asked to referee fagnano s paper including addition formula for the arc length of ellipse and the lemniscate . since then it has been established and developed by jacobi , abel , gauss and others , and it now becomes fundamental and considerable subjects in both mathematics and physics . one of mysterious topics of the theta functions is ramanujan s mock theta function ( see _ e.g. _ refs . ) . it is not modular , but it has a nice asymptotic behavior when @xmath1 is root of unity . watson gave a proof for the third order mock theta functions @xcite , though the systematic understanding and even the meaning of `` order '' are still missing . in his thesis @xcite , zwegers investigated the mock theta functions by use of a real analytic modular form with half - integral weight ( see also ref . ) . he showed that the mock theta functions can be written as a sum of the _ indefinite _ theta functions and the eichler integral of the half - integral weight modular forms . he then obtained the transformation formula for the mock theta functions , which indicates that they have a _ nearly _ modular property . nearly modular property has also revealed in recent studies of the quantum invariant @xcite . therein the su(2 ) witten reshetikhin turaev ( wrt ) invariant @xmath2 @xcite for the poincar homology sphere @xmath3 was identified with a limiting value of the eichler integral of the modular form with half - integral weight in @xmath4 , and an exact asymptotic expansion in @xmath5 was given by use of the modular transformation . this method is further applied to the colored jones polynomial for torus knot @xcite and torus link @xcite , and to the wrt invariant for the seifert manifold @xcite , and topological invariants such as the chern simons invariant , the casson invariants , the reidemeister ray singer torsion , and the ohtsuki invariant are interpreted from the view point of the modular form . purpose of this article is to show that the mock theta functions can be regarded as the wrt invariant @xmath2 for the seifert manifold @xmath6 ( see , _ e.g. _ , ref . for the seifert manifolds ) . precisely the wrt invariant can be written in terms of the mock theta functions as a limiting value @xmath4 from lower half of the complex plane . this coincidence was already pointed out in ref . ( see also ref . ) for the case of the poincar homology sphere which is connected to the fifth order mock theta function , and here we find that such relationship does exist for every order of ramanujan s mock theta functions . an outline of the present article is as follows . in section [ sec : modular ] we introduce the vector - valued modular form with weight @xmath7 . we define the eichler integral thereof , and review a property of the eichler integral . in sections [ sec:5th ] [ sec:10th ] , we discuss separately ramanujan s mock theta functions of order 3 , 5 , 6 , 7 , 10 . therein we consider the mock theta functions when @xmath1 is outside the unit circle , and prove that they are the eichler integral of the vector - valued modular form with weight @xmath7 , and that a limiting value in @xmath8 gives the wrt invariant for the seifert manifold which was studied in ref . . the last section is devoted to discussions . hereafter we set @xmath9 with @xmath10 , and we use a standard notation of @xmath1-calculus ; @xmath11 ( a , b , \cdots ; q)_n = ( a ; q)_n \ , ( b ; q)_n \cdots , \\[2 mm ] \begin{bmatrix } n \\ m \end{bmatrix}_q = \frac{(q)_n}{(q)_{n - m } \ , ( q)_m } .\end{gathered}\ ] ] for our convention , we list some identities ( see , _ e.g. _ , ref . ) ; * @xmath1-binomial theorem , @xmath12 * @xmath1-binomial series , @xmath13 * @xmath1-binomial formula , @xmath14 * the euler identity , which follows from in @xmath5 , @xmath15 * the jacobi triple product identity , @xmath16 we define an odd periodic function @xmath17 with modulus @xmath18 @xmath19 0 , & \text{otherwise , } \end{cases}\ ] ] for @xmath20 . when we introduce the function @xmath21 this becomes a vector - valued modular form with weight @xmath7 satisfying @xmath22 \label{psi_under_t } \psi_p^{(a)}(\tau+1 ) = { \mathrm{e}}^{\frac{a^2}{2 p } \pi { \mathrm{i } } } \ , \psi_p^{(a)}(\tau ) , \end{gathered}\ ] ] where @xmath23 is a @xmath24 matrix @xmath25 the eichler integral of this set of the modular forms with half - integral weight is then defined as a half - integration of @xmath26 with respect to @xmath27 by @xcite @xmath28 the eichler integral of the _ integral_-weight modular form is known to have a nearly modular property ( see , _ e.g. _ , ref . ) . to see this nearly modular property in our case , we introduce another eichler integral following ref . ; @xmath29 which is defined for @xmath30 in the lower half plane , @xmath31 . we see that the modular property of @xmath32 , especially the modular @xmath33-transformation , leads a nearly modular property of the eichler integral as @xcite ( see also refs . ) @xmath34 then taking a limit @xmath35 for @xmath36 in the eichler integrals @xmath37 and @xmath38 , we find both eichler integrals coincide in this limit , and we obtain an asymptotic expansion in @xmath5 as @xmath39 where @xmath40 denotes the dirichlet @xmath41-function associated with @xmath42 defined in . we note that for @xmath36 and @xmath43 we have @xmath44 \widetilde{\psi}_p^{(a)}(n ) = \left ( 1- \frac{a}{p } \right ) \ , { \mathrm{e}}^{\frac{a^2}{2 p } \pi { \mathrm{i}}n } , \end{gathered}\ ] ] where the bernoulli polynomial is @xmath45 . we note that the eichler integral @xmath46 is the spefic value of the @xmath47-colored jones polynomial for torus link @xmath48 @xcite . the nearly modular property resembles with the transformation properties of the mock theta functions @xcite . in fact the right hand side of can be rewritten in terms of the mordell integrals @xcite , and it suggests a connection between the eichler integrals of the weight @xmath7 modular form and the mock theta functions . we start from the fifth order mock theta functions @xcite . ramanujan defined 10 functions , and here we treat 2 functions among them defined by @xmath49 \chi_1(q ) = \sum_{n=0}^\infty \frac{q^n}{(q^{n+1})_{n+1 } } , \label{define_chi1}\end{gathered}\ ] ] other 8 functions can be given by use of these functions and theta functions @xcite . as was noticed in ref . , definitions and can be extended to outside the unit circle @xmath50 , and we define new functions by substituting @xmath51 in place of @xmath1 ; @xmath52 \label{chi1_inverse } \chi_1^*(q ) = q^{-1 } \ , \chi_1(1/q ) = \sum_{n=0}^\infty ( -1)^n \frac{q^{\frac{3}{2 } n ( n+1 ) } } { ( q^{n+1})_{n+1 } } .\end{gathered}\ ] ] @xmath53 \label{chi1_eichler } \chi_1^*(q ) = \sum_{n=0}^\infty \chi_{60}^{(1,1,2)}(n ) \ , q^{\frac{n^2 - 49}{120 } } , \end{gathered}\ ] ] where @xmath54 is an odd periodic function with modulus @xmath55 defined by @xmath56 \chi_{60}^{(1,1,2)}(n ) = \psi_{60}^{(7)}(n ) + \psi_{60}^{(13)}(n ) + \psi_{60}^{(17)}(n ) + \psi_{60}^{(23)}(n ) . \end{gathered}\ ] ] these can be proved by applying the baily chain @xcite ( see also ref . ) ; if for @xmath57 the bailey pair @xmath58 satisfies @xmath59 we have @xmath60 which reduces to @xmath61 by setting @xmath62 , @xmath63 and @xmath5 . eqs . and follow immediately when we use the bailey pairs , a(5 ) and a(8 ) in slater s list @xcite . see also ref . , where these functions are called the _ false _ theta function . according to ref . , these _ false _ theta functions should rather be identified as the eichler integral . namely when we set @xmath64 q^{\frac{49}{120 } } \ , \chi_1^*(q ) \end{pmatrix } , \ ] ] which has a form like due to , it can be regarded as the eichler integral of the vector - valued modular form with weight @xmath7 ; @xmath65 \chi_{60}^{(1,1,2)}(n ) \end{pmatrix } \ , q^{\frac{n^2}{120 } } .\ ] ] the modular @xmath66- and @xmath67-matrices under @xmath68 and @xmath69 are respectively given by @xmath70 \sin \left ( \frac{2 \ , \pi}{5 } \right ) & - \sin \left ( \frac{\pi}{5 } \right ) \end{pmatrix } , & \mathbf{t } & = \begin{pmatrix } { \mathrm{e}}^ { \frac{1}{60 } \pi { \mathrm{i } } } & \\[2 mm ] & { \mathrm{e}}^ { \frac{49}{60 } \pi { \mathrm{i } } } \end{pmatrix } . \end{aligned}\ ] ] this shows that the modular form @xmath71^{-1/5 } \cdot \boldsymbol{\phi}_{2,3,5}(\tau)$ ] with rational weight @xmath72 is on the principal congruence subgroup @xmath73 . intriguing is that the eichler integral @xmath74 gives the wrt invariant for the poincar homology sphere in a limit of @xmath1 being the @xmath47-th root of unity as was proved by lawrence and zagier @xcite . @xmath75 here the wrt invariant @xmath2 for 3-manifold @xmath76 is normalized to be @xmath77 \tau_n(s^2 \times s^1 ) = \sqrt { \frac{n}{2 } } \ , \frac{1 } { \sin\left ( \pi /n \right ) } .\end{gathered}\ ] ] we should remark that we have the icosahedral symmetry of the group @xmath73 and the fundamental group of the poincar sphere . as a consequence of , we see that the eichler integral in @xmath4 has a nearly modular property replacing @xmath23 and @xmath17 with @xmath78 and @xmath54 respectively , and @xmath79 . we can obtain an explicit form of the quantum invariant as a linear combination of by taking a limiting value of the eichler integral @xmath74 in @xmath8 @xcite . to derive this invariant in terms of @xmath1-series , it is useful to rewrite the eichler integral in the form such that the infinite sum terminates at the finite sum in the case of @xmath1 being root of unity . @xmath80 \chi_1^*(q ) = \sum_{n=0}^\infty q^n \ , ( q^{n+1})_n . \end{gathered}\ ] ] from we compute as follows ; @xmath81 \\ & = 1 + q \sum_{m \geq n\geq 0}^\infty q^{2 m } \ , \begin{bmatrix } m \\ n \end{bmatrix}_q \ , ( -q^{m+1})^n \ , q^{\frac{1}{2 } n(n-1 ) } \\ & = 1 + q \sum_{m=0}^\infty q^{2 m } \ , ( q^{m+1})_m \tag{by \eqref{gauss_polynomial } } \displaybreak[0 ] \\ & = 1 + q \ , ( q)_\infty \sum_{m=0}^\infty \frac{q^{2m}}{(q)_m } \ , \frac{1}{(q^{2m+1})_\infty } \\ & = 1 + q \ , ( q)_\infty \sum_{n , m=0}^\infty \frac{q^n}{(q)_n } \ , \frac{q^{2 nm + 2 m}}{(q)_m } \tag{by \eqref{q - binomial } } \displaybreak[0 ] \\ & = 1 + q \sum_{n=0}^\infty q^n \ , ( q^{n+1})_{n+1 } . \tag{by \eqref{q - binomial } } \end{aligned}\ ] ] identity can be proved in the same manner . we see that , by definition of the @xmath1-product , the infinite sum in the expression terminates at the finite order when @xmath1 is root of unity . furthermore the expression exactly coincides with the form given by le in refs . , where the wrt invariant was computed by use of habiro s cyclotomic expansion of the @xmath47-colored jones polynomial for trefoil @xcite @xmath82 where we have used the normalized colored jones polynomial s.t . it is known to be rewritten as @xcite @xmath84 from which , simply applying @xmath85-surgery on this expression following ref . ( see also ref . ) , we obtain another expression @xmath86 this expression also reduces to finite sum when @xmath1 is root of unity . we do not have a direct proof of this @xmath1-series identity at present . to close this section we comment on the @xmath1-hypergeometric type generating function of the @xmath41-function at negative values . studies on such generating functions have been developed since ref . . see refs . . based on , we have the following as a power series in @xmath87 ; @xmath88 \begin{aligned } { \mathrm{e}}^{-49 t/120 } \sum_{n=0}^\infty { \mathrm{e}}^{-n t } \ , ( 1 - { \mathrm{e}}^{-(n+1 ) t } ) \ , ( 1 - { \mathrm{e}}^{-(n+2 ) t } ) & \cdots ( 1 - { \mathrm{e}}^{-2 n t } ) \\ & = \frac{1}{2 } \sum_{k=0}^\infty \frac { l\left(-2 \ , k , \chi_{60}^{(1,1,2)}\right ) } { k ! } \ , \left ( \frac { -t}{120 } \right)^k , \end{aligned } \end{gathered}\ ] ] where @xmath89 2 \,\frac{\cos(5 \ , x ) \ , \cos(3 \ , x ) } { \cos(15 \ , x ) } = \sum_{k=0}^\infty \frac{l \left ( -2 \ , k , \chi_{60}^{(1,1,2 ) } \right ) } { ( 2 \ , k ) ! } \ , ( -1)^k \ , x^{2 k } . \end{gathered}\ ] ] we next study the third order mock theta functions . among 4 ramanujan s mock theta functions of the third order , we study functions defined by @xmath90 \nu(q ) = \sum_{n=0}^\infty \frac{q^{n(n+1 ) } } { \left ( -q ; q^2 \right)_{n+1 } } .\end{gathered}\ ] ] watson gave 3 more functions in ref . , which were in `` lost '' notebook @xcite . other third order mock theta functions can be given from these functions and the theta functions as was proved in ref . . both defining @xmath1-series @xmath91 and @xmath92 converge not only inside , but also outside the unit circle as was noticed in ref . , and we define new functions replacing @xmath1 by @xmath51 as @xmath93 \nu^*(q ) = q^{-1 } \ , \nu(1/q ) = \sum_{n=0}^\infty \frac { q^{n } } { ( -q ; q^2)_{n+1 } } .\end{gathered}\ ] ] as in the case of the 5th order mock theta functions , these are the _ false _ theta functions in a sense of rogers as was proved in ref . . @xmath94 \label{nu_eichler } \nu^*(q ) = \sum_{n=0}^\infty \chi_{24}^{(2)}(n ) \ , q^{\frac{1}{24 } ( n^2 - 16 ) } , \\[2 mm ] \label{phi_eichler_minus } \phi^*(-q ) = \sum_{n=0}^\infty \psi_{6}^{(1)}(n ) \ , q^{\frac{1}{24 } ( n^2 - 1 ) } , \end{gathered}\ ] ] where @xmath95 is an odd periodic function with modulus @xmath96 defined by @xmath97 \chi_{24}^{(2)}(n ) & = \psi_{24}^{(4)}(n ) + \psi_{24}^{(8)}(n ) , \end{aligned}\ ] ] and @xmath17 is defined in . such _ false _ theta functions can be defined based on some of other mock theta functions of the third order . we recall the definitions @xcite ; @xmath98 ^ 2 } , \\[2 mm ] \label{define_omega } \omega(q ) = \sum_{n=0}^\infty \frac { q^{2 n ( n+1 ) } } { \left [ ( q ; q^2)_{n+1 } \right]^2 } , \\[2 mm ] \upchi(q ) = \sum_{n=0}^\infty \frac { q^{n^2 } } { ( 1 -q + q^2 ) \ , ( 1-q^2+q^4 ) \cdots ( 1-q^n + q^{2n } ) } , \\[2 mm ] \varrho(q ) = \sum_{n=0}^\infty \frac { q^{2n(n+1 ) } } { ( 1+q+q^2 ) ( 1+q^3+q^6 ) \cdots ( 1+q^{2n+1 } + q^{4n+2 } ) } .\end{gathered}\ ] ] as was proved in ref . , we have @xmath99 \omega(q ) = \sum_{n=0}^\infty \frac{q^n}{(q;q^2)_{n+1 } } , \\[2 mm ] \upchi(q ) = 1 - { \mathrm{e}}^{2 \pi { \mathrm{i}}/3 } \sum_{n=1}^\infty \frac{{\mathrm{e}}^{\frac{\pi { \mathrm{i}}}{3 } n } \ , q^n } { ( - { \mathrm{e}}^{2 \pi { \mathrm{i}}/3 } \ , q)_n } , \\[2 mm ] \varrho(q ) = \sum_{n=0}^\infty \frac { { \mathrm{e}}^{- \frac{2}{3 } \pi { \mathrm{i}}n } \ , q^n } { ( { \mathrm{e}}^{2 \pi { \mathrm{i}}/3 } \ , q ; q^2)_{n+1 } } .\end{gathered}\ ] ] we can extend these defining @xmath1-series into @xmath50 , and define new functions ; @xmath100 \omega^*(q ) = - q^{-1 } \ , \omega(1/q ) = \sum_{n=0}^\infty ( -1)^n \ , \frac{q^{n(n+1 ) } } { ( q ; q^2)_{n+1 } } , \\[2 mm ] \upchi^*(q ) = \upchi(1/q ) = 1 - { \mathrm{e}}^{2 \pi { \mathrm{i}}/3 } \sum_{n=1}^\infty \frac { { \mathrm{e}}^{- \frac{\pi { \mathrm{i}}}{3 } n } \ , q^{n(n-1)/2 } } { ( { \mathrm{e}}^{\frac{\pi { \mathrm{i}}}{3 } } \ , q)_n } , \\[2 mm ] \varrho^*(q ) = \varrho(1/q ) = \sum_{n=0}^\infty \frac { { \mathrm{e}}^{-\frac{\pi { \mathrm{i}}}{3 } n } \ , q^{n(n+1 ) } } { ( { \mathrm{e}}^{-2 \pi { \mathrm{i}}/3 } \ , q ; q^2)_{n+1 } } .\end{gathered}\ ] ] by applying the bailey chain method with the pairs such as c(7 ) and c(5 ) in slater s list @xcite , we obtain @xcite the following ; @xmath101 \label{omega_star_eichler } \omega^*(q ) = \sum_{n=0}^\infty \left ( \psi_{6}^{(1)}(n ) + \psi_6^{(2)}(n ) \right ) \ , q^{\frac{1}{3}(n^2 - 1 ) } , \\[2 mm ] \upchi^*(q ) = \sum_{n=0}^\infty \psi_6^{(1)}(n ) \ , q^{\frac{1}{24 } ( n^2 - 1 ) } \ , \left ( 1 + { \mathrm{e}}^{- \frac{2}{3 } \pi { \mathrm{i}}n } \right ) , \\[2 mm ] \varrho^*(q ) = \sum_{n=0}^\infty \left ( \psi_6^{(1)}(n ) + \psi_6^{(2)}(n ) \right ) \ , q^{\frac{1}{3 } ( n^2 - 1 ) } \ , { \mathrm{e}}^{\frac{2}{3 } \pi { \mathrm{i}}(1-n ) } . \end{gathered}\ ] ] we remark that , comparing with , we find @xmath102 to study the transformation property of these functions using , it is better to regard these functions as the eichler integrals . we define @xmath103 q^{\frac{1}{3 } } \ , \nu^*(q^{1/2 } ) \\[2 mm ] q^{\frac{1}{48 } } \ , \phi^*(-q^{1/2 } ) \end{pmatrix } .\ ] ] due to , this is the eichler integral of the vector - valued modular form with weight @xmath7 ; @xmath104 \chi_{24}^{(2 ) } ( n ) \\[2 mm ] \psi_{6}^{(1 ) } ( n ) \end{pmatrix } \ , q^{\frac{n^2}{48 } } .\ ] ] the modular @xmath66- and @xmath67-matrices under @xmath105 and @xmath69 are respectively computed by use of and as @xmath106 & 0 & 1 \\[2 mm ] & 1 & 0 \end{pmatrix } , & \mathbf{t } & = \begin{pmatrix } & & { \mathrm{e}}^{\frac{1}{24 } \pi { \mathrm{i } } } \\[2 mm ] & { \mathrm{e}}^{\frac{2}{3 } \pi { \mathrm{i } } } & \\[2 mm ] { \mathrm{e}}^{\frac{1}{24 } \pi { \mathrm{i } } } & & \end{pmatrix } , \end{aligned}\ ] ] which shows that the modular form @xmath107 ^ 2 $ ] is on @xmath108 , and that it has the octahedral symmetry . as a consequence the modular form @xmath109 can be rewritten in terms of the jacobi theta functions @xcite . we then obtain an explicit form of the eichler integral in a limit @xmath8 in terms of the bernoulli polynomial as , and in this case it gives the wrt invariant for the seifert manifold @xmath110 @xcite , whose fundamental group represents the octahedral group ; @xmath111 by definition of @xmath112 , we can conclude that the wrt invariant is a limiting value from outside the unit circle of the ramanujan mock theta function of the third order . concerning the function @xmath113 , we see from that the eichler integral with @xmath114 is written as @xmath115 \omega^ * ( q^{1/4 } ) - \omega^*(- q^{1/4 } ) \end{pmatrix } , \ ] ] and a result of ref . indicates the following ; @xmath116 to rewrite the quantum invariant in terms of the @xmath1-product , it makes sense to give the mock ( false ) theta functions @xmath112 and @xmath117 in the form which terminates at the finite order in the case of @xmath1 being root of unity . @xmath118 \label{nu_star_finite } \nu^*(q ) = \omega^*(q^2 ) = \sum_{n=0}^\infty ( q^2 ; q^4)_n \ , q^{2n } . \end{gathered}\ ] ] when we define the @xmath1-hypergeometric function by @xmath119 we have ( see ( 20.71 ) and ( 20.72 ) in ref . ) @xmath120 ( 1-a ) \ , f \left ( \begin{matrix } -1 \\ 0 \end{matrix } \ , ; q , a \right ) = f \left ( \begin{matrix } 0 \\ a \end{matrix } \ , ; q^2 , a \,q \right ) . \end{gathered}\ ] ] applying these identities , we obtain the statement . alternative computation of @xmath1-series follows from topological fact that the seifert manifold @xmath110 is constructed from @xmath121-surgery on the right - handed trefoil @xcite . applying a surgery formula @xcite to the colored jones polynomial for the trefoil , we get @xmath122 to close this section , we note that and may give formulae as follows as a power series in @xmath87 ; @xmath123 { \mathrm{e}}^{-t/3 } \sum_{n=0}^\infty { \mathrm{e}}^{-n t } \ , ( 1 - { \mathrm{e}}^{-t } ) \ , ( 1 - { \mathrm{e}}^{-3 t } ) \cdots ( 1 - { \mathrm{e}}^{-(2n-1)t } ) = \sum_{k=0}^\infty \frac { l \left ( -2 \ , k , \chi_{24}^{(2 ) } \right ) } { k ! } \ , \left ( - \frac{t}{48 } \right)^k , \end{gathered}\ ] ] where @xmath124 \frac { \cos(x ) } { \cos(3 \ , x ) } = \sum_{k=0}^\infty \frac { l \left ( -2 \ , k , \chi_{24}^{(2 ) } \right ) } { ( 2 \ , k ) ! } \ , ( -1)^k \ , \left ( \frac{x}{2 } \right)^ { 2 k } . \end{gathered}\ ] ] we continue to study the seventh order mock theta functions @xcite . there are 3 ramanujan s mock theta functions , and they are read as @xmath125 \mathcal{f}_1(q ) & = \sum_{n=1}^\infty \frac{q^{n^2}}{(q^{n})_n } , \\[2 mm ] \mathcal{f}_2(q ) & = \sum_{n=0}^\infty \frac{q^{n(n+1)}}{(q^{n+1})_{n+1 } } .\end{aligned}\ ] ] these @xmath1-series have also meaning even when @xmath50 , and for our convention we define @xmath126 \mathcal{f}_1^*(q ) & = \mathcal{f}_1(1/q ) = \sum_{n=1}^\infty ( -1)^n \ , \frac{q^{\frac{1}{2 } n(n-1 ) } } { ( q^{n})_n } , \\[2 mm ] \mathcal{f}_2^*(q ) & = q^{-1 } \ , \mathcal{f}_2(1/q ) = -\sum_{n=0}^\infty ( -1)^n \ , \frac { q^{\frac{1}{2 } n ( n+3 ) } } { ( q^{n+1})_{n+1 } } .\end{aligned}\ ] ] by applying the method of the bailey chain with the bailey pair a(3 ) , a(2 ) , and a(4 ) in slater s list @xcite , we can prove that these are the _ false _ theta function in the sense of rogers @xcite . @xmath127 \mathcal{f}_1^*(q ) = \sum_{n=0}^\infty \chi_{84}^{(1,1,2)}(n ) , q^{\frac{n^2 - 25}{168 } } , \\[2 mm ] \mathcal{f}_2^*(q ) = \sum_{n=0}^\infty \chi_{84}^{(1,1,3)}(n ) \ , q^{\frac{n^2 - 121}{168 } } , \end{gathered}\ ] ] where @xmath128 is an odd periodic function with modulus @xmath129 defined by @xmath130 \chi_{84}^{(1,1,2)}(n ) & = - \psi_{84}^{(5)}(n ) - \psi_{84}^{(19)}(n ) - \psi_{84}^{(23)}(n ) - \psi_{84}^{(37)}(n ) , \\[2 mm ] \chi_{84}^{(1,1,3)}(n ) & = - \psi_{84}^{(11)}(n ) - \psi_{84}^{(17)}(n ) - \psi_{84}^{(25)}(n ) - \psi_{84}^{(31)}(n ) . \end{aligned}\ ] ] this proposition enables us to define @xcite @xmath131 q^{25/168 } \ , \mathcal{f}_1^*(q ) \\[2 mm ] q^{121/168 } \ , \mathcal{f}_2^*(q ) \end{pmatrix } , \ ] ] which can be regarded as the eichler integral of the vector - valued modular form with weight @xmath7 ; @xmath132 \chi_{84}^{(1,1,2)}(n ) \\[2 mm ] \chi_{84}^{(1,1,3)}(n ) \end{pmatrix } \ , q^{\frac{n^2}{168 } } .\ ] ] the modular @xmath66- and @xmath67-matrices under @xmath68 and @xmath133 are respectively given by @xmath134 \sin ( \frac{2 \,\pi}{7 } ) & - \sin ( \frac{3 \ , \pi}{7 } ) & \sin(\frac { \pi}{7 } ) \\[2 mm ] \sin ( \frac{3 \ , \pi}{7 } ) & \sin ( \frac{\pi}{7 } ) & -\sin(\frac{2 \ , \pi}{7 } ) \end{pmatrix } , \\[2 mm ] \mathbf{t } = \begin{pmatrix } { \mathrm{e}}^ { \frac{1}{84 } \pi { \mathrm{i } } } & & \\ & { \mathrm{e}}^ { \frac{25}{84 } \pi { \mathrm{i } } } \\ & & { \mathrm{e}}^ { - \frac{47}{84 } \pi { \mathrm{i } } } \end{pmatrix } . \end{gathered}\ ] ] result in ref . thus proves that the mock ( false ) theta function @xmath135 gives the wrt invariant for the brieskorn homology sphere @xmath136 in a limit that @xmath27 goes to the @xmath47-th root of unity ; @xmath137 an explicit form of the right hand side follows from . correspondingly an exact asymptotic expansion of the wrt invariant in @xmath5 can be computed from the nearly modular property @xcite . by the correspondence with the quantum invariant which is originally defined for @xmath1 being root of unity , it is tempting to get the @xmath1-series expression which terminates at the finite order when @xmath1 is root of unity . we obtain the following from ; @xmath138 another construction of @xmath1-series follows from topological understanding that the brieskorn homology sphere @xmath136 is constructed by @xmath139-surgery on the right - handed trefoil @xcite . applying a surgery formula to the colored jones polynomial for the trefoil we obtain @xmath140 a simpler expression was given in ref . where used was the cyclotomic expansion of the colored jones polynomial for trefoil . there are 7 sixth order mock theta functions in ramanujan s `` lost '' notebook @xcite , and we treat 3 functions among them defined by @xmath141 \psi(q ) = \sum_{n=0}^\infty ( -1)^n \ , q^{(n+1)^2 } \ , \frac{(q ; q^2)_n } { ( -q ; q)_{2 n + 1 } } , \\[2 mm ] \rho(q ) = \sum_{n=0}^\infty q^{n(n+1)/2 } \ , \frac { ( -q)_n } { ( q ; q^2)_{n+1 } } .\end{gathered}\ ] ] other functions can be written in terms of these functions and theta functions as was proved in ref . . these @xmath1-series converge also for @xmath50 , and for our convention we define @xmath142 \psi^*(q ) = \psi(1/q ) = \sum_{n=0}^\infty q^n \ , \frac { ( q ; q^2)_n } { ( -q)_{2n+1 } } , \\[2 mm ] \rho^*(q ) = -q^{-1 } \ , \rho(1/q ) = \sum_{n=0}^\infty ( -1)^n \ , q^n \ , \frac { ( -q)_n } { ( q ; q^2)_{n+1 } } .\end{gathered}\ ] ] these new functions are also the false theta functions _ la _ rogers as follows ; @xmath143 \label{psi_star } \psi^*(q ) = \sum_{n=0}^\infty \psi_{12}^{(3)}(n ) \ , q^{\frac{n^2 - 9}{24 } } , \\[2 mm ] \label{rho_star } \rho^*(q ) = \psi^*(q^2 ) = \sum_{n=0}^\infty \psi_{24}^{(6)}(n ) \ , q^{\frac{n^2 - 36}{48 } } , \end{gathered}\ ] ] where the odd periodic function @xmath17 is defined in . we do not have a simple proof . anyway we recall the transformation formula of the @xmath1-hypergeometric functions as ( see , _ e.g. _ , ref . ) @xmath144 and another identity ( see , _ e.g. _ , ( 25.96 ) in ref . ) @xmath145 to prove , we compute as follows ; @xmath146 ^ 2 } \\ & = \sum_{k=0}^\infty ( -q)^k \ , q^{\frac{3}{2 } k ( k+1 ) } . \tag{by \eqref{fine_formula } with $ \alpha=1 $ , $ \beta=0 $ , $ z = q$ } \end{aligned}\ ] ] a proof of follows in the same manner ; @xmath147 identity can be proved as follows ; @xmath148 which completes the proof . looking at and , we set the vector - valued functions as @xmath149 \psi_{12}^{(3)}(n ) \end{pmatrix } \ , q^{\frac{n^2}{24 } } .\ ] ] by the poisson summation formula we find that this is a modular form with weight @xmath7 , and that @xmath66- and @xmath67-matrices are respectively given by @xmath150 \sqrt{2 } & -1 \end{pmatrix } , & \mathbf{t } & = \begin{pmatrix } { \mathrm{e}}^{\frac{1}{12 } \pi { \mathrm{i } } } & \\[2 mm ] & { \mathrm{e}}^{\frac{3}{4 } \pi { \mathrm{i } } } \end{pmatrix } .\end{aligned}\ ] ] the modular form @xmath71^{-1 } \cdot \boldsymbol{\phi}_{2,3,3}(\tau)$ ] is on @xmath151 , which has the symmetry of the tetrahedral group . by definition , the mock theta functions can be regarded as the eichler integral of this modular form ; @xmath152 q^{\frac{3}{8 } } \ , \psi^*(q ) \end{pmatrix } .\ ] ] the eichler integral @xmath153 has appeared as the wrt invariant for the seifert manifold @xmath154 @xcite whose fundamental group is the tetrahedral group ; @xmath155 as was noticed in ref . , the wrt invariants for the manifold @xmath156 and @xmath154 are related to each other , and we also obtain the wrt invariant for @xmath156 as @xmath157 the function @xmath158 is also regarded as the eichler integral of the weight-@xmath7 modular form @xmath71 ^ 3 $ ] , which , by the jacobi triple product formula , is written in an infinite sum as @xmath159 ^ 3 = \sum_{n=0}^\infty ( -1)^n \ , ( 2 \ , n + 1 ) \ , q^{\frac{1}{8 } ( 2 n + 1)^2 } .\ ] ] recalling a result in ref . , we can conclude that it gives the wrt invariant for the seifert manifold @xmath160 . remaining ramanujan s mock theta functions are the tenth order . there are 4 functions in ref . ( see also ref . ) , and they are defined by @xmath162 \psi(q ) = \sum_{n=0}^\infty \frac{q^{(n+1)(n+2)/2}}{(q;q^2)_{n+1 } } , \\[2 mm ] x(q ) = \sum_{n=0}^\infty ( -1)^n \ , \frac { q^{n^2 } } { ( -q)_{2n } } , \\[2 mm ] \chi(q ) = \sum_{n=0}^\infty ( -1)^n \ , \frac { q^{(n+1)^2 } } { ( -q)_{2n+1 } } .\end{gathered}\ ] ] all these defining @xmath1-series converge also for @xmath50 , and as before we define our functions as follows ; @xmath163 \psi^*(q ) = - \psi(1/q ) = \sum_{n=0}^\infty ( -1)^n \ , \frac { q^{n(n+1)/2 } } { ( q;q^2)_{n+1 } } , \\[2 mm ] x^*(q ) = x(1/q ) = \sum_{n=0}^\infty ( -1)^n \ , \frac{q^{n ( n+1 ) } } { ( -q)_{2n } } , \\[2 mm ] \chi^*(q ) = \chi(1/q ) = \sum_{n=0}^\infty ( -1)^n \ , \frac{q^{n ( n+1 ) } } { ( -q)_{2n+1 } } .\end{gathered}\ ] ] we have the following ; @xmath164 \psi^*(q ) = \sum_{n=0}^\infty \left ( \psi_{10}^{(1)}(n ) + \psi_{10}^{(4)}(n ) \right ) \ , q^{\frac{1}{5}(n^2 - 1 ) } , \\[2 mm ] x^*(q ) = \sum_{n=0}^\infty \psi_{10}^{(1)}(n ) q^{\frac{1}{40}(n^2 - 1 ) } , \\[2 mm ] \chi^*(q ) = \sum_{n=0}^\infty \psi_{10}^{(3)}(n ) q^{\frac{1}{40}(n^2 - 9 ) } . \end{gathered}\ ] ] all these identities follow from with the bailey pair , c(4 ) , c(3 ) , g(3 ) , and g(2 ) in slater s list @xcite respectively . our functions are regarded as the eichler integral of the modular form with @xmath165 . a result in ref . proves that these give the wrt invariant for the seifert manifold @xmath166 . @xmath167 by applying to the defining @xmath1-series , it is straightforward to get the infinite @xmath1-series , which terminates at the finite order for root of unity , of the quantum invariants . we have shown that the ramanujan mock theta functions are related to the quantum invariant for the seifert manifolds @xmath168 . we see that some of the mock theta functions can be defined also even when @xmath1 is outside the unit circle , and that , by replacing @xmath1 by @xmath51 , they become the _ false _ theta functions in a sense of rogers . this substitution @xmath169 seems to correspond to study , not directly the eichler integral @xmath37 , but the eichler integral @xmath38 defined in the lower half plane , which has a nice transformation property . these _ false _ theta functions have already appeared in studies of the wrt invariant for the seifert manifold as the eichler integral of the half - integral weight modular form . namely a limiting value of the eichler integral in @xmath4 for @xmath170 gives the wrt invariant @xmath2 . combining these results , we can deduce a remarkable connection between the ramanujan mock theta functions and the quantum invariants . we summarize this correspondence in table [ tab : mock_order ] . .mock theta functions as the su(2 ) wrt invariant @xmath2 for the seifert manifolds @xmath171 . functions of order 2 , 4 , and 8 are our proposal based on the wrt invariants . [ cols="^,^,^",options="header " , ] unfortunately we do not find such correspondence for the eighth order mock theta functions proposed in ref . . motivated from ref . , we may expect that the functions defined by @xmath172 d_6(q ) = \sum_{n=0}^\infty q^n \ , \frac { ( -q^2 ; q^2)_n } { ( q^{n+1})_{n+1 } } , \\[2 mm ] i_{12}(q ) = \sum_{n=0}^\infty q^{2n } \ , \frac { ( -q ; q^2)_n } { ( q^{n+1})_{n+1 } } , \\[2 mm ] i_{13}(q ) = \sum_{n=0}^\infty q^{n } \ , \frac { ( -q ; q^2)_n } { ( q^{n+1})_{n+1 } } , \end{gathered}\ ] ] will be related to the 2nd/4th/8th order mock theta functions . we can introduce the @xmath1-series from above definitions by replacing @xmath1 with @xmath51 as @xmath173 d_6^*(q ) = -q^{-1 } \ , d_6(1/q ) = \sum_{n=0}^\infty ( -1)^n q^{n ( n+1)/2 } \ , \frac { ( -q^2 ; q^2)_n } { ( q^{n+1})_{n+1 } } , \\[2 mm ] i_{12}^*(q ) = - q^{-1 } \ , i_{12}(1/q ) = \sum_{n=0}^\infty ( -1)^n \ , q^{n ( n+1)/2 } \ , \frac { ( -q;q^2)_{n } } { ( q^{n+1})_{n+1 } } , \\[2 mm ] i_{13}^*(q ) = - q^{-1 } \ , i_{13}(1/q ) = \sum_{n=0}^\infty ( -1)^n \ , q^{n ( n+3)/2 } \ , \frac { ( -q;q^2)_{n } } { ( q^{n+1})_{n+1 } } , \end{gathered}\ ] ] and the bailey pairs , d(5 ) , d(6 ) , i(12 ) , and i(13 ) , in slater s list prove the following ; @xmath174 d_6^*(q ) = \sum_{n=0}^\infty \left ( \psi_8^{(1)}(n ) + \psi_8^{(3)}(n ) \right ) \ , q^{\frac{1}{4 } ( n^2 - 1 ) } , \\[2 mm ] i_{12}^*(q ) = \sum_{n=0}^\infty \left ( \psi_{16}^{(1)}(n ) + \psi_{16}^{(7)}(n ) \right ) \ , q^{\frac{1}{16}(n^2 - 1 ) } , \\[2 mm ] i_{13}^*(q ) = \sum_{n=0}^\infty \left ( \psi_{16}^{(3)}(n ) + \psi_{16}^{(5)}(n ) \right ) \ , q^{\frac{1}{16}(n^2 - 9 ) } . \end{gathered}\ ] ] this proves @xmath175 where @xmath176 and @xmath177 are from the 6-th order , . from computations in ref . , these give the wrt invariants as follows ; @xmath178 \left ( { \mathrm{e}}^{\frac{2 \pi { \mathrm{i}}}{n } } - 1 \right ) \ , \tau_n \left ( m(2,2,4 ) \right ) = \left ( 1 + ( -1)^n \right ) \ , \left ( 1 - d_6^ * ( { \mathrm{e}}^ { \frac{\pi { \mathrm{i}}}{2 n } } ) \right ) , \\[2 mm ] { \mathrm{e}}^{\frac{3 \pi { \mathrm{i}}}{2 n } } \ , \left ( { \mathrm{e}}^{\frac{2 \pi { \mathrm{i}}}{n } } - 1 \right ) \ , \tau_n \left ( m(2,2,8 ) \right ) = \left ( 1 + ( -1)^n \right ) \ , \left ( 1 - i_{12}^ * ( { \mathrm{e}}^ { \frac{\pi { \mathrm{i } } } { n } } ) \right ) . \end{gathered}\ ] ] as the wrt invariant for wide class of the seifert manifolds has a nearly modular property @xcite , we may derive mock ( false ) theta functions from explicit form of these invariants . in such process , the colored jones polynomial for torus knots @xcite and the twist knots @xcite would be helpful . the author would like to thank thang le for explaining his construction of the quantum invariants . he also thanks t. takata for useful discussions . this work is supported in part by the grant - in - aid for young scientists from the ministry of education , culture , sports , science and technology of japan . j. milnor , _ on the 3-dimensional brieskorn manifolds @xmath181 _ , in l. p. neuwirth , ed . , _ knots , groups , and 3-manifolds _ , pp . 175225 , princeton univ . press , 1975 , papers dedicated to the memory of r. h. fox . s. p. zwegers , _ mock @xmath182-functions and real analytic modular forms _ , in b. c. berndt and k. ono , eds . , _ @xmath1-series with applications to combinatorics , number theory , and physics_,_comtemp . _ * 291 * , pp . 269277 , amer . soc . , providence , 2001 .

the properties of the highest velocity ejecta of normal type ia supernovae ( sne ia ) are studied via models of very early optical spectra of 6 sne . at epochs earlier than 1 week before maximum , sne with a rapidly evolving line velocity ( hvg ) have a larger photospheric velocity than sne with a slowly evolving line velocity ( lvg ) . since the two groups have comparable luminosities , the temperature at the photosphere is higher in lvg sne . this explains the different overall spectral appearance of hvg and lvg sne . however , the variation of the and absorptions at the highest velocities ( km s ) suggests that additional factors , such as asphericity or different abundances in the progenitor white dwarf , affect the outermost layers . the line is marginally detected in 3 lvg sne , suggesting that lvg undergo less intense burning . the carbon mass fraction is small , only less than 0.01 near the photosphere , so that he mass of unburned c is only . radioactiveni and stable fe are detected in both lvg and hvg sne . different fe - group abundances in the outer layers may be one of the reasons for spectral diversity among sne ia at the earliest times . the diversity among sne ia at the earliest phases could also indicate an intrinsic dispersion in the width - luminosity relation of the light curve .

0712.2823

type ia supernovae ( sne ia ) are believed to be thermonuclear explosions of co white dwarfs ( wds ) . because of their high luminosity together with a tight relation between the maximum brightness and decline rate of the light curve ( lc ) , sne ia are one of the most accurate distance indicators ( e.g. , phillips 1993 ; riess et al . 1998 ; perlmutter et al . 1999 ; knop et al . 2003 ; riess et al . 2004 ; astier et al . 2006 ; wood - vasey et al . 2007 ) . in addition to the homogeneity of the lc , optical spectra of sne ia are also rather homogeneous . the ratio of two lines in spectra at maximum correlates well with the sn luminosity ( nugent et al . the width of the fe emission line in the nebular spectra also correlates with the lc decline rate ( mazzali et al . these works show that the spectral properties both at maximum brightness and at the nebular phases are determined primarily by the maximum luminosity . however , recent studies using large samples revealed large diversity in the early spectra of sne ia , in particular in the line velocities ( see e.g. , hatano et al . 2000 ; benetti et al . 2005 , hereafter b05 ; branch et al . b05 divided sne ia into three groups according to the post - maximum evolution of the @xmath06355 line velocity . high velocity gradient ( hvg ) sne have a large line velocity that decreases rapidly after maximum . on the other hand , low velocity gradient ( lvg ) sne have a lower velocity with a slower post - maximum evolution ( see fig . 1 in b05 ) . hvg and lvg sne have similar maximum luminosities , implying that line velocity is independent of luminosity or lc shape ( see fig . 3 in hachinger , mazzali & benetti 2006 ) . faint sne are characterized by a lower luminosity and a low line velocity , whose evolution is comparable to that of lvg sne . the origin of the spectral diversity is not yet fully understood , reflecting uncertainties on the explosion mechanism . despite general consensus that burning starts as a subsonic deflagration ( nomoto , sugimoto & neo 1976 ; nomoto , thielemann & yokoi 1984 ) , it is still unclear whether or not a transition to a supersonic detonation ( khokhlov 1991b ) occurs . in recent years , the explosion mechanism has been studied by detailed numerical simulations free from spherically symmetry ( e.g. , reinecke , hillebrandt , & niemeyer 2002 ; gamezo et al . 2003 ; plewa , calder & lamb 2004 ; rpke & hillebrandt 2005 ; gamezo , khokhlov & oran 2005 ) . early phase spectra are an unique tool to investigate the state of the outer layers of the ejecta , i.e. , the final fate of the burning front . furthermore , as diversity in line velocities is largest at pre - maximum phases , modeling spectra at these epochs is likely to yield clues as to the origins of this diversity . although several studies focussed on spectral diversity at maximum light ( e.g. bongard et al . 2006 ; branch et al . 2006 ; hachinger et al . 2006 ) , only few have addressed pre - maximum spectra ( e.g. mazzali et al . 1993 , fisher et al . 1997 ; mazzali 2001 ; branch et al . this is mostly due to the lack of very early data , which stems from the difficulty in discovering sne well before maximum . in order to remedy this situation , the european supernova collaboration ( esc ) has endeavored to gather very early spectra of nearby sne ia . one of the most interesting results of very early sn ia spectroscopy is the possible presence of unburned carbon in the outer layers . this may be a critical signature of the outer extent of burning . the @xmath06578 line has been detected in sn 1998aq ( at @xmath5 11,000 km s@xmath2 , branch et al . 2003 ) and in sn 1999ac ( at @xmath5 16,000 km s@xmath2 , garavini et al . the clearest features were seen in a pre - maximum spectrum of sn 2006d ( at @xmath5 12,000 km s@xmath2 , thomas et al . 2006 ) . 15,000 km s@xmath2 , although the sn is not normal and possibly has a super - chandrasekhar mass progenitor ( hicken et al . 2007 ) , as suggested for sn 2003fg ( howell et al . 2006 ) ] on the other hand , in near - infrared ( nir ) spectra has never been firmly detected ( marion et al . 2006 ) . high - velocity absorption features ( hvfs , @xmath1 km s@xmath2 ) in the ir triplet ( e.g. , hatano et al . 1999 ; mazzali et al . 2005a ; thomas et al . 2004 ; quimby et al . 2006 ) may be a feature of all sne ia at the earliest epochs ( mazzali et al . 2005b ) , and may indicate aspherical ejecta , as suggested by the high polarization level ( wang et al . 2003 ; kasen et al . they may result from the explosion itself ( thomas et al . 2004 ; mattila et al . 2005 ) or from the interaction of the ejecta with circumstellar medium ( csm ; gerardy et al . 2004 ; mazzali et al . the variation and distribution of the strength of hvfs may be explained by line - of - sight effects if hvfs result from aspherical structures like a few blobs or a thick disk at @xmath6 km s@xmath2 ( tanaka et al . 2006 ) . in this paper , we use sne ia pre - maximum optical spectra to study the physical properties of the ejecta , modeling the spectra using a numerical code . details of the method are presented in [ sec : method ] . the results of the modeling are shown in [ sec : results ] . in [ sec : discussion ] , our results are summarized and the implications are discussed . finally , conclusions are in [ sec : conclusions ] . spectra of 6 nearby sne ia observed earlier than 1 week before maximum are studied ( table [ tab : prop ] ) . the data were obtained by the esc campaign . the esc has collected photometric and spectroscopic data of a number of nearby sne ia with very good temporal and wavelength coverage . sne with very early spectra include sn 2001el ( lvg ; mattila et al . 2005 ) , sn 2002bo ( hvg ; benetti et al . 2004 ) , sn 2002dj ( hvg ; pignata et al . 2005 ) , sn 2002er ( hvg ; pignata et al . 2004 ; kotak et al . 2004 ) , sn 2003cg ( lvg ; elias - rosa et al . 2006 ) , and sn 2003du ( lvg ; stanishev et al . our sample does not include faint sne . to investigate the physical conditions of the ejecta , such as temperatures , ionization profiles , and element abundance distribution , we use a monte carlo spectrum synthesis code ( mazzali & lucy 1993 ) , which was used for several sne ( e.g. , mazzali et al . 1993 ) . the code requires as input the bolometric luminosity ( @xmath7 ) , photospheric velocity ( @xmath8 ) , and a model of the density and abundance distribution in the sn ejecta . the code assumes a sharply defined spherical photosphere . the position of the photosphere is expressed in terms of a velocity thanks to the homologous expansion ( @xmath9 ) . the temperature structure of the optically thin atmosphere is solved in radiative equilibrium tracing a large number of energy packets as they propagate in the sn ejecta . starting from a trial temperature distribution ( radiation temperature , @xmath10 ) , the population of the excited level ( @xmath11 ; @xmath12 for the ground state ) is computed as @xmath13 where @xmath14 and @xmath15 are the statistical weight and the excitation energy from the ground level , respectively . here w is the so called dilution factor , which is defined as @xmath16 in eq . ( 1 ) , w is set to unity for metastable levels ( lucy 1999 ) . the ionization regime is computed using a modified nebular approximation ( abbott & lucy 1985 , mazzali & lucy 1993 ) : @xmath17 where @xmath18 is the electron density and @xmath19 the electron temperature . the starred term on the right hand side is the value computed using the saha equation with the temperature @xmath10 ; the partition functions entering the saha equation are , however , computed with eq . ( [ eq : level ] ) . in eq . ( [ eq : ion ] ) , @xmath20 is defined as @xmath21 where @xmath22 is the fraction of recombinations going directly to the ground state . mazzali & lucy ( 1993 ) introduced a correction factor @xmath23 for an optically thick continuum at shorter wavelengths than the ionization edge ( for a detailed definition , see eqs . ( 15 ) and ( 20 ) in mazzali & lucy 1993 ) . the second term in eq . ( 4 ) represents ionization from excited levels ( mazzali & lucy 1993 ) . the monte carlo radiation packets are followed through the atmosphere , where they can undergo electron scattering or interact with spectral lines . for line scattering , sobolev approximation , which is a sound approximation in a rapidly expanding medium , is applied . from the level populations , a sobolev line opacity is obtained : @xmath24 here @xmath25 is the time since the explosion and @xmath26 and @xmath27 are einstein @xmath28-coefficients . the effect of photon branching is considered ( lucy 1999 ; mazzali 2000 ; see also pinto & eastman 2000 ) . the monte carlo experiment gives a flux in each radial point , and then a frequency moment @xmath29 is computed . hence the temperature structure can be determined via @xmath30 here @xmath31 represents for mean energy of blackbody radiation ( @xmath32 ) . the electron temperature is crudely assumed to be @xmath33 to simulate approximately ) computed with @xmath10 , which is a direct result of the monte carlo simulation . ] the situation where the electron temperature is largely controlled by radiation ( klein & castor 1978 ; abbott & lucy 1985 ) . level populations , ionizations and opacities are then updated using the temperature structure derived from the monte carlo simulation . this process is iterated until convergence . lrccccccc sn 2002bo ( hvg ) & @xmath34 & 1.17 & 0.38 & 31.67 & 15.0 & 14.9 & 22.1 & 1 + sn 2002dj ( hvg ) & @xmath35 & 1.12 & 0.10 & 32.98 & 16.8 & 17.7 & 27.6 & 2 + sn 2002er ( hvg ) & @xmath36 & 1.33 & 0.36 & 32.90 & 13.0 & 15.6 & 23.1 & 3 , 4 + sn 2001el ( lvg ) & @xmath37 & 1.15 & 0.22 & 31.26 & 14.0 & 17.1 & 23.8 & 5 + sn 2003cg ( lvg ) & @xmath38 & 1.25 & 1.36 & 31.28 & 12.0 & 12.7 & 22.0 & 6 + sn 2003du ( lvg ) & @xmath35 & 1.02 & 0.01 & 32.79 & 12.0 & 15.5 & 22.5 & 7 + [ tab : prop ] we use the w7 deflagration model ( nomoto et al . 1984 ) as a standard density profile . alternatively , a density structure of a delayed detonation model is used ( model wdd2 in iwamoto et al . 1999 , hereafter , dd model ) . the density structures of these models are most different in layers above @xmath39 km s@xmath2 . the density in the dd model at @xmath40 km s@xmath2 is higher than that of w7 by a factor of @xmath41 ( see fig . 11 in baron et al . thus the `` density enhancement '' referred to in mazzali et al . ( 2005a ) is equivalent to the structure of the dd model if the densities are increased by a factor of @xmath42 . the kinetic energy of the dd model ( @xmath43 ergs ) is slightly larger than that of w7 ( @xmath44 ergs ) . element abundance distributions ( i.e. , mass fractions of elements ) are parameterized for better reproduction of the observed spectra . initially , a homogeneous abundance distribution is assumed , averaging the abundance distribution of w7 above the photospheric velocity , and a general fit of the spectrum is obtained adjusting the abundances . in this process , the input parameters @xmath7 and @xmath8 are also optimized . then a stratified abundance distribution is adopted and optimized to reproduce the details . in this paper , we employ two different - abundance zones except for sn 2003cg ( 3 zones are used ; see [ sec:03cgsi ] ) . the abundances in the outermost ejecta are further checked against the spectra at maximum . although the photosphere recedes with time as the ejecta expand , the abundances at high velocities can still affect the synthetic spectra at later epochs ( stehle et al . therefore we modelled the spectra at maximum using the abundance ratios derived from the pre - maximum spectra for the region at @xmath45 , where @xmath46 is the photospheric velocity of the pre - maximum spectra . if the result is inconsistent ( e.g. , incorrect high - velocity absorptions resulting from the pre - maximum spectra photospheric component ) , the pre - maximum spectra are re - investigated until consistence is achieved . lrccc sn 2002bo ( hvg ) & @xmath47 & 42.78 & 12,900 & 10,900 + sn 2002dj ( hvg ) & @xmath48 & 42.44 & 14,000 & 10,200 + sn 2002er ( hvg ) & @xmath49 & 42.98 & 9,500 & 17,300 + sn 2001el ( lvg ) & @xmath50 & 42.87 & 10,500 & 17,100 + sn 2003cg ( lvg ) & @xmath51 & 42.84 & 9,000 & 17,000 + sn 2003du ( lvg ) & @xmath48 & 42.71 & 10,000 & 18,600 + [ tab : param ] lrcccccccc sn 2002bo ( hvg ) & @xmath52 & & 0.51 & 0.10 & 0.02 & 0.20/0.02 & 0.02/0.006 & 0.020 & 0.045 + sn 2002dj ( hvg ) & @xmath48 & & 0.50 & 0.09 & 0.02 & @xmath530.80/0.08 & 0.02/0.005 & 0.005 & 0.0052 + sn 2002er ( hvg ) & @xmath54 & & 0.76 & 0.12 & 0.05 & 0.40/0.05 & 0.05/0.007 & 0.012 - 0.013 & @xmath55 0.025 + sn 2001el ( lvg ) & @xmath56 & 0.003 & 0.42 & 0.05 & 0.02 & @xmath53 0.80/0.10 & 0.10/0.02 & 0.048 & 0.19 + sn 2003cg ( lvg ) & @xmath57 & @xmath55 0.002 & 0.40 ( 0.08 ) & 0.080 & 0.02 & 0.03/0.005 & 0.001/0.0001 & 0.0033 - 0.0036 & @xmath55 0.0088 + sn 2003du ( lvg ) & @xmath48 & 0.002 & 0.29 & 0.060 & 0.016 & 0.70/0.11 & 0.10/0.015 & 0.003 & 0.005 + [ tab : abun ] in this section , we present the results of the modeling for each sn . the properties of the sne and the spectra are summarized in table [ tab : prop ] . in addition to the parameters @xmath7 and @xmath58 , and the abundances in the ejecta , the distance modulus ( @xmath59 ) and the reddening [ in terms of @xmath60 and @xmath61 are also required to reproduce the observed spectra . thus the luminosity and the epoch , which are often neglected by rescaling the flux arbitrarily , are important because @xmath8 is constrained not only by the line velocities but also by the relation @xmath62 . here @xmath25 is the time since the explosion and @xmath63 the effective temperature . all spectra have been calibrated against the photometry . the epoch is estimated assuming a rise time of 19 days for all objects for simplicity . we take @xmath59 and @xmath60 from the literature ( see table [ tab : prop ] ) . using all the quantities above , the photospheric temperature @xmath64 is computed in the code considering the back - scattering effect , and thus tends to be higher than @xmath63 ( mazzali & lucy 1993 ; mazzali 2000 ) . in order to facilitate the comparison of intrinsic colors , all spectra are displayed after correcting for reddening . the values of @xmath7 and @xmath8 derived from modeling are shown in table [ tab : param ] along with the computed photospheric temperature ( @xmath64 ) , while the abundances of the elements at the photosphere that most affect the synthetic spectra are shown in table [ tab : abun ] . the abundances in the outer shell are not strongly constrained because of the lack of strong lines . only has high velocity lines ( see following sections ) , and the ca mass fraction in the outer shell is shown in the table ( @xmath65(ca ) ) . the iron mass fraction @xmath66(fe ) indicates the sum of stable fe and @xmath4feproduced by @xmath4ni decay at the time of the observation . if only stable fe is considered , this is explicitely denoted as @xmath66(fe)@xmath67 , where the index 0 `` means ' ' at explosion . furthermore , we use @xmath66(@xmath4ni)@xmath67 for the initial mass fraction of @xmath4ni at the time of the explosion . the well - studied sn 2002bo ( stehle et al . 2005 ) is a member of the hvg group . the spectrum at @xmath68 days is shown in figure [ fig:02bo ] , together with our synthetic spectrum . good agreement is obtained with @xmath69 ( erg s@xmath2 ) @xmath70 and @xmath71 km s@xmath2 , giving a @xmath72k . the flux redward of @xmath73 is overestimated ( see also stehle et al . this may be due to a deviation from the blackbody approximation . a modified input radiation field , where the blue region ( @xmath74 ) has greater weight , can reproduce the spectrum better , with only a small difference in the photospheric temperature and ionization states compared to the original case . an absorption feature around 4300 results mainly from @xmath04481 and @xmath04420 , with a smaller contribution by @xmath04553 . around 4800 , many and lines , e.g. , @xmath05169 and @xmath05056 , produce the strong absorption feature . to account for these strong features , the mass fraction of fe , including both stable fe and decay products of @xmath4ni , should be @xmath66(fe)=0.022 at the photosphere . our method of estimating the fe abundance is illustrated in figure [ fig:02bofe ] , right panel . for solar fe abundance , the absorption at 4950 ( mainly @xmath05169 ) is weaker than the observed feature ( red solid line ) . on the other hand , fe - rich ejecta [ @xmath66(fe)@xmath75 produce too strong a feature ( dotted line ) . the abundance of radioactive @xmath4ni synthesized in the explosion can be estimated using the ni and co lines that are present around 2000 - 4000 . since at @xmath76 10 days around 30% of @xmath4ni has not yet decayed , the strong @xmath04067 line is useful to constrain the mass fraction . the procedure is illustrated in figure [ fig:02bofe ] , left panel . for a larger @xmath4ni mass fraction [ @xmath66(@xmath4ni)@xmath77 , the absorption at 3900 appears ( red dotted line ) , which is inconsistent with the observed feature . on the other hand , when only a solar abundance of @xmath4ni is assumed , the flux level at @xmath78 becomes too high ( red solid line ) . this is because of insufficient blocking of ( near ) uv photons by iron group lines . [ our code treats the blocking by iron group elements correctly , considering photon branching ( lucy 1999 ; mazzali 2000 ) ] . a @xmath4ni mass fraction @xmath66(@xmath4ni)@xmath67 = 0.045 is derived . although the observed flux level at @xmath78 is often uncertain , the upper limit of @xmath66(@xmath4ni)@xmath67 is strict because it is derived from the absence of @xmath04067 . therefore , most fe detected above the photosphere must be stable [ @xmath66(fe)@xmath67= 0.020 ] . the ratio of the two absorptions at 5800 and 6000 is reproduced well . a strong @xmath06355 suggests that the photosphere resides in the si - rich layer . the mass fraction of si is estimated as @xmath66(si)@xmath50.50 . the emission peak of @xmath06355 is well developed , and there is no evidence of @xmath06578 . the blue wing of the synthetic absorption is weak , suggesting that the line may be contaminated by a hvf , as are the lines ( see 3.1.4 ) . at 7500 , a weak absorption is seen , which is often attributed to @xmath07774 . however , in the synthetic spectrum the feature is mainly due to @xmath07869 and @xmath07849 . these lines are actually stronger than the line if @xmath04481 and @xmath06355 are reproduced . the spectrum around h&k and the ir triplet is enlarged in the insets of figure [ fig:02bo ] . the thin cyan line shows a synthetic spectrum computed with the original w7 density structure and homogeneous abundances , setting @xmath66(ca)=0.02 . while the observed features have a blue wing extending to @xmath79 km s@xmath2 ( h&k ) and @xmath80 km s@xmath2 ( ir triplet ) , the wing of the synthetic absorption disappears at @xmath81 km s@xmath2 and @xmath82 km s@xmath2 , respectively . this difficulty has been reported in many papers , e.g. , hatano et al . ( 1999 ) for sn 1994d , gerardy et al . ( 2004 ) for sn 2003du , mazzali et al . ( 2005a ) for sn 1999ee and mazzali et al . ( 2005b ) for the objects in this paper . to reproduce these features , we test several cases . first , we modified the abundance distribution , keeping the w7 density structure . although the features can be reproduced with @xmath66(ca ) @xmath83 0.20 at @xmath84 km s@xmath2 , such a large fraction is very unlikely because the ca mass fraction in the most ca - rich layer is @xmath85 0.1 in all explosion models ( see e.g. , iwamoto et al . secondly , when we use the dd density structure , the observed profile can be reproduced with @xmath66(ca)=0.02 at @xmath84 km s@xmath2(insets of fig . [ fig:02bo ] , dotted red lines ) because of the higher density at @xmath86 km s@xmath2 . finally , we test the interesting possibility that interaction with csm leads to some abundance mixing ( gerardy et al . as discussed by mazzali et al . ( 2005a ) , when a hydrogen - rich csm is mixed with the ejecta , the electron density increases , favouring recombination . therefore some fraction of , which is the dominant state ( see [ sec : hvg ] ) , becomes , making the line stronger . here we assume that the interaction causes a density jump by a factor of 4 at @xmath84 km s@xmath2 , corresponding to @xmath87 of material . hydrogen is assumed to be mixed into the shocked region and we find that h@xmath88 does not appear as long as @xmath66(h ) @xmath89 ( @xmath90(h)@xmath91 ) with this model , the mass fraction of ca in the high - velocity layers is reduced to @xmath66(ca)=0.02 ( with the w7 density structure ; dashed blue line in insets of fig . [ fig:02bo ] ) and @xmath66(ca)=0.006 ( with the dd density structure ) , respectively . the deceleration following the interaction is not considered in our computation and a more realistic hydrodynamic calculation is needed to confirm this scenario . sn 2002dj has a spectral sequence and lc similar to sn 2002bo ( pignata et al . 2005 ) , and is also an hvg sn . the spectrum @xmath93 days before b maximum is well reproduced by a synthetic spectrum with @xmath69 ( erg s@xmath2 ) = 42.44 and @xmath94 km s@xmath2 ( fig . [ fig:02dj ] ) . almost all lines are the same as in the spectrum of sn 2002bo , although the velocities are higher because the epoch is earlier . the photospheric temperature , 10,200 k , is also similar . and lines contribute to the features near 4200 and 4800 , as in sn 2002bo , but they are weaker . in fact , the absorption at 4200 is caused mostly by @xmath04481 . utilizing the feature around 4800 ( a combination of and lines ) as in [ sec:02bofe ] , the mass fraction of fe is evaluated as @xmath66(fe ) @xmath95 at the photosphere . the features at 3500 and 3900 require @xmath66(@xmath4ni)@xmath67 = 0.005 . therefore , the fe causing the observed spectral features is again mostly stable fe . although the values are smaller than those of sn 2002bo , they are still larger than the solar abundances by a factor of 3.5 and 70 for fe and ni , respectively . the @xmath05979 line is quite strong , which is consistent with the low photospheric temperature in the synthetic spectrum . the @xmath06355 line in sn 2002dj is the broadest of the 6 sne . in the model , the mass fraction of si is @xmath66(si)@xmath96 at the photosphere . the synthetic profile does not show enough absorption in the blue wing , as was the case for sn 2002bo , suggesting the presence of a hvf . the emission profile of @xmath06355 is well developed and there is no evidence of an absorption by @xmath06578 . the hvf in sn 2002dj is the strongest of the 6 sne . even assuming @xmath66(ca)=0.80 at the corresponding velocity range , the profile is not reproduced with the w7 density structure ( the thin cyan line in insets of fig . [ fig:02dj ] is a homogeneous abundance model with @xmath66(ca)=0.02 ) . the blue dashed line is a synthetic spectrum obtained with a density jump at @xmath97 km s@xmath2 ( adding 0.005 @xmath98 ) , where @xmath66(h ) = 0.30 and @xmath66(ca ) = 0.10 . this hydrogen mass fraction is an upper limit for the absence of h@xmath88 again ( @xmath90(h ) @xmath99 ) . using the dd density structure , the ca mass fraction at high - velocity layers is reduced to @xmath66(ca)=0.02 ( assuming no csm interaction , red dotted lines in the insets of fig . [ fig:02dj ] ) and @xmath66(ca)=0.005 ( assuming csm interaction ) , respectively . the spectra of the hvg sn 2002er were studied by kotak et al . we use the earliest spectrum that covers the ir triplet ( @xmath54 days ) . a synthetic spectrum with @xmath69 ( erg s@xmath2 ) @xmath100 and @xmath101 km s@xmath2provides the best fit ( fig . [ fig:02er ] ) . the lower @xmath8 agrees with the fact that this sn has lower line velocities than other hvg sne ( kotak et al . 2005 ; b05 ) . at the same time , @xmath102 k , higher than in both sne 2002bo and 2002dj . this results from the combination of the higher @xmath7 and lower @xmath8 . lines contribute to the blue absorption near 4300 ( @xmath04396 , 4420 , 4431 ) . at 4800 , lines ( e.g. , @xmath05127 ) dominate over lines ( e.g. , @xmath05018 ) , because the temperature is higher than in sne 2002bo and 2002dj . to get sufficiently strong fe lines , @xmath66(fe)=0.013 near the photosphere . although the lack of spectral coverage at @xmath103 makes it difficult to estimate the abundances of ni and co ( see [ sec:02bofe ] ) , an upper limit for these elements can be estimated as @xmath66(@xmath4ni)@xmath1040.025 considering the absence of @xmath04067 . therefore , the mass fraction of stable fe is @xmath66(fe)@xmath1050.012 - 0.013 . around 4300 , ( @xmath04553 , 4658 ) lines are stronger than in other hvg sne , indicating a higher temperature . the blue wing of @xmath06355 is also reproduced fairly well , and there is no clear evidence of a hvf , unlike sne 2002bo and 2002dj . @xmath06578 is not seen , but the emission peak is weaker than in sne 2002bo and 2002dj . a weak absorption near 7500 is mostly due to @xmath07849 , and the contribution of @xmath07869 is small . the ir triplet is again not accounted for by the density structure of w7 ( thin cyan lines in the insets of fig . [ fig:02er ] show the profile of one - zone abundance model with @xmath66(ca)=0.05 ) . with the original w7 density structure , a ca - dominated layer [ @xmath66(ca ) @xmath5 0.40 ] is required for the feature , which is highly unrealistic . the line depth and width can be reproduced with a density jump at @xmath106 km s@xmath2 , increasing the mass by 0.015 @xmath98 and containing @xmath66(h ) = 0.30 ( @xmath90(h)@xmath91 ) and @xmath66(ca ) = 0.05 , which is similar to sne 2002bo and 2002dj . if the dd model density profile is used instead , the feature is reproduced with @xmath66(ca)=0.05 at @xmath106 km s@xmath2 even without csm interaction ( insets of fig . [ fig:02er ] , dotted red lines ) . if csm interaction is considered , this is further reduced to @xmath66(ca)=0.007 . the lvg sn 2001el was studied in spectroscopy ( mattila et al . 2005 ) and spectropolarimetry ( wang et al . the best - fitting synthetic spectrum has @xmath69 ( erg s@xmath2 ) @xmath107 and @xmath108 km s@xmath2 ( fig . [ fig:01el ] ) . this leads to @xmath109 k , which is similar to sn 2002er , as can be expected since the spectra are similar . as in sn 2002er , the features around 4300 and 4800 are attributed mostly to lines . an iron mass fraction @xmath66(fe ) @xmath110 0.055 at the photosphere is needed to reproduce these features . relatively reliable flux data at @xmath111 makes it possible to estimate the mass fractions of ni and co reliably ( see [ sec:02bofe ] ) : @xmath66(@xmath4ni)@xmath67=0.19 provides the best fit . thus the mass fraction of stable fe is estimated as @xmath66(fe)@xmath67=0.048 . the narrow feature near 4400 has similar strength as in sn 2002er , supporting the similar photospheric velocity and temperature . the @xmath06355 absorption has a boxy profile and is not reproduced well by the synthetic spectrum , suggesting a contribution from hvfs . similar profiles were also seen in sne 1990n ( leibundgut et al . 1991 ; mazzali 2001 ) and 2005cf ( garavini et al . 2007 ) . the @xmath06355 emission peak is not as round and well developed in sn 2001el as in sne 2002bo and 2002dj . the synthetic spectrum computed including c ( fig . [ fig:01el ] , dashed green line ) reproduces the feature well . if c is not included , the emission peak becomes too strong ( fig . [ fig:01el ] , red line ) . this may be a sign of the presence of c. the c abundance required for the feature is , however , only @xmath66(c ) @xmath95 . the mass of c is estimated as m(c ) @xmath112 assuming the w7 density structure and a homogeneous c mass fraction above the photosphere . since @xmath06578 is the strongest optical c line , the presence of this amount of c does not affect other parts of the spectrum . as in other sne , the hvfs can not be reproduced with the w7 density structure ( the thin cyan line in insets of fig.[fig:01el ] shows the profile of a homogeneous abundance model with @xmath66(ca)=0.02 ) . the hvfs in sn 2001el are very strong and are not reproduced , even using @xmath66(ca)=0.80 at high velocities . a better match is obtained using a density jump of a factor of 4 at @xmath113 km s@xmath2 , adding a mass of 0.015 @xmath98 , mixed - in hydrogen ( @xmath66(h)=0.30 , @xmath90(h)@xmath91 ) , and @xmath66(ca)=0.10 ( insets of fig [ fig:01el ] , dashed green line ) , but the synthetic ir triplet is too narrow . it may be made broader by increasing ca mass fraction or using a larger density jump , but both changes seem unlikely and lead to a worse fit of h&k . a large ca mass fraction , @xmath66(ca)=0.10 in the high - velocity layers , is needed even with the dd density structure ( insets of fig . [ fig:01el ] , dotted red lines ) . if csm interaction is considered for the dd model , this is , however , reduced to @xmath66(ca)=0.02 . the lvg sn 2003cg was studied by elias - rosa et al . ( 2006 ) . a synthetic spectrum with @xmath69 ( erg s@xmath2 ) @xmath114 and @xmath115 km s@xmath2 provides the best fit ( fig . [ fig:03cg ] ) . the model has @xmath116k , similar to sne 2002er and 2001el but higher than sne 2002bo and 2002dj . to account for the strength of the and features at 4000 - 4900 , @xmath66(fe)@xmath5 0.036 . we estimate an upper limit for @xmath4ni of @xmath66(@xmath4ni)@xmath117 from the absence of the @xmath04067 line . the mass fraction of stable fe is therefore @xmath66(fe)@xmath118 . a similar result is obtained in elias - rosa et al . ( 2006 ) , but the mass fraction of stable fe they find is only @xmath66(fe)@xmath67=0.015 . the main difference is that @xmath8 is lower in our model , leading to a higher temperature and suppressing . in our model , the fraction of is smaller than in elias - rosa et al . ( 2006 ) by a factor of 3 . sn 2003cg has the narrowest @xmath06355 line of our 6 sne , although the line velocities are typical for lvg ( elias - rosa et al . 2006 ) . to reproduce the narrow feature , we use a layer at @xmath119 km s@xmath2 with a smaller si mass fraction [ @xmath66(si ) @xmath120 than at the photosphere [ @xmath66(si ) @xmath121 . there is no evidence of a hvf , which seems consistent with the weakness of the hvf ( 3.5.4 ) . the emission peak is somewhat suppressed , as in sne 2002er and 2001el . elias - rosa et al . ( 2006 ) identified this as @xmath06578 , who obtained a strong synthetic line useing @xmath66(c)=0.05 . we find that @xmath122 is an upper limit at the photosphere . however , c is not necessarily needed and the identification is not conclusive . while the ir hvf might be present in sn 2003cg , it is the weakest of the 6 sne . in fact , a synthetic spectrum computed with the w7 density structure and homogeneous abundances [ @xmath66(ca)=0.02 ] ( insets of figure [ fig:03cg ] , thin cyan line ) provides a reasonably good match to the observed profile , although a better match is obtained for a slightly higher ca abundance [ @xmath66(ca)=0.03 ] at high velocity than at the photosphere [ @xmath66(ca)=0.02 ] . in the context of csm interaction , i.e. , a density jump by a factor of 4 and @xmath66(h)=0.30 ( @xmath90(h)@xmath91 ) at @xmath123 km s@xmath2 , @xmath66(ca)=0.001 is enough for the hvf . with the dd density structure a smaller fraction of ca is required , as expected [ @xmath66(ca)=0.001 ( 0.0001 ) without ( with ) csm interaction ] . the lvg sn 2003du was studied by stanishev et al . ( 2007 ) . a synthetic spectrum with @xmath69 ( erg s@xmath2 ) @xmath124 and @xmath125 km s@xmath2 gives the best agreement with the observed spectrum ( fig . [ fig:03du ] ) . these parameters lead to @xmath126 k , which is close to that of other lvg sne . the feature near 4300 is weaker than in sne 2002el and 2003cg , but the temperature is not significantly different , suggesting a smaller fe abundance at the photosphere : @xmath66(fe ) = 0.003 ( determined as in [ sec:02bofe ] ) is enough to obtain this feature . we estimate the ni mass fraction as @xmath66(@xmath4ni)@xmath67=0.005 from the feature at 3900 . thus , almost all the fe contributing to the spectrum is likely to be stable . the @xmath06355 line is relatively weak and can be reproduced with a si mass fraction of @xmath66(si)=0.3 . in spite of the low velocity of the absorption minimum , the blue wing of the absorption extends to @xmath127 km s@xmath2 . although this could be a sign of high - velocity material as in sne 2002bo , 2002dj and 2001el , the line profile is different from other sne , and the high - velocity absorption ( @xmath128 km s@xmath2 ) is relatively weak ( mazzali et al . 2005b ) . the emission peak of the line is suppressed , as seems common among lvg sne . including carbon with @xmath66(c ) @xmath5 0.003 improves the fit . if the identification is correct , the mass of c is estimated to be m(c)@xmath129 assuming the w7 density structure and a homogeneous c mass fraction above the photosphere . the ir triplet at day @xmath130 was studied by gerardy et al . the spectrum at day @xmath92 also shows a conspicuous hvf ( fig . [ fig:03du ] ) , with profile and strength similar to those of sn 2001el . a very large high - velocity ca mass fraction [ @xmath66(ca)@xmath530.7 ] is required if w7 is used ( thin cyan lines in the insets of fig . [ fig:03du ] show the profile of a homogeneous abundance model with @xmath66(ca)=0.016 ) . for the csm interaction scenario , we need @xmath66(ca)=0.11 at high velocies ( @xmath131 km s@xmath2 ) , where the density is increased by a factor of 4 , adding 0.012 @xmath98 , and @xmath66(h ) is assumed to be 0.30 ( @xmath90(h)@xmath91 , dashed green line in insets of fig . [ fig:03du ] ) . even if the dd density structure is adopted , a high mass fraction of ca at high velocity is required [ @xmath66(ca)=0.10 ] . this is reduced to @xmath66(ca)=0.015 when csm interaction is assumed(see table [ tab : abun ] ) . we have modelled the pre - maximum spectra of 6 different sne ia ( see table [ tab : prop ] ) . three of these ( 2002bo , 2002dj and 2002er ) are classified as high velocity gradient ( hvg ) , and are characterized by a large si line velocity that evolves rapidly after maximum brightness . the other three sne ( 2001el , 2003cg , and 2003du ) , classified as low velocity gradient ( lvg ) , have a smaller si line velocity that evolves slowly after maximum . the parameters of our models are shown in tables [ tab : param ] ( luminosity and photospheric velocity ) and [ tab : abun ] ( abundances ) . in the following , we summarize the properties of hvg ( [ sec : hvg ] ) and lvg ( [ sec : lvg ] ) sne . their different appearance stems basically from a different photospheric velocity . the presence of iron - peak elements in the outer layers is revisited and its implications are discussed in [ sec : fe ] . then , the origin of the difference between hvg and lvg is discussed in [ sec : hvglvg ] . the variety of high velocity features ( hvfs ) is very large independently of the classification of hvg or lvg . the causes for hvfs are also discussed in [ sec : hvf ] . finally , possible influences to the lc width - luminosity relation are discussed ( [ sec : wlr ] ) . hvg sne 2002bo and 2002dj have the highest line velocities and the lowest photospheric temperatures of the 6 sne . the higher temperature for sn 2002er results from the high @xmath7 and the low @xmath8 . sn 2002er seems to be intrinsically more luminous than sn 2002bo at a similar epoch ( table 2 ) , although it has a larger decline rate . this is also supported by the analysis of nebular spectra ( mazzali et al . the velocity of sn 2002er is intrinsically lower than that of other hvg sne ( b05 ; kotak et al . although a sn may be classified as hvg , its @xmath63 can be as high as that of lvg sne if the sn has both a high enough @xmath7 and low enough @xmath8 . this suggests continuity between two groups ( branch et al . 2006 , 2007 ) . the @xmath06355 lines of sne 2002bo and 2002dj have well - defined p - cygni profiles with blue wings reaching @xmath82 km s@xmath2(blue lines in fig . [ fig : si ] ) . in the spectra of these sne , is not seen below at least @xmath39 km s@xmath2 . this suggests that the abundance distribution of hvg sne resembles that of 1d delayed detonation models ( khokhlov et al . 1991a ; hflich & khokhlov 1996 ; iwamoto et al . 1999 ) rather than that of the 1d deflagration model w7 ( nomoto et al . 1984 ) . the computed physical conditions in the ejecta of a representative hvg sn , 2002bo , and those of a typical lvg sn of similar @xmath132 , and thus presumably similar luminosity , sn 2001el , are shown in figure [ fig : ion ] . in the ionization plot , solid lines represent the lower of two states marked in each panel . the temperature structure , the number density of electrons , and the ionization state of sn 2002bo and sn 2001el computed under the assumptions described in [ sec : method ] are rather similar at all layers with @xmath133 km s@xmath2 . consequently , the ionization fractions are also similar . b05 show the evolution of the parameter @xmath134(si ) , the ratio of the lines at 5800 and 6000 . @xmath134(si ) is thought to be a temperature indicator ( nugent et al . 1995 ) . at pre - maximum epochs ( @xmath135 days ) , @xmath134(si ) is larger in hvg than in lvg sne ( see fig . 2 in b05 ) . as we have shown , a typical hvg sn has a lower photospheric temperature before maximum . the fact that sne 2002bo and 2001el have a similar @xmath132 , and therefore should have a similar luminosity , suggests that the lower temperature of sn 2002bo at pre - maximum epochs reflects its higher photospheric velocity . therefore , @xmath134(si ) is not a good indicator of sn luminosity at pre - maximum phases , when the dispersion of the photospheric velocity among different sne is still very large ( b05 ) . on the other hand , since the diversity in photospheric velocity is small at epochs near maximum , both @xmath134(si ) and the temperature more closely reflects the intrinsic luminosity . the three lvg sne have similar photospheric velocities and temperatures . the temperature structure , electron density , and ionization state of sn 2001el are shown in figure [ fig : ion ] ( green lines ) . the run of all physical quantities is similar to hvg sn 2002bo at comparable velocities , but the values at the photosphere are significantly different because @xmath8 is lower for lvg . in particular , important ions such as , , , and are a factor of @xmath136 less abundant near the photosphere of lvg sne ( at @xmath137 km s@xmath2 ) then near the photosphere of hvg sne ( at @xmath138 km s@xmath2 ) . the @xmath06578 line is likely to be present in the lvg sne 2001el and 2003du , and possibly also in sn 2003cg near the photosphere ( @xmath137 km s@xmath2 ) . the mass fraction of c required to suppress the @xmath06355 emission peak is however only 0.002 - 0.003 . the mass of c in sne 2001el and 2003du is only @xmath139 if we assume a homogeneous , spherically symmetric distribution above the photosphere . even if we consider that we can not estimate the c abundance at @xmath140 km s@xmath2 because of the blend with the line , and assume @xmath66(c)=0.5 at @xmath141 km s@xmath2 , the mass of c is at most @xmath142 . this is still less than predicted by the w7 model ( @xmath143 ) . although is used to estimate c abundance in marion et al . ( 2006 ) , is not the dominant ionization state as shown in figure [ fig : ion ] ( solid line ) , and the exact amount of is very sensitive to the temperature structure . in our calculations , lines never become visible even if makes a deep absorption , suggesting that is a better indicator of carbon abundance in this temperature range . alternatively , the emission peak of @xmath06355 may be suppressed if si is detached from the photosphere ( see branch et al . 2002 for the case of h@xmath88 ) . this is however not the case for the objects presented here because the velocity of decreases to @xmath144 km s@xmath2 at subsequent epochs and si is abundant at the corresponding velocities . in addition to this , the fact that the emission profile of the line is well - peaked at maximum light even in lvg suggests that the suppression is related to the conditions of the outer layers . both and lines are clearly present in pre - maximum spectra . we showed that about a few percent of fe in mass fraction , including both stable fe and decay products of @xmath4ni , are required to match the observed lines ( [ sec:02bofe ] ) . lines of both ions ( and ) are correctly reproduced , which suggests that the computation of the ionization state of fe is reliable . the fact that fe is not highly abundant indicates that the photosphere resides outside the fe - rich layers . in order to distinguish stable fe ( e.g. , @xmath145fe ) from the decay product of @xmath4ni ( i.e. , @xmath4fe ) , the abundances of ni and co were quantified from the spectra ( [ sec:02bofe ] ) . in 4 of our sne , we are able to estimate the @xmath4ni abundance near the photosphere , while only upper limits could be obtained in the remaining 2 cases . if @xmath4ni accounted for all the present fe abundance at such an early epoch ( when only 3% of @xmath4ni has decayed to @xmath4fe ) , the ejecta would be dominated by @xmath4ni . this is , however , clearly inconsistent with the observed spectra and the parameters obtained from the modeling . therefore , we conclude that not only @xmath4ni , but also stable fe must exist at the photosphere . the respective abundances obtained are always larger than solar suggesting that the origin is not the progenitor but nucleosynthesis during the explosion . the metal abundances in the outer layers suggest that the abundances are not sharply stratified with velocity . even in 1d , the region where @xmath4ni is synthesized extends to si - rich layers because the temperature changes smoothly during the explosion . @xmath145fe can also be synthesized in si - rich layers ( e.g. , iwamoto et al . another explanation for fe - group elements in outer layers may be large scale mixing , as seen in recent three - dimensional deflagration models ( e.g. , gamezo et al . 2003 ; rpke & hillebrandt 2005 ; rpke et al . 2006 ) . the presence of fe - group elements in the outer layers and the diversity of their abundances and a low abundance of unburned elements may hinder the use of pre - maximum spectra as indicators of the progenitor metallicity ( lentz et al . in fact , larger diversity than predicted by model computations is seen in uv spectra ( ellis et al . this may reflect diversity in the abundances of fe - group elements in the outer layers , or in the photospheric temperature ( [ sec : hvg ] and [ sec : lvg ] ) . the difference between hvg and lvg sne can be studied comparing the properties of sne with similar decline rates ( @xmath132 ) . for an optimal comparison at a similar epoch , we again take sne 2002bo ( hvg ; @xmath52 days , @xmath146 ) and 2001el ( lvg ; @xmath147 days , @xmath148 ) as examples . the marginal detection of carbon in the emission peak of @xmath06355 in lvg sne suggests that the burning is less powerful in lvg sne than in hvg sne . it is also interesting that other sne where c was detected ( sn 1998aq , branch et al . 2003 ; and possibly sn 1994d , branch et al . 2005 ) are also lvg . in addition , we find that hvg sne do not have carbon at least up to @xmath149 km s@xmath2 . if more burning occurs in hvg than in lvg sne , the former may have more kinetic energy . this would tend to make the photospheric velocity of hvg higher . the kinetic energy difference between hvg and lvg can estimated by a simple analysis . we assume that the outermost layer are fully burned to si in hvg while half of the material at @xmath141 km s@xmath2 is oxygen and carbon in lvg . since the mass above @xmath150 km s@xmath2 is @xmath112 assuming the w7 density structure , the difference in nuclear energy release is @xmath151 erg , which is less than @xmath152 of the typical kinetic energy of sne ia models . the expected difference in photospheric velocity is however only @xmath153% . therefore , it seems unlikely that the difference between hvg and lvg can be accounted for by a different kinetic energy alone . yet , such a difference in kinetic energy can not be ruled out . alternatively , hvg sne may have more massive ejecta than lvg sne . the limiting mass of a spinning wd can exceed the chandrasekhar mass of the static wd , @xmath154 ( uenishi , nomoto & hachisu 2003 ; yoon & langer 2004 , 2005 ; domnguez et al . since the density structure of a rotating wd is flatter than that of a static one , the photospheric velocity may be larger before maximum but then evolve more rapidly . unless the wd mass is extremely large such as the 2@xmath98 suggested for sn 2003fg by howell et al . ( 2006 ) , the expansion kinetic energy would not be significantly reduced by a higher binding energy . however , there is no evidence that hvg sne tend to synthesize more @xmath4ni , which does not support this interpretation . even though the mass and the kinetic energy of the ejecta are similar in hvg and lvg , a different fe - group abundance in the outer region could make their appearance different . some amount of fe - group elements is present in the outer layers of all 6 sne ( see [ sec : fe ] ) . line velocities are expected to be higher if more fe - group elements are present in the outer layers because of the larger opacity of these elements . our results show that the near - photospheric abundance of fe - group elements is not smaller in hvg than in lvg , even though the photospheric velocity is higher in hvg by @xmath155 km s@xmath2 . this might imply that hvg sne have more fe - group elements if the metal abundance is a decreasing function of radius ( velocity ) . therefore , the abundances of fe - group elements in the outer layers can be a cause of spectral diversity , while differences in the kinetic energy or the progenitor mass are less likely . to confirm this hypothesis , the metal abundance in the spectra of lvg with @xmath156 km s@xmath2 should be investigated with earlier spectra . high - velocity features ( @xmath1 km s@xmath2 ) are seen in both h&k and in the ir triplet . except for sn 2003cg , hvfs are not present in synthetic spectra based on w7 . they can be reproduced using the dd density structure and a relatively large ca abundance . alternatively , they can be also explained by an enhancement in density of the outer layer ( @xmath157 km s@xmath2 ) of a factor of 4 from the w7 value assuming some h ( @xmath158% in mass fraction ) in this region . interaction with csm may produce the latter situation ; mixing of h into the ejecta increases the fraction by a factor of 5 - 10 as it favors recombination ( mazzali et al . 2005a ) . the profiles of the ir triplet show variations ( fig . [ fig : ca ] ) . in hvg sne ( blue lines ) , the profile is round and the high - velocity component tends to blend with the photospheric component . in lvg sne , on the other hand , the profiles are less blended because of the weakness of the photospheric component ( green lines , see also sn 2003kf in mazzali et al . this is a natural consequence of the high @xmath159 and the low @xmath8 of lvg sne . the high temperature suppresses and lines near the photosphere , and the low photospheric velocity makes the position of the photospheric component redder . the combination of these two effects results in a weaker and redder photospheric absorption , so that the high - velocity component becomes more detached in lvg sne . the strength of the high - velocity component also varies from sn to sn . as our models show , this is not a temperature effect , because the temperature in the outer region is similar in all cases ( fig . [ fig : ion ] ) . although we could reproduce the strength of the hvf increasing the ca abundance , the abundances that are required show a large spread ( of a factor of 100 ; table [ tab : abun ] ) . this diversity may be a line - of - sight effect if hvfs are caused by aspherical structures , like a torus or clumps ( tanaka et al . 2006 ) , as also suggested by the high polarization level of the feature ( wang et al . 2003 ; kasen et al . 2003 ) . the very strong , broad blue wing of the @xmath06355 absorption in sne 2002bo , 2002dj and 2001el is not perfectly explained by our synthetic spectra ( figures [ fig:02bo ] , [ fig:02dj ] and [ fig:01el ] ) . this suggests the presence of hvfs in as well as in lines , although the velocity of the blue edge of the absorption is lower than that of the lines ( figures [ fig : si ] and [ fig : ca ] ) . on the other hand , sne 2002er , 2003cg and 2003du lack this feature ( figures [ fig:02er ] , [ fig:03cg ] , and [ fig:03du ] ) . the origin of hvfs is debated ( e.g. , hatano et al . 1999 ; wang et al . 2003 ; thomas et al . 2004 ; gerardy et al . 2004 ; mazzali 2005a ; kasen & plewa 2005 ; tanaka et al . the difficulty in using the dd model is that it still requires a high ca abundance in the high - velocity layers ( @xmath160 km s@xmath2 ; table [ tab : abun ] ) despite the absence of a very strong si absorption at similar velocities . one problem with the csm interaction scenario is that it requires the accumulation of a relatively large mass . the estimated mass in the high - velocity shell is @xmath142 . in order to accumulate this mass , a mass loss rate @xmath161 is required for a wind velocity of 10 km s@xmath2 . this mass loss rate is higher than that constrained from x - ray , optical and radio observations ( e.g. , cumming et al . 1996 ; mattila et al . 2005 ; immler et al . 2006 ; hughes et al . 2007 ; panagia et al . a combination of these two scenarios ( i.e. , dd model and csm interaction ) may relax these requirements ( gerardy et al . 2004 ) . alternatively , hvfs may reflect the pre - sn abundances of the progenitor wd . in the single - degenerate scenario , the wd accretes matter from a companion star ( see nomoto et al . 1994 for a review ) . at an accrretion rate suitable for stable h shell burning ( nomoto 1982 ) , he shell flashes are rather weak ( taam 1980 ; fujimoto & sugimoto 1982 ) , and their products can accrete onto the wd . the products of he shell flashes could be ca - rich depending on conditions such as the pressure at the burning shell ( hashimoto , hanawa & sugimoto 1983 ) . after many cycles of he flashes , when the mass of the wd reaches the chandrasekhar mass , the outermost layers of the wd should consist not only of c and o but also of heavier elements . if such matter resides in the outermost layers after the explosion , it could cause hvfs . to investigate this scenario further , the growth of the wd should be studied in detail . if and hvfs are caused by the explosion itself or by csm interaction , they should correlate ( see fig . 12 in tanaka et al . in fact , this seems to be the case for sne 2002bo , 2002dj , 2001el , and 2003cg , but the correlation is not clear for sne 2002er and 2003du ( figs . [ fig : si ] and [ fig : ca ] ) . this may be useful to discriminate the origin of the hvfs . the relation between lc width and luminosity in sne ia indicates that the properties of sne ia are determined by a single dominant parameter . however , dispersion is present in the relation , and it may due to uncertainties in distance and reddening or to intrinsic properties of sne ia ( mazzali & podsiadlowski 2006 ) . understanding the origins of the dispersion is important for a more precise cosmological use of sne ia . the earliest spectra of sne ia are clearly not determined by a single parameter , e.g. , sne with a similar maximum luminosity have different line velocities . therefore , the variation in the early phase spectra could , to some extent , reveal an intrinsic dispersion in the width - luminosity relation . since the variation of properties is larger in hvg than in lvg sne ( b05 ; mazzali et al . 2007 ) , the hvg group at least could be related to the dispersion in the width - luminosity relation . a first parameter to look at in this context is the kinetic energy of the ejecta . hvg sne may have higher kinetic energy than lvg by at most @xmath162% ( [ sec : hvglvg ] ) . since the light curve width ( @xmath163 ) scales roughly as @xmath164 ( arnett 1982 ) , where @xmath165 is the kinetic energy of the ejecta , the lc width of hvg sne may become narrower than that of lvg sne . this effect should not be significant given the dependence of @xmath163 . however , an earlier lc rise , accompanied by a brighter luminosity at maximum even when the ejecta contain the same amount of @xmath4ni , is a change orthogonal to the observed width - luminsity relation . thus , differences in @xmath165 could cause an intrinsic dispersion in the width - luminosity relation . the @xmath4ni distribution could also affect the lc rise time . if hvg sne have an extended @xmath4ni distribution as suggested in [ sec : hvglvg ] , their lc may peak earlier than in the case where @xmath4ni is confined in the innermost layers because the diffusion time is shorter for the outer @xmath4ni . an early lc rise would make the peak luminosity brighter even for the same @xmath4ni mass . consequently , sne with similar @xmath132 may have different peak luminosities in certain conditions of mixing ( e.g. , mazzali & podsiadlowski 2006 , woosley et al . the outermost ejecta of type ia sne are studied by modeling very early spectra . of the 6 sne we studied , 2 ( sne 2002bo and 2002dj ) are classified as hvg , being characterized by a high photospheric velocity and a low photospheric temperature . all three lvg sne ( sne 2001el , 2003cg and 2003du ) have uniform properties and are characterized by a low photospheric velocity and a high photospheric temperature . the properties of sn 2002er , which is classified as hvg , are close to those of lvg , suggesting that there is continuity between the groups . this is consistent with suggestions by branch et al . ( 2006 , 2007 ) hvg sne have a @xmath06355 line with a pronounced emission peak and a broad blue absorption , suggesting that si is present at @xmath86 km s@xmath2 . the abundance distribution in hvg sne is similar to that of a delayed detonation model , i.e. , the burning front reaches the outermost layers . on the other hand , the @xmath06355 emission profile of lvg sne tends to be suppressed , possibly because of the presence of @xmath06578 . this suggests that the burning front in lvg sne is weaker than in hvg sne . however , the mass fraction of carbon at the photosphere of lvg sne is only @xmath66(c ) @xmath166 , and the estimated c mass is less than @xmath142 , which is less than predicted by w7 ( @xmath143 ) . the difference in the photospheric velocity explains the different appearance of hvg and lvg sne at pre - maximum phases through the temperature difference . at the highest velocities ( @xmath1 km s@xmath2 ) , however , the variety of the and features is not explained by this scenario . this may require additional factors such as asphericity or diversity in the element abundance of the progenitor wd . both stable fe and @xmath4ni are detected at the photosphere ( @xmath167 km s@xmath2 ) in all 6 sne , suggesting that fe - group elements are always present in the outer layers . the differece in their abundances may be an important reason for the diversity among sn ia spectra . on the other hand , the difference in kinetic energy between hvg and lvg sne is too small to be the origin of the spectral diversity . the diversity of sne ia seen in early phase spectra could cause the intrinsic dispersion in the lc width - luminosity relation , through the difference in kinetic energy and the amount of iron - group elements in the outer layers . m.t . is supported by the jsps ( japan society for the promotion of science ) research fellowship for young scientists . acknowledges support by the proyecto fondecyt 3070034 . this research was supported in part by the grant - in - aid for scientific research ( 18104003 , 18540231 ) and the 21st century coe program ( quest ) from the jsps and mext of japan

we construct a new set of generalized coherent states , _ the electron - hole coherent states _ , for a ( quasi-)spin particle on the infinite line . the definition is inspired by applications to the bogoliubov - de gennes equations where the quasi - spin refers to electron- and hole - like components of electronic excitations in a superconductor . electron - hole coherent states generally entangle the space and the quasi - spin degrees of freedom . we show that the electron - hole coherent states allow obtaining a resolution of unity and form minimum uncertainty states for position and velocity where the velocity operator is defined using the bogoliubov - de gennes hamiltonian . the usefulness and the limitations of electron - hole coherent states and the phase space representations built from them are discussed in terms of basic applications to the bogoliubov - de gennes equation such as andreev reflection .

1509.06936

coherent states in their multiple variants and generalizations have become an indispensable tool in various branches of quantum mechanics and quantum field theory . the standard coherent states are strongly connected to the quantum description of a one - dimensional harmonic oscillator . they were first mentioned by schrdinger @xcite , who realised that they describe wave packets with minimal uncertainty that follow the classical dynamics without changing their shape . glauber @xcite and sudarshan @xcite later rediscovered these states , gave a very detailed description of their properties and showed their relevance in quantum optics . since then they have also become an important tool in quantum field theory and a central ingredient of phase space formulations of quantum mechanics and semiclassical approximations . + the standard coherent states have been generalised along various lines . e.g. , the group - theoretic approach generalizes the fact that standard coherent states can be obtained by a group action of the heisenberg - weyl group on the harmonic oscillator ground state . generalized coherent states can then be obtained by taking a different initial state or a different group ( and a hilbert space which carries a representation of that group ) @xcite . there are various other approaches to generalize coherent states that are discussed in the literature with many interesting applications @xcite . a property shared by most generalized coherent states is that they are labeled by continuous parameters and allow obtaining a formal resolution of unity ( _ via _ a projector valued measure on the set of continuous parameters ) . + in this contribution we will consider a special case of a construction of coherent states on a hilbert space which is a tensor product @xmath0 . there is an obvious canonical construction if each of the factor spaces already comes with a set of coherent states , say @xmath1 and @xmath2 where @xmath3 and @xmath4 stands for the set of continuous parameters on that space . one may then define the coherent states on the tensor product @xmath5 as the tensor product of coherent states in each factor @xmath6 . in the following we will show that the special case where @xmath7 and @xmath8 one may define generalized coherent states that are ( for almost all values of the continuous parameters ) entangled , i.e. not product states . physically , the tensor coherent states are quite natural for describing a quantum point - particle on the line with spin @xmath9 ( a point particle in @xmath10 is described analogously by appropriate tensor - product coherent states ) . + it is however interesting to see that entangled coherent states , sharing many properties of the standard coherent states can be defined . we adopt the name _ electron - hole coherent states _ for our construction as they may be used in a natural way for electronic excitations in a superconductor or a hybrid superconducting - normalconducting device . these are described by the bogoliubov - de gennes equation and the factor @xmath11 of the hilbert space refers to electron - like and hole - like components of the wave function . in section [ can - cs ] we review the main properties of standard coherent states , @xmath12 coherent states for arbitrary spin , and tensor product states of standard coherent states with the spin coherent states . + in section [ eh - cs ] we construct electron - hole coherent states , derive their main properties , and show that they fit into the group theoretic approach to generalized coherent states if one drops the assumption that the group has to act linearly on the hilbert space . we also give a brief account of the relation of electron - hole coherent states to the bogoliubov - de gennes equation + in section [ eh - cs - app ] we look at generalisations of various coherent state representations of operators and states . we also discuss some limitations of electron - hole coherent states when trying to generalize the phase space formulations of quantum mechanics that is based on husimi functions . these limitations may be less severe in a semiclassical regime of the bogoliubov - de gennes equation . the construction of standard coherent states can be found in various textbooks . as they are one ingredient of the definition of electron - hole coherent states in section [ eh - cs ] we will give a short summary of their construction and their basic properties . we will focus mainly on properties that may be used as alternative definitions each of these has been used as a starting point for generalising the concept of coherent states . usually focussing on one property of coherent states will imply that not all properties of standard coherent states will hold in the generalisation . in section [ eh - cs ] we will analyse how these properties generalise to the electron - hole coherent states defined there ( and what limitations arise ) . one may start from the ( unique up to isomorphisms ) irreducible unitary representation of the heisenberg - weyl group @xmath13 . in this representation the group is generated by the annihilation operator @xmath14 , the creation operator @xmath15 and the identity operator @xmath16 ( if @xmath17 is a scalar we will usually write @xmath18 ) with commutation relations @xmath19 = \hat{1 } \quad\text{and } \quad \left[\hat{1},a\right]=\left[\hat{1},a^\dagger\right]=0 .\ ] ] in the unitary representation the creation and annihilation operators are hermitian conjugates of each other ( in general we will denote the hermitian conjugate of an operator @xmath20 as @xmath21 ) . + the unitary representation of the heisenberg - weyl group is obtained by taking the exponential of anti - hermitian linear combinations of the generators , i.e. a ( representation of a ) group element is of the form @xmath22 where @xmath23 and @xmath24 . the second equality can be established using the baker - campbell - hausdorff formula which also allows one to find the multiplication law @xmath25 which is indeed the multiplication law of the heisenberg - weyl group @xmath13 . one also finds @xmath26 which shows that the representation is indeed unitary . + the infinite dimensional hilbert space @xmath27 of this irreducible representation is spanned by the orthogonal ( and normalised ) number state basis ( aka fock basis ) @xmath28 ( where orthonormality implies @xmath29 ) in which the generators act as @xmath30 the last equation shows that @xmath31 is an eigenstate of the number operator @xmath32 . these states are thus the energy eigenstates the harmonic oscillator which is described by the hamiltonian @xmath33 . for @xmath34 the number state @xmath35 will be called the vacuum . + it is well known that the abstract hilbert space @xmath36 together with the representation of the heisenberg - weyl generators are isomorphic to the hilbert space of square - integrable functions on the line @xmath37 which is used to describe quantum point particle in one dimension . the solution of the harmonic oscillator provides a standard way to identify the two hilbert spaces and representations the isomorphism is however not unique and we will come back to this point . the standard coherent states on @xmath36 can now be defined by acting with elements of the heisenberg - weyl group on the vacuum @xmath38 note that @xmath39 , so by setting @xmath40 in our definition we fixed a phase convention . it is sometimes convenient to define coherent states that omit the normalization factor @xmath41 , that is @xmath42 while @xmath43 is not normalised , it is analytic as a function of @xmath3 . as a manifold the coherent states are equivalent to the complex plane @xmath44 which is equivalent to the coset space @xmath45 ( _ via _ the map @xmath46 ) . one may identify this plane with the classical phase space and the heisenberg - weyl group translates points in phase space , i.e. @xmath47 in the rest of the paper standard coherent states in the hilbert space @xmath36 will always be denoted with the letter @xmath3 ( sometimes with an additional index ) . note that the vacuum is the coherent state with @xmath48 , i.e. @xmath49 ( writing just @xmath35 does thus not lead to ambiguities ) . a general coherent state is a superposition of number states @xmath50 this implies that the probability @xmath51 to find the @xmath52-th number state in a given coherent state is poisson distributed @xmath53 . the probabilities take dominant values when @xmath52 is comparable to @xmath54 . indeed the maximal probability arises when @xmath52 is the largest integer smaller than @xmath54 . it is also straightforward to find the expectation value of the number operator @xmath55 this calculation becomes even simpler if one establishes that standard coherent states are eigenstates of the annihilation operator , @xmath56 this property can be used also as an alternative definition for standard coherent states . using the identities @xmath57 one can derive and show that a definition based on is equivalent to our definition ( up to choice of phase conventions ) . apart from the algebraic structure given so far the probably most important property of coherent states is that they form a complete set in @xmath36 that can be used in a similar way as a basis . using either the expansion in number states or the group multiplication one can establish that coherent states have a non - vanishing overlap , @xmath58 which implies that one can not build an orthonormal basis from coherent states nevertheless , the coherent states are complete ( indeed overcomplete ) as they allow obtaining the resolution of unity in terms of projectors onto coherent states @xmath59 where @xmath60 . this resolution can be derived either by formally reducing it to the resolution @xmath61 in terms of the orthonormal basis of number states or ( more elegantly but also by means of more advanced tools ) using schur s lemma from representation theory . from this identity it follows that we can calculate the trace of an operator @xmath20 as @xmath62 and that we may represent any abstract state @xmath63 in terms of an explicit function @xmath64 on the plane . the scalar product of two states is then given by @xmath65 note that @xmath66 is not an analytic function of @xmath3 . however @xmath67 turns out to be analytic in @xmath3 and this gives rise to the so - called bargmann representation ( see @xcite ) . as @xmath14 and @xmath15 are not hermitian operators they do not qualify as quantum observables . we thus introduce the formal ( dimensionless ) position and momentum operators @xmath68 which are hermitian by definition and obey the commutation relation @xmath69=\hat{1}\ .\ ] ] the operators @xmath70 , @xmath71 and @xmath16 may be used as an alternative set of generators for the heisenberg - weyl group @xmath13 . their expectation values in a coherent state are given by @xmath72 which shows that we may thing of the alpha plane as the classical phase space , such that for each point in phase space there is one coherent state with the corresponding expectation values for momentum and position . + the uncertainty of an observable @xmath20 in the quantum state @xmath73 may be measured by the variance @xmath74= \langle \psi|o^2|\psi\rangle -\langle \psi|o|\psi \rangle^2 $ ] which vanishes if @xmath20 takes a sharp value in the state @xmath73 . quantum mechanics forbids that @xmath70 and @xmath71 both take sharp values at the same time in any quantum state @xmath73 . heisenberg s uncertainty principle states @xmath75 \mathrm{var}\left[q\right ] \ge \frac{1}{4}\ ] ] for any state @xmath73 . standard coherent states obey @xmath75 = \mathrm{var}\left[q\right ] = \frac{1}{2}\ ] ] and thus minimize the uncertainty . it leads to an alternative definition of standard coherent states as the set of minimal uncertainty states with equal variances of @xmath70 and @xmath71 . the minimal uncertainty of coherent states is one of the main reasons why coherent states have become such an important tool for investigating quantum - classical correspondence in the semiclassical regime , when a coherent state deforms mildly during the dynamics for sufficiently small times . if one starts the dynamics in the coherent state @xmath76 then for sufficiently short times the state is well approximated by a trajectory @xmath77 of coherent states , where @xmath78 is a classical trajectory in phase space ( obtained from corresponding classical hamiltonian dynamics ) . if the classical dynamics is not chaotic this approximation may work for quite long times . a special case is the harmonic oscillator @xmath79 , where @xmath80 . in the position representation this leads to a gaussian function whose centre follows the classical oscillations in space without changing its shape otherwise . this is the form in which coherent states were used by schrdinger to investigate quantum - classical correspondence for the harmonic oscillator @xcite . one well - known generalisation of the standard coherent states replaces the heisenberg - weyl group by some other ( physically relevant ) group @xmath81 whose generators are observables of a quantum system . this has been developed mainly by gilmore @xcite and perelomov @xcite . in this case one takes an irreducible representation of @xmath81 on an appropriate hilbert space , chooses an appropriate reference state @xmath82 and then defines the coherent states by @xmath83 where @xmath84 is an element of the group . if the reference state is chosen appropriately one may find coherent states which have analogues of almost all main properties of the standard coherent states ( overcompleteness with an explicit resolution of unity , appropriately defined minimal uncertainties , relation to classical phase - space , etc . ) . the most prominent example of this are the @xmath12 coherent states @xcite , which are also known as spin coherent states or angular momentum coherent states . the generators of the @xmath12 group are the angular momentum or spin operators @xmath85 , @xmath86 and @xmath87 with commutation relations @xmath88=i j_3 , \quad \left [ j_2 , j_3\right]=i j_1 \quad \text{and } \left [ j_3 , j_1\right]=i j_2 .\ ] ] the irreducible representations are characterised by the half - integer number @xmath89 , such that the total angular momentum is @xmath90 and the dimension of the hilbert space is @xmath91 . in the present context only the simplest case @xmath92 is relevant , so we will focus our summary on these ( see @xcite for details ) . in that case the hilbert space @xmath93 is spanned by two orthogonal states , @xmath94 and the @xmath12 generators are represented by the ( hermitian ) pauli matrices , @xmath95 let us introduce the combinations @xmath96 note that contain a factor @xmath9 that is omitted in ( for convenience ) . almost all elements of @xmath12 are covered by the parameterisation @xmath97 where @xmath98 and @xmath99 . the @xmath12 coherent states are now defined by @xmath100 to which one should add the coherent state latexmath:[$|\beta=\infty\rangle \equiv that are analytic in @xmath4 but not normalized by @xmath102 as a manifold the coherent states form the sphere @xmath103 ( see below for more details ) , which is the appropriate classical phase space for the dynamics of an angular momentum vector ( with fixed length ) . in analogy to the standard coherent states the @xmath12 coherent states can be used as an overcomplete basis in the hilbert space @xmath104 of the irreducible @xmath12 representation with spin @xmath89 . for @xmath92 it is straightforward to evaluate the overlap @xmath105 of two coherent states and find an explicit resolution of unity as @xmath106 where the integration over @xmath44 may be considered as an intregral over the classical phase space which is a sphere . this allows us to represent states in @xmath107 as complex functions on phase space in analogy to the standard coherent states described above . + one may see that the coherent states span a manifold equivalent to a sphere by considering the expectation values @xmath108 the three expectation values @xmath109 ( @xmath110 ) build a vector in @xmath111 of unit length , i.e. @xmath112 and it is clear that mapping coherent states to points on the unit sphere with is one - to - one ( including the coherent state @xmath113 which maps to the south pole ) . + to conclude this section let us comment on the minimum uncertainty properties of @xmath12 coherent states . for the @xmath92 representation every normalized state is a coherent state ( up to an irrelevant phase ) , so coherent states are not distinguished by any minimal uncertainty . for other irreducible representations @xmath114 @xmath12 coherent states do fulfill a minimal uncertainty relation ( see @xcite ) . it is straightforward to define coherent states on the tensor product space @xmath115 by taking tensor product of coherent states in the two factors . if @xmath116 is a standard coherent state and @xmath117 is a spin coherent state then one defines a ( tensor-)product coherent states as @xmath118\ , |0\rangle \otimes|+\rangle\ . \label{product_cs_definition}\ ] ] the analytic but non - normalised variant is defined as @xmath119 . the product coherent states are associated with the group @xmath120 , which acts naturally on @xmath121 , as each factor is equipped with an appropriate irreducible representation . this means that acting with any element of @xmath120 on a product coherent state gives a product coherent state up to an additional scalar phase factor . product coherent states of this or analogous forms are often very useful in the analysis of quantum systems that involve tensor product spaces . for instance , the states @xmath122 are often a natural choice to analyse equations of the form @xmath123 where @xmath124 and @xmath125 are hermitian operators on @xmath36 lifted naturally to @xmath126 ( such standard abuse of notation will be used frequently ) . one important example in this class is the pauli equation where @xmath127 with the magnetic potential @xmath128 , and @xmath129 is the magnetic field that couples to the spin variables @xmath130 . in this case in the leading order of the semiclassical asymptotics ( formally @xmath131 ) the spin degrees of freedom decouple from the space degrees of freedom and an initial coherent state @xmath132 will approximately remain a coherent state @xmath133 such that @xmath134 follows the corresponding classical dynamics generated by @xmath135 and the spin rotates along the trajectory . it is interesting to note that is not the only way to define a continuous overcomplete basis on @xmath126 that one may build from the standard coherent states in the factor @xmath36 , as we will show in the next section . let us define _ electron - hole coherent states _ in @xmath136 as @xmath137 where @xmath138 is a standard coherent state in the factor @xmath36 . the corresponding non - normalized variant @xmath139 will also be used as many formulas turn out to be more compact than by using the normalised variants . + for the normalised states we formally add the point @xmath140 by defining @xmath141 . note that the states are generally entangled ( unless @xmath142 or @xmath140 ) . they do not seem to arise in an obvious way from a linear group action . we will show , however , that they possess many properties associated to coherent states . e.g. they form an overcomplete basis that allows obtaining a straightforward resolution of unity with associated coherent state representations of the hilbert space and its operators . we will also show that it is indeed associated with a group action though not a linear one , as will be explained in section [ sec : group_action ] . the name electron - hole coherent states refers to possible applications in superconducting systems described by the bogoliubov - de gennes equation in that setting the basis @xmath143 of the factor @xmath107 refers to electron - like and hole - like excitations in a superconductor . it is straightforward to evaluate the overlap between two different electron - hole coherent states or the overlap of a electron - hole coherent state with a product coherent state @xmath144 for the non - normalised states . the corresponding overlaps for the normalised coherent states follow straightforwardly . the electron - hole states allow obtaining the following resolutions of unity @xmath145 which can be shown straightforwardly by first integrating out @xmath4 and then applying the resolution of unity of standard coherent states . the resolution of unity allows us to represent states and operators in the overcomplete basis of electron - hole coherent states . we will explore such representations in sections and . as we have reviewed above , standard coherent states can be viewed as a mapping from classical phase space to minimal uncertainty states in a quantum hilbert space , such that expectation values of position and momentum give back the original phase space point . in this section we want to discuss the expectation values of the central quantities and see to which extent analogous properties hold for electron - hole coherent states . it is straightforward to evaluate the expectation values of the fundamental observables . writing @xmath146 where @xmath147 and @xmath148 are real one obtains @xmath149 the expectation values of @xmath71 , @xmath150 and @xmath151 do not suggests a clear connection to a phase space description . replacing the momentum operator @xmath71 by the operator @xmath152 makes the picture somewhat nicer as @xmath153 we will see later in section [ sec : bdg ] , that @xmath154 has the interpretation of a velocity operator in the context of the bogoliubov - de gennes equation . note that @xmath155 and @xmath156 . we thus regard @xmath157 as a point in the phase space spanned by position and velocity , rather than position and momentum . the expectation values of the pauli - matrices @xmath150 and @xmath151 imply @xmath158 which , in general , does not describe a unit sphere . if we allow the sphere to deform into an ellipsoid we find indeed @xmath159 altogether this implies that the phase space underlying electron - hole coherent states can be viewed as the plane @xmath160 spanned by position and velocity expectation values , and an ellipsoid of revolution with two axes of length @xmath161 and the third one of the unit length , attached at each point of the plane . let us now show that electron - hole coherent states are minimal uncertainty states with respect to position operator @xmath70 and velocity operator @xmath154 . it is straightforward to show that coherent states have uncertainties @xmath162=\mathrm{var}_{|\alpha \join \beta\rangle}[q]=\frac{1}{2}\ ] ] and thus @xmath162\mathrm{var}_{|\alpha \join \beta\rangle}[q]=\frac{1}{4}\ .\ ] ] now the standard uncertainty relation for two operators @xmath163 and @xmath164 and an arbitrary state @xmath165 in quantum mechanics reads @xmath166\mathrm{var}_{|\psi\rangle}[b]\ge \left| \langle \psi| [ a , b]|\psi \rangle \right|^2/4 $ ] . in our case @xmath167= -i \sigma_3 $ ] what results in a lower bound @xmath168 taking any value between zero and a quarter . the stricter uncertainty relation @xmath169\mathrm{var}_{|\psi\rangle}[q]\ge \frac{1}{4 } \label{vq - uncertainty}\ ] ] can be obtained by a slight modification of the standard derivation by first writing ( using @xmath170 and @xmath171 and defining @xmath172 and @xmath173 ) @xmath169\mathrm{var}_{|\psi\rangle}[q]= \langle \psi| \left(v - v\right)^2|\psi \rangle \langle \psi|\left(q - q\right)^2|\psi\rangle = \langle \psi| \left(p - v\sigma_3\right)^2|\psi \rangle \langle \psi|\left(q - q\right)^2|\psi\rangle\ .\ ] ] we may now introduce operators @xmath174 and @xmath175 and proceed as in the standard derivation , i.e. using hermiticity and schwartz s inequality we write @xmath176 now , as for any complex number @xmath177 one has @xmath178 and @xmath179| \psi \rangle=\frac{1}{2},\ ] ] and the uncertainty follows . while we have found that electron - hole coherent states obey an uncertainty relation in @xmath154 and @xmath70 , they do not obey any uncertainty relation for the quasi - spin variables @xmath130 . with @xmath180 , the uncertainty relation obeyed by spin coherent states is @xmath181 . this lower bound for the sum over uncertainties is equivalent to the upper bound @xmath182 and we have seen above that the quasi - spin expectation values do not lie on the unit sphere but on an ellipsoid inside the sphere . the failure of electron - hole coherent states to obey a minimum uncertainty relation is related to entanglement between the two factor spaces @xmath36 and @xmath107 as we will discuss next . let us first consider how quasi - spin expectation values can measure the amount of mixing for quasi - spin states in @xmath107 . for this note that all pure ( normalised ) states @xmath183 obey @xmath184 . a mixed states in @xmath107 is described by a density matrix @xmath185 where @xmath186 and @xmath187 . the expectation values of a quasi - spin matrix @xmath130 are @xmath188 without loss of generality we may assume that @xmath189 and @xmath190 are orthogonal ( as @xmath191 is hermitian ) . the state is pure if @xmath192 , that is either @xmath193 or @xmath194 . we want to show that the quantity @xmath195 equals unity if and only if the state is pure , and @xmath196 otherwise . since @xmath197 is a rotation invariant quantity we may assume @xmath198 and @xmath199 and thus obtain @xmath200 so @xmath201 if and only if @xmath193 or @xmath194 which is what we wanted to show . moreover @xmath196 for @xmath202 with a minimum at @xmath203 where @xmath204 . we can thus use @xmath197 to measure the amount of mixing in a state @xmath191 . + now let us come back to the product space @xmath205 . a pure state @xmath206 is called separable if @xmath207 with @xmath208 and @xmath209 . a state @xmath206 is called entangled if it is not separable . the product coherent states @xmath210 discussed in section [ sec : product_cs ] are seperable by definition . the electron - hole coherent states @xmath211 are separable if either @xmath142 ( and @xmath3 arbitrary ) or @xmath140 ( and @xmath3 arbitrary ) or @xmath212 ( and @xmath4 arbitray ) otherwise they are entangled . + we may measure the amount of entanglement of a pure state @xmath213 by considering the reduced density matrix @xmath214 where the trace is taken over the factor @xmath36 , so that @xmath215 is an operator in @xmath107 . it is easy to see that @xmath165 is separable if and only if @xmath216 describes a pure state . the quantity @xmath217 defined in applied to @xmath218 can thus be used as a measure of entanglement such that @xmath201 for separable states and @xmath196 for entangled states with maximal entanglement for @xmath204 . a short calculations shows @xmath219 which we have given for electron - hole coherent states in equation . for a fixed value @xmath220 electron - hole coherent states obey @xmath221 where the lower bound ( maximal entanglement ) is obtained when @xmath222 ( the upper bound is obtained for @xmath142 and @xmath140 ) . though there is no obvious group which transforms one electron - hole coherent state into another _ via a linear representation _ in hilbert space we will demonstrate here that there is a _ non - linear _ action of the product group @xmath223 . we have seen in section that the product coherent states are spanned by the action of the linear ( tensor - product ) representation of this group acting on the state @xmath224 . repeated linear actions of the group @xmath225 transform then product coherent states to product coherent states ( up to a phase factor ) . a small twist of this action will render the group action non - linear and lead to an analogous behaviour for electron - hole coherent states . let us first define an anti - unitary time - reversal operator @xmath226 by its action on an arbitrary state in @xmath227 @xmath228 where @xmath31 is the number basis in @xmath36 . anti - linearity and anti - unitarity follow straightforwardly from the definition . acting with @xmath226 on a product coherent state @xmath229 gives just the product coherent state with complex conjugate parameters @xmath3 and @xmath4 ( @xmath140 should be considered as a real number for this purpose ) . + now let us define the operator @xmath230 in @xmath227 , where @xmath231 is the identity operator in the first factor . this is neither a linear nor an anti - linear operator as it behaves linearly in one subspace but anti - linearly in the other . it is also clear from the definition that @xmath232 is its own inverse , @xmath233 moreover , it leaves the norm of any state @xmath234 invariant , @xmath235 however , in general one has @xmath236 , so @xmath232 can not be considered as a unitary operation . moreover there is no way to define a generalised adjoint to @xmath232 , such an operation would need to satisfy at least @xmath237 for two arbitrary states , and this leads to contradictions . + our main interest here stems from the fact that @xmath232 transforms product coherent states into electron - hole coherent states and _ vice versa _ , @xmath238 which implies @xmath239 \mathcal{z } |0\rangle \otimes |+\rangle.\ ] ] we thus define the action of the group @xmath240 in @xmath227 by the operator @xmath241\mathcal{z}$ ] , where @xmath242 and @xmath243 are the corresponding unitary representations of the two groups in the corresponding spaces . it is easy to check that this group action leaves the set of electron - hole coherent states invariant ( up to phase factors ) . note that the conjugation of @xmath244 with the nonlinear operator @xmath232 can no longer be written in terms of two factors acting independently in the two factor spaces . this is indeed necessary to produce the entanglement of electron - hole coherent states that we have discussed in section [ sec : entanglement ] . + finally note that group elements that are represented by a simple phase factor in the linear unitary representation may be less trivial in the nonlinear representation ( and vice versa ) . for instance @xmath245 acting directly on any state is just scalar multiplication with @xmath246 while @xmath247 \mathcal{z } = d(0,0 ) \otimes u(0,\phi)\ ] ] happens to be unitary and the angle @xmath248 has swapped from the @xmath13 to the @xmath12 part . similarly one has @xmath249 \mathcal{z } = d(0,\phi ) \otimes u(0,0)\ .\ ] ] let us now discuss some applications of electron - hole coherent states . we will focus on quasi - particle excitations in a superconductor which can be described in the hilbert space @xmath227 where the first factor carries the observables that correspond to space ( here taken to be just the line ) , and the second factor is spanned by electron - like excitations represented by the basis state @xmath250 and hole - like excitations represented by @xmath251 . quasiparticle excitations in a one - dimensional superconductor are described by the bogoliubov - de gennes equation @xcite which can be written formally as a schrdinger equation @xmath252 with the hamilton operator @xmath253 here , @xmath254 is the fermi energy and @xmath255 is the real - valued pair potential ( replacing the operator @xmath70 by a real eigenvalue @xmath147 ) that couples electron- and hole - like excitations . we have here assumed the simplest setting , generalizations such as adding a magnetic field or taking a complex - valued pair potential are straightforward to implement but will not be discussed here . note that we assume @xmath256 . + let us discuss the dynamics of the bogoliubov - de gennes equation in the heisenberg picture for either an electron - hole or a product coherent state . the heisenberg equations of motion for the fundamental operators are easily found to be @xmath257 where we have added the index @xmath258 to refer to the time dependent operators in the heisenberg pictures . we have included here the operator @xmath259 equation shows the origin of the label velocity operator that we have used before . + at time @xmath260 we start the dynamics with the standard representations of the corresponding operators ( i.e @xmath261 and so on for all other operators ) . now let the system be either in the electron - hole coherent state @xmath211 or the product coherent state @xmath210 with @xmath146 . both states are localized in configuration space at @xmath260 near @xmath147 and we will focus on how the localization properties change for short time . we will assume here that any variation of @xmath255 can be neglected over the extent of the coherent state and its vicinity and replace it by a constant @xmath262 and neglect all terms @xmath263 . using a short - hand @xmath264 for electron - hole and @xmath265 for product coherent states we then have the following expectation values at @xmath260 , @xmath266 their corresponding changes at @xmath260 are @xmath267 for an electron - hole coherent state this implies that the expectation values for short time follow a free trajectory with velocity @xmath148 in the phase space spanned by position and velocity irrespective of the parameter @xmath4 that describes the quasi - spin . for product coherent states the expectation value of the position also follows the free motion but the corresponding velocity can be anything between @xmath268 and @xmath148 depending on quasi - spin configuration . the structure of electron - hole coherent states thus seems more adapted to applications in the bogoliubov - de gennes equation . this statement is supported more profoundly by looking at the variance in position after a short time @xmath269 . for an electron - hole coherent state this variance is @xmath270= & \frac{1}{2 } & \mathrm{var}_\otimes \left[q_h(0)\right]= & \frac{1}{2}\\ \mathrm{var}_{\join } \left [ q_h(\delta t)=q+\delta t v \right]= & \frac{1}{2 } \left(1 + \delta t^2 \right ) & \mathrm{var}_\otimes \left [ q_h(\delta t ) \right]= & \frac{1}{2}\left(1 + \left(1+v^2\frac{8|\beta|^2}{(1+|\beta|^2)^2}\right ) \delta t^2 \right)\ . \end{aligned}\ ] ] thus product coherent states generally disperse much quicker than electron - hole coherent states . the latter follow the short - time dispersion for a free scalar wave packet . the origin of the quick dispersion is obvious , the electron - like component of the product coherent state moves with velocity @xmath148 while the hole - like component moves with opposite velocity @xmath268 and such that the total quasi - spinor wave packet breaks up . this does not happen for the electron - hole coherent state , where both components have the same velocity @xmath148 . + the time scale connected to quasi - spin rotations is @xmath271 . one may expect that this is the time scale where the electron - hole coherent state may change its character and , for instance , ceases to be well localised in the position - velocity phase space . a more detailed analysis of the wave packet dynamics on intermediate and longer time scales in the schrdinger picture reveals an interesting interplay between space and spinor evolutions for electron - hole coherent states where the velocity @xmath148 is close to the fermi velocity @xmath272 ( or momentum close to the fermi momentum @xmath273 which is equivalent to the fermi velocity in value as we have used units such that the mass is unity ) @xcite . the resolution of unity implies that we can use electron - hole coherent states to represent states in @xmath126 as functions of two complex parameters @xmath3 and @xmath4 @xmath274 the hilbert space spanned by such functions @xmath275 under the above map consists of functions of the form @xmath276 where @xmath277 is analytic in @xmath278 , @xmath279 analytic in @xmath3 and both integrals @xmath280 , and @xmath281 it is more convenient to use the functions @xmath282 with a rescaled scalar product @xmath283 using @xmath284 [ eq : operator_action ] one may then rewrite the bogoliubov - de gennes equation as a nonlocal partial differential equation @xmath285f_{|\psi\rangle}(\alpha,\alpha^*,\beta)\\ & \quad-\left[\beta\frac{\partial}{\partial \beta}\left(1-{\alpha}^2 + 2\alpha\frac{\partial}{\partial \alpha}-\frac{\partial^2}{\partial { \alpha}^2}\right ) \right]f_{|\psi\rangle}(\alpha,\alpha^*,\beta)\\ & \quad-\mu\left[1 - 2\beta\frac{\partial}{\partial \beta}\right]f_{|\psi\rangle}(\alpha,\alpha^*,\beta)\\ & \quad+ \delta \left[\beta\left(1-\beta \frac{\partial}{\partial \beta}\right ) + \frac{\partial}{\partial \beta}\right ] f_{|\psi\rangle}(\alpha^*,\alpha,\beta ) \end{split } \label{bdg_csrep}\ ] ] for a scalar ( complex - valued ) function on phase space ( with constant pair potential @xmath286 ) . the nonlocal character of can be seen from the different order of arguments in the last term ( where @xmath3 and @xmath278 have been interchanged ) . + this representation in terms of functions @xmath287 has similarities to the well - known bargmann representations with the drawback that @xmath287 is not an analytic function of @xmath3 . ( top figures with @xmath288 ) and @xmath289 ( bottom figures with @xmath220 ) for product coherent state @xmath290 ( left figures ) and an electron - hole coherent state @xmath291 ( with @xmath292 and @xmath293 ) . the top left ( top right ) figure is identical to the bottom right ( bottom left ) figure apart from the axis description which changes from momentum @xmath294 to velocity @xmath148 . , title="fig:",scaledwidth=45.0% ] ( top figures with @xmath288 ) and @xmath289 ( bottom figures with @xmath220 ) for product coherent state @xmath290 ( left figures ) and an electron - hole coherent state @xmath291 ( with @xmath292 and @xmath293 ) . the top left ( top right ) figure is identical to the bottom right ( bottom left ) figure apart from the axis description which changes from momentum @xmath294 to velocity @xmath148 . , title="fig:",scaledwidth=45.0% ] + ( top figures with @xmath288 ) and @xmath289 ( bottom figures with @xmath220 ) for product coherent state @xmath290 ( left figures ) and an electron - hole coherent state @xmath291 ( with @xmath292 and @xmath293 ) . the top left ( top right ) figure is identical to the bottom right ( bottom left ) figure apart from the axis description which changes from momentum @xmath294 to velocity @xmath148 . , title="fig:",scaledwidth=45.0% ] ( top figures with @xmath288 ) and @xmath289 ( bottom figures with @xmath220 ) for product coherent state @xmath290 ( left figures ) and an electron - hole coherent state @xmath291 ( with @xmath292 and @xmath293 ) . the top left ( top right ) figure is identical to the bottom right ( bottom left ) figure apart from the axis description which changes from momentum @xmath294 to velocity @xmath148 . , title="fig:",scaledwidth=45.0% ] . stationary scattering states at energy @xmath295 for an incoming electron wave are shown in terms of the reduced position - velocity phase space function @xmath289 ( left column ) , the reduced position - momentum phase space function @xmath296 ( middle column ) , and their spinor wave functions ( right column , red : absolute value squared of electron component , blue : absolute value squared of hole component ) . the scale @xmath297 is the penetration depth defined in . + upper row : @xmath298 , @xmath299 , @xmath300 ; middle row : @xmath301 , @xmath302 , @xmath300 ; middle row : @xmath303 , @xmath304 , @xmath305 . , title="fig:",scaledwidth=33.0% ] . stationary scattering states at energy @xmath295 for an incoming electron wave are shown in terms of the reduced position - velocity phase space function @xmath289 ( left column ) , the reduced position - momentum phase space function @xmath296 ( middle column ) , and their spinor wave functions ( right column , red : absolute value squared of electron component , blue : absolute value squared of hole component ) . the scale @xmath297 is the penetration depth defined in . + upper row : @xmath298 , @xmath299 , @xmath300 ; middle row : @xmath301 , @xmath302 , @xmath300 ; middle row : @xmath303 , @xmath304 , @xmath305 . , title="fig:",scaledwidth=33.0% ] . stationary scattering states at energy @xmath295 for an incoming electron wave are shown in terms of the reduced position - velocity phase space function @xmath289 ( left column ) , the reduced position - momentum phase space function @xmath296 ( middle column ) , and their spinor wave functions ( right column , red : absolute value squared of electron component , blue : absolute value squared of hole component ) . the scale @xmath297 is the penetration depth defined in . + upper row : @xmath298 , @xmath299 , @xmath300 ; middle row : @xmath301 , @xmath302 , @xmath300 ; middle row : @xmath303 , @xmath304 , @xmath305 . , title="fig:",scaledwidth=29.0% ] + . stationary scattering states at energy @xmath295 for an incoming electron wave are shown in terms of the reduced position - velocity phase space function @xmath289 ( left column ) , the reduced position - momentum phase space function @xmath296 ( middle column ) , and their spinor wave functions ( right column , red : absolute value squared of electron component , blue : absolute value squared of hole component ) . the scale @xmath297 is the penetration depth defined in . + upper row : @xmath298 , @xmath299 , @xmath300 ; middle row : @xmath301 , @xmath302 , @xmath300 ; middle row : @xmath303 , @xmath304 , @xmath305 . , title="fig:",scaledwidth=33.0% ] . stationary scattering states at energy @xmath295 for an incoming electron wave are shown in terms of the reduced position - velocity phase space function @xmath289 ( left column ) , the reduced position - momentum phase space function @xmath296 ( middle column ) , and their spinor wave functions ( right column , red : absolute value squared of electron component , blue : absolute value squared of hole component ) . the scale @xmath297 is the penetration depth defined in . + upper row : @xmath298 , @xmath299 , @xmath300 ; middle row : @xmath301 , @xmath302 , @xmath300 ; middle row : @xmath303 , @xmath304 , @xmath305 . , title="fig:",scaledwidth=33.0% ] . stationary scattering states at energy @xmath295 for an incoming electron wave are shown in terms of the reduced position - velocity phase space function @xmath289 ( left column ) , the reduced position - momentum phase space function @xmath296 ( middle column ) , and their spinor wave functions ( right column , red : absolute value squared of electron component , blue : absolute value squared of hole component ) . the scale @xmath297 is the penetration depth defined in . + upper row : @xmath298 , @xmath299 , @xmath300 ; middle row : @xmath301 , @xmath302 , @xmath300 ; middle row : @xmath303 , @xmath304 , @xmath305 . , title="fig:",scaledwidth=29.0% ] + . stationary scattering states at energy @xmath295 for an incoming electron wave are shown in terms of the reduced position - velocity phase space function @xmath289 ( left column ) , the reduced position - momentum phase space function @xmath296 ( middle column ) , and their spinor wave functions ( right column , red : absolute value squared of electron component , blue : absolute value squared of hole component ) . the scale @xmath297 is the penetration depth defined in . + upper row : @xmath298 , @xmath299 , @xmath300 ; middle row : @xmath301 , @xmath302 , @xmath300 ; middle row : @xmath303 , @xmath304 , @xmath305 . , title="fig:",scaledwidth=33.0% ] . stationary scattering states at energy @xmath295 for an incoming electron wave are shown in terms of the reduced position - velocity phase space function @xmath289 ( left column ) , the reduced position - momentum phase space function @xmath296 ( middle column ) , and their spinor wave functions ( right column , red : absolute value squared of electron component , blue : absolute value squared of hole component ) . the scale @xmath297 is the penetration depth defined in . + upper row : @xmath298 , @xmath299 , @xmath300 ; middle row : @xmath301 , @xmath302 , @xmath300 ; middle row : @xmath303 , @xmath304 , @xmath305 . , title="fig:",scaledwidth=33.0% ] . stationary scattering states at energy @xmath295 for an incoming electron wave are shown in terms of the reduced position - velocity phase space function @xmath289 ( left column ) , the reduced position - momentum phase space function @xmath296 ( middle column ) , and their spinor wave functions ( right column , red : absolute value squared of electron component , blue : absolute value squared of hole component ) . the scale @xmath297 is the penetration depth defined in . + upper row : @xmath298 , @xmath299 , @xmath300 ; middle row : @xmath301 , @xmath302 , @xmath300 ; middle row : @xmath303 , @xmath304 , @xmath305 . , title="fig:",scaledwidth=29.0% ] . stationary scattering states at energy @xmath295 for an incoming electron wave are shown in terms of the reduced position - velocity phase space function @xmath289 ( left column ) , the reduced position - momentum phase space function @xmath296 ( middle column ) , and their spinor wave functions ( right column , red : absolute value squared of electron component , blue : absolute value squared of hole component ) . the scale is given in terms of the turning point @xmath306 . + first row : @xmath307 , @xmath308 , @xmath309 ; second row : @xmath310 , @xmath305 , @xmath311 ; third row : @xmath307 , @xmath300 , @xmath312 ; fourth row : @xmath313 , @xmath300 , @xmath312 . , title="fig:",scaledwidth=33.0% ] . stationary scattering states at energy @xmath295 for an incoming electron wave are shown in terms of the reduced position - velocity phase space function @xmath289 ( left column ) , the reduced position - momentum phase space function @xmath296 ( middle column ) , and their spinor wave functions ( right column , red : absolute value squared of electron component , blue : absolute value squared of hole component ) . the scale is given in terms of the turning point @xmath306 . + first row : @xmath307 , @xmath308 , @xmath309 ; second row : @xmath310 , @xmath305 , @xmath311 ; third row : @xmath307 , @xmath300 , @xmath312 ; fourth row : @xmath313 , @xmath300 , @xmath312 . , title="fig:",scaledwidth=33.0% ] . stationary scattering states at energy @xmath295 for an incoming electron wave are shown in terms of the reduced position - velocity phase space function @xmath289 ( left column ) , the reduced position - momentum phase space function @xmath296 ( middle column ) , and their spinor wave functions ( right column , red : absolute value squared of electron component , blue : absolute value squared of hole component ) . the scale is given in terms of the turning point @xmath306 . + first row : @xmath307 , @xmath308 , @xmath309 ; second row : @xmath310 , @xmath305 , @xmath311 ; third row : @xmath307 , @xmath300 , @xmath312 ; fourth row : @xmath313 , @xmath300 , @xmath312 . , title="fig:",scaledwidth=29.0% ] + . stationary scattering states at energy @xmath295 for an incoming electron wave are shown in terms of the reduced position - velocity phase space function @xmath289 ( left column ) , the reduced position - momentum phase space function @xmath296 ( middle column ) , and their spinor wave functions ( right column , red : absolute value squared of electron component , blue : absolute value squared of hole component ) . the scale is given in terms of the turning point @xmath306 . + first row : @xmath307 , @xmath308 , @xmath309 ; second row : @xmath310 , @xmath305 , @xmath311 ; third row : @xmath307 , @xmath300 , @xmath312 ; fourth row : @xmath313 , @xmath300 , @xmath312 . , title="fig:",scaledwidth=33.0% ] . stationary scattering states at energy @xmath295 for an incoming electron wave are shown in terms of the reduced position - velocity phase space function @xmath289 ( left column ) , the reduced position - momentum phase space function @xmath296 ( middle column ) , and their spinor wave functions ( right column , red : absolute value squared of electron component , blue : absolute value squared of hole component ) . the scale is given in terms of the turning point @xmath306 . + first row : @xmath307 , @xmath308 , @xmath309 ; second row : @xmath310 , @xmath305 , @xmath311 ; third row : @xmath307 , @xmath300 , @xmath312 ; fourth row : @xmath313 , @xmath300 , @xmath312 . , title="fig:",scaledwidth=33.0% ] . stationary scattering states at energy @xmath295 for an incoming electron wave are shown in terms of the reduced position - velocity phase space function @xmath289 ( left column ) , the reduced position - momentum phase space function @xmath296 ( middle column ) , and their spinor wave functions ( right column , red : absolute value squared of electron component , blue : absolute value squared of hole component ) . the scale is given in terms of the turning point @xmath306 . + first row : @xmath307 , @xmath308 , @xmath309 ; second row : @xmath310 , @xmath305 , @xmath311 ; third row : @xmath307 , @xmath300 , @xmath312 ; fourth row : @xmath313 , @xmath300 , @xmath312 . , title="fig:",scaledwidth=29.0% ] + . stationary scattering states at energy @xmath295 for an incoming electron wave are shown in terms of the reduced position - velocity phase space function @xmath289 ( left column ) , the reduced position - momentum phase space function @xmath296 ( middle column ) , and their spinor wave functions ( right column , red : absolute value squared of electron component , blue : absolute value squared of hole component ) . the scale is given in terms of the turning point @xmath306 . + first row : @xmath307 , @xmath308 , @xmath309 ; second row : @xmath310 , @xmath305 , @xmath311 ; third row : @xmath307 , @xmath300 , @xmath312 ; fourth row : @xmath313 , @xmath300 , @xmath312 . , title="fig:",scaledwidth=33.0% ] . stationary scattering states at energy @xmath295 for an incoming electron wave are shown in terms of the reduced position - velocity phase space function @xmath289 ( left column ) , the reduced position - momentum phase space function @xmath296 ( middle column ) , and their spinor wave functions ( right column , red : absolute value squared of electron component , blue : absolute value squared of hole component ) . the scale is given in terms of the turning point @xmath306 . + first row : @xmath307 , @xmath308 , @xmath309 ; second row : @xmath310 , @xmath305 , @xmath311 ; third row : @xmath307 , @xmath300 , @xmath312 ; fourth row : @xmath313 , @xmath300 , @xmath312 . , title="fig:",scaledwidth=33.0% ] . stationary scattering states at energy @xmath295 for an incoming electron wave are shown in terms of the reduced position - velocity phase space function @xmath289 ( left column ) , the reduced position - momentum phase space function @xmath296 ( middle column ) , and their spinor wave functions ( right column , red : absolute value squared of electron component , blue : absolute value squared of hole component ) . the scale is given in terms of the turning point @xmath306 . + first row : @xmath307 , @xmath308 , @xmath309 ; second row : @xmath310 , @xmath305 , @xmath311 ; third row : @xmath307 , @xmath300 , @xmath312 ; fourth row : @xmath313 , @xmath300 , @xmath312 . , title="fig:",scaledwidth=29.0% ] + . stationary scattering states at energy @xmath295 for an incoming electron wave are shown in terms of the reduced position - velocity phase space function @xmath289 ( left column ) , the reduced position - momentum phase space function @xmath296 ( middle column ) , and their spinor wave functions ( right column , red : absolute value squared of electron component , blue : absolute value squared of hole component ) . the scale is given in terms of the turning point @xmath306 . + first row : @xmath307 , @xmath308 , @xmath309 ; second row : @xmath310 , @xmath305 , @xmath311 ; third row : @xmath307 , @xmath300 , @xmath312 ; fourth row : @xmath313 , @xmath300 , @xmath312 . , title="fig:",scaledwidth=33.0% ] . stationary scattering states at energy @xmath295 for an incoming electron wave are shown in terms of the reduced position - velocity phase space function @xmath289 ( left column ) , the reduced position - momentum phase space function @xmath296 ( middle column ) , and their spinor wave functions ( right column , red : absolute value squared of electron component , blue : absolute value squared of hole component ) . the scale is given in terms of the turning point @xmath306 . + first row : @xmath307 , @xmath308 , @xmath309 ; second row : @xmath310 , @xmath305 , @xmath311 ; third row : @xmath307 , @xmath300 , @xmath312 ; fourth row : @xmath313 , @xmath300 , @xmath312 . , title="fig:",scaledwidth=33.0% ] . stationary scattering states at energy @xmath295 for an incoming electron wave are shown in terms of the reduced position - velocity phase space function @xmath289 ( left column ) , the reduced position - momentum phase space function @xmath296 ( middle column ) , and their spinor wave functions ( right column , red : absolute value squared of electron component , blue : absolute value squared of hole component ) . the scale is given in terms of the turning point @xmath306 . + first row : @xmath307 , @xmath308 , @xmath309 ; second row : @xmath310 , @xmath305 , @xmath311 ; third row : @xmath307 , @xmath300 , @xmath312 ; fourth row : @xmath313 , @xmath300 , @xmath312 . , title="fig:",scaledwidth=29.0% ] the standard coherent states @xmath314 allow constructing a phase space representations of quantum mechanics , where a state of the system @xmath191 ( mixed or pure ) is represented by the husimi function @xmath315 which has all formal properties of a classical probability density function ( non - negative and normalised ) . analogously an operator @xmath163 is represented by its symbol @xmath316 there is a one - to - one correspondence between operators and their symbols , or quantum states and their husimi functions . the dynamics of the state and evaluation of any expectation values can be expressed entirely in terms of symbols and the husimi function ( by adding some further structure such as a non - commutative product for functions on the phase space ) . many generalizations of coherent states ( e.g. the @xmath12 coherent states ) allow similar constructions . on the tensor product space @xmath227 the product coherent states lead to an analogous one - to - one correspondence between operators on @xmath227 and their functions on the phase space @xmath317 . for an operator @xmath163 the corresponding function is again known as the symbol of @xmath163 and is defined as @xmath318 if @xmath319 is a ( possibly mixed ) state of the system then the corresponding husimi function defined via product coherent states is @xmath320 if one tries to repeat these constructions with electron - hole coherent states one finds that there is _ no one - to - one correspondence _ between operators and the corresponding ` symbols ' in electron - hole coherent states . for instance the two hermitian operators @xmath321 and @xmath322 would have the same symbol @xmath323 and this also equals the would - be symbol of the non - hermitian operators @xmath324 and @xmath325 . + analogously the ` husimi function ' @xmath326 does not give a complete description of the state . it can still be useful in order to analyse visually how a given state is distributed in the position - velocity phase space . one may compare this to the distribution in the position - momentum phase space _ via _ the corresponding husimi function @xmath327 in product coherent states ( giving a complete description ) . for this one may use the reduced phase space functions @xmath328 where the quasi - spin variable is integrated out . figure [ fig1 ] shows such reduced phase space distributions if the state is in a particular electron - hole coherent state @xmath329 . the quasi - spin position of the product state @xmath330 can not be seen in the reduced function @xmath296 in position - momentum phase space . comparing it to the reduced function @xmath289 in position - velocity phase space reveals that the original state has both an electron and a hole component . for an electron - hole coherent state @xmath331 the roles of the two reduced phase space functions is interchanged . an interesting physical phenomenon that may be analysed visually in the phase space is the andreev reflection at a boundary between a normal conducting region ( where the pair potential vanishes @xmath332 ) and a superconducting region ( @xmath333 ) . an incoming electron - like state with energy @xmath295 close to the fermi energy @xmath334 is then reflected as a hole - like state with ( almost ) opposite velocity while the momentum has hardly changed . figure [ fig2 ] shows reduced phase space functions for the stationary scattering states at energy @xmath295 from a staircase pair potential @xmath335 together with the corresponding spinor wave function . for @xmath336 the ( envelope of the ) intensity of the spinor wave function decays exponentially @xmath337 where @xmath338 for these states the reduced representation @xmath289 in the position - velocity phase space separates the incoming electron and the reflected hole amplitudes and gives a more detailed picture of the dynamics . however this does not necessarily imply that the electron - hole coherent state representation should always be used on its own . for instance in the middle row of figure [ fig2 ] parameters have been chosen such that the incoming electron has a velocity that is somewhat above the fermi velocity and the velocity of the reflected hole is ( in absolute value ) somewhat below the fermi velocity . this is well resolved in the position - velocity phase space function @xmath289 built from electron - hole coherent states but not in the position - momentum phase space function @xmath296 built on product coherent states , where the incoming and outgoing contributions overlap strongly . on the other side the same parameters also imply an apreciable electron - electron reflection where the reflected electron has opposite momentum to the incoming . this weak effect can only be seen in the position - momentum representation ( the weak stripe near @xmath339 ) . the andreev reflection from an inhomogeneous superconductor may be modelled by a pair potential of the form @xmath340 . the stationary solutions at energy @xmath295 are oscillatory for @xmath341 and decay for @xmath342 . we will choose @xmath343 and @xmath344 . then the corresponding turning point is at @xmath306 . figure [ fig3 ] shows reduced phase space distributions for stationary scattering states for an incoming electronic wave . while generally the phase space function @xmath289 in position - velocity space gives more details than the position - momentum function @xmath296 . however , the position - velocity representation does not offer nice semiclassical descriptions in terms of trajectories . as can be seen in the second row of figure [ fig3 ] these only come about in the position - momentum representation . we have constructed electron - hole coherent states as minimum uncertainty states for position and velocity for the bogoliubov - de gennes equation . we derived and analysed their main properties : they entangle space and quasi - spin degrees of freedom and they form an overcomplete set with an explicit resolution of unity . basic applications to stationary scattering in superconductors revealed both their usefulness and some limitations . in spite of these limitations we have shown that electron - hole coherent states have a potential of describing the dynamics of electron - hole excitations in a superconductor as remain localized in position - velocity phase space for a certain time . a more detailed description of the dynamics of these states may lead to new insights on well - known effects in superconductors and at normalconducting - superconducting interfaces .

we extract parameters relevant for distinguishing among single - field inflation models from the wilkinson microwave anisotropy probe ( wmap ) data set , and from a combination of the wmap data and seven other cosmic microwave background ( cmb ) experiments . we use only cmb data and perform a likelihood analysis over a grid of models including the full error covariance matrix . we find that a model with a scale - invariant scalar power spectrum ( ) , no tensor contribution , and no running of the spectral index , is within the 1- contours of both data sets . we then apply the monte carlo reconstruction technique to both data sets to generate an ensemble of inflationary potentials consistent with observations . none of the three basic classes of inflation models ( small - field , large - field , and hybrid ) are completely ruled out , although hybrid models are favored by the best - fit region . the reconstruction process indicates that a wide variety of smooth potentials for the inflaton are consistent with the data , implying that the first - year wmap result is still too crude to constrain significantly either the height or the shape of the inflaton potential . in particular , the lack of evidence for tensor fluctuations makes it impossible to constrain the energy scale of inflation . nonetheless , the data rule out a large portion of the available parameter space for inflation . for instance , we find that potentials of the form are ruled out to by the combined data set , but not by the wmap data taken alone .

hep-ph0305130

one of the fundamental ideas of modern cosmology is that there was an epoch early in the history of the universe when potential , or vacuum , energy dominated other forms of energy densities such as matter or radiation . during such a vacuum - dominated era the scale factor grew exponentially ( or nearly exponentially ) in some small time . during this phase , dubbed inflation @xcite , a small , smooth spatial region of size of order the hubble radius grew so large that it easily could encompass the comoving volume of the entire presently observable universe . if the universe underwent such a period of rapid expansion , one can understand why the observed universe is homogeneous and isotropic to such high accuracy . one of the predictions of the simplest models of inflation is a spatially flat universe , _ @xmath4 , with great precision . inflation has also become the dominant paradigm for understanding the initial conditions for structure formation and for cosmic microwave background ( cmb ) anisotropy . in the inflationary picture , primordial density and gravity - wave ( tensor ) fluctuations are created from quantum fluctuations and `` redshifted '' out of the horizon during an early period of superluminal expansion of the universe , where they are `` frozen '' as perturbations in the background metric @xcite . metric perturbations at the surface of last scattering are observable as temperature anisotropies in the cmb . the first and most impressive confirmation of the inflationary paradigm came when the cmb anisotropies were detected by the cosmic background explorer ( cobe ) satellite in 1992 @xcite . subsequently , it became clear that the measurements of the spectrum of the cmb anisotropy can provide very detailed information about fundamental cosmological parameters @xcite and other crucial parameters for particle physics . in the past few years , a number of balloon - borne and terrestrial experiments have mapped out the cmb angular anisotropies @xcite , revealing a remarkable agreement between the data and the inflationary predictions of a flat universe with a nearly scale - invariant spectrum of adiabatic primordial density perturbations , ( see _ e.g. , _ @xcite ) . despite the simplicity of the inflationary paradigm , the number of inflation models that have been proposed in the literature is enormous @xcite . this is true even if we limit ourselves to models with only one scalar field ( the _ inflaton _ ) . with the previous data on cmb anisotropies from balloon and terrestrial experiments it has been possible for the first time to place interesting constraints on the space of possible inflation models @xcite however , the quality of the data were not good enough to rule out entire classes of models . a boost along these lines has been very recently provided by the data of the wilkinson microwave anisotropy probe ( wmap ) mission , which has marked the beginning of the precision era of the cmb measurements in space @xcite . the wmap collaboration has produced a full - sky map of the angular variations in the microwave flux , in particular the cosmic microwave background , with unprecedented accuracy . wmap data support the inflationary mechanism for the generation of curvature superhorizon fluctuations and provide a strong bound on the possible admixture of isocurvature modes @xcite . furthermore , consistent with the simplest single - field models of inflation @xcite , no evidence of nongaussianity is found @xcite . the goal of this paper is to use the wmap data to discriminate among the various single - field inflationary models . to obtain some indication of the robustness of our analysis , we also consider a data set consisting of wmap augmented with several other cmb experiments . for single - field inflation models , the relevant parameter space for distinguishing among models is defined by the scalar spectral index @xmath5 , the ratio of tensor to scalar fluctuations @xmath6 , and the running of the scalar spectral index @xmath7 . we employ _ monte carlo reconstruction _ , a stochastic method for `` inverting '' observational constraints to determine an ensemble of inflationary potentials compatible with observation @xcite . in addition to encompassing a broader set of models than usually considered ( large - field , small - field , hybrid and linear models ) , monte carlo reconstruction makes it possible easily to incorporate constraints on the running of the spectral index as well as to include effects to higher order in slow roll . since studies on the implications of wmap data for inflation @xcite have already appeared , we briefly mention the different elements between our analysis and others . ( we will elaborate on these differences later . ) the wmap collaboration analysis @xcite included wmap data , additional cmb data ( cbi @xcite and acbar @xcite ) , large - scale structure data ( 2dfgrs @xcite ) , as well as lyman-@xmath8 power spectrum data @xcite . they used a markov chain monte carlo technique to explore the likelihood surface . et al . _ @xcite , considered wmap data only , but with a top - hat prior on the hubble constant @xmath9 ( @xmath10 100 @xmath9 km sec@xmath11 mpc@xmath11 ) from the hst key project @xcite . also , _ did not consider a running of the scalar spectral index . we only consider cmb data . we first analyze just the wmap results . we then analyze the wmap data set in conjunction with other cmb data sets ( boomerang-98 @xcite , maxima-1 @xcite , dasi @xcite , cbi @xcite , acbar @xcite , vsae @xcite , and archeops @xcite ) . we employ a grid of models in the likelihood analysis , which differs from the method used by the wmap team . the paper is organized as follows . in sec . ii we discuss single - field inflation and the relevant observables in more detail . in sec . iii we discuss the inflationary model space , and in sec . iv we describe the monte carlo reconstruction technique . section v describes the methods used for the cmb analysis . in sec . vi we present the constraints from the cmb anisotropy data sets . in sec . vii we present our conclusions . in this section we briefly review scalar field models of inflationary cosmology , and explain how we relate model parameters to observable quantities . inflation , in its most general sense , can be defined to be a period of accelerating cosmological expansion during which the universe evolves toward homogeneity and flatness . this acceleration is typically a result of the universe being dominated by vacuum energy , with an equation of state @xmath12 . within this broad framework , many specific models for inflation have been proposed . we limit ourselves here to models with `` normal '' gravity ( _ i.e. , _ general relativity ) and a single order parameter for the vacuum , described by a slowly rolling scalar field @xmath13 , the inflaton . these assumptions are not overly restrictive ; the most widely studied inflation models fall within this category , including linde s `` chaotic '' inflation scenario @xcite , inflation from pseudo nambu - goldstone bosons ( `` natural '' inflation @xcite ) , dilaton - like models involving exponential potentials ( power - law inflation ) , hybrid inflation @xcite , and so forth . other models , such as starobinsky s @xmath14 model @xcite and versions of extended inflation , can , through a suitable transformation , be viewed in terms of equivalent single - field models . of course in single - field models of inflation , the inflaton `` field '' need not be a fundamental field at all . also , some `` single - field '' models require auxiliary fields . hybrid inflation models @xcite , for example , require a second field to end inflation . what is significant is that the inflationary epoch be described by a single dynamical order parameter , the inflaton field . a scalar field in a cosmological background evolves with an equation of motion @xmath15 the evolution of the scale factor is given by the scalar field dominated frw equation , @xmath16,\cr \left(\ddot a \over a\right ) & & = { 8 \pi \over 3 m_{\rm pl}^2 } \left[v\left(\phi\right ) - \dot\phi^2\right ] . \label{eqbackground}\end{aligned}\ ] ] here @xmath17 is the planck mass , and we have assumed a flat friedmann - robertson - walker metric , @xmath18 where @xmath19 is the scale factor of the universe . _ inflation _ is defined to be a period of accelerated expansion , @xmath20 . a powerful way of describing the dynamics of a scalar field - dominated cosmology is to express the hubble parameter as a function of the field @xmath13 , @xmath21 , which is consistent provided @xmath13 is monotonic in time . the equations of motion become @xcite : @xmath22 ^ 2 - { 12 \pi \over m_{\rm pl}^2 } h^2(\phi ) = - { 32 \pi^2 \over m_{\rm pl}^4 } v(\phi).\label{eqbasichjequations}\end{aligned}\ ] ] these are completely equivalent to the second - order equation of motion in eq.([eqequationofmotion ] ) . the second of the above equations is referred to as the _ hamilton - jacobi _ equation , and can be written in the useful form @xmath23 = \left({8 \pi \over 3 m_{\rm pl}^2}\right ) v(\phi),\label{eqhubblehamiltonjacobi}\ ] ] where @xmath24 is defined to be @xmath25 the physical meaning of @xmath24 can be seen by expressing eq.([eqbackground ] ) as @xmath26,\ ] ] so that the condition for inflation @xmath27 is given by @xmath28 . the scale factor is given by @xmath29,\ ] ] where the number of e - folds @xmath30 is @xmath31 to create the observed flatness and homogeneity of the universe , we require many e - folds of inflation , typically @xmath32 . this figure varies somewhat with the details of the model . we can relate a comoving scale @xmath33 in the universe today to the number of e - folds @xmath30 before the end of inflation by @xcite @xmath34 here @xmath35 is the potential when the mode leaves the horizon , @xmath36 is the potential at the end of inflation , and @xmath37 is the energy density after reheating . scales of order the current horizon size exited the horizon at @xmath38 . since this number depends , for example , on the details of reheating , we will allow @xmath30 to vary within the range @xmath39 for any given model in order to consider the most general case . ( dodelson and hui have recently argued that the value of @xmath30 corresponding to the current horizon size can be no larger than 60 @xcite , so in this sense we are being more general than is necessary . ) we will frequently work within the context of the _ slow roll _ approximation @xcite , which is the assumption that the evolution of the field is dominated by drag from the cosmological expansion , so that @xmath40 and @xmath41 the equation of state of the scalar field is dominated by the potential , so that @xmath42 , and the expansion rate is approximately @xmath43 the slow roll approximation is consistent if both the slope and curvature of the potential are small , @xmath44 . in this case the parameter @xmath24 can be expressed in terms of the potential as @xmath45 we will also define a second `` slow roll parameter '' @xmath46 by : @xmath47.\end{aligned}\ ] ] slow roll is then a consistent approximation for @xmath48 . inflation models not only explain the large - scale homogeneity of the universe , but also provide a mechanism for explaining the observed level of _ inhomogeneity _ as well . during inflation , quantum fluctuations on small scales are quickly redshifted to scales much larger than the horizon size , where they are `` frozen '' as perturbations in the background metric @xcite . the metric perturbations created during inflation are of two types : scalar , or _ curvature _ perturbations , which couple to the stress - energy of matter in the universe and form the `` seeds '' for structure formation , and tensor , or gravitational wave perturbations , which do not couple to matter . both scalar and tensor perturbations contribute to cmb anisotropy . scalar fluctuations can also be interpreted as fluctuations in the density of the matter in the universe . scalar fluctuations can be quantitatively characterized by perturbations @xmath49 in the intrinsic curvature scalar . as long as the equation of state @xmath24 is slowly varying , the curvature perturbation can be shown to be @xcite @xmath50_{k = a h}.\ ] ] the fluctuation power spectrum is in general a function of wavenumber @xmath33 , and is evaluated when a given mode crosses outside the horizon during inflation , @xmath51 . outside the horizon , modes do not evolve , so the amplitude of the mode when it crosses back _ inside _ the horizon during a later radiation- or matter - dominated epoch is just its value when it left the horizon during inflation . the _ spectral index _ @xmath5 for @xmath49 is defined by @xmath52 so that a scale - invariant spectrum , in which modes have constant amplitude at horizon crossing , is characterized by @xmath53 . some inflation models predict running of the spectral index with scale @xcite or even sharp features in the power spectrum @xcite . we will consider the running of the spectral index in more detail on sec . [ secmontecarlorecon ] . instead of specifying the fluctuation amplitude directly as a function of @xmath33 , it is often convenient to specify it as a function of the number of e - folds @xmath30 before the end of inflation at which a mode crossed outside the horizon . scales of interest for current measurements of cmb anisotropy crossed outside the horizon at @xmath54 , so that @xmath49 is conventionally evaluated at @xmath55 . the power spectrum of tensor fluctuation modes is given by @xcite @xmath56_{n}.\ ] ] the ratio of tensor to scalar modes is then @xmath57 so that tensor modes are negligible for @xmath58 . tensor and scalar modes both contribute to cmb temperature anisotropy . if the contribution of tensor modes to the cmb anisotropy can be neglected , normalization to the cobe four - year data gives @xcite @xmath59 . in the next section , we will describe the predictions of various models in this parameter space . to summarize the results of the previous section , inflation generates scalar ( density ) and tensor ( gravity wave ) fluctuations which are generally well approximated by power laws : @xmath60 in the limit of slow roll , the spectral indices @xmath5 and @xmath61 vary slowly or not at all with scale . we can write the spectral indices @xmath5 and @xmath61 to lowest order in terms of the slow roll parameters @xmath24 and @xmath46 as @xcite : @xmath62 the tensor spectral index is _ not _ an independent parameter , but is proportional to the tensor / scalar ratio , given to lowest order in slow roll by @xmath63 this is known as the _ consistency relation _ for inflation . ( this relation holds only for single - field inflation , and weakens to an inequality for inflation involving multiple degrees of freedom @xcite . ) a given inflation model can therefore be described to lowest order in slow roll by three independent parameters , @xmath49 , @xmath64 , and @xmath5 . if we wish to include higher - order effects , we have a fourth parameter describing the running of the scalar spectral index , @xmath65 . the tensor / scalar ratio is frequently expressed as a ratio of their contributions to the cmb quadrupole , @xmath66 the relation between @xmath6 and ratio of amplitudes in the primordial power spectra @xmath67 depends on the background cosmology , in particular the densities of matter ( @xmath68 ) and cosmological constant ( @xmath69 ) . for the currently favored values of @xmath70 and @xmath71 , the relation is approximately @xmath72 to lowest order in slow roll . conventions for the normalization of this parameter vary widely in the literature . in particular , peiris et al . @xcite use @xmath73 . calculating the cmb fluctuations from a particular inflationary model reduces to the following basic steps : ( 1 ) from the potential , calculate @xmath24 and @xmath46 . ( 2 ) from @xmath24 , calculate @xmath30 as a function of the field @xmath13 . ( 3 ) invert @xmath74 to find @xmath75 . ( 4 ) calculate @xmath49 , @xmath5 , and @xmath76 as functions of @xmath13 , and evaluate them at @xmath77 . for the remainder of the paper , all parameters are assumed to be evaluated at @xmath78 . even restricting ourselves to a simple single - field inflation scenario , the number of models available to choose from is large @xcite . it is convenient to define a general classification scheme , or `` zoology '' for models of inflation . we divide models into three general types : _ large - field _ , _ small - field _ , and _ hybrid _ , with a fourth classification , _ linear _ models , serving as a boundary between large- and small - field . a generic single - field potential can be characterized by two independent mass scales : a `` height '' @xmath79 , corresponding to the vacuum energy density during inflation , and a `` width '' @xmath80 , corresponding to the change in the field value @xmath81 during inflation : @xmath82 different models have different forms for the function @xmath83 . the height @xmath84 is fixed by normalization , so the only free parameter is the width @xmath80 . with the normalization fixed , the relevant parameter space for distinguishing between inflation models to lowest order in slow roll is then the @xmath85 plane . ( to next order in slow - roll parameters , one must introduce the running of @xmath5 . ) different classes of models are distinguished by the value of the second derivative of the potential , or , equivalently , by the relationship between the values of the slow - roll parameters @xmath24 and @xmath46 . model @xcite and the `` dual inflation '' model @xcite are characterized by @xmath86 , but are `` small - field '' in the sense that @xmath87 , with @xmath88 and negligible tensor modes . ] each class of models has a different relationship between @xmath6 and @xmath5 . for a more detailed discussion of these relations , the reader is referred to refs.@xcite . first order in @xmath24 and @xmath46 is sufficiently accurate for the purposes of this section , and for the remainder of this section we will only work to first order . the generalization to higher order in slow roll will be discussed in sec . [ secmontecarlorecon ] . large - field models have inflaton potentials typical of `` chaotic '' inflation scenarios @xcite , in which the scalar field is displaced from the minimum of the potential by an amount usually of order the planck mass . such models are characterized by @xmath90 , and @xmath91 . the generic large - field potentials we consider are polynomial potentials @xmath92 , and exponential potentials , @xmath93 . for the case of an exponential potential , @xmath94 , the tensor / scalar ratio @xmath6 is simply related to the spectral index as @xmath95 this result is often incorrectly generalized to all slow - roll models , but is in fact characteristic _ only _ of power - law inflation . for inflation with a polynomial potential , @xmath96 , we again have @xmath97 , @xmath98 so that tensor modes are large for significantly tilted spectra . we will be particularly interested in models with @xmath99 as a test case for our ability to rule out models . for @xmath99 , the observables are given in terms of the number of e - folds @xmath30 by @xmath100 small - field models are the type of potentials that arise naturally from spontaneous symmetry breaking ( such as the original models of `` new '' inflation @xcite ) and from pseudo nambu - goldstone modes ( natural inflation @xcite ) . the field starts from near an unstable equilibrium ( taken to be at the origin ) and rolls down the potential to a stable minimum . small - field models are characterized by @xmath102 and @xmath101 . typically @xmath24 ( and hence the tensor amplitude ) is close to zero in small - field models . the generic small - field potentials we consider are of the form @xmath103 $ ] , which can be viewed as a lowest - order taylor expansion of an arbitrary potential about the origin . the cases @xmath104 and @xmath105 have very different behavior . for @xmath104 , @xmath106,\ ] ] where @xmath30 is the number of e - folds of inflation . for @xmath105 , the scalar spectral index is @xmath107 _ independent _ of @xmath6 . assuming @xmath108 results in an upper bound on @xmath6 of @xmath109 the hybrid scenario @xcite frequently appears in models which incorporate inflation into supersymmetry . in a typical hybrid inflation model , the scalar field responsible for inflation evolves toward a minimum with nonzero vacuum energy . the end of inflation arises as a result of instability in a second field . such models are characterized by @xmath90 and @xmath110 . we consider generic potentials for hybrid inflation of the form @xmath111.$ ] the field value at the end of inflation is determined by some other physics , so there is a second free parameter characterizing the models . because of this extra freedom , hybrid models fill a broad region in the @xmath85 plane ( see fig . [ figregions ] ) . there is , however , no overlap in the @xmath85 plane between hybrid inflation and other models . the distinguishing feature of many hybrid models is a _ scalar spectral index , @xmath112 . this corresponds to the case @xmath113 . hybrid models can also in principle have a red spectrum , @xmath88 . linear models , @xmath115 , live on the boundary between large - field and small - field models , with @xmath116 and @xmath117 . the spectral index and tensor / scalar ratio are related as : @xmath118 this enumeration of models is certainly not exhaustive . there are a number of single - field models that do not fit well into this scheme , for example logarithmic potentials @xmath119 typical of supersymmetry @xcite . another example is potentials with negative powers of the scalar field @xmath120 used in intermediate inflation @xcite and dynamical supersymmetric inflation @xcite . both of these cases require an auxiliary field to end inflation and are more properly categorized as hybrid models , but fall into the small - field region of the @xmath85 plane however , the three classes categorized by the relationship between the slow - roll parameters as @xmath121 ( large - field ) , @xmath122 ( small - field , linear ) , and @xmath110 ( hybrid ) , cover the entire @xmath85 plane and are in that sense complete . for large - field and @xmath123 for small - field . ] figure 1 @xcite shows the @xmath85 plane divided into regions representing the large field , small - field and hybrid cases . figure 2 shows a `` zoo plot '' of the particular potentials considered here plotted on the @xmath85 plane . in this section we describe _ monte carlo reconstruction _ , a stochastic method for `` inverting '' observational constraints to determine an ensemble of inflationary potentials compatible with observation . the method is described in more detail in refs . in addition to encompassing a broader set of models than we considered in sec . [ seczoology ] , monte carlo reconstruction allows us easily to incorporate constraints on the running of the spectral index @xmath124 as well as to include effects to higher order in slow roll . we have defined the slow roll parameters @xmath24 and @xmath46 in terms of the hubble parameter @xmath125 as @xmath126 these parameters are simply related to observables @xmath127 , and @xmath128 to first order in slow roll . ( we discuss higher order expressions for the observables below . ) taking higher derivatives of @xmath129 with respect to the field , we can define an infinite hierarchy of slow roll parameters @xcite : @xmath130,\cr { } ^\ell\lambda_{\rm h } & \equiv & \left({m_{\rm pl}^2 \over 4 \pi}\right)^\ell { \left(h'\right)^{\ell-1 } \over h^\ell } { d^{(\ell+1 ) } h \over d\phi^{(\ell + 1)}}.\end{aligned}\ ] ] here we have chosen the parameter @xmath131 to make comparison with observation convenient . it is convenient to use @xmath30 as the measure of time during inflation . as above , we take @xmath132 and @xmath133 to be the time and field value at end of inflation . therefore , @xmath30 is defined as the number of e - folds before the end of inflation , and increases as one goes _ backward _ in time ( @xmath134 ) : @xmath135 where we have chosen the sign convention that @xmath136 has the same sign as @xmath137 : @xmath138 then @xmath24 itself can be expressed in terms of @xmath129 and @xmath30 simply as , @xmath139 similarly , the evolution of the higher order parameters during inflation is determined by a set of `` flow '' equations @xcite , @xmath140 \left({}^\ell\lambda_{\rm h}\right ) + { } ^{\ell+1}\lambda_{\rm h}.\label{eqfullflowequations}\end{aligned}\ ] ] the derivative of a slow roll parameter at a given order is higher order in slow roll . a boundary condition can be specified at any point in the inflationary evolution by selecting a set of parameters @xmath141 for a given value of @xmath30 . this is sufficient to specify a `` path '' in the inflationary parameter space that specifies the background evolution of the spacetime . taken to infinite order , this set of equations completely specifies the cosmological evolution , up to the normalization of the hubble parameter @xmath129 . furthermore , such a specification is exact , with no assumption of slow roll necessary . in practice , we must truncate the expansion at finite order by assuming that the @xmath142 are all zero above some fixed value of @xmath143 . we choose initial values for the parameters at random from the following ranges : @xmath144\cr \epsilon & = & \left[0,0.8\right]\cr \sigma & = & \left[-0.5,0.5\right]\cr { } ^2\lambda_{\rm h } & = & \left[-0.05,0.05\right]\cr { } ^3\lambda_{\rm h } & = & \left[-0.025,0.025\right],\cr & \cdots&\cr { } ^{m+1}\lambda_{\rm h } & = & 0.\label{eqinitialconditions}\end{aligned}\ ] ] here the expansion is truncated to order @xmath145 by setting @xmath146 . in this case , we still generate an exact solution of the background equations , albeit one chosen from a subset of the complete space of models . this is equivalent to placing constraints on the form of the potential @xmath147 , but the constraints can be made arbitrarily weak by evaluating the expansion to higher order . for the purposes of this analysis , we choose @xmath148 . the results are not sensitive to either the choice of order @xmath145 ( as long as it is large enough ) or to the specific ranges from which the initial parameters are chosen . once we obtain a solution to the flow equations @xmath149 $ ] , we can calculate the predicted values of the tensor / scalar ratio @xmath6 , the spectral index @xmath5 , and the `` running '' of the spectral index @xmath150 . to lowest order , the relationship between the slow roll parameters and the observables is especially simple : @xmath151 , @xmath152 , and @xmath153 . to second order in slow roll , the observables are given by @xcite , @xmath154,\label{eqrsecondorder}\ ] ] for the tensor / scalar ratio , and @xmath155 for the spectral index . the constant @xmath156 , where @xmath157 is euler s constant . , given by @xmath158 . the effect of this error is significant at second order in slow roll . ] derivatives with respect to wavenumber @xmath33 can be expressed in terms of derivatives with respect to @xmath30 as @xcite @xmath159 the scale dependence of @xmath5 is then given by the simple expression @xmath160 which can be evaluated by using eq . ( [ eqnsecondorder ] ) and the flow equations . for example , for the case of @xmath161 , the observables to lowest order are @xmath162 the final result following the evaluation of a particular path in the @xmath145-dimensional `` slow roll space '' is a point in `` observable parameter space , '' i.e. , @xmath163 , corresponding to the observational prediction for that particular model . this process can be repeated for a large number of models , and used to study the attractor behavior of the inflationary dynamics . in fact , the models cluster strongly in the observable parameter space @xcite . figure [ figzooplot2 ] shows an ensemble of models generated stochastically on the @xmath164 plane , along with the predictions of the specific models considered in sec . [ seczoology ] . figure [ nnk ] shows an ensemble of models generated stochastically on the @xmath165 plane . as one can see , and contrary to what commonly believed , there are single - field models of inflation which predict a significant running of the spectral index . the same can be appreciated in fig . [ figrnk ] , where we plot an ensemble of models generated stochastically on the @xmath166 plane . the reconstruction method works as follows : 1 . specify a `` window '' of parameter space : _ e.g. , _ central values for @xmath167 , @xmath6 , or @xmath168 and their associated error bars . 2 . select a random point in slow roll space , @xmath169 $ ] , truncated at order @xmath145 in the slow roll expansion . 3 . evolve forward in time ( @xmath170 ) until either ( a ) inflation ends ( @xmath171 ) , or ( b ) the evolution reaches a late - time fixed point ( @xmath172 ) . 4 . if the evolution reaches a late - time fixed point , calculate the observables @xmath6 , @xmath173 , and @xmath174 at this point . if inflation ends , evaluate the flow equations backward @xmath30 e - folds from the end of inflation . calculate the observable parameters at that point . if the observable parameters lie within the specified window of parameter space , compute the potential and add this model to the ensemble of `` reconstructed '' potentials . 7 . repeat steps 2 through 6 until the desired number of models have been found . the condition for the end of inflation is that @xmath175 . integrating the flow equations forward in time will yield two possible outcomes . one possibility is that the condition @xmath175 may be satisfied for some finite value of @xmath30 , which defines the end of inflation . we identify this point as @xmath176 so that the primordial fluctuations are actually generated when @xmath177 . alternatively , the solution can evolve toward an inflationary attractor with @xmath178 and @xmath112 , in which case inflation never stops . in reality , inflation must stop at some point , presumably via some sort of instability , such as the `` hybrid '' inflation mechanism @xcite . here we make the simplifying assumption that the observables for such models are the values at the late - time attractor . given a path in the slow roll parameter space , the form of the potential is fixed , up to normalization @xcite . the starting point is the hamilton - jacobi equation , @xmath179.\label{eqhjpotential}\ ] ] we have @xmath180 trivially from the flow equations . in order to calculate the potential , we need to determine @xmath181 and @xmath182 . with @xmath24 known , @xmath181 can be determined by inverting the definition of @xmath24 , eq.([eqepsilonfromn ] ) . similarly , @xmath182 follows from the first hamilton - jacobi equation ( [ eqbasichjequations ] ) : @xmath183 using these equations and eq . ( [ eqhjpotential ] ) , the form of the potential can then be fully reconstructed from the numerical solution for @xmath180 . the only necessary observational input is the normalization of the hubble parameter @xmath129 , which enters the above equations as an integration constant . here we use the simple condition that the density fluctuation amplitude ( as determined by a first - order slow roll expression ) be of order @xmath184 , @xmath185 a more sophisticated treatment would perform a full normalization to the cobe cmb data @xcite . the value of the field , @xmath13 , also contains an arbitrary , additive constant . our analysis method is based on the computation of a likelihood distribution over a fixed grid of pre - computed theoretical models . we restrict our analysis to a flat , adiabatic , @xmath84-cdm model template computed with cmbfast ( @xcite ) , sampling the parameters as follows : @xmath186 , in steps of @xmath187 ; @xmath188 , in steps of @xmath189 and @xmath190 , in steps of @xmath191 . the value of the hubble constant is not an independent parameter , since : @xmath192 and we use the further prior : @xmath193 . we allow for a reionization of the intergalactic medium by varying the compton optical depth parameter @xmath194 in the range @xmath195 in steps of @xmath191 . our choice of the above parameters is motivated by big bang nucleosynthesis bounds on @xmath196 ( both from d @xcite and @xmath197 @xcite ) , from supernovae ( @xcite ) and galaxy clustering observations ( see _ e.g. , _ @xcite ) , and by the wmap temperature - polarization cross - correlation data , which indicate an optical depth @xmath198 ( @xcite ) . our choice for an upper limit of @xmath199 is very conservative respect to the maximum values expected in numerical simulations ( see e.g. , @xcite ) even in the case of non - standard reionization processes . from the grid above , we only consider models with an age of the universe in excess of 11 gyr . variations in the inflationary parameters @xmath5 , @xmath6 and @xmath200 are not computationally relevant , and for the range of values we considered they can be assumed as free parameters . the tensor spectral index @xmath201 is determined by the consistency relation . for the wmap data we use the recent temperature and cross polarization results from ref . @xcite and compute the likelihood @xmath202 for each theoretical model as explained in ref . @xcite , using the publicly available code on the lambda web site ( http://lambda.gsfc.nasa.gov/ ) . we further include the results from seven other experiments : boomerang-98 @xcite , maxima-1 @xcite , dasi @xcite , cbi @xcite , acbar @xcite , vsae @xcite , and archeops @xcite . the expected theoretical gaussian signal inside the bin @xmath203 is computed by using the publicly available window functions and lognormal prefactors as in ref.@xcite . the likelihood @xmath204 from this dataset and for a given theoretical model is defined by @xmath205 where @xmath206 is the gaussian curvature of the likelihood matrix at the peak and @xmath207 is the experimental signal in the bin . we consider @xmath208 , @xmath209 , @xmath210 , @xmath211 , @xmath211 , @xmath211 and @xmath211 gaussian distributed calibration errors for the archeops , boomerang-98 , dasi , maxima-1 , vsae , acbar , and cbi , experiments respectively and include the beam uncertainties using the analytical marginalization method presented in @xcite . we use as combined likelihood just the normalized product of the two likelihood distributions : @xmath212 . in order to constrain a set of parameters @xmath213 we marginalize over the values of the remaining `` nuisance '' parameters @xmath214 . this yields the marginalized likelihood distribution @xmath215 in the next section we will present constraints in several two - dimensional planes . to construct plots in two dimensions , we project ( not marginalize ) over the third , `` nuisance '' parameter . for example , likelihoods in the @xmath164 plane are calculated for given choice of @xmath6 and @xmath5 by using the value of @xmath216 which maximizes the likelihood function . the error contours are then plotted relative to likelihood falloffs of @xmath217 , @xmath218 and @xmath219 as appropriate for @xmath220 , @xmath221 , and @xmath3 contours of a _ three - dimensional _ gaussian . in effect we are taking the shadow of the three - dimensional error contours rather than a slice , which makes clear the relationship between the error contours and the points generated by monte carlo . all likelihoods used to constrain models are calculated relative the the full three - dimensional likelihood function . we will plot the likelihood contours obtained from our analysis on three different planes : @xmath216 vs. @xmath5 , @xmath6 vs. @xmath5 and @xmath6 vs.@xmath216 . presenting our results on these planes is useful for understanding the effects of theoretical assumptions and/or external priors . we do this in fig . [ fig_like ] for two cases : the wmap dataset alone ( left column ) , and wmap plus the additional cmb experiments boomerang-98 , maxima-1 , dasi , cbi , acbar , vsae , and archeops ( right column ) . by analyzing these different datasets we can check the consistency of the previous experiments with wmap . the dots superimposed on the likelihood contours show the models sampled by the monte carlo reconstruction . in the top row of fig . [ fig_like ] , we show the 68% , 95% , and 99% likelihood contours on the @xmath5 vs. @xmath216 plane ( refer to the end of sec . [ seccmbanalysis ] for a discussion of the method used to plot the contours . ) the pivot scale @xmath222 is @xmath223 mpc@xmath11 . as we can see , both datasets are consistent with a scale invariant @xmath0 power law spectrum with no further scale dependence ( @xmath224 ) . a degeneracy is also evident : an increase in the spectral index @xmath5 is equivalent to a negative scale dependence ( @xmath225 ) . we emphasize , however , that this beahavior depends strictly on the position of the pivot scale @xmath222 : choosing @xmath226 mpc@xmath11 would change the direction of the degeneracy . models with @xmath227 need a negative running at about the @xmath3 level . it is interesting also to note that models with lower spectral index , @xmath228 , are in better agreement with the data with a zero or positive running . for @xmath229 , the running is bounded by @xmath230 at @xmath220 . in the center row of fig . [ fig_like ] we plot the 68% , 95% , and 99% likelihood contours on the @xmath6 vs. @xmath5 plane . as we can see , the present data only weakly constrain the presence of tensor modes , although a gravity wave component is not preferred . models with @xmath231 must have a negligible tensor component , while models with @xmath232 can have @xmath6 larger than @xmath233 ( @xmath221 c.l . ) . however , as we can see from the bottom row of fig . [ fig_like ] , there is no correlation between the tensor component and the running of the scalar index . figures [ fig_potentials ] , [ fig_xsigma ] , and [ fig_xsigma_rescaled ] show a subset of 300 reconstructed potentials selected from the sampled set . note in particular the wide range of inflationary energy scales compatible with the observational constraint . this is to be expected , since there was no detection of tensor modes in the wmap data , which would be seen here as a detection of a nonzero tensor / scalar ratio @xmath6 . in addition , the _ shape _ of the inflationary potential is also not well constrained by wmap . figure [ fig_like_zoo ] shows the models sampled by the monte carlo categorized by their `` zoology '' , _ i.e. , _ whether they fit into the category of small - field , large - field , or hybrid . we see that all three types of potential are compatible with the data , although hybrid class models are preferred by the best - fit region . figures [ fig_v_zoo ] , [ fig_v_zoo_type ] , and [ fig_v_zoo_type_rescaled ] show the reconstructed potentials divided by type . perhaps the only conclusion to be drawn here is that the wmap data places no significant constraint on the shape of the inflationary potential ; many `` reasonable '' potentials are consistent with the data . however , significant portions of the observable parameter space are ruled out by wmap , and future observations can be expected to significantly tighten these constraints @xcite . for the particular example of @xmath161 , using eq . [ eqphi4obs ] we find that this choice of potential ruled out to @xmath3 only for @xmath234 for the wmap data set . this constraint is even weaker than that claimed by barger _ et al . _ @xcite , most likely because we allow for a running of the spectral index in our constraint . when augmenting the wmap data set with the data from seven other cmb experiments , the most noticeable improvement in the constraints is a better upper limit to the tensor / scalar ratio @xmath6 , which results in a slightly improved upper limit on the height of the potential . also , the width of the reconstructed potentials in planck units is somewhat less than in the case of the wmap - only constraint , showing that the additional data more strongly limit the form of the inflationary potential . the combined data rules out a potential with @xmath161 for @xmath235 to @xmath3 , effectively killing such models as observationally viable candidates for the inflaton potential . in this paper , we presented an analysis of the wmap data set with an emphasis on parameters relevant for distinguishing among the various possible models for inflation . in contrast to previous analyses , we confined ourselves to cmb data only and performed a likelihood analysis over a grid of models including the full error covariance data from the wmap satellite alone , and in conjunction with measurements from boomerang-98 , maxima-1 , dasi , cbi , acbar , vsae , and archeops . we found that the wmap data alone are consistent with a scale - invariant power spectrum , @xmath53 , with no running of the spectral index , @xmath236 . however , a great number of models indicates a compatibility of the data with a blue spectral index and a substantial negative running . this is consistent with the result published by the wmap team including data from large - scale structure measurements and the lyman-@xmath8 forest . the wmap result is also consistent with previous cmb experiments . the inclusion of previous datasets in the analysis has the effect of reducing the error bars and give a better determination of the inflationary parameters . still , no clear evidence for the running is present in the combined analysis . this result differs from the result obtained in peiris _ et al . _ in the case of combined ( wmap+cbi+acbar ) analysis , where a mild ( about @xmath237 ) evidence for running was reported . the different and more conservative method of analysis adopted here , and the larger cmb dataset used in our paper can explain this difference . in addition , we applied the monte carlo reconstruction technique to generate an ensemble of inflationary potentials consistent with observation . of the three basic classes of inflation model , small - field , large - field , and hybrid , none are conclusively ruled out , although hybrid models are favored by the best fit region . the reconstruction process indicates that a wide variety of smooth potentials for the inflaton are consistent with the data , indicating that the wmap result is too crude to significantly constrain either the height or the shape of the inflaton potential . in particular , the lack of evidence for tensor fluctuations makes it impossible to constrain the energy scale at which inflation takes place . nonetheless , wmap rules out a large portion of the available parameter space for inflation , itself a significant improvement over previous measurements . for the particular case of a potential of the form @xmath238 , wmap rules out all such potentials for @xmath234 at the @xmath3 level , which means that @xmath239 potentials are not conclusively ruled out by wmap alone . the combined data set , however , rules out @xmath239 models for @xmath235 , which kills such potentials as viable candidates for inflation . after this paper first appeared , leach and liddle also released a reanalysis of the wmap data @xcite , which is in general agreement with our results here . w.h.k . is supported by iscap and the columbia university academic quality fund . iscap gratefully acknowledges the generous support of the ohrstrom foundation . e.w.k . is supported in part by nasa grant nag5 - 10842 . we thank richard easther for helpful conversations and for the use of computer code . a. melchiorri and j. silk , astro - ph/0203200 ; s. h. hansen , g. mangano , a. melchiorri , g. miele and o. pisanti , phys . d * 65 * , 023511 ( 2002 ) , astro - ph/0105385 ; a. melchiorri , l. mersini , c. j. odman and m. trodden , astro - ph/0211522 . w. h. kinney , a. melchiorri , a. riotto , phys . d * 63 * , 023505 ( 2001 ) , astro - ph/0007375 ; s. hannestad , s. h. hansen , and f. l. villante , astropart . phys . * 17 * 375 ( 2002 ) ; s. hannestad , s. h. hansen and f. l. villante , astropart . * 16 * , 137 ( 2001 ) ; d. j. schwarz , c. a. terrero - escalante , and a. a. garcia , phys . b * 517 * , 243 ( 2001 ) ; x. wang , m. tegmark , b. jain and m. zaldarriaga , astro - ph/0212417 . p.m. garnavich et al , ap.j . letters * 493 * , l53 - 57 ( 1998 ) ; s. perlmutter et al , ap . j. * 483 * , 565 ( 1997 ) ; s. perlmutter et al ( the supernova cosmology project ) , nature * 391 * 51 ( 1998 ) ; a.g . riess et al , ap . j. * 116 * , 1009 ( 1998 ) ; models generated by monte carlo plotted on the @xmath164 plane ( black dots ) . the colored lines are the same models as in fig . 2 . for comparison with the models , points are plotted to first order in slow roll.,width=384 ] likelihood contours for wmap alone ( left column ) and wmap plus seven other experiments ( right column ) , plotted in the ( @xmath241 ) plane ( top ) , the ( @xmath242 ) plane ( center ) , and the ( @xmath243 ) plane ( bottom ) . the points represent the results of the monte carlo sampling of inflationary models consistent with @xmath220 ( red , dots ) , @xmath221 ( blue , triangles ) , and @xmath3 ( black , stars ) contours . all points in this and subsequent figures are plotted to second order in slow roll.,title="fig:",width=259 ] likelihood contours for wmap alone ( left column ) and wmap plus seven other experiments ( right column ) , plotted in the ( @xmath241 ) plane ( top ) , the ( @xmath242 ) plane ( center ) , and the ( @xmath243 ) plane ( bottom ) . the points represent the results of the monte carlo sampling of inflationary models consistent with @xmath220 ( red , dots ) , @xmath221 ( blue , triangles ) , and @xmath3 ( black , stars ) contours . all points in this and subsequent figures are plotted to second order in slow roll.,title="fig:",width=259 ] likelihood contours for wmap alone ( left column ) and wmap plus seven other experiments ( right column ) , plotted in the ( @xmath241 ) plane ( top ) , the ( @xmath242 ) plane ( center ) , and the ( @xmath243 ) plane ( bottom ) . the points represent the results of the monte carlo sampling of inflationary models consistent with @xmath220 ( red , dots ) , @xmath221 ( blue , triangles ) , and @xmath3 ( black , stars ) contours . all points in this and subsequent figures are plotted to second order in slow roll.,title="fig:",width=259 ] likelihood contours for wmap alone ( left column ) and wmap plus seven other experiments ( right column ) , plotted in the ( @xmath241 ) plane ( top ) , the ( @xmath242 ) plane ( center ) , and the ( @xmath243 ) plane ( bottom ) . the points represent the results of the monte carlo sampling of inflationary models consistent with @xmath220 ( red , dots ) , @xmath221 ( blue , triangles ) , and @xmath3 ( black , stars ) contours . all points in this and subsequent figures are plotted to second order in slow roll.,title="fig:",width=259 ] likelihood contours for wmap alone ( left column ) and wmap plus seven other experiments ( right column ) , plotted in the ( @xmath241 ) plane ( top ) , the ( @xmath242 ) plane ( center ) , and the ( @xmath243 ) plane ( bottom ) . the points represent the results of the monte carlo sampling of inflationary models consistent with @xmath220 ( red , dots ) , @xmath221 ( blue , triangles ) , and @xmath3 ( black , stars ) contours . all points in this and subsequent figures are plotted to second order in slow roll.,title="fig:",width=259 ] likelihood contours for wmap alone ( left column ) and wmap plus seven other experiments ( right column ) , plotted in the ( @xmath241 ) plane ( top ) , the ( @xmath242 ) plane ( center ) , and the ( @xmath243 ) plane ( bottom ) . the points represent the results of the monte carlo sampling of inflationary models consistent with @xmath220 ( red , dots ) , @xmath221 ( blue , triangles ) , and @xmath3 ( black , stars ) contours . all points in this and subsequent figures are plotted to second order in slow roll.,title="fig:",width=259 ] three hundred reconstructed potentials chosen from the sampling shown in fig . [ fig_like ] for the wmap dataset ( left column ) and the wmap data set plus seven other experiments ( right column ) . the potentials are color - coded according to their likelihoods : @xmath220 ( red ) @xmath221 ( blue ) , and @xmath3 ( black ) . the top figure shows the potentials with height and width plotted in units of @xmath244 , and the bottom figure shows the same potentials rescaled to all have the same height and width.,title="fig:",width=288 ] three hundred reconstructed potentials chosen from the sampling shown in fig . [ fig_like ] for the wmap dataset ( left column ) and the wmap data set plus seven other experiments ( right column ) . the potentials are color - coded according to their likelihoods : @xmath220 ( red ) @xmath221 ( blue ) , and @xmath3 ( black ) . the top figure shows the potentials with height and width plotted in units of @xmath244 , and the bottom figure shows the same potentials rescaled to all have the same height and width.,title="fig:",width=288 ] three hundred reconstructed potentials chosen from the sampling shown in fig . [ fig_like ] for the wmap dataset ( left column ) and the wmap data set plus seven other experiments ( right column ) . the potentials are color - coded according to their likelihoods : @xmath220 ( red ) @xmath221 ( blue ) , and @xmath3 ( black ) . the top figure shows the potentials with height and width plotted in units of @xmath244 , and the bottom figure shows the same potentials rescaled to all have the same height and width.,title="fig:",width=288 ] three hundred reconstructed potentials chosen from the sampling shown in fig . [ fig_like ] for the wmap dataset ( left column ) and the wmap data set plus seven other experiments ( right column ) . the potentials are color - coded according to their likelihoods : @xmath220 ( red ) @xmath221 ( blue ) , and @xmath3 ( black ) . the top figure shows the potentials with height and width plotted in units of @xmath244 , and the bottom figure shows the same potentials rescaled to all have the same height and width.,title="fig:",width=288 ] reconstructed potentials from fig . [ fig_potentials ] plotted separately by likelihood : @xmath220 ( red , top ) , @xmath221 ( blue , center ) , and @xmath3 ( black , bottom ) . the wmap results are in the left column , and wmap plus seven other experiments are in the right column.,title="fig:",width=259 ] reconstructed potentials from fig . [ fig_potentials ] plotted separately by likelihood : @xmath220 ( red , top ) , @xmath221 ( blue , center ) , and @xmath3 ( black , bottom ) . the wmap results are in the left column , and wmap plus seven other experiments are in the right column.,title="fig:",width=259 ] reconstructed potentials from fig . [ fig_potentials ] plotted separately by likelihood : @xmath220 ( red , top ) , @xmath221 ( blue , center ) , and @xmath3 ( black , bottom ) . the wmap results are in the left column , and wmap plus seven other experiments are in the right column.,title="fig:",width=259 ] reconstructed potentials from fig . [ fig_potentials ] plotted separately by likelihood : @xmath220 ( red , top ) , @xmath221 ( blue , center ) , and @xmath3 ( black , bottom ) . the wmap results are in the left column , and wmap plus seven other experiments are in the right column.,title="fig:",width=259 ] reconstructed potentials from fig . [ fig_potentials ] plotted separately by likelihood : @xmath220 ( red , top ) , @xmath221 ( blue , center ) , and @xmath3 ( black , bottom ) . the wmap results are in the left column , and wmap plus seven other experiments are in the right column.,title="fig:",width=259 ] reconstructed potentials from fig . [ fig_potentials ] plotted separately by likelihood : @xmath220 ( red , top ) , @xmath221 ( blue , center ) , and @xmath3 ( black , bottom ) . the wmap results are in the left column , and wmap plus seven other experiments are in the right column.,title="fig:",width=259 ] same as fig . [ fig_xsigma ] , but with the potentials rescaled to all have the same height and width.,title="fig:",width=259 ] same as fig . [ fig_xsigma ] , but with the potentials rescaled to all have the same height and width.,title="fig:",width=259 ] same as fig . [ fig_xsigma ] , but with the potentials rescaled to all have the same height and width.,title="fig:",width=259 ] same as fig . [ fig_xsigma ] , but with the potentials rescaled to all have the same height and width.,title="fig:",width=259 ] same as fig . [ fig_xsigma ] , but with the potentials rescaled to all have the same height and width.,title="fig:",width=259 ] same as fig . [ fig_xsigma ] , but with the potentials rescaled to all have the same height and width.,title="fig:",width=259 ] likelihood contours plotted in the ( @xmath241 ) plane ( top ) , the ( @xmath242 ) plane ( center ) , and the ( @xmath243 ) plane ( bottom ) . the points represent the results of the monte carlo sampling , color coded by model type : small - field ( red , dots ) , large - field ( green , triangles ) , and hybrid ( blue , crosses ) . the left column is wmap only while the right column is wmap plus seven other experiments , title="fig:",width=259 ] likelihood contours plotted in the ( @xmath241 ) plane ( top ) , the ( @xmath242 ) plane ( center ) , and the ( @xmath243 ) plane ( bottom ) . the points represent the results of the monte carlo sampling , color coded by model type : small - field ( red , dots ) , large - field ( green , triangles ) , and hybrid ( blue , crosses ) . the left column is wmap only while the right column is wmap plus seven other experiments , title="fig:",width=259 ] likelihood contours plotted in the ( @xmath241 ) plane ( top ) , the ( @xmath242 ) plane ( center ) , and the ( @xmath243 ) plane ( bottom ) . the points represent the results of the monte carlo sampling , color coded by model type : small - field ( red , dots ) , large - field ( green , triangles ) , and hybrid ( blue , crosses ) . the left column is wmap only while the right column is wmap plus seven other experiments , title="fig:",width=259 ] likelihood contours plotted in the ( @xmath241 ) plane ( top ) , the ( @xmath242 ) plane ( center ) , and the ( @xmath243 ) plane ( bottom ) . the points represent the results of the monte carlo sampling , color coded by model type : small - field ( red , dots ) , large - field ( green , triangles ) , and hybrid ( blue , crosses ) . the left column is wmap only while the right column is wmap plus seven other experiments , title="fig:",width=259 ] likelihood contours plotted in the ( @xmath241 ) plane ( top ) , the ( @xmath242 ) plane ( center ) , and the ( @xmath243 ) plane ( bottom ) . the points represent the results of the monte carlo sampling , color coded by model type : small - field ( red , dots ) , large - field ( green , triangles ) , and hybrid ( blue , crosses ) . the left column is wmap only while the right column is wmap plus seven other experiments , title="fig:",width=259 ] likelihood contours plotted in the ( @xmath241 ) plane ( top ) , the ( @xmath242 ) plane ( center ) , and the ( @xmath243 ) plane ( bottom ) . the points represent the results of the monte carlo sampling , color coded by model type : small - field ( red , dots ) , large - field ( green , triangles ) , and hybrid ( blue , crosses ) . the left column is wmap only while the right column is wmap plus seven other experiments , title="fig:",width=259 ] three hundred reconstructed potentials chosen from the sampling shown in fig.[fig_like ] for wmap ( left column ) and wmap plus seven other experiments ( right column ) . the potentials are color - coded according to model type : small - field ( red ) , large - field ( green ) , and hybrid ( blue ) . the top figure shows the potentials with height and width plotted in units of @xmath244 , and the bottom figure shows the same potentials rescaled to all have the same height and width.,title="fig:",width=288 ] three hundred reconstructed potentials chosen from the sampling shown in fig.[fig_like ] for wmap ( left column ) and wmap plus seven other experiments ( right column ) . the potentials are color - coded according to model type : small - field ( red ) , large - field ( green ) , and hybrid ( blue ) . the top figure shows the potentials with height and width plotted in units of @xmath244 , and the bottom figure shows the same potentials rescaled to all have the same height and width.,title="fig:",width=288 ] three hundred reconstructed potentials chosen from the sampling shown in fig.[fig_like ] for wmap ( left column ) and wmap plus seven other experiments ( right column ) . the potentials are color - coded according to model type : small - field ( red ) , large - field ( green ) , and hybrid ( blue ) . the top figure shows the potentials with height and width plotted in units of @xmath244 , and the bottom figure shows the same potentials rescaled to all have the same height and width.,title="fig:",width=288 ] three hundred reconstructed potentials chosen from the sampling shown in fig.[fig_like ] for wmap ( left column ) and wmap plus seven other experiments ( right column ) . the potentials are color - coded according to model type : small - field ( red ) , large - field ( green ) , and hybrid ( blue ) . the top figure shows the potentials with height and width plotted in units of @xmath244 , and the bottom figure shows the same potentials rescaled to all have the same height and width.,title="fig:",width=288 ] reconstructed potentials from fig . [ fig_v_zoo ] plotted separately by model type : small - field ( red , top ) , large - field ( green , center ) , and hybrid ( blue , bottom ) . the wmap results are shown in the left column while wmap plus seven other experiments are shown onthe right column.,title="fig:",width=259 ] reconstructed potentials from fig . [ fig_v_zoo ] plotted separately by model type : small - field ( red , top ) , large - field ( green , center ) , and hybrid ( blue , bottom ) . the wmap results are shown in the left column while wmap plus seven other experiments are shown onthe right column.,title="fig:",width=259 ] reconstructed potentials from fig . [ fig_v_zoo ] plotted separately by model type : small - field ( red , top ) , large - field ( green , center ) , and hybrid ( blue , bottom ) . the wmap results are shown in the left column while wmap plus seven other experiments are shown onthe right column.,title="fig:",width=259 ] reconstructed potentials from fig . [ fig_v_zoo ] plotted separately by model type : small - field ( red , top ) , large - field ( green , center ) , and hybrid ( blue , bottom ) . the wmap results are shown in the left column while wmap plus seven other experiments are shown onthe right column.,title="fig:",width=259 ] reconstructed potentials from fig . [ fig_v_zoo ] plotted separately by model type : small - field ( red , top ) , large - field ( green , center ) , and hybrid ( blue , bottom ) . the wmap results are shown in the left column while wmap plus seven other experiments are shown onthe right column.,title="fig:",width=259 ] reconstructed potentials from fig . [ fig_v_zoo ] plotted separately by model type : small - field ( red , top ) , large - field ( green , center ) , and hybrid ( blue , bottom ) . the wmap results are shown in the left column while wmap plus seven other experiments are shown onthe right column.,title="fig:",width=259 ] same as fig . [ fig_v_zoo_type ] , but with the potentials rescaled to all have the same height and width.,title="fig:",width=259 ] same as fig . [ fig_v_zoo_type ] , but with the potentials rescaled to all have the same height and width.,title="fig:",width=259 ] same as fig . [ fig_v_zoo_type ] , but with the potentials rescaled to all have the same height and width.,title="fig:",width=259 ] same as fig . [ fig_v_zoo_type ] , but with the potentials rescaled to all have the same height and width.,title="fig:",width=259 ] same as fig . [ fig_v_zoo_type ] , but with the potentials rescaled to all have the same height and width.,title="fig:",width=259 ] same as fig . [ fig_v_zoo_type ] , but with the potentials rescaled to all have the same height and width.,title="fig:",width=259 ]

for any finite index normal subgroup of a finitely presented group , we obtain some lower bounds of the rank of the first homology of ( with mod coefficients ) in terms of some invariants of and . using this , we confirm the halperin - carlsson conjecture for any free-torus actions ( is any prime ) on-dimensional finite cw - complexes and any free-torus actions on compact-manifolds .

1210.4068

let @xmath8 be the circle group , and let @xmath4 be the additive group @xmath9 where @xmath2 is a prime integer . to avoid confusion , we use @xmath10 to denote @xmath9 as a multiplicative ring ( field ) . for any topological space @xmath11 , let @xmath12 for any group @xmath1 , let @xmath13.0.2 cm .2 cm * halperin - carlsson conjecture : * if @xmath14 ( @xmath2 is a prime ) or @xmath15 can act freely and continuously on a finite cw - complex @xmath11 , then we must have @xmath16 or @xmath17 respectively .. 2 cm for convenience , we define @xmath18 the above conjecture was proposed in the middle of 1980s by s. halperin in @xcite for the torus case , and by g. carlsson in @xcite for the @xmath4-torus case . it is also called _ toral rank conjecture _ in some literature . .2 cm in the beginning , this conjecture mainly took the form of whether the existence of a free @xmath19-action on a product of spheres @xmath20 implies @xmath21 . many authors have studied this intriguing conjecture and contributed results with respect to different aspects ( see @xcite ) . the reader is referred to see a survey of such results in @xcite and @xcite . but the general case is still open .. 2 cm when @xmath11 is a general finite cw - complex , halperin - carlsson conjecture was proved in @xcite for @xmath22 in the torus and @xmath6-torus cases and @xmath23 in the @xmath4-torus ( @xmath2 is odd prime ) case . some other evidences supporting this conjecture can be found in cao - l @xcite , choi - masuda - oum @xcite , flix - oprea - tanr @xcite , kamishima - nakayama @xcite , ustinovskii @xcite and yu @xcite in various settings . .2 cm in many cases where the halperin - carlsson conjecture is confirmed , the multiplicative structure of cohomology plays an essential role in the proof . but we will see that when the dimension of @xmath11 is @xmath5 or @xmath7 , the estimate on @xmath24 is often enough for us to prove the lower bound of @xmath25 suggested by the conjecture.0.2 cm in this paper , we are mainly interested in the halperin - carlsson conjecture for free @xmath19-actions . notice that a free action of @xmath19 on an @xmath26-dimensional finite cw - complex @xmath11 may not preserve the cell structure of @xmath11 . so it is not very clear to the author whether the orbit space of the action is still an @xmath26-dimensional cw - complex ( or homotopy equivalent to an @xmath26-dimensional cw - complex ) . since we would like to understand @xmath11 as a covering space over a not too wild space , we assume that the orbit space @xmath27 is still an @xmath26-dimensional finite cw - complex . under this assumption , @xmath11 can be thought of as a regular @xmath19-covering space over an @xmath26-dimensional finite cw - complex . in addition , we do not assume @xmath11 is path - connected on priori . the main results of the paper are the following.0.2 cm [ thm : main1 ] let @xmath28 be a @xmath5-dimensional path - connected finite cw - complex . if @xmath11 is a regular covering space over @xmath28 with deck transformation group @xmath19 ( @xmath29 ) , then @xmath30 . in particular , if @xmath11 is path connected and @xmath31 , then @xmath32 must be @xmath33 or @xmath5 and , @xmath11 and @xmath28 must satisfy one of the following conditions:0.1 cm * @xmath34 , @xmath35 , 0.1 cm * @xmath34 , @xmath36 , @xmath37 , @xmath38 . 0.1 cm * @xmath39 , @xmath40 . 0.2 cm all of ( a ) , ( b ) and ( c ) can be realized by concrete free @xmath19-actions on some @xmath5-dimensional cw - complexes @xmath11 . but these conditions do not determine the homotopy type of @xmath28 and @xmath11 . for example,0.1 cm * when @xmath34 , @xmath41 , the @xmath28 and @xmath11 in ( a ) could be both @xmath42 or klein bottle;0.1 cm * when @xmath39 , @xmath36 , the @xmath28 in ( c ) may be either @xmath43 or klein bottle while @xmath11 is @xmath43 . 0.2 cm * problem : * classify all regular @xmath19-covering spaces @xmath11 over @xmath5-dimensional finite cw - complexes with @xmath44 up to homotopy equivalence . 0.2 cm [ thm : main2 ] if @xmath45 ( @xmath29 ) can act freely on a compact @xmath7-manifold @xmath46 , then @xmath46 must satisfy @xmath47 . in particular , if @xmath46 is closed , connected and @xmath48 , we must have @xmath49 and , the orbit space @xmath50 and @xmath46 must satisfy one of the following conditions:0.1 cm * @xmath34 , @xmath51 , @xmath52,0.1 cm * @xmath39 , @xmath53 , @xmath54,0.1 cm * @xmath55 , @xmath56 . 0.2 cm [ thm : main3 ] suppose @xmath19 ( @xmath29 ) acts freely on a compact manifold @xmath46 with orbit space @xmath57 . assume that the deficiency of the fundamental group of @xmath58 is at least @xmath33 . then we must have @xmath59 and @xmath60 . in particular , if @xmath46 is connected and @xmath61 , we must have @xmath62 or @xmath5 , @xmath41 , and @xmath46 and @xmath58 must satisfy one of the following conditions : * @xmath34 , @xmath41 , @xmath63.0.1 cm * @xmath39 , @xmath41 , @xmath64 @xmath65 . 0.2 cm in the settings of theorem [ thm : main2 ] and theorem [ thm : main3 ] , we observe that if @xmath19 ( @xmath29 ) can act freely on a closed connected @xmath7-manifold @xmath46 with @xmath61 , the homology groups @xmath66 agree with @xmath67 for some @xmath68 . this motivates us to ask the following question.0.2 cm * question : * if there exists a free @xmath19-action ( @xmath29 ) on a closed connected manifold @xmath46 where @xmath61 , then does @xmath66 necessarily agree with @xmath67 for some integer @xmath68?0.2 cm note that the dimension of @xmath69 may be less than the dimension of @xmath46 when @xmath41 . so far , all the examples known to the author ( including higher dimensional ones ) give positive answer to this question . moreover , we can strengthen the question by replacing the `` closed connected manifold '' by `` path - connected finite cw - complex '' and ask whether the same conclusion holds.0.2 cm the key piece of machinery used in the proof of all the above results is the following theorem . 0.2 cm * theorem [ thm : homology - estimate - h ] . * suppose a group @xmath1 admits finite presentation of deficiency @xmath70 , i.e. the deficiency of @xmath1 is at least @xmath70 . then for any prime @xmath2 and any finite index normal subgroup @xmath0 of @xmath1 with @xmath71 , @xmath72 where @xmath73 , @xmath74 is the augmentation ideal of the group ring @xmath75 $ ] . note that @xmath76 . 0.2 cm in particular , when @xmath77 , we obtain the following results.0.2 cm * theorem [ thm : homology - estimate ] . * let @xmath1 be a finitely presentable group with deficiency at least @xmath70 . then for any prime @xmath2 and any normal subgroup @xmath0 of @xmath1 with @xmath78 , @xmath79 where @xmath80 is the coefficient of @xmath81 in the polynomial @xmath82 . in particular , if @xmath83 , @xmath84 indeed , theorem [ thm : homology - estimate ] gives us a collection of inequalities , one for each integer @xmath85 between @xmath86 and @xmath87 . in practice , one chooses a proper @xmath85 to obtain the strongest possible inequality . theorem [ thm : homology - estimate - h ] and theorem [ thm : homology - estimate ] ought to be useful for the study of homology growth and subgroup growth of a group . 0.1 cm ( * ? ? ? * theorem 1.6 ) gave another set of lower bounds of @xmath88 where @xmath0 is a normal subgroup of a finitely generated group @xmath1 with @xmath78 . but it seems to the author that theorem [ thm : main1 ] and theorem [ thm : main2 ] can not be derived from ( * ? ? ? * theorem 1.6 ) . 0.2 cm the paper is organized as follows . in section [ sec2 ] , we will study finitely presented groups and their presentation complexes . we prove that any finite presentation of a group @xmath1 can be transformed to another finite presentation of @xmath1 with some special property . in section [ sec3 ] , we analyze the cellular structure of a finite - sheeted regular covering space @xmath11 over a @xmath5-dimensional finite cw - complex @xmath28 . we will see that the boundary map of the cellular chain complex of @xmath11 has a rather special form . in section [ sec4 ] , we will study a special type of square matrices whose rows and columns are indexed by all the elements of an ordered finite group @xmath89 . in particular , we will study the relations between such matrices and the group ring @xmath75 $ ] . in section [ sec5 ] , we investigate the properties of the family of integers @xmath80 and compare them with binomial coefficients . knowing these properties is very important for the proof of theorem [ thm : main1 ] and theorem [ thm : main2 ] . in section [ sec6 ] , we prove theorem [ thm : homology - estimate - h ] and theorem [ thm : homology - estimate ] . in addition , we will investigate some special cases of theorem [ thm : homology - estimate ] . in section [ sec7 ] , we prove theorem [ thm : main1 ] . in section [ sec8 ] , we prove theorem [ thm : main2 ] and theorem [ thm : main3 ] . finally , in section [ sec9 ] , we use our results to give a new proof of a proposition in @xcite . 0.2 cm to avoid confusion , we first address some conventions used in this paper.0.2 cm * conventions : * * the elements of @xmath10 are treated as integers in some occasions ( though they are different from the actual integers ) . but the meaning should be clear from the context . 0.2 cm * for a set @xmath90 , we use @xmath91 to denote the number of elements in @xmath90 . 0.2 cm suppose @xmath1 be a finitely presentable group . let @xmath92 be a finite presentation of @xmath1 . the integer @xmath93 is called the _ deficiency _ of @xmath94 . when @xmath95 , @xmath94 is called a _ balanced presentation_. the _ deficiency _ of @xmath1 is the maximum over all its finite presentations , of the deficiency of each presentation.0.2 cm any finite presentation @xmath94 canonically determines a @xmath5-dimensional cw - complex @xmath96 called the _ presentation complex _ of @xmath94 . 0.2 cm * @xmath96 has a single vertex @xmath97 , and one oriented @xmath33-cell @xmath98 attached to @xmath97 for each generator @xmath99 ( @xmath100 ) . so the @xmath33-skeleton of @xmath96 is a bouquet of @xmath26 circles attached to @xmath97.0.2 cm * @xmath96 has one oriented @xmath5-cell @xmath101 for each relator @xmath102 ( @xmath103 ) , where @xmath101 is attached to the @xmath33-skeleton of @xmath96 via a map defined by @xmath102.0.2 cm here , we think of a @xmath33-dimensional cw - complex as a @xmath5-dimensional cw - complex with no @xmath5-cells which corresponds to @xmath1 being a free group . + * fact : * @xmath104 g^p \cong h_1(k_{\mathcal{p}};{\mathbb f}_p)$ ] for any prime @xmath2 . + let @xmath105 be the cellular chain complex of @xmath96 with @xmath10-coefficients . @xmath106 @xmath107 where @xmath108 and @xmath109 are the subspaces of @xmath110 and @xmath111 spanned by @xmath112 and @xmath113 , respectively . the map @xmath114 can be represented by a matrix @xmath115 , @xmath116 , so that @xmath117 it is clear that @xmath118 and so @xmath119 . in addition , by the euler characteristic of @xmath96 , we have : @xmath120 so we get @xmath121 , and so @xmath122 it is easy to see that we can use two types of elementary transformations to turn @xmath123 into its _ smith normal form _ ( we do not require the nonzero entries in the smith normal form to be @xmath124 ) . 0.2 cm 1 . multiply one row ( or column ) of @xmath123 by a nonzero element of @xmath10 and then add it to another row ( or column),0.2 cm 2 . switch two rows ( or two columns ) of @xmath123.0.2 cm in other words , there exists a sequence of elementary transformation matrices @xmath125 and @xmath126 over @xmath10 so that @xmath127 where @xmath128 is a diagonal matrix of size @xmath129 whose diagonals are nonzero elements of @xmath10 . @xmath130 are matrices of one of the following forms : @xmath131 @xmath1320.2 cm notice that @xmath133 and @xmath134.0.2 cm in the process of turning @xmath123 into its smith normal form , we do not need to multiplies one row ( or column ) by a nonzero scalar in @xmath10 since we do not require the nonzero entries in the smith normal form to be @xmath124 . 0.2 cm [ lemma : presentation ] for any presentation @xmath135 of a group @xmath1 , there exists another presentation @xmath136 of @xmath1 , so that in the cellular chain complex @xmath137 of the presentation complex @xmath138 , the boundary map @xmath139 is represented by @xmath140 where @xmath128 is a diagonal matrix of size @xmath129 whose diagonals are some nonzero elements of @xmath10 . denote by @xmath141 the free group generated by @xmath142 . for a sequence of words @xmath143 on @xmath142 , let @xmath144 be the normal subgroup of @xmath141 generated by @xmath145 . for any @xmath146 and @xmath147 , we define @xmath148 @xmath1490.2 cm it is easy to see that @xmath150 next , we return to the discussion of presentation of @xmath1 and start from . let @xmath151 . then we have a new presentation @xmath152 of @xmath1 which has the same number of generators and relators as @xmath94 . @xmath153 let @xmath154 be the presentation complex of @xmath152 . then in the cellular chain complex @xmath155 , the map @xmath156 is represented by @xmath157.0.2 cm moreover , let @xmath158 @xmath159 @xmath160 if we replace the generators @xmath142 in the presentation by @xmath161 , replace the @xmath162 ( @xmath163 ) occurring in @xmath164 by @xmath165 and replace @xmath166 in @xmath164 by @xmath167 , we will get a new presentation of @xmath1 , denoted by @xmath168 similarly , if we replace the generators @xmath142 in the presentation by @xmath169 and change the relators accordingly , we will get a new presentation of @xmath1 , denoted by @xmath170 let @xmath171 and @xmath172 be the presentation complex of @xmath173 and @xmath174 respectively . then the boundary map @xmath175 and @xmath176 in the cellular chain complexes of @xmath171 and @xmath172 are represented by @xmath177 and @xmath178 respectively.0.2 cm by iterating the above process of transforming the presentation of @xmath1 according to the matrices @xmath125 , we get a new presentation of @xmath1 , denoted by @xmath179 where the generators @xmath180 . let @xmath138 be the presentation complex of @xmath181 . then in the cellular chain complex @xmath137 of @xmath138 , the map @xmath139 is represented by @xmath182 ( see ) . so the lemma is proved . 0.2 cm in lemma [ lemma : presentation ] , the two group presentations @xmath94 and @xmath181 of @xmath1 have the same number of generators and relators . so @xmath94 and @xmath181 have the same deficiency . this fact is important for our proof of theorem [ thm : main1 ] and theorem [ thm : main2 ] later . let @xmath89 be a finite group and @xmath11 be a regular covering space over a @xmath5-dimensional finite cw - complex @xmath28 with deck transformation @xmath89 . let @xmath183 be the covering map . up to homotopy equivalence , we can assume that @xmath28 has a single @xmath86-cell @xmath97 . let the set of @xmath33-cells of @xmath28 be @xmath184 and the set of @xmath5-cells of @xmath28 be @xmath185 . then @xmath11 has a natural cell structure induced from @xmath28 by @xmath186 which is invariant under the action of @xmath89 . 0.2 cm next , we choose 0.2 cm * a point @xmath187;0.1 cm * an oriented @xmath33-cell @xmath188 of @xmath11 attached to @xmath189 so that @xmath190 ( @xmath191);0.1 cm * an oriented @xmath5-cell @xmath192 of @xmath11 attached to @xmath189 so that @xmath193 ( @xmath194).0.2 cm then the sets of @xmath86-cells , @xmath33-cells and @xmath5-cells in @xmath11 are0.2 cm * @xmath86-cells : @xmath195 denoted by @xmath196;0.1 cm * @xmath33-cells : @xmath197 denoted by @xmath198;0.1 cm * @xmath5-cells : @xmath199 denoted by @xmath200.0.2 cm we label @xmath189 , @xmath188 and @xmath192 by @xmath201 ( the identity of @xmath89 ) , and label @xmath202 , @xmath203 and @xmath204 by @xmath205 for any @xmath206 . then every cell @xmath207 in @xmath11 is labeled by a unique element @xmath208 so that @xmath209 for all @xmath210 . 0.2 cm let @xmath211 be the cellular chain complex of @xmath11 with @xmath10-coefficients . @xmath212 @xmath213 where @xmath214 and @xmath215 are the free submodules of @xmath216 and @xmath217 generated by the set @xmath198 and @xmath200 over @xmath10 , respectively . clearly , @xmath218,\ ] ] where @xmath75 $ ] is the group ring of @xmath89 over @xmath10 . 0.2 cm for the sake of self - containness , we review some basic facts of @xmath75 $ ] below . the group ring @xmath75 $ ] is a free module over @xmath10 generated by all the elements of @xmath89 . so we can think of the elements in @xmath89 forming a basis of @xmath75 $ ] over @xmath10 , denoted by @xmath219 . then any element @xmath220 of @xmath75 $ ] can be written as @xmath2210.1 cm * conventions : * * we use @xmath222 to denote the identity @xmath223 for brevity when there is no ambiguity in the context.0.1 cm * we use @xmath224 to denote the zero element of @xmath75 $ ] to distinguish it from the scalar @xmath225 . 0.2 cm the product @xmath226 on the group ring @xmath75 $ ] is defined by @xmath227 @xmath228 the product @xmath226 is commutative if and only if @xmath89 itself is commutative.0.2 cm notice that @xmath229 for all @xmath206 . so @xmath230 can be identified with the scalar @xmath231 , and the field @xmath10 can be embedded in @xmath75 $ ] via the map @xmath232 \\ k \ , & \longmapsto\ , k\delta_e \end{aligned}\ ] ] there also exists a canonical homomorphism going the other way , called the _ augmentation_. it is the map @xmath233 \rightarrow { \mathbb f}_p$ ] , defined by @xmath234 the kernel of @xmath235 is called the _ augmentation ideal _ of @xmath75 $ ] , denoted by @xmath74 . indeed , @xmath74 is a free @xmath10-module generated by the set @xmath236 and @xmath237 = \delta_{{\mathbb f}_p}(h ) \oplus { \mathbb f}_p .\ ] ] the reader to referred to @xcite and @xcite for more information of group rings and their augmentation ideals . 0.2 cm although we can identify @xmath231 with @xmath230 , we would rather write @xmath238 instead of @xmath239 in this paper.0.2 cm there is a natural filtration of @xmath75 $ ] as following . @xmath240 \supset \delta_{{\mathbb f}_p}(h ) \supset \delta^2_{{\mathbb f}_p}(h ) \supset\cdots \supset \delta^k_{{\mathbb f}_p}(h ) \supset \delta^{k+1}_{{\mathbb f}_p}(h ) \supset \cdots\ ] ] by abuse of notation , let @xmath241 $ ] . in addition , define @xmath242 so we have @xmath243 since @xmath89 is a finite group , the filtration becomes stable after finitely many steps . in particular , @xmath244 for some @xmath85 if and only if @xmath74 is nilpotent . a well known fact ( see ( * ? ? ? * theorem 9 ) ) asserts that for any nontrivial group @xmath1 , the augmentation ideal of the group ring @xmath245 $ ] over a unital ring @xmath246 is nilpotent if and only if @xmath1 is a finite @xmath2-group and @xmath2 is nilpotent in @xmath246 . 0.2 cm next , let us see what @xmath247 looks like with respect to the @xmath89-invariant cell structure of @xmath11 . first of all , we choose a total order of all the elements of @xmath89 , denoted by @xmath248 , and call @xmath249 an _ ordered group_. then we order the basis @xmath250 of @xmath75 $ ] according to @xmath248 , too.0.2 cm a total order of elements in @xmath251 is @xmath252 where @xmath253 is the additive unit of @xmath251 . then the corresponding ordered basis of @xmath254 $ ] is @xmath255 . suppose @xmath256 note that @xmath257 may be greater than @xmath33 because the @xmath33-cells in the set @xmath258 may appear more than once in the boundary of a @xmath5-cell @xmath192 ( see example [ exam : torus]).0.2 cm then for any @xmath210 , we have @xmath259 next , we order the elements of the sets @xmath198 and @xmath200 according to the order @xmath248 on @xmath89 . then we get an ordered basis of @xmath216 and @xmath217 , denoted by @xmath260 @xmath261 @xmath262 each @xmath263 ( @xmath100 , @xmath194 ) is a @xmath264 matrix which tells us the weight of each @xmath33-cell @xmath265 in the boundary of any @xmath5-cell @xmath266 . we prefer to use row vectors in our discussion , so we define @xmath267 note that the rows and columns of each @xmath268 are both indexed by elements of @xmath89 in order @xmath248 . let @xmath269 denote the entry of @xmath268 with row index @xmath210 and column index @xmath206 , and let @xmath270 be the row of @xmath268 with index @xmath271 . we can identify @xmath270 with the element @xmath272 $ ] . then by and the definition of @xmath273 ( see ) , we have @xmath274 @xmath275 therefore , all the row vectors of @xmath268 are determined by only one row @xmath276 . we will see that this is the essential reason why theorem [ thm : main1 ] and theorem [ thm : main2 ] could hold . to make our discussion convenient , let us introduce the following notion.0.2 cm ( equivariant matrix ) let @xmath277 be a @xmath264 matrix over @xmath10 whose rows and columns are indexed by the elements of @xmath89 in a total order @xmath248 . we denote by @xmath278 the entry of @xmath277 with row index @xmath210 and column index @xmath206 , and denote by @xmath279 the row vector of @xmath277 indexed by @xmath210 . we identify each row @xmath279 with the element @xmath280 $ ] . we call @xmath277 an _ equivariant matrix indexed by @xmath249 _ if @xmath281 for any @xmath282 ( see and ) . by the above definition , each @xmath268 in is an equivariant matrix indexed by @xmath249 . we will study more properties of this kind of matrices in the next section . before that , let us see an example of regular covering space from our viewpoint.0.2 cm [ exam : torus ] let @xmath11 be the @xmath5-dimensional torus @xmath283 there is a free @xmath284 action on @xmath11 defined by @xmath285 it is easy to see that the orbit space @xmath286 is homeomorphic to @xmath287 . we can decompose @xmath28 as the union of one @xmath86-cell @xmath97 , two @xmath33-cells @xmath288 and one @xmath5-cell @xmath289 . the induced @xmath290-invariant cell decomposition on @xmath11 and a choice of labeling of cells are shown in figure [ p : torus ] . there are four @xmath86-cells @xmath291 , eight @xmath33-cells @xmath292 , @xmath293 and four @xmath5-cells @xmath294 in @xmath11 . choose a total order on @xmath290 to be @xmath295 , where @xmath253 is the additive unit of @xmath290 . then the boundary map @xmath247 is represented by the following matrix . @xmath296 @xmath297 and @xmath298 are equivariant matrices indexed by @xmath290 . they encode the weight of each @xmath33-cell in the boundary of the four @xmath5-cells @xmath294 in @xmath11 . an element @xmath299 $ ] is called _ balanced _ if @xmath220 lies in @xmath74 , otherwise @xmath220 is called _ unbalanced_. an equivariant matrix @xmath277 indexed by an ordered group @xmath249 is called _ balanced _ ( or _ unbalanced _ ) if every row of @xmath277 is a balanced ( or unbalanced ) element of @xmath75 $ ] . the following lemma is immediate from our definitions.0.2 cm for an equivariant matrix @xmath277 indexed by an ordered group @xmath249 , the following three statements are equivalent.0.1 cm * @xmath277 is balanced ( or unbalanced).0.1 cm * there exists some @xmath206 so that @xmath300 is balanced ( or unbalanced).0.1 cm * @xmath301 is balanced ( or unbalanced ) . so all the equivariant matrices indexed by @xmath249 are divided into two classes balanced vs. unbalanced.0.2 cm in the rest of the paper , we will not distinguish an equivariant matrix @xmath277 indexed by an ordered group @xmath249 and the linear transformation on @xmath75 $ ] defined by @xmath277 with respect to the ordered basis @xmath302 of @xmath75 $ ] . in addition , we think of any @xmath299 $ ] as a row vector and use @xmath303 to denote the image of @xmath220 under @xmath277.0.2 cm [ lem : v - a ] let @xmath277 be an equivariant matrix indexed by an ordered group @xmath249 . then @xmath304 for any @xmath305 $ ] . suppose @xmath306 where @xmath307 . then @xmath308 [ lem : a - send - general ] for an equivariant matrix @xmath277 indexed by an ordered group @xmath249 , the linear map on @xmath75 $ ] defined by @xmath277 preserves @xmath309 for any @xmath310 . in particular , when @xmath277 is balanced , @xmath277 maps @xmath309 into @xmath311 for any @xmath310.0.2 cm note that @xmath309 ( @xmath312 ) is linearly spanned by the following set over @xmath10 @xmath313 , \ \text{where } \ g_1,\cdots , g_k \in h\}\ ] ] note that here @xmath314 are not necessarily different . for any tuple @xmath315 of elements in @xmath89 , let @xmath316 in particular , @xmath317 for any @xmath206 and @xmath318 . by this notation , @xmath319 \ , ; \ , g_1,\cdots , g_k\in h \ } , \,\ k\geq 1.\ ] ] 0.2 cm let @xmath320 , @xmath307 . for brevity , let @xmath321 . then @xmath277 is balanced if and only if @xmath322 . @xmath323 so for any tuple @xmath324 of elements in @xmath89 , we have @xmath325 since @xmath326 , @xmath327 , so @xmath328 . this implies that @xmath277 preserves @xmath309 . 0.2 cm in particular , when @xmath277 is balanced , i.e. @xmath329 , we have @xmath330 hence @xmath277 maps @xmath309 into @xmath311 . the lemma is proved . 0.2 cm if @xmath277 is a balanced equivariant matrix indexed by an ordered group @xmath249 , then @xmath331 for all @xmath332 . by lemma [ lem : a - send - general ] , @xmath277 maps @xmath309 into @xmath311 for all @xmath332 . so we have @xmath333 in general , @xmath334 is not so easy to calculate for an arbitrary finite group @xmath89 . but when @xmath335 , the calculation has been done in @xcite.0.2 cm [ thm : jenning ] for any @xmath29 , @xmath332 , @xmath336 equals the coefficient of @xmath337 in the polynomial @xmath82 . 0.1 cm in other words , @xmath336 equals the order @xmath338 of the following set @xmath339 . @xmath340 0.2 cm obviously , @xmath341 if and only if @xmath342 , and we have @xmath343 in addition , since @xmath344 , we get @xmath345 @xmath346 in addition , let @xmath347 be a set of additive generators of @xmath19 and let @xmath253 be the additive unit . then a basis of @xmath348 is given by : @xmath349 @xmath350 0.2 cm the augmentation ideal @xmath351 of @xmath352 $ ] is also well studied ( see @xcite ) . in particular , @xmath353 s are all free abelian groups of the same rank . so @xmath354 is torsion for all @xmath312 . in this section , we investigate the properties of @xmath80 s and compare them with binomial coefficients . knowing these properties is very important for the proof of theorem [ thm : main1 ] and theorem [ thm : main2].0.2 cm it is easy to see that @xmath356 , @xmath357 for any prime @xmath2 , and @xmath358 moreover , we have the following recursive relation @xmath359 an easy way to prove this recursive relation is to understand @xmath355 as the number of different ways of putting @xmath360 indistinguishable balls into @xmath32 different bags where each bag can hold at most @xmath361 balls . if there are exactly @xmath362 balls in the first bag , then the remaining @xmath363 balls must be put in other @xmath364 bags , which gives the term @xmath365 on the right hand side of . 0.2 cm when @xmath39 , @xmath366 for @xmath367 . so we get @xmath368 notice that @xmath369 for any @xmath370 , and @xmath371 reaches its maximum at @xmath372 . in general , @xmath355 has the following properties which are similar to the properties of binomial coefficients.0.2 cm [ lem : omega - prop ] for any prime @xmath2 and any @xmath373 , @xmath29,0.1 cm * @xmath374;0.1 cm * @xmath375 for any @xmath376 and @xmath377$];0.1 cm * @xmath378 for any @xmath376 and @xmath379 \leq m \leq [ \frac{(r+1)(p-1)}{2}]$].0.1 cm * @xmath355 reaches its maximum at @xmath380 $ ] . there is a bijection between the set @xmath381 and @xmath382 which sends any @xmath383 to @xmath384 . this proves ( i).0.2 cm the formula implies that ( ii ) and ( iii ) holds for @xmath39 . assume ( ii ) and ( iii ) both hold for @xmath364 . then by , we have @xmath385 if @xmath386 $ ] , the induction hypothesis of ( ii ) and equation imply @xmath387 . if @xmath388 \leq m \leq [ \frac{r(p-1)}{2}]$ ] , the induction hypothesis of ( iii ) and imply @xmath387 . so the induction step of ( ii ) is finished . as for ( iii ) , we observe that when @xmath379 \leq m \leq [ \frac{(r+1)(p-1)}{2}]$],0.1 cm @xmath3890.2 cm * when @xmath390 $ ] , i.e. @xmath391 $ ] , @xmath392 * when @xmath393 $ ] , since @xmath394 , @xmath395 so when @xmath379 \leq m \leq [ \frac{(r+1)(p-1)}{2}]$ ] , we have @xmath396 for each @xmath397 . using this , we pair @xmath398 with @xmath399 in the right hand side of and get @xmath400 this finishes the induction step of ( iii ) . finally , ( iv ) follows easily from ( i)(iii ) . so the lemma is proved . 0.2 cm to have some idea of how fast @xmath401 increases with respect to @xmath360 , let us compare @xmath401 with binomial coefficients . but the formula of @xmath401 is an alternating sum , which is not convenient for our purpose . so we give another way to compute @xmath401 as following . let @xmath402 and @xmath403 . then @xmath404 where @xmath405 is the @xmath360-th derivative of @xmath406 . to write a concise formula of @xmath405 , let us first introduce some notations . let @xmath407 be the set of all partitions of @xmath360 into positive integers no more than @xmath361 . any element of @xmath407 can be represented by an integral vector @xmath408 where @xmath409 , @xmath410 and @xmath411 . @xmath412 @xmath413 @xmath414 where @xmath415 is the @xmath85-th derivative of @xmath416 . by the fact that @xmath417 when @xmath418 , we can show ( by inductively taking the derivative of @xmath406 using leibniz s rule ) @xmath419 since @xmath420 , @xmath421 , we have @xmath422 . so we get @xmath423 @xmath424 [ lem : omega - prop-2 ] for any prime @xmath2 and any @xmath425 , @xmath426 moreover , the above equality holds only when @xmath427 . we can embed @xmath428 into @xmath407 by mapping @xmath429 to @xmath430 @xmath431 this map is obviously injective . note that @xmath432 , @xmath433 @xmath434 since @xmath435 , we have @xmath436 . so @xmath437 then by , we get @xmath438 in particular , when @xmath439 , the partition @xmath440 can not be written as @xmath441 for any @xmath442 , which implies @xmath443 in the above argument . so if @xmath444 , @xmath360 must equal @xmath33 . 0.2 cm 0.2 cm in this section , we will prove the following theorem which is the driving force behind the proofs of all the main results of this paper.0.2 cm [ thm : homology - estimate - h ] suppose a group @xmath1 admits finite presentation of deficiency @xmath70 , i.e. the deficiency of @xmath1 is at least @xmath70 . then for any prime @xmath2 and any finite index normal subgroup @xmath0 of @xmath1 with @xmath71 , @xmath445 where @xmath73 , @xmath74 is the augmentation ideal of the group ring @xmath75 $ ] . note that @xmath76 . let @xmath446 be a deficiency-@xmath70 presentation of @xmath1 . let @xmath28 be the presentation complex of @xmath94 . then @xmath447.0.2 cm the euler characteristic of @xmath28 is @xmath448 @xmath449 by lemma [ lemma : presentation ] , we can assume that in the cellular chain complex of @xmath28 , @xmath450 where the map @xmath451 is represented by @xmath452 . so all the entries of @xmath453 are @xmath86 except @xmath454 , @xmath455 . 0.2 cm let @xmath97 be the single @xmath86-cell in @xmath28 and let the set of @xmath33-cells and @xmath5-cells of @xmath28 be @xmath184 and @xmath456 . 0.2 cm for a finite index normal subgroup @xmath0 of @xmath1 with @xmath71 , let @xmath457 be the covering space of @xmath28 determined by @xmath0 . we label the cells of @xmath457 in the same way as we do for @xmath11 in section [ sec3 ] . notice that @xmath458 in addition , we choose a total order @xmath248 of the elements of @xmath89 ( the choice of the total order is not essential for our proof ) . 0.2 cm by the discussion in section [ sec3 ] , the map @xmath459 in the cellular chain complex @xmath460 is represented by a block matrix @xmath461 with respect to the basis @xmath462 and @xmath463 of @xmath464 and @xmath465 . @xmath466 each block @xmath268 in @xmath461 is an equivariant matrix indexed by the ordered group @xmath249 which tells us the weight of each @xmath33-cell @xmath467 in the boundary of any @xmath5-cell @xmath204 . notice that @xmath268 is balanced if and only if the sum of the weight of all the @xmath33-cells from @xmath468 in the boundary of the @xmath5-cell @xmath192 is a multiple of @xmath2 , which is equivalent to say @xmath469 . so for any @xmath470 , @xmath100 , @xmath471 0.2 cm now we think of @xmath461 as a linear map @xmath472\oplus \cdots \oplus { \mathbb f}_p[h ] } } \ , \longrightarrow \ , \overset{n}{\overbrace{{\mathbb f}_p[h]\oplus \cdots \oplus { \mathbb f}_p[h]}},\ ] ] where @xmath268 is a linear map from the @xmath362-th copy of @xmath75 $ ] on the left of to the @xmath473-th copy of @xmath75 $ ] on the right . although it is difficult to compute the @xmath474 precisely , we can get some lower bound of @xmath475 as following.0.2 cm note that when @xmath268 is balanced , it maps @xmath309 into @xmath311 ; and when @xmath268 is unbalanced , it still maps @xmath309 into @xmath309 for all @xmath332 ( see lemma [ lem : a - send - general ] ) . and by and our assumption on @xmath453 , @xmath476 ( @xmath455 ) is unbalanced while all other @xmath268 s are balanced . so in particular , @xmath477 we claim that : for all @xmath332 , @xmath478 indeed , for any @xmath479 , the image of @xmath480 under @xmath461 is @xmath481 by lemma [ lem : a - send - general ] , each @xmath482 belongs to @xmath309 . in particular , implies @xmath483 @xmath4840.2 cm from , we can get the following inequality for each @xmath332 . @xmath485 in addition , the euler characteristic of @xmath457 is @xmath486 @xmath487 plugging into , we get @xmath488 when @xmath489 , theorem [ thm : homology - estimate - h ] and theorem [ thm : jenning ] together imply the following.0.2 cm [ thm : homology - estimate ] let @xmath1 be a finitely presentable group with deficiency at least @xmath70 . then for any prime @xmath2 and any normal subgroup @xmath0 of @xmath1 with @xmath78 , @xmath79 where @xmath80 is the coefficient of @xmath81 in the polynomial @xmath82 . in particular , if @xmath83 , @xmath84 to have more quantitative idea of the lower bounds of @xmath490 in theorem [ thm : homology - estimate ] , let us investigate two special cases below.0.2 cm [ thm : homology - estimate-2 ] let @xmath1 be a finitely presentable group with deficiency at least @xmath33 . then for any normal subgroup @xmath0 of @xmath1 with @xmath491 , @xmath29 , @xmath492 in particular , if @xmath493 , we must have @xmath494 . set @xmath495 in theorem [ thm : homology - estimate ] , we have for any @xmath496 , @xmath497 to prove @xmath498 , we need to show that there exists some @xmath496 so that @xmath499 0.1 cm * claim 1 : * @xmath500 for any prime @xmath501 and @xmath502 $ ] , and the equality holds only when @xmath36 or @xmath503.0.2 cm the claim is trivial when @xmath36 . when @xmath504 , by lemma [ lem : omega - prop ] ( i ) , @xmath505 note that @xmath506 for all prime @xmath2 , so we have @xmath507 moreover , lemma [ lem : omega - prop-2 ] implies that for each @xmath508 $ ] , @xmath509 @xmath510 the last `` @xmath511 '' is because @xmath512 when @xmath513 $ ] . 0.2 cm 0.1 cm * claim 2 : * for a fixed @xmath29 , @xmath514 reaches the maximum only at @xmath515 $ ] . 0.2 cm indeed , since @xmath516 , @xmath517 . so @xmath518 so when @xmath519 $ ] , @xmath520 . similarly , we can show when @xmath521 + 1 $ ] , @xmath522 . therefore , @xmath514 reaches the maximum at and only at @xmath515 $ ] . the claim 2 is proved . 0.2 cm by the above two claims , we have @xmath523}_{p , r } \geq \pi^{[\frac{r+1}{2}]}_{2,r}$ ] . next , we show that @xmath524}_{2,r } = ( r-1)\binom{r}{[\frac{r+1}{2 } ] } - \sum^r_{i=[\frac{r+1}{2 } ] + 1 } \binom{r}{i } \geq 2^{r-1}-1 , \ \,\text{for}\ \forall\ , r\geq 1.\ ] ] 0.1 cm * when @xmath525 , @xmath526 , @xmath527 = t+1 $ ] , @xmath528}_{2,r } = 2 t \binom{2t+1}{t+1 } - \sum^{2t+1}_{j = t+2 } \binom{2t+1}{j } = ( 2t+1 ) \binom{2t+1}{t+1 } - 2^{2t}.\ ] ] so to prove @xmath529}_{2,r } \geq 2^{r-1}-1 = 2^{2t}-1 $ ] , it is enough to prove @xmath530 0.1 cm * when @xmath531 , @xmath532 , @xmath527 = t$ ] , @xmath528}_{2,r } = ( 2t-1 ) \binom{2t}{t } - \sum^{2t}_{i = t+1 } \binom{2t}{i } = ( 2t-\frac{1}{2 } ) \binom{2t}{t } - 2^{2t-1}.\ ] ] so to prove @xmath529}_{2,r } \geq 2^{r-1}-1 = 2^{2t-1}-1 $ ] , it is enough to prove @xmath533 0.1 cm it is an elementary exercise to verify and , so we leave it to the reader . in particular , we find that0.1 cm * if the equality in holds , @xmath534 must be @xmath86 , hence @xmath34.0.1 cm * if the equality in holds , @xmath534 must be @xmath33 , hence @xmath39 . 0.1 cm so when the equality @xmath535 holds , the inequality in implies that @xmath536 or @xmath5 . the theorem is proved . 0.2 cm notice that @xmath537}=\binom{r}{[\frac{r}{2}]}$ ] . so is equivalent to the following inequality @xmath538 } \geq 2^{r-1 } + \sum^r_{i=[\frac{r+1}{2 } ] } \binom{r}{i } -1,\ , \ \forall\ , r\geq 1,\ ] ] where the equality holds only when @xmath34 or @xmath5 . 0.2 cm [ thm : homology - estimate-3 ] suppose a group @xmath1 admits a balanced finite presentation . then for any normal subgroup @xmath0 of @xmath1 with @xmath491 , @xmath539}_{p , r}| - p^r$ ] . in particular when the equality holds , we must have @xmath540 . set @xmath541 in theorem [ thm : homology - estimate ] , we get for any @xmath542 , @xmath543 in addition , by lemma [ lem : omega - prop ] ( iv ) , @xmath338 reaches its maximum at @xmath544 $ ] . 0.2 cm suppose @xmath28 is a @xmath5-dimensional path - connected cw - complex with finitely many cells and @xmath183 is a regular covering space whose deck transformation group is @xmath19 , @xmath29 . 0.2 cm up to homotopy equivalence , we may assume that @xmath28 has a single @xmath86-cell @xmath97 . let the set of @xmath33-cells of @xmath28 be @xmath184 and the set of @xmath5-cells of @xmath28 be @xmath185 . we will use the same notations and labels for the cells of @xmath11 as in section [ sec3 ] . note that @xmath545 and @xmath546.0.2 cm 0.1 cm 0.1 cm when @xmath11 is path connected , @xmath547 and we must have @xmath548 note that the fundamental group @xmath549 of @xmath28 has a natural presentation @xmath550 defined by the cellular structure of @xmath28 so that @xmath28 is the presentation complex of @xmath550 . the deficiency of @xmath550 is @xmath93 . in addition , since @xmath551 is a monomorphism , we can identify @xmath552 with its image @xmath553 . in the rest , we think of @xmath552 as a normal subgroup of @xmath549 , and so we have @xmath554 this allows us to use the theorems in section [ sec6 ] to estimate @xmath555.0.2 cm the euler characteristic of @xmath28 and @xmath11 are @xmath556 * case 1 * : when @xmath557 , @xmath558 . this implies @xmath559 so @xmath560 . so in this case , @xmath25 must be strictly greater than @xmath561 . 0.2 cm * case 2 * : when @xmath562 , we get @xmath563 , @xmath564 and @xmath565 . so the natural presentation @xmath550 of @xmath549 has deficiency @xmath33 . then theorem [ thm : homology - estimate-2 ] implies @xmath566 . so we get @xmath567 moreover , when @xmath568 , we have @xmath569 . so theorem [ thm : homology - estimate-2 ] implies that @xmath570 or @xmath5 , and we must have:0.2 cm * @xmath571 , @xmath572 , @xmath573 . then @xmath574 0.1 cm * @xmath575 , @xmath576 , @xmath577 . then @xmath578 0.2 cm * case 3 * : when @xmath579 , @xmath580 in particular , if @xmath581 , we must have @xmath582 then and implies @xmath95 and @xmath583 . so the natural presentation @xmath550 of @xmath549 has deficiency @xmath86 . then by theorem [ thm : homology - estimate-3 ] , we get @xmath584}_{2,r}| - 2^r & = 1 + r \binom{r}{[\frac{r}{2 } ] } - 2^r \\ ( \text{by}~\eqref{equ : pi - inequ-2})\ \ , & \geq \sum^r_{i=[\frac{r+1}{2 } ] } \binom{r}{i } -2^{r-1 } \geq 0 . \end{aligned}\ ] ] * the first inequality holds only when @xmath585;0.2 cm * the second inequality holds only when @xmath586 or @xmath5 ( see ) ; 0.2 cm * the third inequality holds only when @xmath32 is odd.0.2 cm so if @xmath587 , we must have @xmath588 , and @xmath589 . in this case , @xmath590 this finishes the proof of theorem [ thm : main1 ] when @xmath11 is path - connected . + 0.1 cm in this case , it is possible that @xmath591 . so we have two cases . 0.2 cm * when @xmath591 , @xmath11 is a trivial covering space over @xmath28 , i.e. @xmath11 is the disjoint union of @xmath561 copies of @xmath28 . so @xmath592 in particular , @xmath593 , i.e. @xmath594.0.2 cm * when @xmath595 , there exists an @xmath596 so that0.1 cm * * @xmath11 has @xmath597 path components @xmath598 which are pairwise homeomorphic . so @xmath599.0.1 cm * * each component @xmath600 of @xmath11 is a regular @xmath601-covering space of @xmath28 . so by our discussion in the previous section , @xmath602 . 0.1 cm + so we have @xmath603.0.2 cm so we finish the whole proof of theorem [ thm : main1 ] . + let us only prove theorem [ thm : main2 ] and theorem [ thm : main3 ] when @xmath46 is connected . if @xmath46 is not connected , the proof follows easily from the connected case by the same argument as in section [ sec7 - 2].0.2 cm suppose we have a free @xmath19-action on a compact connected @xmath7-manifold @xmath46 . then the orbit space @xmath604 is also a compact connected @xmath7-manifold , denoted by @xmath58 . the fundamental group of @xmath58 is finitely presentable . let @xmath605 be the orbit map . choose a basepoint @xmath606 and let @xmath607 . 0.2 cm if @xmath46 has boundary , so does @xmath58 . since any compact connected @xmath7-manifold with boundary is homotopy equivalent to a finite @xmath5-dimensional cw - complex , so @xmath46 is homotopy equivalent to a regular @xmath19-covering over a finite @xmath5-dimensional cw - complex . then by theorem [ thm : main1 ] , @xmath608 . 0.2 cm so in the rest of this section , we assume that @xmath46 has no boundary , and so @xmath58 is a closed connected @xmath7-manifold . the following well known result asserts that @xmath609 always admits a balanced presentation . 0.2 cm the fundamental group of any closed connected @xmath7-manifold @xmath58 admits a balanced presentation . in fact , if @xmath58 possesses a heegaard decomposition of genus @xmath85 , then the fundamental group of @xmath58 can be presented by @xmath85 generators and @xmath85 defining relators . indeed , suppose @xmath58 has a heegaard decomposition of genus @xmath26 . then there is a canonical cell decomposition of @xmath58 associated to this heegaard decomposition which consists of a single @xmath86-cell , @xmath26 @xmath33-cells , @xmath26 @xmath5-cells and a single @xmath7-cell . so the fundamental group @xmath609 has a presentation @xmath610 whose presentation complex is exactly the @xmath5-skeleton of @xmath58.0.2 cm since @xmath46 is a regular covering space of @xmath58 with deck transformation group @xmath19 , @xmath611 can be thought of as a normal subgroup of @xmath609 with @xmath612 so by the fact that @xmath609 admits a balanced presentation and theorem [ thm : homology - estimate-3 ] , @xmath613}_{p , r}| - p^r \notag \\ & \geq 1 + r |\omega^{[\frac{r(p-1)}{2}]}_{p , r}| -p^r . \end{aligned}\ ] ] 0.2 cm _ * proof of theorem [ thm : main2 ] . * _ when @xmath36 , the inequality reads @xmath614}_{2,r}| - 2^r \geq 1 + r \binom{r}{[\frac{r}{2 } ] } - 2^r.\ ] ] in addition , @xmath615 by poincar duality . so to prove @xmath616 we need to show @xmath617 . then by , it is sufficient to show @xmath618 } \geq 3\cdot 2^{r-1 } -2.\ ] ] but it turns out that the inequality in holds if and only if @xmath619 ( we leave it as an exercise to the reader ) . so only when @xmath619 can we use our method to prove @xmath47 . note that the @xmath34 case is trivial . but for @xmath620 , our estimate of @xmath621 from theorem [ thm : homology - estimate ] is insufficient ! fortunately , when @xmath49 , ( * ? ? ? * theorem 1.1 ) has already shown @xmath622 . so we have @xmath47 for all @xmath29.0.2 cm next , we discuss when @xmath48 . notice that when @xmath619 , the inequality in is always strict . so @xmath48 implies @xmath22 and @xmath623 ( see ) . then by the assumption @xmath624 , we get @xmath625 . in addition , since @xmath58 is a closed connected manifold , we have @xmath626 . all the possible cases are:0.2 cm * @xmath34 , @xmath627 and @xmath628 . so @xmath629 * @xmath39 , @xmath630 and @xmath631 . so @xmath632 * @xmath55 , @xmath633 and @xmath634 . so @xmath635 so theorem [ thm : main2 ] is proved . note that all of the above three cases can be realized by concrete free actions of @xmath45 on some @xmath7-manifolds . in theorem [ thm : main2 ] , if @xmath41 , we can not prove @xmath636 via our estimates in section [ sec6 ] . so some more accurate estimate of @xmath637 is needed to deal with those cases . + _ * proof of theorem [ thm : main3 ] . * _ since we assume that the deficiency of the fundamental group @xmath609 of @xmath58 is at least @xmath33 , so in particular , @xmath609 admits a deficiency-@xmath33 presentation . then by the fact @xmath638 and theorem [ thm : homology - estimate-2 ] , we can conclude that @xmath639 and , only when @xmath640 or @xmath5 can we have @xmath641.0.1 cm * if @xmath46 is orientable , @xmath642.0.1 cm * if @xmath46 is not orientable,0.1 cm * * @xmath36 , @xmath643.0.1 cm * * @xmath41 , @xmath644 . 0.1 cm so in all cases , @xmath645 . in particular , @xmath61 implies0.1 cm * @xmath646 , so @xmath34 or @xmath5,0.1 cm * @xmath41 , i.e. @xmath2 is an odd prime,0.1 cm * @xmath46 is non - orientable , so @xmath58 is also non - orientable.0.1 cm then @xmath58 and @xmath46 must be one the following cases:0.1 cm * @xmath34 , @xmath41 , @xmath647 , @xmath648 and @xmath649 , @xmath650 . so @xmath651 * @xmath39 , @xmath41 , @xmath652 , @xmath653 , @xmath654 and @xmath655 , @xmath656 , @xmath657 . so @xmath658 so theorem [ thm : main3 ] is proved . note that the above two cases can both be realized by concrete free actions of @xmath19 on some @xmath7-manifolds . we use theorem [ thm : homology - estimate-2 ] to give a new proof of the following proposition in @xcite.0.2 cm let @xmath661 h^p$ ] . then we have @xmath662 . so by theorem [ thm : homology - estimate-2 ] , we conclude that @xmath663 . repeating the argument for @xmath664 , and so on , we obtain a sequence @xmath665 of finite index subgroup of @xmath1 , such that @xmath666 tends to infinity . in the above statement , the hypothesis that the deficiency of @xmath1 is at least @xmath33 is necessary . for example , the deficiency of @xmath667 is @xmath86 , but any finite index subgroup of @xmath1 is isomorphic to @xmath668 . in addition , the condition @xmath659 can not be loosen either . for example , the deficiency of @xmath669 is @xmath33 and @xmath670 , but any finite index subgroup of @xmath669 is isomorphic to @xmath669 .

we present additional magic wavelengths ( ) for the clock transitions in the alkaline - earth metal ions considering circular polarized light aside from our previously reported values in [ j. kaur et al . , phys . rev . a * 92 * , 031402(r ) ( 2015 ) ] for the linearly polarized light . contributions from the vector component to the dynamic dipole polarizabilities ( ) of the atomic states associated with the clock transitions play major roles in the evaluation of these , hence facilitating in choosing circular polarization of lasers in the experiments . moreover , the actual clock transitions in these ions are carried out among the hyperfine levels . the values in these hyperfine transitions are estimated and found to be different from for the atomic transitions due to different contributions coming from the vector and tensor part of . importantly , we also present values that depend only on the scalar component of for their uses in a specially designed trap geometry for these ions so that they can be used unambiguously among any hyperfine levels of the atomic states of the clock transitions . we also present values explicitly at the 1064 nm for the atomic states associated with the clock transitions which may be useful for creating `` high - field seeking '' traps for the above ions using the nd : yag laser . the tune out wavelengths at which the states would be free from the stark shifts are also presented . accurate values of the electric dipole matrix elements required for these studies are given and trends of electron correlation effects in determining them are also highlighted .

1703.08969

atomic clocks based on optical lattices are capable of proffering outstanding stable and accurate time keeping devices . a fundamental feature of an optical lattice clock is that it interrogates an optical transition with the controlled atomic motion @xcite . at present , the most stable clock is based on the optical lattices of @xmath2sr atoms with accuracy below @xmath3 @xcite . the unique feature of this clock is that the atoms are trapped at the wavelengths of an external electric field at which the differential light shift of an atomic transition nullifies . these wavelengths are commonly known as the magic wavelengths ( @xmath0 ) @xcite . however , ions provide more accurate atomic clocks since various systematics in the ions can be controlled better @xcite . as a result a number of ions , such as , @xmath4al@xmath5 @xmath6hg@xmath5 , @xmath7yb@xmath5 , @xmath8in@xmath5 , @xmath9sr@xmath5 , @xmath10ca@xmath5 , @xmath11cd@xmath5 etc . are under consideration for building accurate clocks . among the various ions proposed for frequency standards @xcite , the alkaline earth metal ions possess an advantage that transitions required for cooling and re - pumping of ions and clock frequency measurement can be easily accessed using non - bulky solid state or diode lasers @xcite . moreover presence of metastable @xmath12 states in these ions , whose lifetimes range from milliseconds to several seconds , assist in carrying out measurements meticulously . the recent development on measurement of @xmath0 in a singly charged @xmath10ca@xmath5 ion has now open - up a platform for the possibility of building up pure optical trapped ion clocks @xcite . optical lattices blended with unique features of optical transitions in an ionic system can revolutionize the secondary as well as primary frequency standards . however , the potential of the optical dipole trap perturb the energy levels of the ion unevenly and the consistency of an ion optical clock depreciates . therefore , knowledge of @xmath0 values in these ions would be instrumental for constructing all - optical trapped ion clocks . these wavelengths can be found out using accurate values of the dynamic dipole polarizabilities @xmath1 of the states associated with the clock transitions . also , information on the dynamic ( @xmath1 ) values , especially at which the ions are being trapped will be of great significance . improved atomic clocks will obviously ease the widely used technologies including precise determination of fundamental constants @xcite , accurate control of quantum states @xcite and advancement in communication , global positioning system @xcite etc . following the measurement of @xmath0 in the @xmath13 transition of @xmath14ca@xmath5 , we had investigated these values in the @xmath15 and @xmath16 transitions of the alkaline - earth ions @xcite for the ground state principal quantum number @xmath17 . however , those studies were focused mainly on the linearly polarized light limiting the choice for experimental measurements . application of circular polarized light to atomic systems introduce contributions from the vector polarizabilities in the stark shifts of the energy levels , which is linearly proportional to the angular frequency of the applied field . this can help in manipulating the stark shifts in the energy levels and can lend to more degrees of freedom to attain further @xmath0 values as per the experimental stipulation . moreover , it is advantageous to consider hyperfine transitions in certain isotopes of the singly charged alkaline - earth ions , giving rise to zero hyperfine angular momentum , to get - rid off the systematics due to electric quadrupole shifts @xcite . since @xmath18 values of atomic and hyperfine states in an atomic system are different when vector and tensor components of @xmath18 contribute , the @xmath0 values also differ between the atomic and hyperfine transitions . thus , it would be pertinent to investigate @xmath0 both in the atomic and hyperfine transitions before the experimental consideration of the proposed ions . in fact , it could be more convenient to have @xmath0 that are independent of choice of both magnetic and hyperfine sub - levels in a given clock transition . in fact , the nd : yag lasers at 1064 nm is often used for trapping atoms and ions because of their relatively high power and low intensity noise @xcite . traps built with long wavelength lasers are generally `` high - field - seeking '' where the atoms are attracted to the intensity maxima . dynamic polarizabilities for the considered ions at 1064 nm will be of immense interest to the experimentalists since these polarizabilities will be immediately useful for operating optical traps at 1064 nm light fields . lccccccc + & & + & @xmath19 & & @xmath19 & @xmath20 & & @xmath19 & @xmath20 + + @xmath21 & 24.38(6 ) & @xmath21 & 19.24(5 ) & @xmath2219.24(5 ) & @xmath23 & 22.77(8 ) & @xmath2222.77(8 ) + @xmath24 & 0.07 & @xmath25 & 0.011 & @xmath220.011 & @xmath26 & 0.016 & @xmath220.016 + @xmath27 & 0.01 & @xmath28 & 3.76 & 3.009 & @xmath29 & 0.12 & 0.14 + @xmath23 & 48.4(17 ) & @xmath26 & 0.0027 & 0.002 & @xmath30 & 0.056 & 0.06 + @xmath31 & 0.01 & @xmath32 & 2.49(1 ) & @xmath220.49 & @xmath29 & 2.39(14 ) & @xmath220.85 + @xmath33 & 0.019 & @xmath34 & 0.81 & @xmath220.16 & @xmath35 & 0.76 & @xmath220.27 + & & @xmath30 & 0.36 & @xmath220.07 & @xmath36 & 0.001 & @xmath220.0005 + main"@xmath37 & 72.88 & & 26.67 & @xmath2216.96 & & 26.12 & @xmath2223.71 + @xmath38 & 3.25 & & 3.25 & & & 3.25 & + tail " @xmath37 & 5.86@xmath39 & & 3.75 & @xmath220.75 & & 3.75 & @xmath221.08 + @xmath40 & @xmath228.85@xmath39 & & @xmath227.94@xmath41 & & & @xmath221.02@xmath39 & + @xmath42(present ) & 76.1(2 ) & & 33.67(1.8 ) & @xmath2217.71 & & 33.11(1.8 ) & @xmath2224.78(4 ) + @xmath42(other ) @xcite & 75.28 & & 32.99 & @xmath2217.88 & & 32.81 & @xmath2225.16 + @xmath42(other ) @xcite & 73.0(1.5 ) & & 28.5(1.0 ) & @xmath2215.8(7 ) & & 29.5(1.0 ) & @xmath2222.45(5 ) + @xmath42(other ) @xcite&75.49 & & 32.73 & @xmath2217.64 & & 32.73 & @xmath2225.20 + @xmath42(expt . ) @xcite & 75.3(4 ) & & & & + + & & + & @xmath19 & & @xmath19 & @xmath20 & & @xmath19 & @xmath20 + + @xmath24 & 29.47(10 ) & @xmath24 & 38.67(2 ) & @xmath2238.67(2 ) & @xmath31 & 44.16(3 ) & @xmath2244.16(3 ) + @xmath43 & 0.0008 & @xmath44 & 0.0063 & @xmath220.0063 & @xmath45 & 0.015 & @xmath220.015 + @xmath46 & 0.071 & @xmath31 & 7.024(2 ) & 5.62(2 ) & @xmath29 & 0.33 & 0.37 + @xmath31 & 56.93(27 ) & @xmath45 & 0.003 & 0.002 & @xmath30 & 0.12 & 0.13 + @xmath47 & 0.0015 & @xmath29 & 6.63(7 ) & @xmath221.33 & @xmath29 & 6.51(5 ) & @xmath222.32(2 ) + @xmath48 & 0.0058 & @xmath34 & 1.76 & @xmath220.35 & @xmath35 & 1.72 & @xmath220.61 + & & @xmath30 & 0.72 & @xmath220.14 & @xmath36 & 0.69 & @xmath220.25 + main"@xmath37 & 86.47 & & 54.81 & @xmath2234.88 & & 53.55 & @xmath2246.85 + @xmath38 & 4.98 & & 4.98 & & & 4.98 & + tail"@xmath37 & 1.96@xmath39 & & 4.94 & @xmath221.0 & & 4.96 & @xmath221.44 + @xmath40 & @xmath220.19 & & @xmath221.77@xmath41 & & & @xmath222.78@xmath39 & + @xmath42(present ) & 91.23(0.3 ) & & 64.7(2.5 ) & @xmath2235.88(5 ) & & 63.5(2.5 ) & @xmath2248.29(7 ) + @xmath42(other ) @xcite & 89.99 & & 61.77 & - & & 61.77 & - + @xmath42(other ) @xcite&88.29(1.0 ) & & 61.43(52 ) & 35.42 & & 62.87(75 ) & @xmath2248.83(30 ) + @xmath42(expt . ) @xcite & 93.3(9 ) & & 57.0 & - & & 57.0 & - + + & & + & @xmath19 & & @xmath19 & @xmath20 & & @xmath19 & @xmath20 + + @xmath43 & 40.23(14 ) & @xmath43 & 22.51(13 ) & @xmath2222.51(13 ) & @xmath47 & 25.66(1 ) & @xmath2225.66(1 ) + @xmath49 & 0.005 & @xmath46 & 0.075 & @xmath220.074 & @xmath48 & 0.13 & @xmath220.13 + @xmath50 & 0.009 & @xmath47 & 3.86 & 3.095 & @xmath29 & 0.668(5 ) & 0.763(2 ) + @xmath47 & 73.93(4 ) & @xmath48 & 0.023 & 0.018 & @xmath51 & 0.0955(1 ) & 0.109(2 ) + @xmath52 & 0.011 & @xmath29 & 11.85(3 ) & @xmath222.37 & @xmath53 & 13.34(13 ) & @xmath224.76(5 ) + @xmath54 & 0.013 & @xmath51 & 1.76 & @xmath220.35 & @xmath35 & 2.863(34 ) & @xmath221.022(17 ) + main"@xmath37 & 114.19 & & 40.07 & @xmath2222.19 & & 42.76 & @xmath2230.71 + @xmath38 & 9.35 & & 9.35 & & & 9.35 & + tail"@xmath37 & 1.66@xmath39 & & 4.75 & @xmath221.0 & & 4.79 & @xmath221.46 + @xmath40 & @xmath220.37 & & @xmath222.34@xmath39 & & & @xmath223.87@xmath39 & + @xmath42(present ) & 123.7(5 ) & & 54.17(2.5 ) & @xmath2222.19(4 ) & & 56.87(2.4 ) & @xmath2232.17(3 ) + @xmath42(other ) @xcite & 124.26(1.0 ) & & 48.81(46 ) & @xmath2224.62(28 ) & & 50.67(58 ) & @xmath2230.85(31 ) + @xmath42(other ) @xcite & 123.88(5 ) + in the present work , we aim to search for the @xmath0 for the @xmath55 optical clock transitions , both in the atomic and hyperfine levels , in @xmath14ca@xmath5 ( with nuclear spin @xmath56=7/2 ) , @xmath2sr@xmath5 ( @xmath56=9/2 ) and @xmath57ba@xmath5 ( @xmath56=3/2 ) ions using circularly polarized light . these values can be compared with the values for the linearly polarized light reported in ref . @xcite for the experimental consideration to trap the above ions . also , we had demonstrated in a recent work how a trap geometry can be chosen in such a way that stark shifts observed by the energy levels can be free from the contributions from the vector and tensor components of the @xmath18 of the atomic states @xcite . assuming such trapping geometries for the considered alkaline - earth ions , we also give @xmath0 values using only the scalar polarizability contributions . moreover , we identify the tune - out wavelengths ( @xmath58 ) of the respective states for which the dynamic dipole polarizability of these ions vanishes . comprehension of these @xmath58 values are needed for the sympathetic cooling of other possible singly and multiply charged ions in two - species mixtures with the considered alkaline earth ions @xcite . we also present dynamic polarizability of these states at the 1064 nm wavelength of the applied external electric field . contributions from various electric dipole ( e1 ) matrix elements in determining the @xmath18 values and the role of the electron correlation effects for the evaluation of accurate values of the e1 matrix elements are also discussed . unless stated otherwise , all the results are given in atomic unit ( a.u . ) throughout this paper . the stark shift in the energy of @xmath59 level of an atom placed in an electric field is given by @xcite @xmath60 where @xmath61 is the amplitude of the external electric field due to the applied laser in an atomic system and @xmath62 is the dynamic dipole polarizability for the state @xmath63 with its magnetic projection @xmath64 . in tensor decomposition , @xmath65 can be expressed as @xmath66 where @xmath67 with @xmath68 are the scalar , vector and tensor components of @xmath62 respectively . in the specific cases , @xmath63 can be either the atomic angular momentum @xmath69 or hyperfine angular momentum @xmath70 . the terms @xmath71 and @xmath72 are defined as @xcite @xmath73 and @xmath74 , \label{gamma}\ ] ] with the quantization axis unit vector @xmath75 and polarization unit vector @xmath76 . the differential stark shift of a transition between states @xmath63 to @xmath77 can be formulated as @xmath78 { \cal e}^2 , \label{eqmag1 } \end{aligned}\ ] ] to obtain null differential stark shift , it is obvious from eq . ( [ eqmag1 ] ) that either the independent components of the polarizabilities cancel out each other or the net resultant nullifies which depend upon the choice of @xmath71 , @xmath72 , @xmath79 and @xmath80 magnetic sublevels . moreover , as we have demonstrated recently , the differential stark shift can be independent of the vector and tensor components of the states involved in a transition for a certain trap geometry @xcite . such trapping scheme is useful for @xmath81 , @xmath70 and @xmath82 insensitive trapping and can be suitably applied for considered clock transitions in alkaline earth metal ions . lccccccccccc + & & + & @xmath19 & @xmath83 & & @xmath19 & @xmath83 & @xmath20 & & @xmath19 & @xmath83 & @xmath20 + + @xmath21 & 28.33(7 ) & @xmath2221.141(5 ) & @xmath21 & 57.12(2 ) & @xmath22139.55(3 ) & @xmath2257.12(2 ) & @xmath23 & 64.13(2 ) & @xmath22154.55(5 ) & @xmath2264.13(2 ) + @xmath24 & 0.0072 & @xmath220.002 & @xmath25 & 0.011 & @xmath220.007 & @xmath220.011 & @xmath26 & 0.016 & @xmath220.009 & @xmath220.016 + @xmath27 & 0.011 & @xmath220.0027 & @xmath28 & 8.59(1 ) & @xmath228.24 & 6.87 & @xmath29 & 0.12 & @xmath220.018 & 0.14 + @xmath23 & 56.03(20 ) & 20.719 & @xmath26 & 0.0028 & @xmath220.0006 & 0.0020 & @xmath51 & 0.056 & @xmath220.006 & 0.06 + @xmath31 & 0.01 & 0.002 & @xmath32 & 2.57(1 ) & 0.79 & @xmath220.51 & @xmath29 & 2.46(12 ) & 0.91 & @xmath220.88 + @xmath33 & 0.019 & 0.003 & @xmath34 & 0.82 & 0.22 & @xmath220.16 & @xmath35 & 0.78 & 0.24 & @xmath220.27 + & & & @xmath30 & 0.37 & 0.09 & @xmath220.07 & @xmath36 & 0.001 & 0.0004 & @xmath220.0005 + main"@xmath37 & 84.4 & @xmath220.423 & & 69.5 & @xmath22146.69 & @xmath2250.9 & & 67.5 & @xmath22153.37 & @xmath2265.1 + @xmath38 & 3.25 & & & 3.25 & & & & 3.25 & & + tail"@xmath37 & 5.90@xmath39 & @xmath223.49@xmath84 & & 3.79 & 0.73 & @xmath220.76 & & 3.80 & 0.81 & @xmath221.09 + @xmath40 & @xmath228.85@xmath39 & & & @xmath227.94@xmath41 & & & & @xmath221.02@xmath39 & & + @xmath85 & 87.6(2 ) & @xmath220.423 & & 76.5(1.9 ) & @xmath22145.96(0.36 ) & @xmath2251.7(3 ) & & 74.6(1.9 ) & @xmath22152.56(0.41 ) & @xmath2266.0(5 ) + & & + + & & + & @xmath19 & @xmath83 & & @xmath19 & @xmath83 & @xmath20 & & @xmath19 & @xmath83 & @xmath20 + + @xmath24 & 34.96(12 ) & @xmath2227.714(9 ) & @xmath24 & @xmath22730.9(4 ) & 2250.05(1.3 ) & 730.9(4 ) & @xmath31 & 769.4(5 ) & @xmath222241.08(1.6 ) & @xmath22769.4(5 ) + @xmath43 & 0.0008 & @xmath220.0003 & @xmath44 & 0.007 & @xmath220.005 & @xmath220.007 & @xmath45 & 0.016 & @xmath220.01 & @xmath220.016 + @xmath46 & 0.007 & -0.002 & @xmath31 & 64.03(2 ) & @xmath2272.49(2 ) & 51.22(2 ) & @xmath29 & 0.34 & @xmath220.059 & 0.38 + @xmath31 & 66.74(32 ) & 25.58(1 ) & @xmath45 & 0.003 & @xmath220.0008 & 0.002 & @xmath51 & 0.12 & @xmath220.02 & 0.5 + @xmath47 & 0.0046 & 0.0003 & @xmath29 & 6.91(7 ) & 2.52 & @xmath221.38 & @xmath29 & 6.79(5 ) & 2.96 & @xmath222.42(2 ) + @xmath48 & 0.0029 & 0.0008 & @xmath34 & 1.81 & 0.54 & @xmath220.36 & @xmath35 & 1.76 & 0.63 & @xmath220.63 + & & & @xmath30 & 0.73 & 0.22 & @xmath220.15 & @xmath36 & 0.71 & 0.23 & @xmath220.25 + main"@xmath37 & 101.7&@xmath222.13 & & @xmath22657.4 & 2180.82 & 780.2 & & 779.1 & @xmath222237.34 & @xmath22771.8 + @xmath38 & 4.98 & & & 4.98 & & & & 4.98 & & + tail"@xmath37 & 1.96@xmath39&@xmath228.13@xmath84 & & 5.05 & 1.11 & @xmath221.02 & & 5.05 & 1.24 & @xmath221.46 + @xmath40 & @xmath220.19 & & & @xmath221.77@xmath41 & & & & @xmath222.79@xmath39 & & & + @xmath85 & 106.52(3 ) & -2.14 & & @xmath22647.4(3 ) & 2181.92(1.79 ) & 779.2(6 ) & & 789.2(2.6)&@xmath222236.10(2.82 ) & @xmath22773.(9 ) + & & + & + & & + & @xmath19 & @xmath83 & & @xmath19 & @xmath83 & @xmath20 & & @xmath19 & @xmath83 & @xmath20 + + @xmath43 & 51.26(18 ) & @xmath2247.55(2 ) & @xmath43 & 35.89(2 ) & @xmath2265.77(4 ) & @xmath2235.89(2 ) & @xmath47 & 38.5(2 ) & @xmath2266.08(3 ) & @xmath2238.5(2 ) + @xmath49 & 0.006 & @xmath220.002 & @xmath46 & 0.077 & @xmath220.05 & @xmath220.077 & @xmath48 & 0.13 & @xmath220.08 & @xmath220.13 + @xmath50 & 0.009 & @xmath220.003 & @xmath47 & 5.55 & @xmath223.66(2 ) & 4.44 & @xmath29 & 0.70 & @xmath220.13 & 0.80 + @xmath47 & 90.52(4 ) & 38.75(2 ) & @xmath48 & 0.024 & @xmath22 - 0.005 & 0.019 & @xmath51 & 0.107(1 ) & @xmath220.02 & 0.122(2 ) + @xmath52 & 0.011 & 0.002 & @xmath29 & 12.44(3 ) & 4.85 & @xmath222.48 & @xmath53 & 14.02(14 ) & 6.59 & @xmath225.007(5 ) + @xmath54 & 0.013 & 0.002 & @xmath51 & 1.83 & 0.5 & @xmath220.36 & @xmath35 & 3.41(2 ) & 1.29 & @xmath221.22 + main"@xmath37 & 141.81 & @xmath228.80 & & 55.8 & @xmath2264.05 & @xmath2234.35 & & 56.86 & @xmath2259.02 & @xmath2243.9 + @xmath38 & 9.35 & & & 9.35 & & & & 9.35 & + tail"@xmath37 & 1.67@xmath39 & @xmath221.24@xmath84 & & 3.29 & 0.7 & @xmath220.71 & & 3.31 & 0.77 & @xmath221.04 + @xmath40 & @xmath220.37 & & & @xmath222.35@xmath39 & & & & @xmath223.88@xmath39 & + @xmath85 & 150.6(5 ) & @xmath228.80 & & 68.44(1.7)&@xmath2263.55(0.35 ) & @xmath2235.05(4 ) & & 69.52(1.7 ) & @xmath2258.25(0.38 ) & @xmath2244.97(5 ) + conveniently the expressions for polarizabilities of the hyperfine and atomic levels can be expressed as @xcite @xmath86 , \label{totalf } \ \ \ \ \\end{aligned}\ ] ] and @xmath87 , \label{totalj } \ \ \ \ \\end{aligned}\ ] ] where @xmath88 with @xmath68 are the scalar , vector and tensor components of the respective polarizabilities , @xmath89 represents degree of polarization , @xmath90 is the angle between the quantization axis and wave vector , and @xmath91 is the angle between the quantization axis and direction of polarization of the field . in the presence of magnetic field , @xmath92 and @xmath93 can have any values depending on the direction of applied magnetic field . in the absence of magnetic field , @xmath94 and @xmath95 for the linearly polarized light , where polarization vector is assumed to be along the quantization axis . however it yields @xmath96 and @xmath97 for the circularly polarized light , where wave vector is assumed to be along the quantization axis . polarizabilities of the hyperfine states can be evaluated from the atomic state results using the relations @xcite @xmath98 and @xmath99 further , we can evaluate the atomic dipole polarizabilities using the expressions @xmath100 and @xmath101 where @xmath102s are the angular momentum of the intermediate states , @xmath103 and @xmath104 are energies of the corresponding states , @xmath105 are the reduced e1 matrix elements . lcccccccccccc & & & & & + resonances & @xmath106 & @xmath0 & @xmath107 & @xmath0 & @xmath107 & @xmath0 & @xmath107 & @xmath0 & @xmath107 + + @xmath14ca@xmath5 & + @xmath108 & 393.366 + & & 394.6(4 ) & @xmath109 & 394.6(9 ) & @xmath110 & 394.6(10 ) & 11.68 & 394.6(10 ) & 22.40 + @xmath111 & 396.847 & + @xmath112 & 849.802 & + & & 850.9(2 ) & 96.33 & 853.1(2 ) & 96.20 & - & - & - & - + @xmath113 & 866.214 & + & & 1467.8(4 ) & 81.31 & 1013.4(5 ) & 89.31 & 870.7(3 ) & 95.15 & - & - + @xmath2sr@xmath5 & + @xmath114 & 407.7714 + & & 412.5(9 ) & @xmath115 & 412.4(9 ) & 3.99 & 412.4(9 ) & 27.18 & 412.3(8 ) & 48.55 + @xmath116 & 421.5524 + @xmath117 & 1003.6654 + & & 1009.3(2 ) & 109.74 & 1019.7(3 ) & 109.29 & 1062.5(3 ) & 107.65 & - & - + @xmath118 & 1091.4874 & + & & - & - & 1577.2(3 ) & 98.17 & - & - & - & - + @xmath57ba@xmath5 & + @xmath119 & 455.4033 + & & 468.8(4 ) & @xmath120 & 468.1(4 ) & @xmath121 & 467.6(5 ) & 30.14 & 467.3(4 ) & 59.27 + @xmath122 & 493.4077 + @xmath123 & 585.3675 + & & 587.6(9 ) & 373.89 & 589.5(3 ) & 368.60 & 589.6(5 ) & 368.35 & - & - + @xmath124 & 649.6898 + & & 841.7(5 ) & 182.54 & 690.7(7 ) & 237.64 & - & - & - & - + + lcccccccccccc & & & & & + resonances & @xmath106 & @xmath0 & @xmath107 & @xmath0 & @xmath107 & @xmath0 & @xmath107 & @xmath0 & @xmath107 + + @xmath14ca@xmath5 & + @xmath112 & 849.802 & + & & 851.2(3 ) & 95.63 & 853.5(4 ) & 95.49 & - & - & - & - + @xmath113 & 866.214 & + & & 1549.9(5 ) & 80.95 & 1031.4(4 ) & 88.28 & 875.56(2 ) & 94.27 & - & - + @xmath2sr@xmath5 & + @xmath114 & 407.7714 + & & 416.9(7 ) & @xmath2221.20 & 416.8(5 ) & 3.93 & 416.8(8 ) & 27.12 & 416.7(8 ) & 48.44 + @xmath116 & 421.5524 + & & 442.5(5 ) & @xmath2225.36 & 442.3(6 ) & 1.94 & 442.1(2 ) & 26.46 & 441.9(4 ) & 48.03 + & & 478.2(5 ) & @xmath2231.49 & 479.2(4 ) & @xmath220.96 & 480.1(2 ) & 25.56 & 481.1(1 ) & 47.83 + @xmath117 & 1003.6654 + & & 1009.2(1 ) & 106.69 & 1019.8(6 ) & 106.29 & 1064.8(5 ) & 104.79 & - & - + @xmath118 & 1091.4874 & + & & & - & - 1593.1(5 ) & 96.54 & - & - & - & - + @xmath57ba@xmath5 & + @xmath119 & 455.4033 + & & 479.2(5 ) & @xmath125 & 478.5(4 ) & @xmath126 & 477.9(1 ) & 28.76 & 477.5(2 ) & 58.89 + @xmath122 & 493.4077 + & & 505.4(1 ) & @xmath127 & 505.1(4 ) & @xmath128 & 504.8(2 ) & 24.05 & 504.7(2 ) & 57.93 + & & 571.4(4 ) & @xmath129 & 574.1(2 ) & @xmath130 & 577.3(4 ) & @xmath131 & - & - + @xmath123 & 585.3675 + & & 589.4(4 ) & 47.82 & 595.5(2 ) & 89.44 & 602.4(2 ) & 120.417 & 590.5(2 ) & 56.46 + @xmath124 & 649.6898 + & & 899.9(5 ) & 153.13 & 710.5(4 ) & 176.48 & - & - & - & - + + we define the tune - out wavelength @xmath58 as the wavelength where the dynamic polarizability of the state is zero . we have determined the tune out wavelengths for the ground , @xmath132 and @xmath133 states of @xmath14ca@xmath5 , @xmath2sr@xmath5 and @xmath57ba@xmath5 ions . the detailed description about these calculations have been given in the refs . as discussed in our earlier works @xcite , each component of @xmath134 can be conveniently evaluated in the considered alkaline earth ions , in which many of the low - lying states have electronic configurations as a common closed core of inert gas atoms and a well defined valence orbital , by classifying into three different contributions such as @xmath135 where @xmath38 , @xmath40 and @xmath37 are referred to as the core , core - valence and valence electron correlation contributions , respectively . since the valence electron correlation effects are mainly estimated by @xmath37 , this gives majority contribution followed by @xmath38 . again , accuracies in the _ ab initio _ results of these quantities mainly suffer due to the uncertainties associated with the calculated energies of the atomic states . therefore , we calculate only the e1 matrix elements of as many as low - lying states of the considered ions employing a relativistic coupled - cluster ( rcc ) method , which is an all order perturbative method , and combine them with the experimental energy values from the national institute of science and technology ( nist ) database @xcite to determine the dominant contributions to @xmath37 . we refer this as `` main '' contribution to @xmath37 ; while the smaller contributions coming from the high - lying excited states are estimated in the _ ab initio _ formalism using the dirac - hartree - fock ( dhf ) method and are mentioned as `` tail '' contribution to @xmath37 . to estimate the other two contributions , it is not possible to use sum - over - states approach , so we determine @xmath38 in the random - phase approximation ( rpa ) . this included core - polarization effects to all orders . it has been found that rpa can give these values reliably for the inert gas configured atomic systems @xcite . again as demonstrated later , the @xmath40 contributions are very small in these ions . thus , we evaluate them using the dhf method . a small number of e1 matrix elements of the @xmath136 transitions of @xmath57ba@xmath5 are borrowed from the work of sahoo _ @xcite , while the rest e1 matrix elements for the evaluation of `` main '' contribution to @xmath137 are obtained considering the singles and doubles excitation approximation in the rcc ( sd ) method as described in refs . @xcite . in the sd method , the wave function of the state with the closed - core with a valence electron @xmath138 is represented as an expansion @xmath139 |\phi_v\rangle,\nonumber \\ & & \label{expansion}\end{aligned}\ ] ] where @xmath140 is the dhf wave function of the state . in the above expression , @xmath141 and @xmath142 are the creation and annihilation operators with the indices @xmath143 and @xmath144 designating the virtual and core orbitals of @xmath140 , @xmath145 and @xmath146 are the corresponding single core and valence excitation coefficients , and @xmath147 and @xmath148 are the double core and valence excitation coefficients . to obtain the dhf wave function , we use a finite size basis set consisting of 70 b - splines constrained to a large cavity of radius r = @xmath149 a.u . and solved self consistently using the roothan equation on a nonlinear grid . lccccccccccccccccc & & & & & & & + resonances & @xmath106 & @xmath0 & @xmath107 & @xmath0 & @xmath107 & @xmath0 & @xmath107 & @xmath0 & @xmath107 & @xmath0 & @xmath107 & @xmath0 & @xmath107 + + @xmath14ca@xmath5 & + @xmath108 & 393.366 + & & 394.64(2 ) & @xmath150 & 394.64(3 ) & @xmath151 & 394.64(2 ) & 1.0 & 394.63(4 ) & 6.33 & 394.63(3 ) & 17.0 & 394.63(3 ) & 22.39 + @xmath111 & 396.847 & + @xmath152 & 854.209 + & & - & - & 1150.39(2 ) & 85.95 & 975.6(4 ) & 90.54 & 891.4(3 ) & 94.09 & -&-&-&- + @xmath2sr@xmath5 & + @xmath114 & 407.7714 + & & 412.5(3 ) & @xmath153 & 412.46(2 ) & @xmath154 & 412.42(4 ) & 6.98 & 412.38(4 ) & 24.16 & 412.35(3 ) & 37.15 & 412.31(3 ) & 54.61 + @xmath116 & 421.5524 & + @xmath155 & 1032.7309 & + & & - & - & - & - & 1379.4(3 ) & 100.44 & 1130.8(4 ) & 105.46 & - & - & - & - + @xmath57ba@xmath5 & + @xmath119 & 455.4033 + & & 469.1(5 ) & @xmath156 & 468.5(1 ) & @xmath157 & 467.9(4 ) & @xmath158 & 467.6(4 ) & 29.01 & 467.3(1 ) & 53.75 & 467.1(1 ) & 70.27 + @xmath122 & 493.4077 & + @xmath159 & 614.1713 + & & 810.1(8 ) & 189.71 & 691(2 ) & 237.75 & 644.7(5 ) & 275.99 & 623.3(5 ) & 302.79 & - & - & - & - + + lccccccccccccccccc & & & & & & & + resonances & @xmath106 & @xmath0 & @xmath107 & @xmath0 & @xmath107 & @xmath0 & @xmath107 & @xmath0 & @xmath107 & @xmath0 & @xmath107 & @xmath0 & @xmath107 + + @xmath14ca@xmath5 & + @xmath111 & 396.847 & + @xmath152 & 854.209 + & & - & - & 1173.5(2 ) & 85.12 & 982.7(2 ) & 89.77 & 893.4(3 ) & 93.37 & -&-&-&- + @xmath2sr@xmath5 & + @xmath114 & 407.7714 + & & 416.93(4 ) & @xmath160 & 416.88(3 ) & @xmath161 & 416.85(4 ) & 7.66 & 416.81(4 ) & 23.82 & 416.77(1 ) & 38.63 & 416.74(3 ) & 52.53 + @xmath116 & 421.5524 & + & & 442.49(4 ) & @xmath162 & 442.36(3 ) & @xmath163 & 442.25(3 ) & 6.12 & 442.13(4 ) & 22.84 & 442.03(8 ) & 38.18 & 441.93(3 ) & 52.07 + & & 477.99(5 ) & @xmath164 & 478.85(1 ) & @xmath165 & 479.32(7 ) & 3.55 & 479.98(9 ) & 21.77 & 480.65(1 ) & 37.83 & 481.28(7 ) & 51.73 + @xmath155 & 1032.7309 & + & & - & - & - & - & 1391.4(3 ) & 98.46 & 1134.5(4 ) & 102.92 & - & - & - & - + @xmath57ba@xmath5 & + @xmath119 & 455.4033 + & & 479.5(2 ) & @xmath166 & 478.9(2 ) & @xmath167 & 478.4(2 ) & @xmath168 & 477.9(4 ) & 27.52 & 477.6(4 ) & 53.33 & 477.4(7 ) & 69.73 + @xmath122 & 493.4077 & + & & 505.5(2 ) & @xmath169 & 505.3(4 ) & @xmath170 & 505.04(41 ) & @xmath171 & 504.8(4 ) & 25.56 & 504.76(4 ) & 52.23 & 504.69(7 ) & 68.53 + & & 568.8(2 ) & @xmath172 & 572.14(2 ) & @xmath173 & 576.52(2 ) & @xmath174 & 582.4(2 ) & @xmath175 & 589.7(4 ) & 50.45 & 591.7(2 ) & 65.99 + @xmath159 & 614.1713 + & & 894.3(8 ) & 153.62 & 729.9(2 ) & 173.67 & 665.9(3 ) & 180.26 & 632.6(5 ) & 171.84 & - & - & - & - + + in order to verify contributions from the higher level excitations , we also estimate leading order contributions from the triple excitations perturbatively in the sd method framework ( sdpt method ) by expressing the atomic wave functions as @xcite @xmath176 where @xmath177 is the perturbed triple valence excitation amplitudes . after obtaining wave functions employing the sd and sdpt method , we determine the e1 matrix element for a given transition between the states @xmath178 and @xmath179 by evaluating the expression @xmath180 lccccccc + & & & + @xmath181 @xmath182,@xmath183 @xmath184,@xmath185 & @xmath0 & @xmath186 & @xmath181 @xmath70,@xmath183 @xmath184,@xmath185 & @xmath0 & @xmath186 + @xmath1813,0 @xmath187 & 395.79(2 ) & 8.69 & @xmath1813,0@xmath1881,0@xmath189 & 395.79(3 ) & 3.35 + & 852.5(2 ) & 95.86 & & 992(4 ) & 89.71 + & 999(4 ) & 89.46 & @xmath1813,0@xmath1882,0@xmath189 & 395.79(3 ) & 3.35 + @xmath1813,0@xmath1883,0@xmath189 & 395.79(2 ) & 3.35 & & 1077(9 ) & 87.31 + & 851.3(2 ) & 95.95 & @xmath1813,0@xmath1883,0@xmath189 & 395.79(3 ) & 3.35 + & 1089(7 ) & 87.02 & & 1064(9 ) & 87.61 + @xmath1813,0@xmath1884,0@xmath189 & 395.79(4 ) & 3.35 & @xmath1813,0@xmath1884,0@xmath189 & 395.78(3 ) & 3.35 + & 851.7(2 ) & 95.92 & & 1032(7 ) & 88.48 + & 1049(4 ) & 87.99 & @xmath1813,0@xmath1885,0@xmath189 & 395.79(3 ) & 3.35 + @xmath1813,0@xmath1885,0@xmath189 & 395.79(2 ) & 8.69 & & 992(4 ) & 89.71 + & 853.4(2 ) & 95.82 & @xmath1814,0@xmath189 @xmath190 @xmath1816,0@xmath189 & 395.79(3 ) & 8.69 + & 970(3 ) & 90.44 & & 949(3 ) & 91.21 + + + & & & + @xmath181f,@xmath191 @xmath190 @xmath192,@xmath185 & @xmath0 & @xmath186 & @xmath181f,@xmath191 @xmath190 @xmath192,@xmath185 & @xmath0 & @xmath186 + @xmath1814,0@xmath189 @xmath190 @xmath1813,0@xmath189 & 416.99(8 ) & 14.79 & @xmath1814,0@xmath189 @xmath190 @xmath1813,0@xmath189 & 416.99(8 ) & 13.72 + & 1017.4(3 ) & 108.24 & @xmath1814,0@xmath189 @xmath190 @xmath1814,0@xmath189 & 416.99(7 ) & 13.72 + & 1534(30 ) & 97.96 & @xmath1814,0@xmath189 @xmath190 @xmath1815,0@xmath189 & 416.99(8 ) & 14.14 + @xmath1814,0@xmath189 @xmath190 @xmath1814,0@xmath189 & 416.99(8 ) & 13.87 & @xmath1814,0@xmath189 @xmath190 @xmath1816,0@xmath189 & 416.99(8 ) & 14.56 + & 1010.7(5 ) & 108.52 & & 1486(32 ) & 98.45 + @xmath1814,0@xmath189 @xmath190 @xmath1815,0@xmath189 & 416.99(7 ) & 14.15 & @xmath1815,0@xmath189 @xmath190 @xmath1817,0@xmath189 & 416(3 ) & 14.56 + & 1012.4(6 ) & 108.45 & & 1301(14 ) & 100.88 + @xmath1814,0@xmath189 @xmath190 @xmath1816,0@xmath189 & 416.99(7 ) & 14.99 + & 1020.6(2 ) & 108.12 + & 1429(20 ) & 99.08 + + + & & & + @xmath181 @xmath182,@xmath183 @xmath184,@xmath185 & @xmath0 & @xmath186 & @xmath181 @xmath70,@xmath183 @xmath184,@xmath185 & @xmath0 & @xmath186 + @xmath1811,0@xmath1880,0@xmath193 & 480.6(7 ) & @xmath194 & @xmath1811,0@xmath1881,0@xmath189 & 480.53(3 ) & 5.36 + & 588.3(2 ) & 329.33 & & 641.2(5 ) & 256.97 + & 695.7(9 ) & 219.39 + @xmath1811,0@xmath1881,0@xmath189 & 480.7(4 ) & @xmath195 & @xmath1811,0@xmath1882,0@xmath189 & 480.58(4 ) & 0.91 + & 587.3(3 ) & 331.55 & & 647.8(6 ) & 251.11 + & 720.47(2 ) & 207.98 + @xmath1811,0@xmath1882,0@xmath189 & 480.68(4 ) & @xmath196 & @xmath1811,0@xmath1883,0@xmath189 & 480.55(4 ) & 3.39 + & 588.3(5 ) & 329.33 & & 644.1(5 ) & 254.42 + & 695.7(5 ) & 219.40 + @xmath1811,0@xmath1883,0@xmath189 & 480.56(4 ) & 3.55 & @xmath1812,0 @xmath1884,0@xmath189 & 480.56(5 ) & 7.84 + & 589.7(4 ) & 329.62 & & 637.7(3 ) & 260.25 + & 675.4(5 ) & 231.02 + to find out the uncertainties with the calculated e1 matrix elements , we have carried out semi - empirical scaling of the wave functions that accounts for evaluation of missing correlations contributions to the wave functions from the approximated sd and sdpt methods . this procedure involves scaling of the excitation coefficients and reevaluation of the e1 matrix elements . the scaling factors are decided from the correlation energy trends in the sd and sdpt methods . details regarding this scaling procedure are given in ref . @xcite . and @xmath197 states of @xmath14ca@xmath5 with @xmath198.,width=321,height=226 ] .[tuneoutj ] the @xmath58 values ( in nm ) for all possible magnetic sublevels @xmath199 of the @xmath14ca@xmath5,@xmath2sr@xmath5 and @xmath57ba@xmath5 alkaline earth - metal ions using circularly polarized light . [ cols="<,^,^,<,^,^,<,^,^ " , ] * @xmath14ca@xmath5 : * as evident from fig . ( [ casd ] ) , the dynamic polarizabilities for the @xmath200 state are small except in the vicinity of the resonant @xmath200-@xmath21 and @xmath200-@xmath23 transitions around 396.847 nm and 393.366 nm respectively . since the @xmath197 states have significant contributions from the resonances in the interested wavelength range , the @xmath0 are expected to lie in between these resonances . we found a total of nine @xmath0 for all possible magnetic sublevels of the @xmath200-@xmath201 transition in between four resonances . the first four @xmath0 are located around 395 nm between the resonant @xmath200-@xmath21 and @xmath200-@xmath23 transitions for all the @xmath199 magnetic sublevels of the @xmath202 state . out of these , two @xmath0 support blue - detuned trap , whereas the other two support red - detuned trap . the next five @xmath0 are identified at 850.9(2 ) nm , 853.1(2 ) nm , 1467.8(4 ) nm , 1013.4(5 ) nm and 870.7(3 ) nm , which lie in the infrared region . for some of the @xmath199 sublevels , the @xmath0 is missing . in such case , it would be imperative to consider a geometry where @xmath0 will be independent of the magnetic sublevels as mention in sec . [ introsec ] and discussed elaborately in our previous work @xcite . in table [ magicd521 ] , we present the @xmath0 for the @xmath203 transition considering all the magnetic sublevels of the @xmath204 state . as seen , most of these @xmath0 in the @xmath14ca@xmath5 ion support red detuned trap , indicated by small positive values of the polarizabilities for the corresponding @xmath0 values . similarly , we tabulate @xmath0 for the @xmath205 ( @xmath81 = @xmath206 ) - @xmath207 transition in table [ magicd322 ] and [ magicd522 ] . it can be evidently seen from the table that @xmath0 are red shifted from the @xmath0 for @xmath205 ( @xmath208 ) - @xmath207 . similarly in table [ magicca ] , we list the @xmath0 values between 300 - 1300 nm for the @xmath209 transitions . the @xmath70-dependent polarizabilities values at the respective @xmath0 are listed as well . for this wavelength range , total twenty four @xmath0 are located . out of which , fourteen @xmath0 are identified in the infrared region . from this table , it can be found that for the @xmath210 transitions , all @xmath0 support red - detuned traps . and @xmath211 states of @xmath2sr@xmath5 with @xmath212.,width=321,height=226 ] * @xmath2sr@xmath5 : * the dynamic polarizabilities for the @xmath213 , @xmath202 and @xmath214 states of @xmath2sr@xmath5 calculated by us are plotted in fig . ( [ srsd ] ) . a number of @xmath0 are identified by the intersections of the polarizability curves of the @xmath213 and @xmath211 states for all their magnetic sublevels of the @xmath211 states in the @xmath215 transitions and are presented in table [ magicd321 ] along with their resonant lines . four @xmath0 are found to be around 413 nm between the @xmath213-@xmath24 and @xmath213-@xmath31 , resonant transitions . these values belong to the visible region , while the other five @xmath0 are located at 1009.3(3 ) nm , 1019.7(3 ) nm , 1062.5(3 ) nm and 1577.2(3 ) nm that lie in the infrared region . all the @xmath0 values mentioned for the @xmath213-@xmath202 transition , except the one at 412.5(9 ) nm , support the red - detuned trapping scheme . similarly , for the @xmath213-@xmath214 transition total eight @xmath0 are appearing for all possible @xmath199 sublevels . among them two @xmath0 are located at 1379.4(3 ) nm and 1130.8(4 ) nm appear after the @xmath214-@xmath31 resonance for the @xmath199=1/2 and @xmath199=@xmath206 magnetic sublevels respectively . use of these @xmath0 are recommended to carry out experiments selectively for the corresponding magnetic sublevels . similarly , we present @xmath0 for the @xmath216 ( @xmath81 = @xmath206 ) - @xmath217 transition in table [ magicd322 ] and [ magicd522 ] . it can be noticed from these tables that @xmath0 are red shifted from the @xmath0 for @xmath216 ( @xmath208 ) - @xmath217 transition . we have also determined total sixteen extra @xmath0 between @xmath213-@xmath24 and @xmath202-@xmath31 resonance transition . in table [ magicca ] , @xmath0 values above 300 nm are listed in the case of @xmath2sr@xmath5 for the @xmath218 transitions . it is found that @xmath0 around 417 nm with very small polarizabilities for the @xmath219 and @xmath220 transitions . therefore , it will be challenging to trap the @xmath2sr@xmath5 ion at these wavelengths . however , the @xmath0 values in the infrared region for these transitions may be useful for trapping the above ion in the experiments . and @xmath221 states of @xmath57ba@xmath5 with @xmath198.,width=321,height=226 ] * @xmath57ba@xmath5 : * total nine @xmath0 are found for the @xmath222-@xmath223 transition of @xmath57ba@xmath5 , among which four @xmath0 are around 468 nm in the vicinity of the @xmath222-@xmath47 resonant transition . the next @xmath0 at 587.6(9 ) nm , 589.5(3 ) nm and 589.6(5 ) nm are located at the sharp intersection of polarizability curves close to the @xmath223-@xmath47 and @xmath223-@xmath43 resonances , as seen in fig . ( [ ba1 ] ) . the last two @xmath0 are located at 841.7(5 ) nm and 690.7(7 ) nm for the @xmath199=@xmath224 and @xmath225 sublevel respectively , have positive polarizabilities . hence these @xmath0 could provide sufficient trap depth at the reasonable laser power . in fact , some of the expected @xmath0 are missing for the @xmath199= @xmath226 and @xmath206 sublevels . similarly , several @xmath0 are also located for the @xmath222-@xmath227 transitions , as seen from fig . ( [ ba1 ] ) , in the wavelength range 300 - 800 nm which are listed in table [ magicd521 ] . the expected trend of locating @xmath0 between the resonances in this transition is similar to the previous two ions . for the @xmath228 ( @xmath229 ) - @xmath230 transition , we list the @xmath0 in table [ magicd322 ] and [ magicd522 ] . these magic wavelengths are slightly red shifted to those demonstrated for @xmath228 ( @xmath229 ) - @xmath230 transition . we also found total fifteen @xmath0 between @xmath222-@xmath43 and @xmath223-@xmath43 resonance transitions in both the tables . we have also determined an extra @xmath0 at 590.5 nm in table [ magicd322 ] , which supports a red detuned trap . in table [ magicca ] , we also list @xmath0 for the @xmath231 transitions which lie within the wavelength range of 300 - 800 nm . we correspondingly locate twenty @xmath0 in the visible region . we notice that all the @xmath0 values , except around 480 nm , are expected to be more promising for experiments . an ion trap at these wavelengths can have sufficient trap depth at the reasonable laser power . dependent dynamic dipole polarizabilities ( in a.u . ) for the @xmath232 ( @xmath233 3 , 4 ) and @xmath234 @xmath235 states of @xmath14ca@xmath5 with @xmath212.,width=321,height=226 ] dependent dynamic dipole polarizabilities ( in a.u . ) for the @xmath236 ( @xmath70= 4 , 5 ) and @xmath237 @xmath238 states of @xmath2sr@xmath5 with @xmath212.,width=321,height=226 ] dependent dynamic dipole polarizabilities ( in a.u . ) for the @xmath239 ( @xmath70 = 1 , 2 ) and @xmath240 @xmath238 states of @xmath57ba@xmath5 with @xmath212.,width=321,height=226 ] * tune - out wavelengths : * table [ tuneoutj ] and [ tuneoutf ] illustrate the identified tune out wavelengths of the @xmath241 , @xmath132 and @xmath133 states in the @xmath242 and @xmath243 levels of the @xmath14ca@xmath5 , @xmath2sr@xmath5 and @xmath57ba@xmath5 alkaline earth metal ions . to locate these tune out wavelengths , we have calculated the dynamic polarizabilities of the above states for a particular range of wavelength in the vicinity of relevant resonances for the corresponding ion and find out values of @xmath244 for which the polarizability values tend to zero . we have also used our the frequency dependent scalar polarizability results of the @xmath14ca@xmath5 , @xmath2sr@xmath5 and @xmath57ba@xmath5 ions to find out @xmath0 that are independent of the magnetic sublevels @xmath81 of the atomic states ; so also hyperfine sublevel independent . table [ magicsd ] lists of the @xmath0 values for the @xmath241-@xmath245 transitions , which lie within the wavelength range of 300 - 1500 nm . we are also able to locate the tune out wavelengths of the ground , @xmath132 and @xmath133 states of the considered alkaline ions that are independent of the @xmath70 , @xmath246 and @xmath199 values of the respective ion and given them in table [ tunesd ] . occurrences of these @xmath0 and @xmath247 for the considered ions can offer pathways to carry out many high precision measurements with minimal systematics . we have determined scalar , vector and tensor polarizabilities of the @xmath241 , @xmath132 and @xmath133 states in the @xmath14ca@xmath5 , sr@xmath5 and @xmath57ba@xmath5 alkaline earth - metal ions with the ground state principal quantum number @xmath17 . we used very precise values of the electric dipole matrix elements that are obtained them by employing a relativistic all - order method . non - dominant contributions in the adopted sum - over - states approach for the evaluation of the polarizabilities are estimated using lower order perturbation methods . the obtained static polarizability values are compared with the available other theoretical results and experimental values to gauge their accuracies . dynamic polarizabilities at the 1064 nm are given explicitly for the @xmath241 and @xmath248 states of the considered alkaline earth - metal ions , which could help in creating high - field seeking traps using the nd : yag laser . furthermore using the dynamic polarizabilities for a wide range of wavelengths , we have located a number of tune out wavelengths @xmath58 of the above states and the magic wavelengths @xmath0 for the @xmath249 clock transitions due to the circularly polarized light in the @xmath14ca@xmath5 , @xmath2sr@xmath5 and @xmath57ba@xmath5 alkaline earth ions . we have located a significant number of @xmath0 for these clock transitions , which can help the experimentalists to trap the above ions to reduce uncertainties in the clock transitions due to stark shifts . this knowledge would also be of immense interest to carry out other high precision studies using the considered ions . in addition , we have also determined the @xmath0 and @xmath58 values that are independent of the choice of magnetic and hyperfine sublevels of the above clock transitions . the work of b.a . is supported by department of science and technology , india and work of j. k. is supported by ugc - bsr grant no . f.7 - 273/2009/bsr , india . s.s . acknowledges the financial support from ugc - bsr . a part of the computations were carried out using vikram-100 hpc cluster of physical research laboratory and the employed sd method was developed in the group of professor m. s. safronova of the university of delaware , usa .

we present a stellar population synthesis study of a type i luminous infrared galaxy ( lirg ) : irasf13308 + 5946 . it is a quasar with absolute magnitude and has a spectral feature of a seyfert 1.5 galaxy . optical images show characteristics of later stages of a merger . with the help of the stellar synthesis code ( cid fernandes et al . 2005 ) and both calzetti et al . ( 2000 ) and leitherer et al.s ( 2002 ) extinction curves , we estimate the past infrared ( ir ) luminosities of the host galaxy and find it may have experienced an ultraluminous infrared galaxy ( ulirg ) phase for nearly 300myr , so this galaxy has probably experienced a type i ulirg phase . both nuclear starburst and active galactic nuclei ( agn ) contribute to the present ir luminosity budget , and starburst contributes% . the mass of supermassive black - hole ( smbh ) is and the eddington ratio is 0.12 , which both approximate to typical values of pg qsos . these results indicate that irasf13308 + 5946 is probably at the transitional phase from a type i ulirg to a classical qso .

1005.3909

luminous infrared galaxies ( lirgs ) and ultraluminous infrared galaxies ( ulirgs ) are galaxies with infrared ( ir ) luminosities @xmath4@xmath5 and @xmath6@xmath5 , respectively . their ir luminosities and space densities are both comparable to qsos ( sanders & mirabel 1996 ) . these galaxies always show signs of tidal interaction and merger , and the interaction rate is observed to increase with ir luminosity ( zou et al . 1991 ; clements et al . 1996 ; murphy et al . 1996 ; wu et al . 1998 ; veilleux et al . nearly all ulirgs are strong interaction or merger systems ( kim et al . 2002 ) . it is widely accepted that galaxy interactions and mergers trigger extreme nuclear activity , as well as more widespread starburst ( toomre & toomre 1972 ; larson & tinsley 1978 ) . the molecular gas concentrations of ulirgs are in their central kpc regions ( downes & solomon 1998 ) and have the ability to form stars with densities comparable to those in elliptical galaxies ( tacconi et al . kormendy & sanders ( 1992 ) proposed that ulirgs may evolve into elliptical galaxies through merger induced , dissipative collapse . structural , kinematic and photometric properties of ulirgs show that they are originated from major mergers and they averagely fall on the fundamental plane of moderate - mass ellipticals ( stellar mass @xmath7@xmath8 ) , but are well offset from giant ellipticals , suggesting that ulirg mergers are ellipticals in formation ( genzel et al . 2001 ; tacconi et al . 2002 ; dasyra et al . 2006a ) . observations revealed that a large fraction of lirgs / ulirgs show spectral characteristics as seyfert galaxies ( wu et al . 1998 ; kewley et al . 2001 ) . the fraction increases dramatically with ir luminosity , and among ulirgs , the fraction is about 50% ( kim et al . 1998a ; veilleux et al . 1999 ; cao et al . 2006 ; yuan et al . 2010 ; nardini et al . 2010 ) , or even higher ( @xmath170% ; nardini et al . 2010 ) . though many of these sources are dominated by agn in their bolometric luminosity ( boller et al . 2002 ; nandra & iwasawa 2007 ) , in most cases the predominance of starburst over agn is proposed ( gu et al . 1997 ; lutz et al . 1999 ; franceschini et al . 2003 ) , and nardini et al . ( 2008 ) proposed a fraction of @xmath9% . from millimeter - wave co observations and optical spectra analyse of 10 ulirgs , sanders et al . ( 1988 ) proposed that ulirgs represent the dust - enshrouded stage of qsos . they proposed a classical evolution scenario that two gas - rich spirals merge ( or interact ) first and the funneled gas toward the merger center triggers nuclear starburst before the ignition of a dust - enshrouded agn . when dust has been consumed or swept away under the radiative pressure of agn and supernovae , an optical quasar would appear . canalizo & stockton ( 2000 , 2001 ) selected a sample of 9 low - redshift qsos which fall onto the intermediate position between the regions occupied by ulirgs and qsos in the far - infrared ( fir ) color - color diagram . all these 9 transition qsos are undergoing tidal interactions and 8 are major mergers . all of them also show strong recent star formation activity within 300myr . they proposed a model involving a dust cocoon that initially surrounds the qso nuclear regions , which can account for all the observed and derived properties in transition objects , and it is consistent with the idea of sanders et al . they suggested that either at least some ulirgs evolve to become classical qsos , or some qsos are born under the conditions as ulirgs and their lifetimes as qsos last @xmath10 300myr . most of these transition qsos are also ulirgs . as kawakatu ( 2006 ) , we refer to qsos and seyfert 1 galaxies that selected from ulirgs as type i ulirgs ( or ir qsos in zheng et al . kawakatu et al . ( 2006 ) and hou et al . ( 2009 ) find that type i ulirgs are the early phase of black - hole ( bh ) growth and they are qsos in formation . the case may be as colina et al . ( 2001 ) had proposed that high - luminosity qsos would be the end point in the merging process of massive ( @xmath11 * ) disk galaxies , while cool ulirgs , which would be the end product in the merging of two or more low - mass ( 0.3@xmath12*0.5@xmath12 * ) disk galaxies , would not evolve into qsos . this scenario is supported by many studies ( mcleod & rieke 1994 ; arribas et al . 2003 ; dasyra et al . 2006a ) . in this paper we present the result of stellar population synthesis of irasf13308 + 5946 and estimations of past ir luminosities . irasf13308 + 5946 is a lirg at present and shows a spectral characteristic of a seyfert 1.5 galaxy , which together suggest a potential transition phase between type i ulirgs and classical qsos . it shows clear absorption lines and young stellar population features for spectral fitting . observations from _ iras _ , 2mass , sdss and first are presented in 2 . we describe the process of extracting the star formation history ( sfh ) , emission - line fitting , aperture correction and the extinction curve in 3 . in 4 we present some estimations like the current star formation rate ( sfr ) , the bh mass , the eddington ratio , etc . , based on the emission - lines fitting . the past ir luminosities are estimated in 5 . conclusions and discussions are given in 6 . we summarize in 7 . through out this paper we adopt a cosmology of @xmath13=71 km s@xmath14 mpc@xmath14 , @xmath15=0.29 and @xmath16=0.71 . irasf13308 + 5946 was observed by the infrared astronomical satellite ( _ iras _ ) , two micron all sky survey ( 2mass ) all - sky extended source survey ( recorded as 2masx j13323783 + 5930538 ) , sloan digital sky survey ( sdss , recorded as sdss j133237.94 + 593053.7 ) and very large array ( vla ) faint images of the radio sky at twenty - centimeters ( first ) survey . it hasnt been explored by uv or x - ray observations by far . the mid - to - far ir fluxes of irasf13308 + 5946 are taken from the _ iras _ faint source catalog ( fsc ) version 2.0 . flux densities at three wavebands , 12 , 25 and 100 @xmath17 m , are upper limits . only the one at 60 @xmath17 m , which is 0.2615 jy , has a record with high quality ( fqual=3 ) . therefore , the ir luminosity ( @xmath18 ) is calculated by @xmath19 ( lawrence et al . 1989 ; bushouse et al . 2002 ; arribas et al . 2004 ; wang et al . 2008 ) , where @xmath20 is the luminosity at 60 @xmath17 m . the derived ir luminosity is @xmath21 @xmath22 . for comparison , we also calculate ir luminosity using an ir spectral energy distribution ( sed ) model . since irasf13308 + 5946 is not only a starburst(sb ) galaxy , it also has a type 1.5 agn ( see 3.4 ) , we adopt the sed model established by siebenmorgen et al . ( 2004 ) for galaxies with both agn and starburst components . after the sed model was scaled to the observed 60@xmath17 m flux , the integrated ir luminosity is calculated to be @xmath23 , exactly the same as the one above . we also calculate the ir luminosity in the range 11000 @xmath17 m through 60 @xmath17 m and 100 @xmath17 m fluxes ( lonsdale - persson & helou 1987 ; calzetti et al . 2000 ) : @xmath24\ ] ] @xmath25 @xmath26 the derived ir luminosity @xmath27 can be used as upper limit , and it is consistent with the result only adopting @xmath28 . therefore , we adopt @xmath29 as the ir luminosity throughout this paper . we do nt do _ k_-corrections on ir fluxes , because the correction factor is small ( @xmath15% adopting sb sed model from siebenmorgen et al . 2007 ) enough to neglect , compared with the uncertainty of ir luminosity calculation and the scatter among different _ k_-correction methods . irasf13308 + 5946 was also observed by sdss imaging camera and spectrograph . the _ u- , g- , r- , i- , z_-band images and the spectrum are from the sdss data release 5 ( dr5 ) . each band was observed with exposure time of 53.9s , and the pixel size is 0.396 on the sky . the spectrum has been corrected for galactic reddening of @xmath30 and redshift of @xmath31 , so the rest - frame wavelength coverage is 32587852 . the sdss spectrum shows a typical feature of a type i agn . the _ i_-band absolute magnitude of @xmath32 satisfies the quasar criterion of @xmath33 defined by schneider et al . ( 2007 ) . however , the _ b_-band absolute magnitude is @xmath34 ( transferred from @xmath35 which adopts @xmath36 km s@xmath14 mpc@xmath14 , vron - cetty & vron 2006 ) , and it does nt satisfy the criterion for quasars of @xmath37 ( @xmath38 km s@xmath14 mpc@xmath14 ) defined by schmidt & green ( 1983 ) . the larger extinction at _ b_-band could explain the inconsistency between the two band ( see also 3.6 ) . there is obvious deviation between sdss and _ iras _ positions . the _ iras _ 2.45-@xmath39 uncertainty ellipse ( corresponding to 95% probability enclosure ) with semi - major and semi - minor axises of 21 and 7 does not embrace the optical ( sdss ) position ( see fig . [ f1eps ] ) . the optical source only locates within the 5-@xmath39 uncertainty ellipse of _ iras _ position . since there is not any other bright sources within _ iras _ uncertainty ellipse in either optical and near - infrared ( 2mass ) bands , and ir phenomenons are most related with galactic activities , we identify the object detected by _ iras _ , 2mass and sdss as the same one . in 2.3 , the well - known radio - fir correlation will confirm such identification . hereafter , we refer to the object as irasf13308 + 5946 . confidence for each axis , which is plotted on sdss _ r_-band image . blue _ circles represent sdss galaxies and _ magenta _ ones sdss stars . the _ cyan _ circle marks the first source in the field of view , and coincides with the galaxy irasf13308 + 5946.,width=272 ] fig . [ f2eps ] ( _ left _ ) gives the _ u_-band image and the 3 diameter fiber of the spectrograph , which covers @xmath110 kpc of the galaxy . two color - map images of _ @xmath40 _ and _ @xmath41 _ are also presented . the field of view ( fov ) of these images are all 25@xmath4225 . these images are made after world coordinate system matching , galactic reddening correction and _ k_-correction ( idl routine from blanton et al . 2003 , v4_1_4 ) . because the seeings at _ u- , g- , r_-bands are 1.3 , 1.3 and 1.1 respectively , we have to match the seeings by convolving the _ r_-band image with a 2-d gaussian kernel with @xmath43 before making @xmath41 image . the @xmath41 image shows a color gradient that the central region of galaxy is quite blue . it could be explained by circumnuclear starburst and/or agn activity . after converting sdss @xmath41 color to color ( jester et al . 2005 ) , we obtained the color varies from @xmath10.25 at the center , to @xmath10.5 at the edge of the fiber aperture and to @xmath441 at the outskirt of the galaxy . if these variations are due to different stellar populations , the ages of the stellar populations may range from @xmath1300myr to @xmath1800 ( assuming solar metallicity ) inside the aperture , and to @xmath4410gyr at the galaxy outskirt , adopting the stellar populations evolution model of bruzual & charlot ( 2003 ; hereafter bc03 ) . considering local dust extinction , the @xmath41 color may be reddened , so the ages obtained above should be somewhat younger . generally , the @xmath40 color distribution is consistent with @xmath41 color according to bc03 . however , unlike @xmath41 color , the bluest region on the @xmath40 image deviates from the galaxy center . it could be attributed to the merger process . from fig . [ f1eps ] and fig . [ f2eps ] we can find characteristics of later stages of a merger . adopting 2mass standard isophotal photometry , h_-band absolute magnitude is @xmath45 , which is 4 times as luminous as @xmath12 * galaxies ( @xmath46 ; colina et al . therefore , the galaxy could be a merger system of massive ( @xmath47 * ) galaxies . the only one first source in the fov of fig . [ f1eps ] is marked with a _ cyan _ circle . the radio source coincides with the optical quasar with a spacial separation of 0.63 . the 1.4 ghz continuum integrated flux density detected by first is 0.93 mjy . it follows the well - known radio - fir correlation which is confirmed by yun et al . ( 2001 ) by the overwhelming majority ( @xmath4898% ) of 1809 ir - selected galaxies ( @xmath49 ) . moreover , the derived parameter @xmath50 ( sramek & weedman 1980 ) is much lower than @xmath51 , indicating that it belongs to radio - quiet populations ( stocke et al . the coincidence between optical and radio position and its radio - fir relation support that the optical and the ir source is the same one . we use the spectral synthesis code ( cid fernandes et al . 2005 ) to derive stellar populations of the host galaxy . fits the observed spectrum @xmath52 with a model spectrum @xmath53 which is made up of a pre - defined set of base spectra . there is no need to give the value of parameters initial guess . it carries out the fitting with a simulated annealing plus metropolis scheme to yield the minimum @xmath54 ^ 2 $ ] , where @xmath55 is the error in @xmath52 at each wavelength . it models @xmath53 by a combination that @xmath56 where @xmath57 , @xmath58 is the normalized flux of the @xmath59 spectrum , @xmath60 is the @xmath59 component of base spectrum , @xmath61 is the value of the @xmath59 base spectrum at the normalization wavelength @xmath62 , @xmath63 is the gaussian distribution centered at velocity @xmath64 and with dispersion @xmath65 of the line - of - sight stellar velocity , @xmath66 is the fraction of light due to component _ j _ at @xmath62 , @xmath67 is the global extinction term and represented by @xmath68 . the nebular and agn emission lines can be obtained by subtracting the model spectrum from the observed one as @xmath69 . uses random markov chains which needs to be appointed an integer seed to generate random numbers during the fitting , and if it uses different seeds , it will give different results even if the parameters are configured identically . such difference will not change the overall populations distribution significantly , but individual component @xmath66 may vary @xmath110 percentage in some cases . to obtain a statistically reliable result , we carried out fittings with a set of seeds generated by monte carlo sampling and adopt the mean value over all seeds for every parameter . we take simple stellar populations ( ssps ) from bc03 and a power - law spectrum of agn ( if needed ) @xmath70 as our base spectra . to minimize the parameter space and to make the result more reliable , we adopt spectral templates with 10 ages ( 0.005 , 0.025 , 0.1 , 0.29 , 0.5 , 0.9 , 1.4 , 2.5 , 4 and 10gyr ) computed with padova 1994 " evolutionary tracks ( alongi et al . 1993 ; bressan et al . 1993 ; fagotto et al . 1994a , b ; girardi et al . 1996 ) and chabrier ( 2003 ) initial mass function ( imf ) . as a type i galaxy , the spectrum of irasf13308 + 5946 is possible to contain a power - law component , though it is not obvious in the observed spectrum . since can include a power - law spectrum in the template base , we can carry out test fittings to confirm whether it exists and find out the value of spectral index @xmath71 . a widely adopted agn power - law slope ( @xmath72 ) over the optical and uv region is @xmath73 ( richstone & schmidt 1980 ) , while other studies had given the values from @xmath741 to 0 ( natali et al . 1998 , and references therein ) , corresponding to @xmath75 . we carry out test fittings with @xmath71 varies from @xmath743.0 to @xmath740.1 at intervals of 0.1 to search for the power - law index , and additional non - power - law fittings with @xmath76 . we optimize the fitting with single metallicity ssps . to search for the best - fit metallicity , we carry out metallicity test with the six metallicities ( z = 0.0001 , 0.0004 , 0.004 , 0.008 , 0.02 , 0.05 ) of bc03 model for every power - law index . after that , we adopt the best - fit metallicity and power - law index for the formal fitting . [ f3eps ] shows the test fitting results with different power - law indices @xmath77 and the six metallicities . we evaluate the fitting quality by the minimum @xmath78 , which is averaged over 25 fittings with different seeds ( randomly and uniformly distributed between @xmath74100000 and 100000 ) . in fig . [ f3eps ] , the minimum @xmath78 set reaches its bottom at z=0.008 and @xmath79 . this is consistent with other studies that metallicities near solar value prevail in starburst galaxies ( tadhunter et al . , 2005 ; pellerin & robert 2007 ) , and also that the power - law continuum slope falls around @xmath80 ( francis 1996 ; vanden berk et al . 2001 ; letawe et al . moreover , the power - law index does nt affect the synthesis very much . we learn from fig . [ f3eps ] that if we fix the metallicity , the variation of @xmath78 among different power - law indices is not significant . its influence on power - law component fraction is showed in fig . [ f4eps ] . these data points are all from the test fits above , but only with subsolar metallicity ( @xmath81 ) . the _ red unfilled diamonds _ with a fixed power - law index represent the agn component fraction at 4020 of 25 different integer seeds . we can see from the figure that in the normal range of power - law index in literature , say , @xmath82 , the average agn fraction keeps almost constant , and the dispersion of the average agn fraction is far less than the one from different seeds . this result indicates that the form of agn component does nt affect the synthesis very much , so neither it affect the stellar component significantly . therefore , we take the parameters @xmath79 , @xmath83 , and 10 ssp ages for formal fitting . the formal fitting involves 100 independent fittings with different seeds . stellar absorption lines are given three times the weight larger than the continuum to emphasize the detail of stellar component . the recognized emission lines like [ o @xmath84 @xmath853726 + 3729 ( [ o @xmath84 @xmath86 ) , ne @xmath873869 , h@xmath88+[o @xmath89 @xmath854959 + 5007 , h@xmath90+[n @xmath84 @xmath856548 + 6583 , etc . , together with nai d5890 ism absorption line are all masked before fitting . the blended fe @xmath91 emission - line regions are also excluded , which are assigned to be 42704700 and 51005600 . these masked regions are illustrated in fig . [ f5eps ] represented by the gaps of the blue curve on the _ left - top _ panel . the blue curve is the error of the observed spectrum , and the inverse of error is the weight at each wavelength in spectral fitting . the observed spectrum @xmath52 and all ssp templates are normalized at wavelength @xmath92 before fitting . the outcome fraction for each stellar component varies within @xmath110 percentage point . such variation is insignificant for main populations . we take the mean value for each parameter as the formal fitting result , such as @xmath93 , @xmath64 , @xmath65 , @xmath78 , etc . these parameters are all summarized in fig . [ f5eps ] . the _ left - top _ panel shows the observed spectrum @xmath52 ( _ green _ ) , the model @xmath53 ( _ red _ ) and the error ( _ blue _ ) . the _ left - bottom _ panel gives the residual spectrum @xmath69 . light - weighted stellar population fractions @xmath93 are shown on the _ right - top _ panel , on which the _ left black _ bar represents the power - law fraction without a concrete age . mass - weighted population fractions @xmath94 are shown on the _ right - bottom _ panel . we learn from fig . [ f5eps ] that the stellar populations of the galaxy consist of two main populations : a younger one @xmath95500myr ( whose 500myr population is insignificant ) and a mediate - to - old @xmath441gyr ( we define young stellar populations as those with ages @xmath95300myr , old populations with ages @xmath441gyr , and intermediate - age populations between them ) . follow cid fernandes et al . ( 2005 ) , we calculated the mean ages weighted by light , @xmath96 and by stellar mass @xmath97 the derived mean ages are @xmath98 and @xmath99 . these figures demonstrate again that the galaxy mass is dominated by old populations with ages of @xmath7 years , and a latest starburst took place about @xmath100 years ago whose massive stars contribute a large fraction of luminosity . the mean velocity dispersion is @xmath101 km s@xmath14 , which is typical of giant ellipticals . the residual spectrum @xmath102 ( the _ purple _ spectrum on the _ left - bottom _ panel ) in fig . [ f5eps ] can be used to measure emission lines , since the stellar component and the agn continuum have been removed . fe @xmath91 emission regions are not included in emission - line fitting . emission lines are modeled by the specfit task in the iraf - stsdas package . 13 components are fitted simultaneously , including the single emission line [ o @xmath84 , [ n @xmath84 and [ s @xmath84 doublets , narrow and broad components of h@xmath88 , h@xmath90 , [ o @xmath89 @xmath103 and [ o @xmath89 @xmath104 ( represented by h@xmath105 , h@xmath106 , and so on ) . each line as well as the narrow and broad components are fitted by a single gaussian profile . the flux ratios of [ o @xmath89@xmath107 , [ o @xmath89@xmath108 and [ n @xmath84 doublets are fixed at their theoretical values . the relative positions of [ o @xmath89 , [ n @xmath84 and [ s @xmath84 doublets are constrained by their laboratory values . fwhms are constrained to be same for each doublets like [ o @xmath89@xmath107 , [ o @xmath89@xmath108 , [ n @xmath84 , [ s @xmath84 and also h@xmath88 , h@xmath90 narrow and broad components . the emission - line properties without aperture or extinction correction are displayed in table 1 . the uncertainties in columns ( 2 ) and ( 3 ) are given by the specfit task . fitting results are shown in two wavelength ranges in fig . [ f6eps ] . the narrow components of h@xmath90 and h@xmath88 have fwhms @xmath1091000 km s@xmath110 , and their broad components are @xmath447000 km s@xmath110 . it can be classified as a quasar with spectral type of seyfert 1.5 galaxy , for both narrow and broad components of the two balmer lines are significant . greene & ho ( 2005 ) found a tight correlation between balmer emission - line luminosities and agn optical continuum luminosity ( little host galaxy contamination ) at 5100 ( @xmath111 at @xmath112 ) . we use this correlation to examine whether our spectral decomposition is successful . we measure @xmath113 from the power - law spectrum which is obtained from spectral synthesis . the model agn spectrum gives @xmath114 . the observed h@xmath115 and h@xmath116 luminosities are @xmath117 and @xmath118 . using greene & ho s correlations , @xmath113 is calculated to be @xmath119 and @xmath120 , respectively . they are both consistent with the 5100 luminosity we modeled , so the spectral decomposition of the agn and host galaxy is reasonable . lccccc line identification & flux & fwhm@xmath121 + @xmath122\lambda$]3727 & 3.96@xmath123 & 725.85@xmath124 + h@xmath105 & 1.96@xmath123 & 587.42@xmath125 + h@xmath106 & 12.17@xmath126 & 7403.59@xmath127 + h@xmath128 & 12.01@xmath129 & 587.42@xmath125 + h@xmath130 & 55.57@xmath131 & 7403.59@xmath127 + @xmath132\lambda4959_{\rm{n}}$ ] & 4.05@xmath133 & 779.80@xmath134 + @xmath132\lambda4959_{\rm{b}}$ ] & 2.83@xmath135 & 2879.82@xmath136 + @xmath132\lambda5007_{\rm{n}}$ ] & 12.26@xmath137 & 779.80@xmath134 + @xmath132\lambda5007_{\rm{b}}$ ] & 8.57@xmath138 & 2879.82@xmath136 + @xmath139\lambda6548 $ ] & 4.29@xmath140 & 543.60@xmath141 + @xmath139\lambda6583 $ ] & 12.69@xmath142 & 543.60@xmath141 + @xmath143\lambda$]6716 & 2.41@xmath123 & 516.52@xmath144 + @xmath143\lambda$]6731 & 1.91@xmath145 & 516.52@xmath144 + we have showed in fig . [ f2eps ] that the observed spectrum is the flux from the 3 diameter fiber of sdss spectrograph . precise calculations must include the flux come from the whole galaxy . table 2 lists the galactic reddening corrected fiber magnitude ( _ fibermag _ , measured within the aperture of a fiber ) , petrosian magnitude ( _ petromag _ , measured by the modified form of the petrosian system ) and _ k_-correction factor for each band . the aperture effect is significant , for the difference between _ fibermag _ and _ petromag _ can be larger than 1 magnitude at some bands . since there s no lager aperture spectrum of the galaxy available by now , aperture correction is necessary in this work . a rough estimation of the aperture effect can be derived via @xmath146 ( hopkins et al . the correction factors for _ u- , g- , r- , i- , z_-bands are 1.68 , 2.2 , 2.58 , 2.60 and 2.57 , respectively . at _ u_-band , the aperture effect is the least , but it is still a factor of @xmath147 . from the spectral synthesis in 3.3 , it is known that the luminosity of the galaxy is dominated by young stellar populations and the agn , which both can be better traced by _ u_-band rather than other four bands , so we carry out aperture correction using _ u_-band s factor . lccccc mag & u & g & r & i & z + fibermag & 18.46 & 17.86 & 17.37 & 17.06 & 16.84 + petromag & 17.90 & 17.00 & 16.34 & 16.02 & 15.82 + k - correction & 0.028 & 0.166 & 0.109 & 0.0213 & 0.086 + equation ( [ aperture ] ) is a simple aperture correction , because it does nt distinguish different origins of the observed light . the 3 fiber is of two orders larger than typical agn scale ( for narrow - line region , @xmath1100 pc ) , so the 3 fiber and the petrosian radius may both embrace the whole agn region . however , the stellar content varies in the two apertures ( we see petrosian radius a larger aperture ) . if we want to derive the flux of a starburst powered emission - line coming from the whole galaxy and adopt equation ( [ aperture ] ) , we will amplify the agn component and therefore underestimate the contribution from stars . the spectral synthesis method provides us an opportunity to improve the aperture correction method above . having decomposed the observed spectrum into agn power - law and stellar component , we can convert their fluxes at a given band to apparent magnitudes . in combination with @xmath148 and @xmath149 , we give a new method for purely stellar component aperture correction . the derived formula is : @xmath150 where @xmath151 and @xmath152 are fluxes of stars at a given band within petrosian and fiber aperture , respectively ; @xmath153 is the apparent magnitude of agn at a given band converted from the flux density of the power - law spectrum . large extinction happens in ulirgs / lirgs and re - emits uv - to - optical emission to fir . the proper dust extinction law we use here is the one given by calzetti et al . ( 1994 ; 1997 ; 2000 ) and leitherer et al . calzetti curve is only applicable to wavelengths longer than 1200 . leitherer et al . ( 2002 ) extended the curve to 970 . at even shorter wavelengths , extinction curve has larger systematic uncertainties , but we extrapolate it a little to 912 to cover the lyman series limit . their curves are identical at 1500 , and the differences at shortward wavelengths are minor . we adopt leitherer et al.s curve for 912@xmath154 1800 , and adopt calzetti curve for 1800@xmath155 9000 . the intrinsic emission @xmath156 can be recovered through the extinction curve @xmath157 via @xmath158}\ ] ] where @xmath159 is the observed spectrum , is the color excess for the stellar and agn continuum spectrum , @xmath160}$ ] corresponds to the global extinction term @xmath161 in equation ( [ mlambda ] ) . is calculated from @xmath162 obtained from the spectral synthesis ( @xmath163 ) and it is @xmath164 . the color excess for nebular gas emission - lines is denoted by and is directly estimated through balmer decrement , the line ratio @xmath165 , whose intrinsic flux ratio is @xmath166 assuming temperature @xmath167 k and case b recombination ( osterbrock 1989 ) . is derived to be @xmath168 . when applying the calzetti curve to b - band , the extinction correction factor is @xmath14 , corresponding to 1.5 magnitude brighter , such the corrected absolute magnitude is @xmath169 , which satisfies the quasar criterion @xmath170 . we can compare the observed ir luminosity with the current sfr . because the h@xmath90 line contains emission from the agn , we use [ o @xmath84 @xmath86 to estimated the sfr . we take color excess @xmath168 for gas reddening and calzetti curve for spectral line extinction correction . the current sfr is calculated by @xmath171)}(m_\odot\,{\rm yr^{-1 } } ) & = & ( 6.58 \pm 1.65 ) \nonumber \\ & & \times 10^{-42}\,l_{[{\rm o~_{ii } } ] } ( { \rm { ergs~s^{-1}}})\end{aligned}\ ] ] ( kewley et al . 2004 ) , where @xmath172}$ ] is the extinction corrected ( a factor of 45.6 ) luminosity of [ o @xmath84 @xmath86 . the derived sfr is [email protected]@xmath174yr@xmath14 . we do nt take aperture correction here , because we have learnt that the starburst takes place mainly at the center region of the galaxy ( see 2.2 ) , where the fiber aperture covers . considering the star - forming region can be also found outside the aperture , we take such sfr as a lower - limit and the _ u_-band aperture corrected sfr as an upper - limit , which is [email protected]@xmath174yr@xmath14 . for starburst galaxies , sfr can be also estimated through ir luminosity ( kennicutt 1998 , and references therein ) . as for irasf13308 + 5946 , since it contains both starburst and agn components , the total ir luminosity @xmath175 only suggests an upper limit of sfr , which is 62.9@xmath174yr@xmath14 . the spectral synthesis has given the starburst fraction of the galaxy , so the sfr can be estimated only by the ir luminosity from the starburst . such calculation will be given in 5 . the bh virial mass is calculated by means of h@xmath115 via @xmath176 ( greene & ho 2005 ) . where @xmath177 is the observed h@xmath115 luminosity . we do nt carry out aperture correction on @xmath177 , because the 3 fiber aperture covers @xmath110 kpc , including the broad - line region definitely . the derived bh mass is @xmath178 . the eddington ratio @xmath3 is calculated adopting the bolometric luminosity @xmath179(5100 ) ( kaspi et al . 2000 ) , where @xmath180(5100 ) is the agn optical continuum luminosity at 5100 . the inferred eddington ratio is @xmath181 , which is typical of pg qsos ( median value 0.24 ; hao et al . 2005 , hereafter hao05 ) and far less than the one of ir qsos ( 1.73 ) and nls1s ( 1.27 ) . these results indicate the smbh is at the late stage of growth . if the smbh grew from a single seed , the duration of the growth can be estimated by the @xmath182-folding timescale , @xmath183 ( haiman & loeb 2001 ) , where @xmath184 is the mass - to - radiation conversion efficiency and @xmath185 is the eddington ratio . for efficient energy production , @xmath186 and the @xmath182-folding timescale is @xmath187yr . it is close to the duration of the starburst phase . this means that before the starburst activity took place , the bh mass was less by a factor of @xmath12.7 , which was still massive enough to support a broad - line region . therefore , the galaxy was probable to harbor a type i agn during the starburst phase . we have little information to determine the origin of the bh , that is , whether it grew from a single seed or from bhs merger when galaxies merged . therefore , the agn emission in the past is hard to estimate . comparing with the uncertainty of agn , we have better defined stellar evolution models ( kurucz 1992 ; lejeune et al . 1997 , 1998 ; bc03 ) . combining with the star formation history , we are able to trace back to the past and calculate the luminosity of the stellar component at a given epoch . then , we could estimate the ir luminosity by calculating the uv - to - optical flux absorbed by dust . because there is no uv observation data available for this galaxy and the shortward end of the rest - frame spectrum is only limited to 3258 , we use model spectrum of uv - band for such calculation . to test the feasibility of our method , we firstly calculate the present fir luminosity by reconstructing the uv - to - optical spectrum . because bc03 ssp templates are in units of solar luminosity per angstrom per solar mass , we can accurately convert the present stellar mass @xmath188 to a spectrum by means of the mass weighted fraction @xmath94 . we reconstruct the model spectrum @xmath189 covering the whole uv - to - optical wavelengths in our calculation ( from 912 to 9000 ) through @xmath190 10^{-0.4(a_\lambda - a_v)}}\ ] ] where @xmath191 is the present time , @xmath188 is the present stellar mass obtained from the spectral synthesis and is derived to be @xmath192 after aperture correction through equation ( [ aperturefine ] ) ( does nt need extinction correction ) , @xmath193 are bc03 ssp templates without normalization , @xmath194 is a double power - law spectrum of agn . the spectral indices of power - law we use here are given by hatziminaoglou et al . ( 2008 ) that @xmath195 for @xmath196 and @xmath197 for @xmath198 . [ f7eps ] shows the reconstructed model spectrum ( _ red - solid _ ) superimposed by the observed one ( _ green - solid _ ) . this model spectrum is used as @xmath159 in equation ( [ intrinflux ] ) . the light in uv - to - optical ( 9129000 ) absorbed by dust and re - emitted to ir is approximately calculated by @xmath199{\rm d}\lambda\ ] ] where @xmath200 and @xmath201 have similar meaning as @xmath202 and @xmath203 in equation ( [ intrinflux ] ) , but they represent the spectra at look - back time _ t_. after _ u_-band aperture correction , the derived ir luminosity is @xmath204 . it approximates to the observed @xmath205 . it demonstrates that our method is feasible in recovering the present fir luminosity , so we can apply the method to the past . meanwhile , it is convenient to calculate the contribution fractions from starburst and agn . we only need to calculate ir luminosities from the two components separately . the starburst and agn contribute @xmath206 and @xmath207 , respectively . the starburst contributes @xmath208 of the fir luminosity , nearly two thirds of the total fir budget . anyway , two components are both important . the current sfr can be estimated through the relation between @xmath209 and sfr ( kennicutt 1998 ) . since the host galaxy contributes @xmath208 of the fir luminosity , we obtain @xmath210yr@xmath14 . it is consistent with the one from [ o @xmath84 luminosity inside the aperture . therefore , the derived stellar populations and agn spectra , the dust extinction and the observed emission line fluxes are concordant with each other . the estimation of the past ir luminosity is carried out in a similar way . we just need to supplement the stellar mass loss of each population at a given look - back time _ t_. because the agn emission in the past is unknown , we just estimate the stellar component contribution . the stellar spectrum ( i.e. the host galaxy ) in the past without extinction is produced through @xmath211 where @xmath212 is the present ( @xmath191 ) fraction of the remaining stellar mass to the initial mass of population _ j _ , @xmath213 is such fraction at a given time _ t _ , @xmath212 and @xmath213 are in the range 0 @xmath214 . from fig . [ f5eps ] ( the _ bottom - right _ panel ) since the recent starburst happened in the past 500myr , we estimate the ir luminosity of the past 25 , 100 , 290 and 500myr separately . take 25myr ago as an example . when the 25 old population was just newly born , it was assigned an age of 1myr and the parameter @xmath213 is equal to 1 . the 100myr old population was 75 old then with @xmath215 . other populations are also configured in the same way . here , the reconstructed spectrum through equation ( [ youngp ] ) is used as the intrinsic spectrum @xmath202 in equation ( [ intrinflux ] ) . the extinction term @xmath216 is assigned to be as the present . the attenuated spectrum through @xmath217 is used as @xmath203 in equation ( [ intrinflux ] ) . because dust extinction increases from optical to uv , we adopt the aperture correction at _ u_-band derived from equation ( [ aperturefine ] ) . calculated through equations ( [ intrinflux ] ) and ( [ lirt ] ) , the ir luminosity transferred from 9129000 to ir at 25myr ago was @xmath218 , which is ulirg luminosity . if we reckon in the agn s contribution , the ir luminosity should be higher . when applying to 100 , 290 and 500myr ago , the estimated ir luminosities from the host galaxy were @xmath219 , @xmath220 and @xmath221 . these ir luminosities are shown as a history in fig . [ f8eps ] . as has been demonstrated in section 4 , the galaxy was probable to harbor a type i agn during the starburst . therefore , a type i ulirg might appeared during the starburst phase and lasted for @xmath1300myr . many ir qsos are extremely strong fe @xmath91 emitters ( zheng et al . 2002 ; lpari et al . 2003 ) . in irasf13308 + 5946 spectrum , we also find fe @xmath91 pseudo - continuum emission in wavelength region 42704700 and 51005600 . we have masked these two wavelength regions before spectral synthesis . actually , we have attempted to subtract fe @xmath91 lines from @xmath52 before fitting . however , as the case of irasz11598@xmath740112 , irasf02065 + 4705 , etc . ( zheng et al . 2002 ) , fe @xmath91 multiplets 37 , 38 ( 45004680 ) are relatively stronger than the fe @xmath91 multiplets 48 , 49 ( 51005400 ) compared with boroson & green ( 1992 , hereafter bg92 ) fe @xmath91 template , and also the vron - cetty et al . ( 2004 ) template . thus , the fe @xmath91 multiplets 37 , 38 are left when the multiplets 48 , 49 are removed . we have also attempted to remove fe @xmath91 lines from the residual spectrum , but the result was not improved . furthermore , including fe @xmath91 template as a component to fit @xmath52 does not bring any improvement at all . therefore , we masked the fe @xmath91 pseudo - continuum regions before spectral synthesis , as have mentioned in 3.4 . the fe @xmath91 @xmath2224570 ( 44344684 ) flux is measured during the procedures above and the line ratio fe @xmath2234570/h@xmath88 is 0.81 . this ratio is typical of optical selected qsos ( bg92 ) . we find that the fe @xmath91 continuum extension to [ o @xmath84 @xmath86 , h@xmath88 , etc . lines is hard to detect , so we measure these lines fluxes directly . in this paper we do nt adopt cardelli et al.s ( 1989 ) extinction curve , because it is for cases which are similar to milky way and not applicable to starburst galaxies . we use the calzetti curve and leitherer et al.s ( 2002 ) extension form , because they are derived from star - forming regions and starburst galaxies . dust in different components and stellar populations vary in temperature , ingredient , amount , geometry , etc . , so different attenuation for each component should be adopted if we could distinguish them . however , as the spectra from the host galaxy and central agn are coupled , and also there is only one extinction term in , the extinction obtained from spectral synthesis is a combination value from both regions . also , there may exist some optically thick star formation regions , which may not be detected even by optical or near - infrared , but they still contribute to ir flux . in this case , the actual extinction could be underestimated . from balmer decrement , the line ratio h@xmath224h@xmath88 , the derived color excess of stellar continuum is @xmath225 , which is twice as the one obtained from spectral fitting . this indicates differential attenuation between the dusty starburst region and older populations . the spectral synthesis gives a 500myr starburst history , which implies a triggering mechanism took place during such epoch . galaxy interactions and mergers have timescales of a few @xmath226yr ( binney & tremaine 1987 ) . the similar timescales imply that the starburst was possibly triggered by galaxies merger . in 4 and 5 , we demonstrated that irasf13308 + 5946 has possibly experienced an ir qso phase since @xmath1300myr ago . follow hao05 , we plot irasf13308 + 5946 on the diagram which shows the relation between ir luminosity and the bolometric luminosity measurement ( represented by @xmath227 ; kaspi et al . 2000 ) in fig . [ f9eps ] . irasf13308 + 5946 locates at a transitional position between ir qsos and pg qsos . the tight correlation followed by pg qsos and nls1s suggests their ir luminosities are associated with the optical through central agns , while ir qsos have ir excess powered by other mechanisms rather than agn . as for iras f13308 + 5946 , we attribute the ir excess to the ongoing star formation . in fig . [ f10eps ] , which shows the ir spectral index @xmath228 ( defined as @xmath229 \over { \rm log}(\lambda_2/\lambda_1)}$ ] ) versus the ir excess @xmath230 , it also locates between ir qsos and pg qsos . the ir excess increase as the dust temperature decrease , suggesting that star formation activities become more and more important in powering ir luminosity of ir qsos and ulirgs , because the temperature of dust heated by stars is lower than that by agn . on the other hand , hao05 gives statistical median @xmath231 values of ir qsos and pg qsos , which are 4.9@xmath232 and 2.1@xmath233 . thus , irasf13308 + 5946 has both bh mass and eddington ratio similar to those of pg qsos . furthermore , as a type 1.5 galaxy , the featureless continuum and emission - lines from the agn may be partially attenuated by the dusty torus , and the agn itself may be powering a classical qso , for the galaxy is an _ i_-band quasar . therefore , irasf13308 + 5946 may evolve into an optical qso when the starburst ceases and the nuclear dust is dissipated by the radiative pressure of agn . in fig . [ f9eps ] and [ f10eps ] , the positional arrangement of ulirgs , ir qsos and classical qsos implies an evolutionary sequence : ulirgs @xmath234 type i ulirgs / ir qsos @xmath234 ( irasf13308 + 5946 ) @xmath234 pg qsos . as the _ h_-band luminosity is @xmath14@xmath12 * , it support the evolutionary scenario proposed by colina et al . ( 2001 ) that ulirgs generated by two or more massive ( @xmath235 ) galaxies would evolve into qsos ( also supported by sanders et al . 1988 and lutz et al . 1999 ) . finally , through _ r_-band surface photometry , we find the outskirt surface brightness profile follows the de vaucouleurs law and the central region follows a point spread function ( psf ) . it is consistent with other studies which show that some ( maybe even higher fraction ) ulirgs may undergo qsos phase in their evolutionary history before they settle down as ellipticals ( zheng et al . 1999 ; arribas et al . 2004 ; dasyra et al 2006b ) . we carry out a study based on stellar population synthesis result of a type i lirg , irasf13308 + 5946 . we find that : \3 . we estimate the past ir luminosity during the starburst epoch . we find it has probably experienced a type i ulirg phase from @xmath237myr ago . when the star formation activity weakened recently , the ir luminosity decrease to the present level ( @xmath238 ) as a lirg . nuclear starburst and agn activity both contribute to the ir luminosity budget , with @xmath170% from starburst . it locates at the transitional position between ir qsos and pg qsos on the plots @xmath240 versus @xmath241 and @xmath228 versus @xmath242 . combining the results above , we conclude that irasf13308 + 5946 is probable an evolutionary transition object from a type i ulirg to a pg qso . the authors are grateful to dr . cai - na hao for her data through private communication . we thank r. cid fernandes for his warm - hearted explanation of and very valuable advices to the paper . we thank dr . stijn wuyts for valuable discussions . this project is supported by nsfc grant 10833006 , 10773014 and the 973 program grant 2007cb815406 . this work is based on the data from first , iras , 2mass and sdss observations . 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 web site is http://www.sdss.org/. the 2mass project is a collaboration between the university of massachusetts and the infrared processing and analysis center ( jpl / caltech ) . funding is provided primarily by nasa and the nsf . the university of massachusetts constructed and maintained the observatory facilities , and operated the survey . all data processing and data product generation is being carried out by ipac . survey operations began in spring 1997 and concluded in spring 2001 . alongi m. , bertelli g. , bressan a. , chiosi c. , fagotto f. , greggio l. , nasie . , 1993 , a&as , 97 , 851 arribas , s. , & colina , l. 2003 , apj , 591 , 791 arribas , s. , bushouse , h. , lucas , r. a. , colina , l. , & borne , k. d. 2004 , aj , 127 , 2522 binney , j. , & tremaine , s. 1987 , galactic dynamics ( princeton : princeton univ . press ) blanton , m. r. , et al . 2003 , aj , 125 , 2348 boller , t. , gallo , l. c. , lutz , d. , & sturm , e. 2002 , mnras , 336 , 1143 boroson , t. a. , & green , r. f. 1992 , apjs , 80 , 109 ( bg92 ) bressan a. , fagotto f. , bertelli g. , chiosi c. , 1993 , a&as , 100 , 647 bruzual g. , & charlot s. , 2003 , mnras , 344 , 1000 ( bc03 ) bushouse h. a. , borne k. d. , colina l. et al . 2002 , apjs , 138 , 1 calzetti , d. 1997 , in aip conf . 408 , the ultraviolet universe at low and high redshift : probing the progress of galaxy evolution , eds . w. h. waller , m. n. fanelli , j. e. hollis , & a. c. danks ( woodbury : aip ) , 403 calzetti , d. , armus , l. , bohlin , r.c . et al . , 2000 . apj 533 , 682 calzetti , d. , kinney , a.l . , storchi - bergmann , t. , 1994 . apj 429,582 canalizo , g. , & stockton , a. 2000 , aj , 120 , 1750 canalizo , g. , & stockton , a. 2001 , apj , 555 , 719 cao , c. , wu , h. , wang , j. l. , hao , c. n. , deng , z. g. , xia , x. y. , & zou , z. l. 2006 , chinese j. astron . astrophys . , 6 , 197 cardelli , j. a. , clayton , g. c. , & mathis , j. s. 1989 , apj , 345 , 245 chabrier g. , 2003 , pasp , 115 , 763 cid fernandes r. , mateus a. , sodr l. , stasiska g. , gomes j. m. , 2005 , mnras , 358 , 363 clements , d. l. , sutherland , w. j. , mcmahon , r. g. , & saunders , w. 1996 , mnras , 279 , 477 colina , l. , et al . 2001 , apj , 563 , 546 dasyra , k. m. , et al . 2006a , apj , 651 , 835 dasyra , k. m. , tacconi , l. j. , davies , r. i. , genzel , r. , lutz , d. , naab , t. , sanders , d. b. , veilleux , s. , & baker , a. j. 2006b , new astronomy review , 50 , 720 downes , d. , & solomon , p. m. 1998 , apj , 507 , 615 fagotto f. , bressan a. , bertelli g. , chiosi c. , 1994a , a&as , 104 , 365 fagotto f. , bressan a. , bertelli g. , chiosi c. , 1994b , a&as , 105 , 29 franceschini , a. , et al . 2003 , mnras , 343 , 1181 francis , p. j. 1996 , publ . astron . soc . australia , 13 , 212 genzel , r. , tacconi , l. j. , rigopoulou , d. , lutz , d. , & tecza , m. 2001 , apj , 563 , 527 girardi l. , bressan a. , chiosi c. , bertelli g. , nasi e. , 1996 , a&as , 117 , 113 greene , j. e. , & ho , l. c. , 2005 , apj , 630 , 122 gu , q. s. , huang , j. h. , su , h. j. , & shang , z. h. 1997 , a&a , 319 , 92 haiman , z. & loeb , a. 2001 , apj , 552 , 459 hao c. n. , xia x. y. , mao s. , wu h. , deng z. g. , 2005 , apj , 625 , 78 hatziminaoglou , e. et al . 2008 , mnras , 386 , 1252 hopkins , a. m. , et al . 2003 , apj , 599 , 971 hou , l. g. , wu , xue - bing , & han , j. l. , 2009 , apj , 704 , 789 jester , s. , et al . 2005 , aj , 130 , 873 kaspi , s. , smith , p. s. , netzer , h. , maoz , d. , jannuzi , b. t. , & giveon , u. 2000 , apj , 533 , 631 kawakatu , n. , anabuki , n. , nagao , t. , umemura , m. , & nakagawa , t. 2006 , apj , 637 , 104 kennicutt , r. c. , jr . 1998 , ara&a , 36 , 189 kewley , l. j. , geller , m. j. , & jansen , r. a. 2004 , aj , 127 , 2002 kewley , l. j. , heisler , c.a . , dopita , m.a . , lumsden , s. , 2001 , apjs , 132 , 37 kim , d .- c . & sanders , d. b. 1998b , apjs , 119 , 41 kim , d .- c . , veilleux , s. , & sanders , d. b. 1998a , apj , 508 , 627 kim , d. c. , veilleux , s. , & sanders , d. b. 2002 , apjs , 143 , 277 kormendy , j. , & sanders , d. b. 1992 , apj , 390 , l53 kurucz r.l . , 1992 , in barbuy b. , renzini a. , eds , proc . iau symp . 149 , the stellar populations of galaxies . dordrecht , kluwer , p. 225 larson rb , & tinsley bm . 219:46 lawrence , a. , rowan - robinson , m. , leech , k. j. , jones , d. h. p. , & wall , j. v. 1989 , mnras , 240 , 329 leitherer , c. , li , i .- h . , calzetti , d. , & heckman , t. m. , 2002 , apjs , 140 , 303 lejeune t. , cuisinier f. , buser r. , 1997 , a&as , 125 , 229 lejeune t. , cuisinier f. , buser r. , 1998 , a&as , 130 , 65 letawe , g. , magain , p. , courbin , f. , jablonka , p. , jahnke , k. , meylan , g. , & wisotzki , l. 2007 , mnras , 378 , 83 lpari , s. , terlevich , r. , daz , r. j. , taniguchi , y. , zheng , w. , tsvetanov , z. , carranza , g. , & dottori , h. , 2003 , mnras , 340 , 289 lonsdale - persson , c. j. , & helou , g. 1987 , apj , 314 , 513 lutz d. , veilleux . s. , & genzel , r. 1999 , apj , 517 , l13 mcleod , k. k. , & rieke , g. h. 1994 , apj , 431 , 137 murphy , t. w. , et al . 1996 , aj , 111 , 1025 nandra k. , iwasawa k. , 2007 , mnras , 382 , l1 nardini , e , et al . 2008 , mnras , 385 , l130 nardini , e. , risaliti , g. , watabe , y. , salvati , m. , sani , e. , 2010 , mnras , inpress ( arxiv:1003.0858 ) natali , f. , giallongo , e. , cristiani , s. , & la franca , f. 1998 , aj , 115 , 397 osterbrock , d. e. 1989 , astrophysics of gaseous nebulae and active galactic nuclei ( mill valley : university science books ) pellerin , a. , & robert , c. , 2007 , mnras , 381 , 228 richstone , d. o. , & schmidt , m. 1980 , apj , 235 , 361 sanders , d. b. , & mirabel , i. f. 1996 , ara&a , 34 , 749 sanders , d. b. , soifer , b. t. , elias , j. h. , madore , b. f. , matthews , k.,neugebauer , g. , & scoville , n. z. 1988 , apj , 325 , 74 schmidt , m. , & green , r. f. 1983 , apj , 269 , 352 schneider , d. p. et al . 2007 , aj , 134 , 102 siebenmorgen , r. , freudling , w. , kr'ugel , e. , & haas , m. 2004 , a&a , 421 , 129 siebenmorgen , r. & kr'ugel , e. 2007 , a&a , 461 , 445 sramek , r. , & weedman , d. 1980 , apj , 238 , 435 stocke , j. s. , morris , s. l. , weymann , r. j. , & foltz , c. b. 1992 , apj , 396 , 487 tacconi , l. j. , genzel , r. , lutz , d. , rigopoulou , d. , baker , a. j. , iserlohe , c. , & tecza , m. 2002 , apj , 580 , 73 tadhunter c. n. , robinson t. g. , gonz@xmath243lez delgado r. m. , wills k. , morganti r. , 2005 , mnras , 356 , 480 toomre a , & toomre j. 1972 . j. 178:623 vanden berk , d. e. et al . 2001 , aj , 122 , 549 veilleux , s. , kim , d .- c . , sanders , d.b . , 1999 , apj , 522 , 113 veilleux , s. , kim , d .- c . & sanders , d. b. 2002 , apjs , 143 , 315 vron - cetty , m .- , joly , m. , & vron , p. 2004 , a&a , 417 , 515 vron - cetty , m. & vron , p. 2006 , a&a , 455 , 773 wang , j. l. 2008 , chinese j. astron . astrophys . , 8 , 643 wu , h. , zou , z. l. , xia , x. y. , deng , z. g. 1998 , a&as , 132 , 181 yuan , t .- , kewley l. j. , sanders d. b. , 2010 , apj , 709 , 884 yun , m. s. , reddy , n. a. , & condon , j. j. 2001 , apj , 554 , 803 zheng , x. z. , xia , x. y. , mao , s. , wu , h. , & deng , z. g. 2002 , aj , 124 , 18 zheng , z. , wu , h. , mao , s. , xia , x .- y . , deng , z .- g . , & zou , z .- 1999 , a&a , 349 , 735 zou , z. , xia , x. , deng , z. , & su , h. 1991 , mnras , 252 , 593

we report the extremely high magnification ( ) binary microlensing event ogle-2007-blg-514 . we obtained good coverage around the double peak structure in the light curve via follow - up observations from different observatories . the binary lens model that includes the effects of parallax ( known orbital motion of the earth ) and orbital motion of the lens yields a binary lens mass ratio of and a projected separation of in units of the einstein radius . the parallax parameters allow us to determine the lens distance kpc and total mass ; this leads to the primary and secondary components having masses of and , respectively . the parallax model indicates that the binary lens system is likely constructed by the main sequence stars . on the other hand , we used a bayesian analysis to estimate probability distributions by the model that includes the effects of xallarap ( possible orbital motion of the source around a companion ) and parallax ( , ) . the primary component of the binary lens is relatively massive with and it is at a distance of kpc . given the secure mass ratio measurement , the companion mass is therefore . the xallarap model implies that the primary lens is likely a stellar remnant , such as a white dwarf , a neutron star or a black hole .

1207.1244

gravitational microlensing is one of the methods that can be employed to search for planetary systems . current microlensing surveys focus primarily on searching for extrasolar planets by monitoring dense star fields in the direction of the galactic bulge . however , stellar binary systems are also detectable using this method and their identification is undoubtedly easier than for planetary binaries . @xcite . stellar binary systems detected via high magnification microlensing events have been studied by @xcite and @xcite , and these events are particularly useful for statistical studies of binary systems @xcite . this is because most high magnification candidates are detected before the peak ( or peaks ) of the light curve and the high magnification regions of the light curve are subsequently monitored by the follow - up networks . this usually leads to comprehensive coverage . therefore , the physical details of the lens system will be revealed by the modeling . gravitational microlensing depends on the mass of the lens and is not a function of lens brightness . therefore , microlensing can readily detect faint stellar remnants such as white dwarfs , neutron stars and black holes . due to their low luminosity , these objects are difficult to detect . isolated white dwarf stars are only really observable optically in the nearby galactic neighborhood , while neutron stars are detected in the radio frequency domain as pulsars . stellar mass black holes have been discovered as part of binary systems via x - ray and optical observations . the macho collaboration have interpreted several microlensing events in the direction of the galactic bulge as due to isolated black holes with stellar masses : macho-96-blg-5 , macho-98-blg-6 @xcite and macho-99-blg-22 @xcite , which is the same event as ogle-1999-bul-32 @xcite . see also @xcite . in this paper , we report the high magnification microlensing event ogle-2007-blg-514 . we describe the observations and data sets in section [ sec : observation ] . the light curve modeling with various effects is presented in section [ sec : modeling ] . we discuss the measurement of the source magnitude and color , and derive the angular einstein radius in section [ sec : correction ] . in section [ sec : blending ] , the blending brightness for this event is discussed . the lens properties including mass and distance are estimated in section [ sec : properties ] . finally , we discuss our results and conclusions in section [ sec : conclusion ] . ogle-2007-blg-514 was announced as a likely microlensing event by the optical gravitational lensing experiment ( ogle ) early warning system ( ews ; @xcite ) on 1 september 2007 ( @xmath12 ) and independently by the microlensing observations in astrophysics ( moa ) collaboration @xcite as moa-2007-blg-464 on 26 september 2007 . the equatorial coordinates for this event are @xmath13 , which yields the galactic coordinates @xmath14 . forty days later ( 11 october ) , the ogle collaboration issued a high - magnification alert . two days after the high - magnification alert , the light curve reached its first peak allowing follow - up observations to be conducted by ten telescopes all around the world . the double peak light curve for this event is shown in figure [ fig : lightcurve ] . the first peak was covered by ogle and ctio ( hjd@xmath15 4386.5 ) , the second peak was covered by bronberg and irsf ( hjd@xmath15 4387.2 ) . the valley between the peaks was covered by two telescopes , canopus and faulkes telescope north . the caustic curve and source trajectory in the best - fit model is shown in figure [ fig : caustic ] . the event was also observedusing high - resolution spectroscopy in order to identify the source star properties @xcite . we utilize this information to estimate the limb darkening coefficients ( section [ sec : ld ] ) . the data set for ogle-2007-blg-514 consists of observations from 12 different observatories representing the ogle , moa , the microlensing follow up network ( @xmath16fun ; @xcite ) , robonet @xcite , the probing lensing anomalies network ( planet ; @xcite ) , as well as the infrared survey facility ( irsf ) . specifically , the data set includes data from the following telescopes , locations and passbands : ogle - iii 1.3 m warsaw telescope at las campanas observatory ( chile ) @xmath17-band , moa - ii 1.8 m telescope at mt john university observatory ( new zealand ) wide-@xmath18-band , moa 0.61 m b@xmath19c telescope at mt john university observatory ( new zealand ) @xmath17 and @xmath20 bands , @xmath16fun 0.4 m telescope at auckland observatory ( new zealand ) @xmath18-band , @xmath16fun 0.35 m telescope at bronberg observatory ( south africa ) unfiltered , @xmath16fun 1.3 m smarts telescope at ctio ( chile ) @xmath20 , @xmath17 , and @xmath21 bands , @xmath16fun 0.5 m telescope at campo catino austral observatory ( cao , chile ) unfiltered , @xmath16fun 0.35 m telescope at farm cove observatory ( new zealand ) unfiltered , robonet 2.0 m faulkes telescope north ( ftn ) at haleakala observatory ( hawaii ) @xmath18 band , robonet 2.0 m faulkes telescope south ( fts ) at siding spring observatory ( australia ) @xmath18 band , planet 1.0 m telescope at canopus observatory ( australia ) @xmath17 band , and irsf 1.4 m telescope at saao ( south africa ) @xmath22 , @xmath21 and @xmath23 bands . the various data sets were reduced using several methods . the ogle data were reduced using the standard ogle dia pipeline @xcite . the moa , @xmath16fun and irsf data were reduced using the moa dia pipeline @xcite . @xmath16fun ctio @xmath24 band data were also reduced using dophot @xcite which we used for the source brightness and color in section [ sec : correction ] . the robonet data were reduced by the dandia pipeline @xcite , and the planet data using pysis version 3.0 reduction pipeline @xcite . the bronberg unfiltered data were not included in the modeling because it may require airmass corrections . fortunately , irsf @xmath25-band data covered the same region as the bronberg data . we have checked that the result was not affected by whether the bronberg data were included or not . the moa b&c data were also not included in the modeling because all of the three day s worth of data points are poor quality due to the weather conditions . moreover , the same region was covered by the moa - ii data . both telescopes are located at mt john , and the moa - ii telescope yields better quality data . the error bars for the data points have been re - normalized such that the reduced @xmath26 of the best fit model @xmath27 . we used the formula , @xmath28 where @xmath29 is the original error value of the @xmath30th data point in magnitude units , and the re - normalizing parameters are @xmath31 and @xmath32 . this non - linear formula operates so that the error bars at high magnification are affected by the parameter @xmath32 , which can be described flat - fielding errors . for the parameter of @xmath32 , we plot the cumulative distribution of @xmath26 where the data points were sorted by error in magnitude , and we choose values of @xmath32 such that the cumulative distribution is a straight line with slope of 1 . then , the parameter of @xmath31 is chosen to be @xmath27 . the @xmath31 and @xmath32 are determined separately for each data set and are shown in table [ tb : norm ] . it is clear from figure [ fig : lightcurve ] that ogle-2007-blg-514 is not a single - lens event due to the double peak structure in the light curve . from the light curve shape alone we can deduce that either the source passed in close proximity to two caustic cusps or it passed across the caustic itself . moreover , the extremely high magnification ( @xmath0 ) of the event indicates that the source star passed near the central caustic . since the ogle and ctio telescopes covered the first peak of the light curve , finite source effects must be included in the modeling . thus , our strategy began by searching for best fit parameters using a binary lens model and finite source parameters . we subsequently introduced parameters to incorporate higher order effects . to properly model finite source effects we need to allow for limb darkening , which accounts for the changing brightness between the source disk center and the rim . we adopted a linear limb darkening law with one parameter for the source brightness @xmath33 . \label{eq : ld}\ ] ] here , @xmath34 is the limb darkening coefficients , @xmath35 is the central surface brightness of the source , and @xmath36 is the angle between the normal to the stellar surface and the line of sight , i.e. , @xmath37=@xmath38/@xmath39 , where @xmath38 is the angular distance from the center of the source as measured in the plane of the sky . from @xcite , the source star ( ogle-2007-blg-514s ) is a g dwarf in the galactic bulge , @xmath40 = 5644 @xmath41 130 k , @xmath42 = 4.10 @xmath41 0.28 cm s@xmath43 , and high - metallicity log[fe / h ] = 0.27 @xmath41 0.09 dex . therefore , we fix the limb darkening coefficients selected from @xcite with effective temperature @xmath40 = 5500 k , surface gravity @xmath42 = 4.0 cm s@xmath43 and metallicity @xmath44[m / h ] = 0.3 ( table [ tb : ldparameter ] ) . the best fit binary lens model parameters were searched for using a markov chain monte carlo ( mcmc ) approach that frequently changes the jump function " in order to efficiently locate the minimum @xmath26 value . for example , refer to @xcite , @xcite and @xcite . a single lens microlensing model has three parameters : the time of peak magnification , @xmath45 , the einstein radius crossing time , @xmath46 , and the minimum impact parameter , @xmath47 . a binary lens model requires three additional parameters : the mass ratio , @xmath48 , which is the mass of the companion relative to the mass of the primary lens ; the binary lens separation , @xmath49 , which is the separation of the binary components projected on to the lens plane and normalized to the einstein radius ; and the angle of the source trajectory relative to the binary lens axis , @xmath50 . an additional required parameter , @xmath51 , is the source radius relative to the angular einstein radius which in combination with the limb darkening law is used to model the finite source effects . furthermore , there are two parameters for each data set and passband that are required to describe the individual unmagnified source and background fluxes . we conducted a broad parameter search with initial parameters distributed over the ranges @xmath52 and @xmath53 . the best models from this search involved a binary lens with a mass ratio of @xmath54 and the trajectory of the source star making a close approach to the central caustic ( figure [ fig : caustic ] ) . given that the primary lens has a stellar mass , the lens companion is also a star with a stellar mass and not a planet . two models with approximately equal @xmath26 values yielded lens component separations of @xmath55 and @xmath56 in units of the einstein radius . this degeneracy in the component separation @xmath57 for a central caustic event was predicted by @xcite . the central caustic has a diamond - like shape . so , we expect that the source trajectory angle @xmath50 will have four possible solutions involving approaches to two caustic cusps . the general parameter search yielded two preferred solutions with @xmath58 and @xmath59 , which correspond to the binary degeneracy @xcite . the other two expected angles @xmath60 and @xmath61 do not generate good models compared with the @xmath58 and @xmath59 models . this is because the magnification contours near the caustic are not perfectly symmetric . as a result , we get four binary lens models with close / wide separation degeneracies and impact parameter @xmath47 degeneracies . the best fit parameters are listed in table [ tb : parallax ] , and are denoted as close1 and wide1 . from the binary modeling ( section [ sec : binary ] ) , the einstein radius crossing time , @xmath46 , is very long , @xmath62 days , implying that there is a good chance to detect earth s orbital parallax effects in the light curve . microlens parallax falls into two different categories , the earth s orbital motion around the sun and the difference of telescope location on the earth . these parallaxes are called orbital parallax " and terrestrial parallax " , respectively . we searched for a parallax model including these effects both separately ( models 2 " and 3 " in the table [ tb : parallax ] ) and combined ( models 4 " in the table [ tb : parallax ] ) . from now on models with parallax " will mean with both effects taken into account . two additional parameters , @xmath63 and @xmath64 , express the parallax effect on the light curve @xcite . these are the two components of the microlens parallax amplitude , @xmath65 . the parallax amplitude @xmath66 is also represented by the lens - source relative parallax , @xmath67 , and the angular einstein radius @xmath68 , @xmath69 where @xmath70 is the einstein radius projected onto the observer plane . from the parallax amplitude parameter , the degeneracy of three physical parameters : lens mass , distance and transverse velocity in @xmath46 , is broken , allowing us to determine the properties of the lens . in our parallax model search , we explored four classes of models with close and wide separations and also in case of impact parameter @xmath71 and @xmath72 based on the results of the binary lens model search . the parameters for the parallax model are listed in table [ tb : parallax ] . we found that the @xmath26 of the best fit parallax model was improved by 19 over the non - parallax model , and these models are denoted as 4 " in the table . most of the improvement of @xmath26 is from the moa and ogle data . this is reasonable because the moa and ogle data cover a large portion of the light curve rendering orbital parallax effects significant for the light curve . the @xmath71 model is a better fit than the @xmath72 model , but since the improvement is only @xmath73 , the @xmath47 sign degeneracy is not decisively resolved . aside from the microlens parallax effects , it is possible that the lens orbital motion effect also have influenced the light curve . the orbital motion of the lens has strong effects when the source passes through or approaches the caustics . for the orbital motion of the lens , we require two additional orbital motion parameters , @xmath74 and @xmath75 , @xcite . these parameters indicate the binary rotation rate , and the uniform expansion rate in binary separation , @xmath49 . therefore , the new ones of @xmath76 and @xmath77 are described as @xmath78 the results of the lens orbital motion modeling are shown in table [ tb : om ] . the close separation model is better fit than the wide separation model . but , the @xmath47 sign degeneracy is not decisively resolved . then , we also performed fits including both parallax and lens orbital motion effects . the results are shown in table [ tb : om ] . the best fit model with the parallax and lens orbital motion indicates that the companion of the lens have the orbital motion parameters of @xmath79 rad days@xmath80 and @xmath81 days@xmath80 . it is already known that @xmath74 is often degenerate with @xmath82 , which is the component of @xmath83 perpendicular to the apparent acceleration of the sun projected on the sky @xcite . for this event , we could break this degeneracy . for composition of the lens system , the projected velocity of the lens companion should be smaller than the escape velocity of the lens system : @xmath84 @xcite , where , @xmath85 and where @xmath86 . we confirmed that the results for each model was not over the escape velocity of the lens system , using the lens mass and distance calculated by the combined parallax and lens orbital motion model ( section [ sec : properties ] ) . the orbital parallax effect is caused by the earth s orbit around the sun . on the other hand , if a companion star is orbiting about the source star , the light curve is affected in the same way with microlens parallax . this effect is called xallarap " . it has been discussed that the earth s orbital parallax effect can be degenerate with the xallarap effect @xcite . the xallarap model has five additional parameters : the two components of xallarap amplitude , @xmath87 and @xmath88 , which represent the xallarap amplitude , @xmath89 , the direction of observer relative to the source orbital axis , r.a.@xmath90 and decl.@xmath90 , and the orbital period , @xmath91 . for an elliptical orbit , two additional parameters are required : the orbital eccentricity , @xmath92 and time of periastron , @xmath93 . in our xallarap model fit , the two parameters for an elliptical orbit are fixed as the parameters for the earth s orbit . note that we assumed that the brightness from the companion of the source is low , so we did not include the additional parameter for the source companion brightness . we also impose on our xallarap the kepler constraint . the @xmath94 is represented by the following equation , estimated from kepler s third law : @xmath95 where @xmath96 is the einstein radius projected onto the source plane ( @xmath97 ) , @xmath98 is the separation of the source companion , @xmath99 and @xmath100 are the mass of the primary and companion source , respectively . to find the best xallarap model that is allowed by kepler s third law , we have done mcmc runs with an additional constraint to @xmath26 @xcite given by @xmath101 where @xmath102 is evaluated by equation [ eq : xallarap_amp ] with parameters in each step of the mcmc and fixed values of @xmath103 and 50% error in @xmath102 , which depend only weakly on other parameters . here , @xmath104 is the heaviside step function . first , we prepared the initial parameters of the orbital period as fixed parameters in the fits in interval of 20 days from 10 to 400 days , and the others as free parameters . the @xmath26 distribution for the orbital period is shown in figure [ fig : xallarap ] . we found that a short orbital period produces a good model , so we restarted the fit around @xmath105 days without any fixed parameters . note that the result is not changed even if the model includes the kepler constraint , due to the small xallarap amplitude ( which is shown by the dashed line in figure [ fig : xallarap ] ) . the best fit xallarap model has @xmath106 , which is significantly better than the only parallax effect model , giving @xmath107 . the best fit parameters are shown in table [ tb : xallarap ] . the parameter @xmath91 is not consistent with the earth s orbital period , so this xallarap model is different than the parallax model . the best fit xallarap @xmath26 value is smaller than the parallax model and the combined parallax and lens orbital motion model . moreover , the xallarap model is different than the parallax model . this implies that the xallarap effect is considered more significant than the parallax effect on the light curve . however , spurious xallarap signals at this level are common in microlensing @xcite . and , the data on this event has high signal - to - noise ratio ( s / n ) , so we expect that stronger false signals are caused by systematics . therefore , the parallax and lens orbital motion model is the most plausible solution , but we can not absolutely excluded the xallarap model . thus , we consider two solutions of the parallax model and the xallarap model for the estimation of the lens properties . for the estimation of the angular einstein radius , @xmath68 , in section [ sec : correction ] , we used the best fit parameters of the model with both parallax and lens orbital motion effects ( close separation and the @xmath72 model ) . the best fit model has a binary mass ratio of @xmath1 and a separation of @xmath2 in a unit of einstein radius with @xmath108 . in contrast , the probability distribution of the lens properties was also estimated by using the xallarap model parameters in section [ sec : properties ] . we performed fits including both parallax and xallarap effects because parallax amplitude is related to the lens mass . the results are listed in table [ tb : xallarap ] . the xallarap and parallax model has a binary mass ratio of @xmath7 and a separation of @xmath8 in a unit of einstein radius with @xmath109 , and @xmath110 days . since the orbital period of the source companion is very short , it may be that parallax and xallarap do not affect the same part of the light curve . it is possible that the parallax results are unaffected by xallarap effects . therefore , the parallax signal has not been detected on the light curve significantly , so we set the upper limit of the parallax amplitude @xmath111 from the @xmath112 contour map of parallax parameters in figure [ fig : parallax ] . this upper limit will imply a lower limit of the lens mass . the source star angular radius @xmath39 was determined using source magnitude and color . the source star magnitudes and colors estimated from the light curve fit need to be corrected for extinction and reddening due to the interstellar dust in the line of sight . the red clump giant ( rcg ) is the standard candle to estimate extinction and reddening . the color magnitude diagram ( cmd ) was made from ctio @xmath17- and @xmath20- band stars within 2@xmath113 of the source star calibrated to the ogle - iii calalog ( figure [ fig : cmd ] ) . in this cmd , we find the rcg centroid to be , @xmath114 we adopt the galactic bulge rcg magnitude @xmath115 from @xcite and color @xmath116 from @xcite . according to @xcite , the clump in this field is 0.1 mag brighter than the galactic center , which we take to be at @xmath117 kpc @xcite . hence , the distance modulus of the clump is dm=@xmath118 . thus , the dereddened rcg centroid is expected to be @xmath119 comparing these centroids , we find the extinction value @xmath120 and reddening value @xmath121 to be , ( @xmath122 ) = ( 2.24 , 2.07 ) @xmath41 ( 0.11 , 0.06 ) . the source star magnitudes from the light curve fit are @xmath123 . applying @xmath124 and @xmath121 , the dereddended source color and magnitude @xmath125 is calculated to be @xmath126 we estimate @xmath127 magnitude from @xmath128 and @xcite color - color relation , and so @xmath129 . from @xcite , the relationship between color / brightness and a star angular radius is @xmath130 so the source angular radius is @xmath131 thus , the angular einstein radius @xmath68 is calculated by the source angular radius @xmath39 and @xmath132 to be @xmath133 and finally the lens - source relative proper motion @xmath134 is @xmath135 blending magnitudes obtained from the fits are listed in table [ tb : flux ] . the result of the fit indicates that the blending flux in the ctio @xmath20-band is @xmath136 adu , however the calculated flux is consistent with zero flux to within 2 sigma . the blending flux in the ctio @xmath17-band is also consistent with zero flux to within 2 sigma . thus , we estimate the upper limit of the brightness by using baseline images from the ctio @xmath137-band and irsf @xmath25-band , by taken about 4 years after the peak so that the source star is not magnified by the microlensing . the @xmath17- and @xmath20-band magnitude were calibrated to the ogle - iii catalog magnitude , and the @xmath25 band magnitude were calibrated to the 2mass catalog . the result is that @xmath138 , and other results are listed in table [ tb : flux ] . these upper limits are estimated from the equation , @xmath139 , where @xmath140 is sky flux at this target position , @xmath141 is a psf area , @xmath142 is the gain value , @xmath143 is the scale factor to calibrate to catalog magnitude . these parameters are listed in table [ tb : upperlimit ] . for @xmath17-band upper limit on the brightness , we use the results of the modeling in the ctio data to estimate , @xmath144 . in section [ sec : modeling ] and [ sec : correction ] , we obtained the best fit model parameters and the angular einstein radius from the source brightness . from them , we estimated the lens properties for the two solutions of the parallax model and the xallarap model . the both of finite source and parallax effects allow us to determine the physical parameters of the lens properties as the lens total mass @xmath145 , distance @xmath146 and velocity @xmath147 . the lens total mass is described by the angular einstein radius @xmath68 and microlens parallax amplitude @xmath66 as given by @xmath148 where @xmath149 . the distance is derived from the equation [ eq : parallax ] . we calculate the lens properties from the parameters of the combined parallax and lens orbital motion model ( close separation and @xmath72 ) . the lens total mass is @xmath150 , as the components of the lens , the primary lens has @xmath151 with a companion @xmath6 . the distance to the lens is @xmath152 . the upper limit of the brightness derives the upper limit of the lens mass . the absolute magnitude @xmath153 is calculated by @xmath154 where @xmath155 is the apparent magnitude of the blend , @xmath156 and @xmath157 are the extinction toward the source and lens , respectively . since the lens should be in front of the source , the extinction toward the source is larger than the one of the lens , ( @xmath158 . therefore , the upper limit of the brightness is @xmath159 . note that this upper limit is quite conservative because there is a possibility that some of the dust is behind the lens . the upper limit of the brightness implies that the upper limit of the lens mass is @xmath160 , assuming the primary lens is a main sequence star . the main sequence star is the dominant component in the galaxy , and the upper limit of the lens mass , assuming that the primary lens is a main sequence star , is consistent within the error bars of the primary lens mass calculated by the parallax model parameters . thus , the parallax model indicates that the binary lens system is likely constructed by the main sequence stars . we also estimated the probability distribution of the lens properties using the combined xallarap and parallax model ( close separation and @xmath71 ) . from the fit parameter for the finite source effects , @xmath51 , we were able to calculate the angular einstein radius . in section [ sec : correction ] , the source angular radius , @xmath39 , and the angular einstein radius , @xmath68 , are estimated by the combined parallax and lens orbital motion model . on the other hand , for the combined xallarap and parallax model , the @xmath161 were invariant with the different model . therefore , we can estimate these quantities using the equation @xcite , @xmath162 where @xmath163 is the source flux as determined from the model , and @xmath164 is the remaining set of factors . the @xmath165 and @xmath166 are calculated by the model parameters of @xmath167 days and @xmath168 that @xmath169 , and @xmath170 mas . note that the index of xa " indicates the parameter of the combined xallarap and parallax model . we estimate the probability distribution from bayesian analysis by combining the equation [ eq : mass ] and the measured values of @xmath166 and @xmath171 . the mass function adopted for the calculation based on @xcite table s3 , model # 1 . the main sequence stars and brown dwarf stars form a power - law function , @xmath172 where @xmath173 for ( @xmath174 ) , @xmath175 for ( @xmath176 ) and @xmath177 for ( @xmath178 ) . the stellar remnant stars are assumed to be white dwarfs ( wds ; @xmath179 ) , neutron stars ( nss ; @xmath180 ) and black holes ( bhs ; @xmath181 ) . these remnants are represented by gaussians . the fraction of initial numbers of these objects in the four classes mss ( which include main sequence and brown dwarf stars ) , wds , nss and bhs are distributed as 88.8 , 10.0 , 1.0 , and 0.2 , respectively . the distance to the galactic center is assumed to be 8 kpc and the upper limit of microlens parallax amplitude , @xmath182 , is adopted in the calculation , which affects the lower limit of the lens mass . the upper limit of the brightness , @xmath183 , is also included . the probability distribution from the bayesian analysis is shown in figure [ fig : likelihood ] . the distribution described in blue indicates the results for all mass functions , and the curves in black and red represent the results for only from mss and wds , respectively . in the bottom - right panel , the probability distribution for the brightness is derived using only main sequence stars and excludes remnant stars as these should be very faint . the solid and dashed curve indicate the probability of the brightness with and without the upper limit of the brightness @xmath184 , respectively , and the vertical dashed line represents the upper limit of the brightness . the probability ratio for each class of mss , wds , nss , and bhs is 0.35 , 0.23 , 0.09 , and 0.33 , respectively , which indicates that the primary lens is likely a stellar remnant 65% of the time , such as a white dwarf , a neutron star or a black hole . as a result , the lens properties were derived from the probability distribution as follows : the primary star is a massive star with a mass of @xmath9 , distance of @xmath10 kpc , and the companion has a mass of @xmath11 . the einstein radius is @xmath185 au , which means that the projected separation is @xmath186 au and the physical three - dimensional separation is @xmath187 au on the close and wide separation model , respectively . the physical three - dimensional separation was estimated by putting a planetary orbit at random inclination , eccentricity and phase @xcite . recently , the mass functions of stellar remnants have been updated by improvements of observational methods and detections of stellar remnants ( white dwarfs : @xcite , neutron stars : @xcite , and black holes : @xcite ) . however , the qualitative conclusions are not affected by these uncertain mass functions . thus , we can conclude that the primary lens is most likely to be a stellar remnant such as a white dwarf , a neutron star or a black hole . we have reported the binary microlensing event ogle-2007-blg-514 . this is an extremely high - magnification event ( @xmath0 ) , which enabled follow - up observations of this event . from the binary lens model search , we found that the lens has a stellar binary mass ratio @xmath188 . the well - known degeneracies of the close / wide separation and impact parameter @xmath71 and @xmath72 both remained unresolved in our light - curve analysis . we also search for the higher order effects of microlens parallax , lens orbital motion , and xallarap ( source orbital motion ) for this event . the xallarap model @xmath26 is better than the parallax and lens orbital motion model , respectively . however , spurious xallarap signals at this level are common in microlensing @xcite . and , the data on this event has high s / n , so we expect that stronger false signals can be caused by systematics . therefore , the combined parallax and lens orbital motion model is the most plausible solution , but we can not absolutely exclude the xallarap model . thus , we consider two solutions of the parallax model and xallarap model for the estimation of the lens properties . the best fit model with parallax and lens orbital motion has a binary mass ratio of @xmath1 and a separation of @xmath2 in a unit of einstein radius with @xmath108 . on the other hand , the xallarap and parallax model has a binary mass ratio of @xmath7 and a separation of @xmath8 einstein radius . the parallax model parameters allow us to determine the lens properties . according to the model with parallax and lens orbital motion effects , the lens total mass is @xmath150 , as the components of the lens , the primary lens has @xmath189 with a companion @xmath190 . the distance to the lens is @xmath152 . the upper limit of the brightness implies that the upper limit of the lens mass is @xmath160 , assuming the primary lens is a main sequence star . the main sequence star is the dominant component in the galaxy , and the upper limit of the lens mass , assuming that the primary lens is a main sequence star , is consistent within the error bars of the primary lens mass calculated by the parallax model parameters . thus , the parallax model indicates that the binary lens system is likely constructed by the main sequence stars . on the other hand , the probability distribution of the lens properties was estimated by the bayesian analysis using the xallarap and parallax model . the parallax amplitude , @xmath111 , is adopted for the upper limit of the lens mass . also , the upper limit of the brightness in @xmath17 band was included in the analysis . as a results , the primary star is a massive star with a mass of @xmath9 , a distance of @xmath10 kpc , and a companion mass of @xmath11 . the probability ratio for each class of mss , wds , nss , and bhs is 0.35 , 0.23 , 0.09 , 0.33 , respectively , which indicates that the primary lens is likely a stellar remnant 65% of the time , such as a white dwarf , a neutron star or a black hole . moreover , based on the ogle astrometry , we found that the blended light is associated with the event to high precision ( the difference of the position of the source and lens is estimated to be 23 mas , i.e. , within the astrometric error . ) . therefore , we considered three possibilities of where the blended light was coming from . the first is the primary lens . this suggestion is consistent with the parallax model that the primary lens is likely a main sequence star , which has a solar mass . the second is the companion of the lens . this is possible to explain by the results of the xallarap model . for example , if the primary lens is a black hole , the companion of the lens have a mass close to a solar mass . the last is the companion of the source in the xallarap model . but , it is difficult to interpret that the blended light is only coming from the companion of the source because the companion of the source is far from the earth and very low mass , so the companion is very faint . as a final conclusion , we found two solutions for this event . the one is the parallax model , which indicates that the binary lens system is likely constructed by the main sequence stars . the another is the xallarap model , which indicates that the primary lens is likely the stellar remnant , such as a white dwarf , a neutron star or a black hole . in the future , follow - up observations by radio , optical , or x - ray space telescopes will identify whether the primary lens is a main sequence star or a stellar remnant . high spatial resolution observations with large ground - based telescope and adaptive - optics instruments , such as the subaru telescope or the very large telescope ( vlt ) , could estimate more precise brightness from the lens using the source brightness obtained from the modeling . in addition , it has been 4 years since the peak magnification of this event , and the lens - source relative proper motion is @xmath191 mas year@xmath80 , indicating that the separation of the source and lens on the sky plane is about 10 mas by now . high resolution observation with the space telescope , hubble space telescope ( hst ) , could detect an elongated psf blended with the lens and source , and verify the nature of the lens . microlensing searches are going on throughout the world . in particular , the ogle group began using a new wide field of view camera , the ogle - iv ( 1.4 deg@xmath192 fov ) , in march 2010 and the number of microlensing events dramatically increased after the ogle - iv upgrade . of course , the moa group is continuing to operate its microlensing search with the moa - ii telescope with its wide field of view ( 2.2 deg@xmath192 ) moa - cam3 ccd camera @xcite . in 2011 , two survey groups found about 1800 microlensing events , which constitutes a two fold increase over last year . more and more binary lens objects varying from planetary systems to massive binary systems can be discovered by further microlensing observation . we acknowledge the following support : the ogle project has received funding from the european research council under the european community s seventh framework programme ( fp7/2007 - 2013 ) / erc grant agreement no . 246678 to au . the moa project is supported by the grant - in - aid for scientific research ( jsps17340074 , jsps18253002 ) , jsps research fellowships and the global coe program of nagoya university quest for fundamental principles in the universe " from jsps and mext of japan . is supported by jsps research fellowships for young scientists . robonet ( kh , db , md , ras , cs , yt ) acknowledges support from the qatar foundation via qnrf grant nprp-09 - 476 - 1 - 78 . abe , f. , et al . 2003 , , 411 , l493 agol , e. , kamionkowski , m , koopmans , l. v. e. , & blandford , r. d. 2002 , , 576 , l131 albrow , m. , et al . 1998 , , 509 , 687 albrow , m. d. , et al . 2009 , , 397 , 2099 an , j. h. , albrow , m. d. , beaulieu , j .- p . , et al . 2002 , , 572 , 521 gould , a. 2000 , , 542 , 785 gould , a. 2004 , , 606 , 319 gould , a. , et al . 2010 , , 720 , 1073 han , c. , @xmath19 gould , a. 2003 , , 592 , 172 han , c. , & hwang , k .- h . 2009 , , 707 , 1264 hwang , k .- h . , 2011 , , 413 , 1244 kepler , s. o. , kleinman , s. j. , nitta , a. , koester , d. , castanheira , b. g. , giovannini , o. , costa , a. f. m. , & althaus , l. 2007 , , 375 , 1315 kervella , p. , thvenin , f. , di folco , e. , & sgransan , d. 2004 , , 426 , 297 kiziltan , b. , kottas , a. , & thorsett , s. e. 2010 , submitted to the , arxiv:1011.4291 mao , s. , et al . 2002 , , 329 , 349 nataf , d. 2012 , in preparation nishiyama , s. , et al . 2005 , , 621 , l105 zel , f. , psaltis , d. , narayan , r. , & mcclintock , j. e. 2010 , , 725 , 1918 poindexter , s. , afonso , c. , bennett , d. p. , glicenstein , j - f . , gould , a. , szymanki , m , k. , & udalski , a. 2005 , , 633 , 914 rattenbury , n. j. , et al . 2005 , , 439 , 645 sako , t. , et al . 2008 , exp . astron . , 22 , 51 skowron , j. , et al . 2011 , , 738 , 87 sumi , t. , et al . 2010 , , 710 , 1641 sumi , t. , et al . 2011 , , 473 , 349 tsapras , y. , et al . 2009 , astron . nachr . , 330 , 4 udalski , a. , et al . 1994 , , 436 , l103 udalski , a. 2003 , acta astron . , 53 , 291 verde , l. , et al . 2003 , , 148 , 195 yee , j. c. , et al . 2009 , , 703 , 2082 yelda , s. , ghez , a. m. , lu , j. r. , do , t. , clarkson , w. , & matthews , k. 2011 , in asp conf . 439 , the galactic center : a window to the nuclear environment of disk galaxies , ed . m. r. morris , q. d. wang , & f. yuan ( san francisco , ca : asp ) , 167 of the best xallarap model as a function of the orbital period of the source star and its companion . the dashed and solid lines indicate the model with and without the kepler constraint , respectively . the cross represents the best fit model with parallax effect . ] color magnitude diagram of stars within 2@xmath113 of the ogle-2007-blg-514 source using @xmath16fun ctio data calibrated to the ogle - iii catalog . the triangle and dot indicate the source star and the center of the red clump giant , respectively . ] , distance , @xmath146 , einstein radius @xmath194 and @xmath17-band magnitude of primary lens brightness . the vertical solid lines in yellow indicate the median values . the distribution described in blue indicates the results with the entire mass function and the curves in black and red represent the results for only main sequence stars and white dwarfs , respectively . the dark and light blue shaded regions indicate the 68% and 95% confidence limits , respectively . in the bottom - right panel , the upper limit of brightness , @xmath184 is shown by the vertical dotted line . the dashed curve indicates the rejected area due to the lens brightness constraint . note that @xmath195 is not fixed in the actual bayesian analysis . ] lcc ogle & 1.16 & 0.01 + moa - ii & 0.85 & 0.02 + @xmath16fun auckland & 0.93 & 0.00 + @xmath16fun smarts ctio @xmath20-band & 3.00 & 0.00 + @xmath16fun smarts ctio @xmath17-band & 1.81 & 0.00 + @xmath16fun smarts ctio @xmath21-band & 1.00 & 0.01 + @xmath16fun campo catino austral & 0.60 & 0.00 + @xmath16fun farm cove & 0.92 & 0.00 + robonet faulkes telescope north & 0.73 & 0.01 + robonet faulkes telescope south & 0.48 & 0.00 + planet canopus & 0.72 & 0.00 + irsf @xmath22-band & 0.76 & 0.01 + irsf @xmath21-band & 0.83 & 0.01 + irsf @xmath196-band & 0.76 & 0.01 + crrrrrrrrrr close1.+ & 4386.784 & 198.96 & 1.60 & 0.249 & 0.0777 & 0.690 & 2.93 & & & 3601.37 + @xmath197 & 0.001 & 3.14 & 0.02 & 0.007 & 0.0007 & 0.002 & 0.05 & & & + & & & & & & & & & & + close1.- & 4386.784 & 199.80 & -1.59 & 0.238 & 0.0787 & 5.595 & 2.91 & & & 3601.50 + @xmath197 & 0.001 & 3.85 & 0.03 & 0.009 & 0.0008 & 0.002 & 0.06 & & & + & & & & & & & & & & + close2.+ & 4386.782 & 254.56 & 1.27 & 0.227 & 0.0709 & 0.679 & 2.29 & -0.206 & 0.021 & 3580.37 + @xmath197 & 0.001 & 3.40 & 0.02 & 0.007 & 0.0009 & 0.002 & 0.05 & 0.013 & 0.006 & + & & & & & & & & & & + close2.- & 4386.783 & 205.34 & -1.57 & 0.251 & 0.0765 & 5.603 & 2.85 & 0.221 & 0.042 & 3590.89 + @xmath197 & 0.001 & 4.81 & 0.03 & 0.008 & 0.0008 & 0.002 & 0.07 & 0.023 & 0.010 & + & & & & & & & & & & + close3.+ & 4386.787 & 200.05 & 1.58 & 0.251 & 0.0773 & 0.690 & 2.95 & -0.142 & -0.642 & 3593.49 + @xmath197 & 0.001 & 2.29 & 0.02 & 0.008 & 0.0009 & 0.003 & 0.03 & 0.203 & 0.159 & + & & & & & & & & & & + close3.- & 4386.784 & 195.45 & -1.63 & 0.241 & 0.0790 & 5.599 & 2.99 & -0.079 & -0.493 & 3598.00 + @xmath197 & 0.001 & 2.74 & 0.03 & 0.005 & 0.0007 & 0.003 & 0.04 & 0.232 & 0.250 & + & & & & & & & & & & + close4.+ & 4386.782 & 246.01 & 1.31 & 0.219 & 0.0729 & 0.680 & 2.37 & -0.194 & 0.018 & 3583.96 + @xmath197 & 0.001 & 3.58 & 0.02 & 0.009 & 0.0008 & 0.002 & 0.04 & 0.021 & 0.008 & + & & & & & & & & & & + close4.- & 4386.784 & 198.00 & -1.62 & 0.262 & 0.0770 & 5.600 & 2.96 & 0.215 & 0.045 & 3590.66 + @xmath197 & 0.001 & 3.38 & 0.03 & 0.009 & 0.0007 & 0.002 & 0.05 & 0.026 & 0.010 & + & & & & & & & & & & + & & & & & & & & & & + wide1.+ & 4386.782 & 232.96 & 1.36 & 0.421 & 17.3863 & 0.686 & 2.51 & & & 3601.18 + @xmath197 & 0.001 & 3.58 & 0.02 & 0.016 & 0.2971 & 0.002 & 0.04 & & & + & & & & & & & & & & + wide1.- & 4386.782 & 235.81 & -1.35 & 0.450 & 17.8030 & 5.596 & 2.48 & & & 3601.03 + @xmath197 & 0.001 & 4.04 & 0.02 & 0.026 & 0.3277 & 0.002 & 0.04 & & & + & & & & & & & & & & + wide2.+ & 4386.781 & 277.00 & 1.17 & 0.395 & 18.5607 & 0.677 & 2.11 & -0.174 & 0.019 & 3578.93 + @xmath197 & 0.001 & 2.77 & 0.01 & 0.018 & 0.2442 & 0.002 & 0.02 & 0.012 & 0.008 & + & & & & & & & & & & + wide2.- & 4386.781 & 238.43 & -1.35 & 0.479 & 18.1461 & 5.604 & 2.45 & 0.182 & 0.039 & 3590.21 + @xmath197 & 0.001 & 2.84 & 0.01 & 0.018 & 0.1547 & 0.003 & 0.02 & 0.022 & 0.006 & + & & & & & & & & & & + wide3.+ & 4386.786 & 236.20 & 1.33 & 0.460 & 17.9249 & 0.688 & 2.49 & -0.137 & -0.527 & 3592.68 + @xmath197 & 0.001 & 3.16 & 0.02 & 0.014 & 0.1940 & 0.003 & 0.03 & 0.135 & 0.120 & + & & & & & & & & & & + wide3.- & 4386.782 & 237.32 & -1.34 & 0.402 & 17.3515 & 5.602 & 2.47 & -0.079 & -0.400 & 3597.90 + @xmath197 & 0.001 & 4.11 & 0.02 & 0.013 & 0.1737 & 0.003 & 0.05 & 0.136 & 0.196 & + & & & & & & & & & & + wide4.+ & 4386.781 & 264.15 & 1.22 & 0.418 & 18.4004 & 0.680 & 2.21 & -0.170 & 0.020 & 3582.38 + @xmath197 & 0.001 & 3.06 & 0.01 & 0.011 & 0.1652 & 0.002 & 0.02 & 0.017 & 0.006 & + & & & & & & & & & & + wide4.- & 4386.781 & 241.60 & -1.33 & 0.477 & 18.2562 & 5.605 & 2.43 & 0.180 & 0.035 & 3590.36 + @xmath197 & 0.001 & 3.32 & 0.02 & 0.014 & 0.1797 & 0.002 & 0.03 & 0.019 & 0.005 & + & & & & & & & & & & + crrrrrrrrrrrr & & & & & & & & & & & & + + & & & & & & & & & & & & + close.+ & 4386.784 & 170.62 & 1.94 & 0.376 & 0.076 & 0.647 & 3.53 & & & -4.9390 & 1.00 & 3553.26 + @xmath197 & 0.001 & 1.60 & 0.02 & 0.014 & 0.001 & 0.003 & 0.03 & & & 0.1753 & 0.20 & + & & & & & & & & & & & & + close.- & 4386.785 & 178.76 & -1.85 & 0.353 & 0.075 & 5.633 & 3.37 & & & 4.9103 & 0.87 & 3551.59 + @xmath197 & 0.001 & 1.91 & 0.02 & 0.008 & 0.001 & 0.003 & 0.04 & & & 0.1955 & 0.15 & + & & & & & & & & & & & & + wide.+ & 4386.746 & 225.95 & 1.38 & 0.427 & 17.552 & 0.674 & 2.48 & & & -0.0015 & -0.57 & 3601.19 + @xmath197 & 0.009 & 3.72 & 0.03 & 0.024 & 0.246 & 0.010 & 0.03 & & & 0.0006 & 0.16 & + & & & & & & & & & & & & + wide.- & 4386.712 & 215.09 & -1.24 & 0.447 & 17.874 & 5.570 & 2.45 & & & 0.0070 & -0.89 & 3600.81 + @xmath197 & 0.006 & 2.67 & 0.02 & 0.015 & 0.152 & 0.009 & 0.02 & & & 0.0006 & 0.10 & + & & & & & & & & & & & & + & & & & & & & & & & & & + + & & & & & & & & & & & & + close.+ & 4386.781 & 250.29 & 1.32 & 0.311 & 0.066 & 0.651 & 2.38 & -0.141 & -0.010 & -3.5485 & 1.02 & 3549.82 + @xmath197 & 0.001 & 3.42 & 0.03 & 0.012 & 0.001 & 0.002 & 0.06 & 0.012 & 0.003 & 0.0051 & 0.02 & + & & & & & & & & & & & & + close.- & 4386.781 & 202.40 & -1.64 & 0.321 & 0.072 & 5.644 & 2.98 & 0.131 & 0.004 & 4.8568 & 0.94 & 3548.29 + @xmath197 & 0.001 & 2.28 & 0.02 & 0.007 & 0.001 & 0.002 & 0.04 & 0.015 & 0.002 & 0.0035 & 0.01 & + & & & & & & & & & & & & + wide.+ & 4386.872 & 292.38 & 0.96 & 0.399 & 18.358 & 0.759 & 2.17 & -0.164 & 0.032 & -0.0004 & 1.42 & 3583.05 + @xmath197 & 0.007 & 3.14 & 0.02 & 0.019 & 0.207 & 0.006 & 0.03 & 0.015 & 0.007 & 0.0001 & 0.08 & + & & & & & & & & & & & & + wide.- & 4386.943 & 288.83 & -1.26 & 0.447 & 17.936 & 5.582 & 2.42 & 0.180 & 0.040 & -0.0066 & 1.85 & 3590.67 + @xmath197 & 0.004 & 3.64 & 0.02 & 0.019 & 0.185 & 0.007 & 0.03 & 0.016 & 0.005 & 0.0004 & 0.05 & + & & & & & & & & & & & & + & & & & & & & & & & & & + + & & & & & & & & & & & & + close.+ & 4386.789 & 153.18 & 2.12 & 0.283 & 0.0860 & 0.683 & 3.83 & & & -1.27 & 4.32 & 210.6 & 13.95 & 9.992 & 3546.78 + @xmath197 & 0.001 & 1.09 & 0.01 & 0.006 & 0.0005 & 0.004 & 0.02 & & & 0.09 & 0.11 & 10.1 & 11.28 & 0.007 & + & & & & & & & & & & & & + close.- & 4386.784 & 157.59 & -2.11 & 0.291 & 0.0842 & 5.616 & 3.73 & & & 4.10 & 3.17 & 250.4 & -63.43 & 9.992 & 3547.27 + @xmath197 & 0.001 & 1.08 & 0.02 & 0.006 & 0.0005 & 0.003 & 0.02 & & & 0.15 & 0.18 & 13.3 & 9.08 & 0.011 & + & & & & & & & & & & & & + wide.+ & 4386.786 & 187.96 & 1.74 & 0.530 & 16.4576 & 0.675 & 3.12 & & & -3.67 & 2.55 & 203.0 & -6.14 & 9.737 & 3546.42 + @xmath197 & 0.001 & 1.24 & 0.01 & 0.014 & 0.1173 & 0.003 & 0.01 & & & 0.32 & 0.09 & 14.2 & 6.12 & 0.009 & + & & & & & & & & & & & & + wide.- & 4386.786 & 185.51 & -1.75 & 0.493 & 16.0743 & 5.608 & 3.17 & & & 2.91 & 2.74 & 16.4 & 1.14 & 9.366 & 3544.88 + @xmath197 & 0.001 & 1.04 & 0.01 & 0.012 & 0.0994 & 0.003 & 0.02 & & & 0.21 & 0.13 & 10.3 & 7.27 & 0.006 & + & & & & & & & & & & & & + & & & & & & & & & & & & + + & & & & & & & & & & & & + close.+ & 4386.788 & 169.39 & 1.93 & 0.270 & 0.0827 & 0.677 & 3.45 & 0.212 & -0.091 & -2.66 & 2.86 & 247.6 & -7.36 & 9.139 & 3535.83 + @xmath197 & 0.001 & 1.02 & 0.01 & 0.005 & 0.0007 & 0.003 & 0.02 & 0.025 & 0.015 & 0.11 & 0.14 & 14.1 & 7.84 & 0.006 & + & & & & & & & & & & & & + close.- & 4386.788 & 162.58 & -2.04 & 0.264 & 0.0850 & 5.608 & 3.61 & 0.213 & -0.064 & 6.75 & 3.93 & 447.1 & -163.07 & 9.158 & 3536.47 + @xmath197 & 0.001 & 1.36 & 0.01 & 0.006 & 0.0007 & 0.004 & 0.03 & 0.032 & 0.023 & 0.15 & 0.20 & 15.2 & 9.75 & 0.008 & + & & & & & & & & & & & & + wide.+ & 4386.785 & 189.06 & 1.73 & 0.519 & 16.4318 & 0.670 & 3.10 & 0.200 & -0.096 & -1.58 & 4.41 & 344.4 & 5.79 & 9.347 & 3537.02 + @xmath197 & 0.001 & 1.27 & 0.01 & 0.011 & 0.0779 & 0.003 & 0.02 & 0.024 & 0.018 & 0.11 & 0.15 & 3.4 & 5.90 & 0.003 & + & & & & & & & & & & & & + wide.- & 4386.786 & 193.58 & -1.68 & 0.454 & 16.0947 & 5.605 & 3.04 & 0.195 & -0.084 & 1.67 & 4.34 & 173.2 & 15.35 & 9.979 & 3539.78 + @xmath197 & 0.001 & 1.74 & 0.02 & 0.010 & 0.0946 & 0.003 & 0.03 & 0.031 & 0.018 & 0.16 & 0.10 & 5.4 & 9.18 & 0.005 + & & & & & & & & & & & & + lrrrrrrrr ogle @xmath17 & 12.63 & 0.10 & 62.4 & 3.0 & 21.25 & 0.01 & 19.51 & 0.04 + ctio @xmath20 & 0.83 & 0.02 & -16.9 & 8.7 & 24.20 & 0.03 & & + ctio @xmath17 & 10.58 & 0.13 & 50.3 & 27.9 & 21.44 & 0.01 & 19.75 & 0.60 + ctio @xmath21 & 103.16 & 1.95 & 2012 & 402 & 18.97 & 0.02 & 15.74 & 0.22 + irsf @xmath22 & 58.37 & 0.30 & 103 & 128 & 19.59 & 0.01 & 18.97 & 1.35 + irsf @xmath21 & 113.28 & 0.51 & 382 & 211 & 18.87 & 0.01 & 17.54 & 0.60 + irsf @xmath196 & 146.48 & 0.69 & 1301 & 294 & 18.59 & 0.01 & 16.21 & 0.25 + lrrrrrr ogle @xmath17 & & & & & & 19.47 + ctio @xmath20 & 379.2 & 5 & 2.3 & [email protected] & 20.90 & + ctio @xmath17 & 609.4 & 5 & 2.3 & [email protected] & 19.59 & 19.15 + irsf @xmath22 & 1297.7 & 4 & 5.0 & [email protected] & 17.33 & 17.62 + irsf @xmath21 & 5427.7 & 4 & 5.0 & [email protected] & 16.61 & 16.96 + irsf @xmath196 & 5955.6 & 3 & 5.0 & [email protected] & 15.97 & 15.96 +

in this paper , we study central production , in the inclusive case , to look for graviton signals in large extra dimensional model using dilepton and diphoton channels . we carefully analyze signal and possible standard model background processes and study the feasibility of such new physics searches in a relatively clean environment as in central production where the proton fragments are mostly emitted in the forward direction , and there is a clear rapidity gap between them and the centrally produced system . our analysis shows that the lhc with 14 tev center of mass energy and 100 fb ( 300 fb ) of integrated luminosity can probe the effective gravity scale up to 3.6 tev ( 4.4 tev ) in both the dilepton and diphoton channels .

1305.3763

the standard model ( sm ) has been extremely compatible with all the experimental tests so far . however , there are many theoretical attempts to go beyond the sm in order to answer some fundamental issues in the sm namely the hierarchy problems , number of generations etc . theories with extra spatial dimensions have been proposed to answer the hierarchy problems of the sm @xcite . in what follows , in this paper we look for the graviton signals as predicted in the large extra dimensional model ( proposed by arkani - hamed , dimopoulos and dvali @xcite ) in the central diffractive processes at the lhc . forward physics has been used with great success in studying the standard model ( sm ) physics at colliders . in particular , hera ( hadron - electron ring accelerator ) has used forward detectors to greatly enhance our knowledge of perturbative quantum chromodynamics ( qcd ) and its effects in deep inelastic scattering ( dis ) experiments . moreover , the observation of diffractive events , those with large rapidity gaps between the final state particles , have been one of the most interesting aspects of hera physics . in the last decade the importance of forward physics has been recognized for lhc ( large hadron collider ) physics too . this has resulted in a proposal to install forward particle detectors at atlas and cms in order to detect protons emitted in the forward direction , implying the existence of a centrally produced system with a rapidity gap between it and the protons @xcite . more specifically , we are talking of a process of the kind @xmath1 , also known as central exclusive production ( cep ) . in this case , the outgoing protons are emitted with barely any loss of energy ( typically less than 2% of their longitudinal momenta ) along with a centrally produced system @xmath2 with a rapidity gap between it and the outgoing protons on either side . this is a clean production of the central system since it corresponds to just two forward protons in the final state along with the decay products used to identify the central particle @xmath2 and no hadronic activity between them . this has therefore been described as being like a vacuum fluctuation as the two protons pass each other , producing the central particle . some good review on the phenomenology of exclusive processes can be found in @xcite . recently , in @xcite a general framework of exclusive double diffractive processes and its prospects are discussed . one of the main aim of the forward physics at the lhc is to look for the exclusive higgs production events . first in @xcite , it was pointed out that higgs can be produced in diffractive processes with rapidity gaps on either side at hadron - hadron colliders . later this interesting scenario has been discussed in a series of papers @xcite by several authors . recently , to compute the background for higgs searches , in @xcite authors have estimated the exclusive @xmath3 pair production cross section at the lhc . in exclusive processes higgs would be produced almost at rest through gluon - gluon fusion . the final state contains two protons and a higgs and , since the process is exclusive , the invariant mass of the higgs is directly related to the energy loss of the outgoing protons @xcite . in exclusive production , measuring the energy loss of the protons one can determine the mass of the central system without looking at its decay products . it was discussed in @xcite , if the outgoing protons scatter through small angles , the two - gluon system is in a @xmath4 c - even , p - even state , where the @xmath5-axis is the proton beam axis . this means that any new resonance must carry the @xmath6 quantum number in cep . in other words , this enables a clean determination of the quantum numbers of any new resonance @xcite . with the possibility of tagging the forward outgoing protons , the lhc can be turned into an effective gluon - gluon , photon - proton and photon - photon collider , giving rise to a major qcd and electroweak physics program @xcite . in addition , the possibility of producing quarkonium states like @xmath7 , @xmath8 , is something being envisaged at the alice detector . as discussed in @xcite , the measurement of the two - body decay channels of @xmath9 to light mesons in exclusive process can be used to study the dynamics of heavy quarkonia and to test the qcd framework of cep . however all these studies require a qcd based model that allows us to couple the di - gluon system to the protons . the preferred model to describe this is the so- called durham model which we briefly describe in the next section . before we do that however , we should point out that various other processes in the cep scenario have recently been observed in the tevatron at cdf which have given physicists the confidence that the durham model is on the right track . even though the durham model was proposed almost a decade ago , it was not until 2006 that cdf saw diphoton production in the central rapidity region which was consistent with the gluon - gluon fusion durham model @xcite . subsequently cdf also found another exclusive process - @xmath10 and @xmath11 production produced by two photons @xcite . here the tevatron ( and in future the lhc ) acts as a photon photon collider . these processes also allow the calibration of the forward proton detectors proposed by cms and atlas . quarkonium production of @xmath8 and @xmath12(2s ) as well as @xmath13 have also been observed through their decay products @xcite . exclusive dijet production has been observed , at cdf @xcite and d0 @xcite in line with the predictions of the durham model . these processes already seen at the tevatron have given confidence that we can use digluon fusion is accordance with the durham model to observe the signatures of new physics . it was indicated in @xcite that new physics like susy , extra dimensions etc . can be searched for in exclusive processes . in the exclusive configuration various bsm signatures namely mssm higgs @xcite , charged higgs @xcite , triplet higgs @xcite , radions @xcite , massive gravitons @xcite , long lived gluinos @xcite , effects of quantum gravity @xcite etc . have been discussed in the literature . in this work we study inclusive double diffractive process as a probe of large extra dimensions ( add model ) using dilepton and diphoton channels via the exchange of kk graviton . in sec . [ sec : addintro ] we briefly introduce the add model . in sec . [ sec : diffrac_process ] we discuss diffractive processes and summarize the durham model . in sec . [ sec : diffrac_produc ] we consider dilepton and diphoton production via kk graviton . we discuss about important sm backgrounds in sec . [ sec : sm_bkgrnd ] . in sec . [ sec : numerical_result ] we summarize our numerical results and present lhc discovery potential of extra dimensions . finally , in sec . [ sec : conclusion ] we offer our conclusions . in this section we give a very brief introduction of the add model @xcite . this model assumes that space is @xmath14 dimensional where @xmath15 is the number of extra spatial dimensions . all the sm particles are confined to the usual @xmath16d spacetime which is called `` brane '' . whereas , gravity can propagate in the extra dimensions . in this model , the @xmath17d planck scale ( @xmath18 gev ) is a derived scale which is related to the fundamental planck scale ( @xmath19 tev ) by @xmath20 where @xmath21 is the extra dimensional volume and @xmath22 is the compactification radius . an important consequence of the add model is the appearance of a tower of kaluza - klein ( kk ) modes as a solution to the linearized einstein equation in @xmath14 dimensions . these kk modes are almost degenerate and separated in mass by @xmath23 ( @xmath24 @xmath25 ev to @xmath26 mev for @xmath27 ) terms . after kk decomposition , we have massive spin-2 kk gravitons ( @xmath28 ) which have interactions with the brane localized sm states via the energy momentum tensor @xmath29 of the sm @xmath30 where @xmath31 and the summation runs over all kk modes . the add model has been studied extensively both theoretically as well as experimentally . recently , the extra dimensions are being searched for using dilepton and diphoton channels at the lhc , both by atlas @xcite and cms @xcite . these searches has already put the lower bound on @xmath32 quite high . we consider two types of double diffractive ( dd ) mechanisms at the lhc as follows : 1 . central exclusive process ( cep ) : @xmath33 as shown in fig . [ fig : dd_process](a ) 2 . central inclusive process ( cip ) : @xmath34 as shown in fig . [ fig : dd_process](b ) here @xmath2 denotes the centrally produced system ( we also denote the invariant mass of the central system with the same notation `` @xmath2 '' ) and the ` @xmath35 ' signs denote the presence of rapidity gaps between the system @xmath2 and the outgoing protons ( in exclusive process ) or proton dissociated products @xmath36 ( in inclusive process ) . the differential cross sections for exclusive and inclusive dd processes shown in fig . [ fig : dd_process ] can be expressed in the factorized form @xcite , @xmath37 where @xmath38 is the cross section of the hard subprocess which produces the color singlet system of invariant mass @xmath2 ( @xmath39 ) and @xmath40 is the effective luminosity for the production of a system of invariant mass @xmath2 at rapidity @xmath41 . the total cross section @xmath42 can be calculated using @xmath43 in our calculation , we keep @xmath40 and @xmath38 in a factorized form as the effective luminosity is independent of the subprocess and for different processes we need to compute just the subprocess cross sections . although we keep @xmath40 and @xmath38 in a factorized form , we must remember that there might exist specific selection rules for different configurations . in the exclusive configuration , the fusing @xmath44 state obeys a special selection rule in the limit that the protons scatter through zero angle . the @xmath5-component of the angular momentum of a centrally produced system vanishes as the incoming state which consists of the fusing hard gluons has @xmath45 with positive @xmath46 and @xmath47 parity @xcite . for inclusive production on the other hand , there is no such selection rule and therefore , the production rate is larger than that for the exclusive production . although the signals are very clear for cep , the luminosity and the event rates are very small due to the restricted kinematics . a much larger phase space is available due to the dissociation of the incoming protons in the inclusive dd case . the luminosities for exclusive and inclusive processes can be computed using the known parton distribution functions ( pdf ) of the protons . to estimate @xmath40 we shall use what is known as the `` durham model '' , described in refs . the @xmath40 depends on off - diagonal or skewed gluon distribution functions ( off - diagonal , since we need to consider the coupling of two gluons with different momentum fractions , @xmath48 and @xmath49 , with a proton ) . however , as we see , for @xmath50 and @xmath51 , the relevant kinematic region for cep , it is possible to approximate the off - diagonal distributions using the integrated gluon distribution function @xcite . for cip the dominant ( leading logarithm ) contribution comes from the asymmetric configuration , @xmath52 . at the partonic level , the inclusive production is equivalent to the exclusive production and the unintegrated gluon distributions in partons may be calculated perturbatively in terms of non - forward bfkl amplitudes . the effective luminosity depends also on two survival probabilities of the rapidity gaps - one to the qcd radiation and another to the non - perturbative soft rescattering of the protons . the first is expressed as a sudakov factor which ensures the fusing gluons remain intact up to the hard scale @xmath53 . the second factor is included as an explicit multiplicative factor @xmath54 , known as the `` gap survival factor '' . it accounts for the probability that apart from the perturbative processes , no other particle is produced via soft interactions between the protons @xcite . in general @xmath54 is expected to depend on the kinematics of the process and can be calculated using various qcd based models @xcite . typically , for lhc energies , @xmath54 for cep is estimated to be a few percent @xcite ( e.g. , in @xcite authors found @xmath55 for 14 tev ( 8 tev ) lhc ) . for cip , relatively smaller absorption cross section leads to a larger @xmath54 . for simplicity , in our analysis we keep @xmath56 for cip , as a constant multiplicative factor . we use the mstw2008 lo parton density @xcite in our computation . since mstw parametrization for the gluon distribution function @xmath57 does not go below @xmath58 gev , for @xmath59 gev , we take the following extrapolation function , @xmath60}\end{aligned}\ ] ] so that for @xmath61 , @xmath62 @xcite . here @xmath63 and @xmath64 are the two functions of @xmath65 and can be computed using the following relations @xmath66}}\end{aligned}\ ] ] where @xmath67 and @xmath68 are evaluated at @xmath69 gev . to compute the sudakov factor , we use the lo @xmath70 ( strong coupling ) with @xmath71 mev and @xmath72 . the full formalism to calculate effective luminosity for exclusive and inclusive configuration is given in ref . @xcite . in appendix [ app : inclu_lum ] we briefly present the computation of @xmath40 for cip . with diffractive processes being dominated by two - gluon exchange as shown in fig . [ fig : dd_process ] , it is obvious that these processes would be observed only for the final states with a substantial coupling to a gluon pair . in this work , we consider the central production of kk gravitons ( @xmath73 ) via dd processes . the spin-2 gravitons can have three possible spin projections @xmath74 0 , 1 and 2 . due to the presence of special @xmath75 selection rule in the exclusive configuration , graviton can not be produced via leading order ( lo ) exclusive processes point - like @xmath44 coupling to a @xmath76 graviton , it can still be produced at next to lo ( nlo ) in the exclusive configuration . ] @xcite . in the unitary gauge , the coupling of the spin-0 projection of @xmath73 with two gluons is zero as the lo coupling of spin-0 component with any two vector bosons is proportional to the square of the mass of the vector boson and gluons are massless in this case @xcite . by the landau - yang theorem @xcite , it is impossible to produce a spin-1 particle via the fusion of two on - shell vector bosons but the production is not forbidden if those vector bosons are off - shell . though the gluons coming from protons are not purely on - shell , the degree of virtuality of the gluons is very small . thus , the contribution in any process from spin-1 projection of graviton is subdominant @xcite . thus , the @xmath77 is the only possible spin projection that can couple to a gluon pair at lo . therefore , in this paper we discuss the lhc signatures of @xmath73 in inclusive configuration . in what follows we analyze the diphoton and the dilepton channels via the exchange of @xmath73 arising in add type models @xcite as a signal of extra dimension . we carefully consider these signal processes and the relevant background processes for them and look for regions of parameter space where the signal would dominate . since these processes can only occur in inclusive configuration the event rates could be substantial to search for at the lhc . moreover , the nature of the intermediate particle and the coupling with sm particles , reflects in the angular distribution of the final state particles . subprocess feynman diagram for dilepton production . ] we consider the process where dilepton pair is produced from gluon fusion via the exchange of @xmath73 as shown in fig . [ fig : dilepton ] . feynman rules for this process are taken from ref . @xcite where one can find definitions of all the terms which appear here . we also show these feynman rules in appendix [ app : feynrules ] . the matrix element for this process is given by @xmath78 where the coupling @xmath79 is related to the newton constant @xmath80 as @xmath81 . since the reduced planck mass @xmath82 ( @xmath83 gev ) , each kk mode coupling to the sm particles is planck mass suppressed . a summation over high multiplicity of kk modes lying below the uv cutoff scale @xmath32 compensates the suppression and give rise to a substantial effective coupling strength . therefore , we replace the graviton propagator in eq . [ eq : m_dilep ] by the effective propagator @xmath84}\ ] ] where @xmath85 sums over all kk towers below @xmath32 and @xmath86 and `` hat '' notation is used for subprocess quantities . here @xmath87 is the invariant mass of the lepton pair . since , kk modes are quasi - continuous , this summation can be done by defining kk state density as shown in appendix [ app : feynrules ] @xcite . for @xmath15 number of extra dimensions , considering the contributions from resonant and nonresonant kk states , the @xmath88 is given by @xmath89\ ] ] where the real part comes from the summation over all resonant contributions below @xmath32 and the imaginary part is the summed contribution coming from all the nonresonant states . the definition of the @xmath90 function can be found in appendix [ app : feynrules ] . after summing over the spins of the final state leptons the effective squared matrix element takes the form @xmath91\delta_{ab}\delta^{ab}\end{aligned}\ ] ] in the zero mass limit of the produced lepton , the angular dependence of @xmath92 takes a very simple form as follows : @xmath93 is the scattering angle of the leptons in the center of mass ( cm ) frame of the @xmath73 . averaging @xmath92 over eight colors and two polarizations of the initial gluons , one can evaluate the subprocess differential cross section after including the phase space factor as @xmath94 if we consider two types of leptons ( i.e. electron and muon ) are contributing in the dilepton final state , we should multiply the above differential cross section by an extra factor of two . subprocess feynman diagram for diphoton production . ] we consider the process where a photon pair is produced from gluon fusion via the exchange of @xmath73 as shown in fig . [ fig : diphoton ] . as mentioned earlier , feynman rules for this process are taken from ref . @xcite where one can find definitions of all the terms which appear here and also can be found in appendix [ app : feynrules ] . the matrix element for this process is given by @xmath95 after summing over all kk states up to @xmath32 contributing to a physical process and summing over two polarizations of the final state photons , the effective squared matrix element takes the form @xmath96 after doing some tedious algebra , the angular dependence of @xmath92 takes the form @xmath93 is the scattering angle of the photons in the cm frame of the @xmath73 . averaging @xmath92 over eight colors and two polarizations of the initial gluons , one can evaluate the subprocess differential cross section after including the phase space factor as @xmath97 here we have included an extra 1/2 factor because of the presence of two identical particles in the final state . in the diffractive configuration , there are possibilities to produce a central system from @xmath44 , @xmath98 , @xmath99 or @xmath100 fusion processes . in a same kinematic region , the effective luminosity for the @xmath44 fusion ( @xmath40 ) is the largest at the lhc among all the alternatives . whereas , the effective luminosities for @xmath98 , @xmath99 or @xmath100 fusion processes are much smaller compared to @xmath40 . estimations of effective luminosities for the @xmath44 , @xmath98 and @xmath100 can be found in @xcite and for the @xmath99 in @xcite . while computing the sm backgrounds for the dilepton and the diphoton signal channels , we can sometime neglect @xmath98 , @xmath99 and @xmath100 initiated background processes due to the small luminosities of the fusing particles . + in fig . [ fig : dilep_bkgrnd ] , we show feynman diagrams of some important background subprocesses for the dilepton channel proceed through @xmath44 , @xmath98 , @xmath99 and @xmath100 initial states . when one considers lepton pair production within the sm in the diffractive configuration , one would not expect a significant rate for a lepton pair unaccompanied by other particles ( like radiation jets / photons or other sm particles ) . although there is always a chance to miss those extra particles other than the lepton pair or to misidentify other particles as leptons . in our background computation we do not include those reducible backgrounds as these are expected to be very small . using landau - yang theorem , dual considerations of angular momentum conservation and bose symmetry prevents a pair of gluons couple to an on - shell @xmath101 ( via a quark loop , see figs . [ fig : dilep_bkgrnd](a ) and [ fig : dilep_bkgrnd](b ) ) or a @xmath100 pair couple to an on - shell @xmath101 ( see figs . [ fig : dilep_bkgrnd](g ) and [ fig : dilep_bkgrnd](h ) ) . while such a coupling , although very small , is indeed permitted for an off - shell @xmath101 . the pieces in the corresponding vertex function are proportional to the degree of virtuality of the initial particles or to the momentum of the @xmath101 itself . moreover , @xmath102 processes involves a quark loop and @xmath103 processes are suppressed due to very low @xmath100 luminosity . because of these reasons we are not including @xmath44 and @xmath100 initiated processes in our computation . we also neglect @xmath99 initiated processes in figs . [ fig : dilep_bkgrnd](d ) and [ fig : dilep_bkgrnd](e ) due to very low luminosity for @xmath99 . consequently , these processes lead only to subdominant contributions . the dominant sm background for the dilepton channel comes from the inelastic photo - production of the lepton pair as shown in fig . [ fig : dilep_bkgrnd](c ) where photons are emitted from the quarks inside protons . this contributions can be very large in the kinematic region where the invariant mass of the lepton pair is small . this is because , the collinear emission of equivalent photons from quark lines can give large contribution . the inelastic @xmath104 cross section has been computed in ref . @xcite for weizscker - willams photon pairs and an outline of the computation is presented in appendix [ app : dilep_bkgrnd ] . the contributions from the interference terms among various background processes might be important at low energies . since we are only interested in the high energy regions , we do not include them for simplicity . + in fig . [ fig : dipho_bkgrnd ] , we show possible background subprocesses for the diphoton channel initiated from @xmath44 , @xmath98 , @xmath99 and @xmath100 initial states . the dominant background for the diphoton channel comes from the @xmath105 box diagram ( via quark loop ) shown in fig . [ fig : dipho_bkgrnd](a ) and cross section is proportional to @xmath106 ( @xmath107 is the qed coupling ) . there are other higher order diagrams that contribute to @xmath105 process as shown in figs . [ fig : dipho_bkgrnd](b ) and [ fig : dipho_bkgrnd](c ) where higgs is involved . the higgs is produced from @xmath44 fusion via a top quark loop and then decayed to @xmath98 via another top quark or a @xmath108 loop . there are also delbruck type ( light - by - light scattering ) backgrounds possible shown in figs . [ fig : dipho_bkgrnd](d ) and [ fig : dipho_bkgrnd](e ) . in this case quarks from the dissociated protons , radiate two photons which then give two final state photons via a fermion or @xmath108 in the loop . this contribution is much suppressed as cross section is proportional to @xmath109 . the other backgrounds which are initiated from @xmath99 ( see fig . [ fig : dipho_bkgrnd](f ) ) and @xmath100 ( see figs . [ fig : dipho_bkgrnd](g ) and [ fig : dipho_bkgrnd](h ) ) initial states are negligible as the corresponding effective luminosities are very low . we note that all the @xmath44 and @xmath98 initiated background processes occur only via loop diagrams and they are usually very small . if the signal rate is substantial in the diphoton channel , searching the signatures of extra dimensions in the diphoton channel might be very promising . we estimate the dominant @xmath105 box diagram background cross section using @xmath110 helicity amplitudes taken from @xcite . relevant formulae are given in appendix [ app : bkgrnd_yy ] . at low energy , the fermion loop contribution in the box diagram dominates , but above a few hundred gev , this contribution starts falling rapidly . whereas , the @xmath108 loop contribution dominates in the total cross section of the @xmath110 process at high energies . other higher order @xmath111 processes similar to figs . [ fig : dipho_bkgrnd](b ) and [ fig : dipho_bkgrnd](c ) are not considered here . we do not include @xmath99 initiated process in our background computation as the contribution originates from @xmath112 @xmath113-channel exchange in fig . [ fig : dipho_bkgrnd](f ) is roughly two order of magnitude lower than the @xmath44 contributions @xcite . as mentioned earlier , we neglect @xmath100 initiated processes for low @xmath100 luminosity and all background interference terms for simplicity . in all our numerical computations we use the mstw2008 lo parton densities to obtain the unintegrated parton distributions . here we once again give the numerical values of some quantities which are used . to compute the sudakov form factors , we use nlo @xmath70 with @xmath114 mev and number of `` active '' flavor @xmath115 . we take a fixed value of the `` gap survival factor '' , @xmath56 in our analysis . our results depend on two model parameters , the number of extra dimensions @xmath15 and the ultra violet cutoff @xmath32 of the theory and four kinematical variables , invariant mass @xmath2 and rapidity @xmath41 of the centrally produced system , rapidity gap @xmath116 between the central system and the outgoing protons or proton dissociated products , and the cm energy @xmath117 of two colliding protons at the lhc . the cutoff scale @xmath32 is of the order of few tev for low scale gravity theories . the lower limit of @xmath32 is bounded from experiments and the bounds are already quite high as stated in sec . [ sec : addintro ] . if we increase @xmath32 , the signal cross sections decrease as usual . in our case the signal falls off very rapidly with increasing @xmath32 because of the presence of @xmath118 in the denominator of the signal cross sections ( see eqs . [ eq : dcsdcos_ll ] and [ eq : dcsdcos_yy ] ) . this makes the add model very difficult to search for experimentally . in fig . [ fig : luminosity](a ) we show the dependence of the effective luminosity @xmath40 on the invariant mass @xmath2 of the central system produced at rapidity region @xmath119 at the 14 tev lhc . the @xmath40 shows an decreasing nature with increasing @xmath2 since it is harder to produce more massive objects while keeping other parameters fixed . in fig . [ fig : luminosity](b ) we plot the dependence of @xmath40 on @xmath41 with @xmath120 gev at the 14 tev lhc . in the diffractive processes it is more likely to produce an object centrally i.e. near @xmath119 region . thus , a system which is produced away from the central region is less probable and we see a decreasing behavior of @xmath40 with increasing @xmath41 . in fig . [ fig : tcs_sqrts_ll ] we show the total signal cross section in the dilepton ( @xmath121 where @xmath122 ) channel as a function of @xmath117 . in fig . [ fig : tcs_sqrts_ll](a ) we show @xmath121 for different @xmath116 taking @xmath123 and @xmath124 tev . the cross section increases as we increase @xmath117 as expected . if we increase @xmath116 , the @xmath121 decreases as the effective luminosity decreases with increasing @xmath116 ( see figs . [ fig : luminosity](a ) and [ fig : luminosity](b ) ) . in fig . [ fig : tcs_sqrts_ll](b ) we show the @xmath121 for different @xmath15 varied from two to seven . we see that as we increase @xmath15 , the @xmath121 decreases . this is because @xmath125 for @xmath126 , in the limit @xmath127 ( see eq . [ eq : apprx_ds ] ) . here we compute the total cross section after applying the following selection cuts : @xmath128 is the scattering angle of @xmath12 ( where @xmath129 ) in the rest frame of the centrally produced system and @xmath130 is the invariant mass of the @xmath131 pair . we have collected events with the rapidity of the central system @xmath132 . the upper limit of @xmath130 can go to @xmath32 up to which the theory is valid . here we have taken the upper limit on @xmath130 slightly smaller than @xmath32 as the @xmath90 function in eq . [ eq : d_eff ] diverges at @xmath32 . the cross sections are only few @xmath133 and we need higher cm energy at the lhc to observe these events . in fig . [ fig : tcs_ms_ll ] we show the total signal cross section in the dilepton ( @xmath121 ) channel as a function of @xmath32 at the 14 tev lhc . in fig . [ fig : tcs_ms_ll](a ) we show @xmath121 for different @xmath116 taking @xmath123 and in fig . [ fig : tcs_ms_ll](b ) we show @xmath121 for different @xmath15 taking @xmath134 . here we compute the total cross section after applying the kinematical cuts defined in eq . [ eq : tcs_cuts1 ] . the cross section decreases rapidly as we increase @xmath32 since @xmath118 is present in the denominator of the total cross section . in fig . [ fig : dcs_ll](a ) we show signal and background invariant mass distributions ( @xmath135 ) of the lepton pair after applying the selection cuts defined in eq . [ eq : dcs_cuts1 ] at the 14 tev lhc . to compute @xmath135 we have not integrated @xmath136 in the full range from @xmath137 to @xmath138 because the background cross section diverges as @xmath139 ( see eq . [ eq : dcsdcosl_bkgrndll ] ) . we plot signal @xmath135 for @xmath140 1.5 , 2.0 and 2.5 tev taking @xmath123 and , @xmath134 for both signal and background . the signal @xmath135 is a monotonically increasing function of @xmath141 ( @xmath142 ) because the subprocess cross section increases roughly as @xmath143 ( see eq . [ eq : dcsdcos_ll ] ) . the background cross section is computed only considering the dominant @xmath144 channel . for small values of @xmath141 the background is quite large but falls off rapidly as we increase @xmath141 . thus , to observe an excess over the background one should collect lepton pairs with sufficiently high @xmath141 . @xmath145(b ) we show signal and background angular distributions ( @xmath146 ) of the lepton pair after applying the selection cuts defined in eq . [ eq : dcs_cuts2 ] at the 14 tev lhc . to compute @xmath146 we apply a very high invariant mass cut on the lepton pair to reduce large background in the small @xmath141 region . we show signal @xmath146 for @xmath140 1.5 , 2.0 and 2.5 tev taking @xmath123 and , @xmath134 for both signal and background . the angular distribution carry spin information of the intermediate particle . for dilepton production , the angular distributions for signal and background show a very contrasting behavior . for signal @xmath147 ( see eq . [ eq : dcsdcos_ll ] ) which attains its maximum at @xmath148 . on the other hand , background attains its minimum at @xmath148 since it goes as @xmath149 and diverges when @xmath150 ( see eq . [ eq : dcsdcosl_bkgrndll ] ) . this contrasting feature can be used as a unique signature of the extra dimensions . @xmath151 we show the total signal cross section in the diphoton channel ( @xmath152 ) as a function of @xmath117 after applying the cut defined in eq . [ eq : tcs_cuts1 ] . in fig . [ fig : tcs_sqrts_yy](a ) we show @xmath152 for different @xmath116 taking @xmath123 and @xmath124 tev . in fig . [ fig : tcs_sqrts_yy](b ) we show @xmath152 for different @xmath15 varied from two to seven . in fig . [ fig : tcs_ms_yy ] we show the @xmath152 as a function of @xmath32 at the 14 tev lhc . in fig . [ fig : tcs_ms_yy](a ) we show @xmath152 for different @xmath116 taking @xmath123 and in fig . [ fig : tcs_ms_yy](b ) we show @xmath152 for different @xmath15 taking @xmath134 . here we compute the @xmath152 after applying the kinematical cuts defined in eq . [ eq : tcs_cuts1 ] . the behavior of these plots are very similar to the corresponding plots for dilepton channel and we are not repeating those features here . an interesting point to notice that @xmath121 and @xmath152 are similar even quantitatively . this is because surprisingly @xmath121 and @xmath152 are equal after @xmath153 integration from @xmath137 to @xmath138 as follows : @xmath154 in fig . [ fig : dcs_yy](a ) we show signal and background invariant mass distributions ( @xmath155 ) of the diphoton pair after applying the selection cuts defined in eq . [ eq : dcs_cutsyy1 ] at the 14 tev lhc . latexmath:[\[\label{eq : dcs_cutsyy1 } signal @xmath155 for @xmath140 1.5 , 2.0 and 2.5 tev taking @xmath123 and , @xmath134 both for signal and background . the signal @xmath155 is a monotonically increasing function of @xmath157 ( @xmath158 ) because the subprocess cross section increases roughly as @xmath143 ( see eq . [ eq : dcsdcos_yy ] ) . the background cross section is computed including @xmath44 and @xmath98 initiated processes as shown in fig . [ fig : dipho_bkgrnd ] . although the background in fig . [ fig : dcs_yy](a ) can beat signal in the small @xmath157 region , it dies out very rapidly in the high @xmath157 region . this is because the diphoton background goes roughly as @xmath159 and becomes almost negligible at large @xmath157 . in fig . [ fig : dcs_yy](b ) we show signal and background angular distributions ( @xmath160 ) of the diphoton pair after applying the selection cuts defined in eq . [ eq : dcs_cutsyy2 ] at the 14 tev lhc . @xmath161 , has a minimum at @xmath162 and attains its maximum value at @xmath163 . whereas , the @xmath160 for the dominant @xmath105 background attains its maximum value at @xmath162 as shown in fig . [ fig : dcs_yy](b ) . this is because of the nature of angular dependent part in eq . [ eq : dcsdcos_bkgrnd ] . the similar angular behavior is true for subdominant @xmath164 background also . whereas , there is no angular dependency of other subdominant @xmath165 backgrounds because of the spin-0 nature of the intermediate higgs . we define the luminosity requirement for the discovery of kk gravitons at the lhc as following : @xmath166 where @xmath167 denotes the luminosity required to attain @xmath168 statistical significance and @xmath169 is the luminosity required to observe 10 signal events . we compute @xmath170 after applying some kinematical cuts which we call the `` discovery cuts '' as defined below . motivated by the invariant mass and angular distributions in figs . [ fig : dcs_ll ] and [ fig : dcs_yy ] we construct some kinematical cuts to separate the signal from the background in the dilepton and diphoton channels as follows : 1 . dilepton channel : @xmath171 2 . diphoton channel : @xmath172 we call these cuts as the `` discovery cuts '' for the discovery of kk graviton at the lhc . in fig . [ fig : ld_ll_yy ] , @xmath170 goes as @xmath169 for both dilepton and diphoton channels in the whole parameter space we have displayed . in case of the diphoton channel the background is already quite small compared to the signal . whereas , for the dilepton channel , background becomes under control after the `` discovery cuts '' . with the `` discovery cuts '' one can probe @xmath32 roughly up to 3.6 tev ( 4.4 tev ) with @xmath173 fb@xmath0 ( @xmath174 fb@xmath0 ) integrated luminosity at the 14 tev lhc . in this paper we have discussed the possibility of looking for new physics , in particular the signatures of large extra dimensions , in diffractive processes at the lhc . we have concentrated on the central inclusive production where the outgoing protons breakup , but the fragments are mostly emitted in the forward direction . we have studied in detail the signal of kk gravitons via the dilepton and the diphoton channels . from angular momentum conservation , the spin-2 graviton can not be produced at lo in exclusive processes as discussed in the text , but it can be produced in an inclusive configuration . although the backgrounds are larger for inclusive production , the larger cross section and event rate for signals make the collider search for extra dimensional theories very promising . since the intermediate particle is spin-2 in nature , we also get a unique angular distribution which can be used to distinguish signals from backgrounds . we have shown that it is very unlikely to produce a lepton or a photon pair in the sm with very high invariant mass in the inclusive configuration . thus , experimentally if we find a reasonable number of lepton or photon pairs with high invariant mass , it would clearly indicate the signature of new physics possibly the signature of large extra dimensions . in this work , we have indicated the possibility of using dilepton and diphoton in the final state to signal the presence of extra dimensions . we have carefully estimated the sizes of the signal and the background and , found that the signal rates are not too big . nevertheless , it is possible within reasonable parameter ranges as suggested by present day studies , to use central production , with forward detectors for the proton fragments , to look for kk gravitons as a signal for add like theories . the great advantage of using such channels for the study of new physics is that they are clean and the signals are easily distinguished from the backgrounds . on the flip side , the event rates are considerably smaller than other standard new physics searches . this can be compensated by runs at higher energies and greater luminosities at some future date . there is another general issue with high luminosity run at the lhc - the presence of huge `` pile - up '' background . this background can overwhelm the rapidity gap signatures . if we consider inclusive processes and sacrifice the forward proton tagging ( applicable only for exclusive processes ) , we can detect rapidity gap events at the lhc @xcite . all higher dimensional models are valid up to a cutoff scale . in our case , the signal cross sections depend on the eighth power of @xmath175 ( @xmath32 is the cutoff scale of the theory ) . thus , numerical results are very sensitive to the uv cutoff . the experimental lower bound of @xmath176 tev for @xmath123 . future experiments can push up the lower bound and make the signatures less visible . we show that lhc with @xmath177 tev and 100 fb@xmath0 ( 300 fb@xmath0 ) of integrated luminosity can probe @xmath32 up to 3.6 tev ( 4.4 tev ) in both the dilepton and the diphoton channels . we are grateful to subhadip mitra for many fruitful discussions and assistance in certain numerical aspects of this work . the inclusive double diffractive amplitude multiplied by its complex conjugate . a central system @xmath2 is produced with rapidity gaps @xmath178 and @xmath179 between proton dissociated products @xmath180 and @xmath181 respectively . ] an outline of the derivation of effective luminosity for the inclusive configuration is presented in @xcite . here , we briefly give some useful formulae for computing inclusive luminosity . at the partonic level the inclusive process in fig . [ fig : inclu_mm ] can be expressed as the exclusive production of the system @xmath2 with rapidity gaps on either side i.e. , @xmath182 the probabilities to find partons @xmath183 ( @xmath184 ) in the protons is given by the effective parton densities @xmath185 evaluated at scale @xmath186 with momentum fraction @xmath48 as given by @xmath187.\ ] ] where @xmath188 , @xmath189 and @xmath190 in the above equation are the usual pdf s . the rapidity gaps can be filled up due to gluon bremsstrahlung effect . the mean number of gluons emitted in transverse momentum interval , @xmath191 ( also @xmath192 , asymmetric @xmath113-channel gluon exchange ) with rapidity gap @xmath193 is given by @xmath194 therefore , the amplitude for no emission in the rapidity interval @xmath193 can be expressed as @xmath195 . the amplitude @xmath196 is called the non - forward bfkl amplitude which is computed resumming the double logarithms in eq . [ eq : mean_glu ] . the factor @xmath197 accounts for the longitudinal bfkl logarithm where @xmath198 . for rapidity gap @xmath199 we have @xmath200 . in the asymmetric region , @xmath192 , it is sufficient to keep only the @xmath201 term , i.e. @xmath202 @xcite . there is another suppression comes in the form of sudakov factor @xmath203 which accounts for the survival probability of a gluon in the interval @xmath204 where @xmath205 is related to the hard scale as @xmath206 . the sudakov factor reads as @xmath207dz\right)\ ] ] where @xmath208 and @xmath209 are the lo dglap splitting functions and @xmath210 . we obtain the effective luminosity ( @xmath40 ) by multiplying the inclusive amplitude by its complex conjugate as shown in fig . [ fig : inclu_mm ] . to compute @xmath40 we carefully account for all the factors i.e. non - forward bfkl amplitudes , sudakov factors , gap survival factor and appropriate color factors , and finally arrive at the expression @xmath211 @xmath212 here all the primed quantities are arising from the complex conjugate of the amplitude . the minimum of the momentum fraction @xmath213 maintains the rapidity gap @xmath193 and given by @xmath214 the momentum of the screening gluon @xmath215 is very small . therefore , in the limit @xmath216 , we have @xmath217 . after performing @xmath218 and @xmath219 integrations @xmath220 takes the simplified form as follows @xmath221 where @xmath222 if @xmath223 , but @xmath224 if @xmath225 . in our computation of @xmath40 we have used this simplified form of @xmath220 taking @xmath226 . in this appendix we give the feynman rules for @xmath227 ( where @xmath228 is a vector boson with mass @xmath229 ) and @xmath230 ( where @xmath231 is a fermion with mass @xmath232 ) vertices @xcite @xmath233 at ( 0,0 ) { \includegraphics[width=3.5cm]{figs / vvg_fr.eps } } ; \node [ ] at ( 0,0 ) { $ \quad\equiv$ } ; \node[right ] at ( 0,0 ) { $ \quad \displaystyle -i\frac{\kappa}{2}\delta^{ab}\left[(m_v^2+k_1\cdot k_2)c_{\mu\nu,\rho\sigma } + d_{\mu\nu,\rho\sigma}(k_1\cdot k_2 ) + \frac{1}{\xi}e_{\mu\nu,\rho\sigma}(k_1\cdot k_2)\right]$ } ; \end{tikzpicture}\end{aligned}\ ] ] @xmath233 at ( 0,0 ) { \includegraphics[width=3.5cm]{figs / ffg_fr.eps } } ; \node [ ] at ( 0,0 ) { $ \quad\equiv$ } ; \node[right ] at ( 0,0 ) { $ \quad \displaystyle -i\frac{\kappa}{8}\left[\gamma_{\mu}(k_1 - k_2)_{\nu}+\gamma_{\nu}(k_1 - k_2)_{\mu } - 2\eta_{\mu\nu } ( \kern+0.30em /\kern-0.70em{k_1}-\kern+0.30em /\kern-0.70em{k_2}-2m_f)\right]$ } ; \end{tikzpicture}\end{aligned}\ ] ] where @xmath81 . the @xmath234 is the @xmath235 dimensional newton constant . all the tensors appear in the feynman rules are defined as @xmath236 @xmath237\end{aligned}\ ] ] @xmath238\end{aligned}\ ] ] the propagator for spin-2 kk graviton with mass @xmath239 and decay width @xmath240 is given by @xmath241 @xmath242 mass of the @xmath85-th kk mode is @xmath243 and mass separation between two adjacent kk modes is a @xmath23 term . therefore , kk modes become quasi - continuous and it is convenient to define kk state density . the number of kk states in the mass interval @xmath244 and @xmath245 for @xmath14 dimensional space can be expressed as @xmath246 where @xmath247 the effective interaction due to all kk states contributing to a physical process can be obtained after summing over all the propagators as follows @xmath248 where @xmath249 and @xmath250 . the effective propagator @xmath251 valid up to @xmath32 looks @xmath252~~\textrm{where}~~\ ] ] @xmath253 in the limit @xmath254 , the effective propagator @xmath251 becomes @xmath255 in our analysis we have used the exact form of @xmath251 given in eq . [ eq : exact_propagator ] . inelastic photo - production of a lepton pair at the lhc . the photons are coming from the quarks inside protons . ] we consider the inelastic photo - production of a lepton pair at the lhc shown in fig . [ fig : inelastic_yy ] as follows @xmath256 a photon with momentum fraction @xmath257 ( @xmath258 ) w.r.t . the quark momentum is emitted from the quark @xmath259 ( @xmath260 ) with momentum fraction @xmath261 ( @xmath262 ) of a proton . the total inelastic @xmath263 cross section can be computed ( this has been thoroughly computed in @xcite ) using weizscker - williams equivalent photon approximation @xcite as given by @xmath264 where @xmath265 are the quarks ( we denote @xmath266 ) , @xmath267 is the mass of the lepton , @xmath268 is the em charge of @xmath189 in the unit of an electron s charge and @xmath38 is the subprocess cross section with cm energy @xmath269 ( @xmath270 , @xmath271 is the lhc cm energy ) . the quark density inside the proton is denoted as @xmath272 . the @xmath273 is the photon spectrum inside a quark and is given by @xmath274}{z}\ln\left(\frac{q_1 ^ 2}{q_2 ^ 2}\right)\ ] ] we choose the scale @xmath275 and @xmath276 gev . the subprocess cross section for @xmath144 is given by @xmath277\ ] ] where @xmath278 . we have used @xmath279 and @xmath280 in our analysis . the differential cross section of the subprocess in the high energy limit is @xmath281 where @xmath282 is scattering angle of the lepton in the cm frame . we see @xmath283 if @xmath284 . we compute the dominant background cross section for the diphoton channel @xmath105 via a quark loop using the @xmath285 helicity amplitudes given in @xcite . the differential cross section of @xmath105 background in the high energy limit and for small @xmath282 ( scattering angle of @xmath286 in the cm frame ) is given by @xmath287 origin of some factors in the above equation are once explained in sec . [ sebsec : diphoton ] and we are not explaining them again . the factor @xmath288 is the number of possible quarks that can contribute in the loop . the @xmath92 is the sum of two helicity amplitudes , which are the only two surviving amplitudes in the high energy limit , @xmath289 n. arkani - hamed , s. dimopoulos and g. r. dvali , phys . b * 429 * , 263 ( 1998 ) [ hep - ph/9803315 ] . l. randall and r. sundrum , phys . lett . * 83 * , 3370 ( 1999 ) [ hep - ph/9905221 ] . c. royon [ rp220 collaboration ] , arxiv:0706.1796 [ physics.ins-det ] . m. g. albrow _ et al . _ [ fp420 r and d collaboration ] , jinst * 4 * , t10001 ( 2009 ) [ arxiv:0806.0302 [ hep - ex ] ] . l. a. harland - lang , v. a. khoze , m. g. ryskin and w. j. stirling , eur . phys . j. c * 72 * , 2110 ( 2012 ) [ arxiv:1204.4803 [ hep - ph ] ] . m. g. albrow , t. d. coughlin and j. r. forshaw , prog . phys . * 65 * , 149 ( 2010 ) [ arxiv:1006.1289 [ hep - ph ] ] . a. d. martin , m. g. ryskin and v. a. khoze , acta phys . polon . b * 40 * , 1841 ( 2009 ) [ arxiv:0903.2980 [ hep - ph ] ] . r. a. ryutin , arxiv:1211.2105 [ hep - ph ] . y. l. dokshitzer , v. a. khoze and t. sjostrand , phys . b * 274 * , 116 ( 1992 ) . r. s. fletcher and t. stelzer , phys . d * 48 * , 5162 ( 1993 ) [ hep - ph/9306253 ] . h. j. lu and j. milana , phys . d * 51 * , 6107 ( 1995 ) [ hep - ph/9407206 ] . t. l. lungov and c. o. escobar , phys . d * 53 * , 4857 ( 1996 ) [ hep - ph/9510209 ] . j. -r . cudell and o. f. hernandez , nucl . b * 471 * , 471 ( 1996 ) [ hep - ph/9511252 ] . v. a. khoze , a. d. martin and m. g. ryskin , phys . b * 401 * , 330 ( 1997 ) [ hep - ph/9701419 ] . v. a. khoze , a. d. martin and m. g. ryskin , eur . j. c * 14 * , 525 ( 2000 ) [ hep - ph/0002072 ] . b. cox , j. r. forshaw and b. heinemann , phys . b * 540 * , 263 ( 2002 ) [ hep - ph/0110173 ] . v. a. petrov and r. a. ryutin , eur . j. c * 36 * , 509 ( 2004 ) [ hep - ph/0311024 ] . j. r. forshaw , hep - ph/0508274 . r. maciula , r. pasechnik and a. szczurek , phys . d * 82 * , 114011 ( 2010 ) [ arxiv:1006.3007 [ hep - ph ] ] . m. g. albrow and a. rostovtsev , hep - ph/0009336 . v. a. khoze , a. d. martin and m. g. ryskin , eur . j. c * 19 * , 477 ( 2001 ) [ erratum - ibid . c * 20 * , 599 ( 2001 ) ] [ hep - ph/0011393 ] . a. b. kaidalov , v. a. khoze , a. d. martin and m. g. ryskin , eur . j. c * 31 * , 387 ( 2003 ) [ hep - ph/0307064 ] . v. a. khoze , a. d. martin and m. g. ryskin , eur . j. c * 24 * , 581 ( 2002 ) [ hep - ph/0203122 ] . b. e. cox , aip conf . proc . * 753 * , 103 ( 2005 ) [ hep - ph/0409144 ] . p. lebiedowicz , r. pasechnik and a. szczurek , phys . b * 701 * , 434 ( 2011 ) [ arxiv:1103.5642 [ hep - ph ] ] . l. a. harland - lang , v. a. khoze , m. g. ryskin and w. j. stirling , eur . j. c * 71 * , 1714 ( 2011 ) [ arxiv:1105.1626 [ hep - ph ] ] . t. aaltonen _ et al . _ [ cdf collaboration ] , phys . * 99 * , 242002 ( 2007 ) [ arxiv:0707.2374 [ hep - ex ] ] . a. abulencia _ et al . _ [ cdf collaboration ] , phys . rev . lett . * 98 * , 112001 ( 2007 ) [ hep - ex/0611040 ] . t. aaltonen _ et al . _ [ cdf collaboration ] , phys . * 102 * , 242001 ( 2009 ) [ arxiv:0902.1271 [ hep - ex ] ] . t. aaltonen _ et al . _ [ cdf collaboration ] , phys . d * 77 * , 052004 ( 2008 ) [ arxiv:0712.0604 [ hep - ex ] ] . v. m. abazov _ et al . _ [ d0 collaboration ] , phys . lett . b * 705 * , 193 ( 2011 ) [ arxiv:1009.2444 [ hep - ex ] ] . v. a. khoze , a. d. martin and m. g. ryskin , eur . j. c * 23 * , 311 ( 2002 ) [ hep - ph/0111078 ] . s. heinemeyer , v. a. khoze , m. g. ryskin , w. j. stirling , m. tasevsky and g. weiglein , eur . j. c * 53 * , 231 ( 2008 ) [ arxiv:0708.3052 [ hep - ph ] ] . r. enberg and r. pasechnik , phys . d * 83 * , 095020 ( 2011 ) [ arxiv:1104.0889 [ hep - ph ] ] . m. chaichian , p. hoyer , k. huitu , v. a. khoze and a. d. pilkington , jhep * 0905 * , 011 ( 2009 ) [ arxiv:0901.3746 [ hep - ph ] ] . v. p. goncalves and w. k. sauter , phys . d * 82 * , 056009 ( 2010 ) [ arxiv:1007.5487 [ hep - ph ] ] . _ `` production of radions and massive gravitons at lhc '' _ a. v. kisselev , v. a. petrov and r. a. ryutin , conference : c05 - 06 - 22 , p. 49 - 58 proceedings . p. j. bussey , t. d. coughlin , j. r. forshaw and a. d. pilkington , jhep * 0611 * , 027 ( 2006 ) [ hep - ph/0607264 ] . a. v. kisselev , v. a. petrov and r. a. ryutin , phys . b * 630 * , 100 ( 2005 ) [ hep - ph/0506034 ] . g. aad _ et al . _ [ atlas collaboration ] , phys . rev . d * 87 * , 015010 ( 2013 ) [ arxiv:1211.1150 [ hep - ex ] ] . g. aad _ et al . _ [ atlas collaboration ] , new j. phys . * 15 * , 043007 ( 2013 ) [ arxiv:1210.8389 [ hep - ex ] ] . s. chatrchyan _ et al . _ [ cms collaboration ] , phys . b * 711 * , 15 ( 2012 ) [ arxiv:1202.3827 [ hep - ex ] ] . s. chatrchyan _ et al . _ [ cms collaboration ] , phys . lett . * 108 * , 111801 ( 2012 ) [ arxiv:1112.0688 [ hep - ex ] ] . a. g. shuvaev , k. j. golec - biernat , a. d. martin and m. g. ryskin , phys . d * 60 * , 014015 ( 1999 ) [ hep - ph/9902410 ] . j. d. bjorken , phys . d * 47 * , 101 ( 1993 ) . e. gotsman , e. m. levin and u. maor , phys . b * 309 * , 199 ( 1993 ) [ hep - ph/9302248 ] . v. a. khoze , a. d. martin and m. g. ryskin , eur . j. c * 18 * , 167 ( 2000 ) [ hep - ph/0007359 ] . a. b. kaidalov , v. a. khoze , a. d. martin and m. g. ryskin , eur . j. c * 21 * , 521 ( 2001 ) [ hep - ph/0105145 ] . v. a. khoze , a. d. martin and m. g. ryskin , jhep * 0605 * , 036 ( 2006 ) [ hep - ph/0602247 ] . m. g. ryskin , a. d. martin and v. a. khoze , eur . j. c * 54 * , 199 ( 2008 ) [ arxiv:0710.2494 [ hep - ph ] ] . e. gotsman , e. levin , u. maor and j. s. miller , eur . j. c * 57 * , 689 ( 2008 ) [ arxiv:0805.2799 [ hep - ph ] ] . m. g. ryskin , a. d. martin and v. a. khoze , eur . j. c * 60 * , 265 ( 2009 ) [ arxiv:0812.2413 [ hep - ph ] ] . a. d. martin , w. j. stirling , r. s. thorne and g. watt , eur . j. c * 64 * , 653 ( 2009 ) [ arxiv:0905.3531 [ hep - ph ] ] . t. han , j. d. lykken and r. -j . zhang , phys . d * 59 * , 105006 ( 1999 ) [ hep - ph/9811350 ] . l. d. landau , dokl . nauk ser . 60 * , 207 ( 1948 ) . c. -n . yang , phys . * 77 * , 242 ( 1950 ) . v. a. khoze , a. d. martin , m. g. ryskin and w. j. stirling , eur . j. c * 38 * , 475 ( 2005 ) [ hep - ph/0409037 ] . m. drees , r. m. godbole , m. nowakowski and s. d. rindani , phys . rev . d * 50 * , 2335 ( 1994 ) [ hep - ph/9403368 ] . g. jikia and a. tkabladze , phys . b * 323 * , 453 ( 1994 ) [ hep - ph/9312228 ] . j. r. forshaw and m. g. ryskin , z. phys . c * 68 * , 137 ( 1995 ) [ hep - ph/9501376 ] . c. f. von weizsacker , z. phys . * 88 * , 612 ( 1934 ) . e. j. williams , phys . rev . * 45 * , 729 ( 1934 ) .

* abstract . * we study an abstract class of autonomous differential inclusions in hilbert spaces and show the well - posedness and causality , by establishing the operators involved as maximal monotone operators in time and space . then the proof of the well - posedness relies on a well - known perturbation result for maximal monotone operators . moreover , we show that certain types of nonlinear boundary value problems are covered by this class of inclusions and we derive necessary conditions on the operators on the boundary in order to apply the solution theory . we exemplify our findings by two examples . evolutionary inclusions , well - posedness , causality , maximal monotonicity , impedance type boundary conditions , frictional boundary conditions . 34g25 , 35f30 , 35r20 , 46n20 , 47j35

1212.2061

in this article we study differential inclusions of the form @xmath0 where @xmath1 denotes the derivative with respect to time and @xmath2 is assumed to be a maximal monotone relation in time and space . the bounded linear operator @xmath3 , also acting in time and space , is a so - called linear material law , as it was introduced in @xcite . since we are dealing with the autonomous case , the operator @xmath3 and the relation @xmath2 are assumed to commute with the temporal translation operator . we will show that under suitable conditions on @xmath3 and @xmath2 the problem is well - posed , i.e. hadamard s requirements on existence and uniqueness of a solution @xmath4 and its continuous dependence on the given data @xmath5 are satisfied . moreover the issue of causality is addressed , meaning that the behaviour of the solution up to a certain time @xmath6 just depends on the behaviour of the given right hand side up to the same time @xmath6 . + originally this class of problems was discussed in @xcite , where the relation @xmath2 was given by a skew - selfadjoint spatial operator . there , well - posedness and causality were shown under a positive - definiteness constraint on the operator @xmath3 . moreover , the abstract solution theory was applied to several examples of linear equations in mathematical physics . this solution theory was generalized by the author to the case of @xmath2 being a maximal monotone relation in space in @xcite , where @xmath3 was of the particular form @xmath7 we will adopt the proof for the solution theory presented in @xcite to show the well - posedness of problems of the form . the proof mainly relies on the realization of the derivative @xmath1 as a maximal monotone operator on an exponentially weighted @xmath8-space ( see e.g. @xcite or subsection 2.1 of this article ) and the application of a well - known perturbation result for maximal monotone relations ( see ) . for the theory of maximal monotone relations we refer to the monographs @xcite . inclusions of the form cover a large class of possible evolutionary problems , such as integro - differential equations @xcite , delay and neutral differential equations @xcite , certain types of control problems @xcite and equations of mathematical physics involving hysteresis @xcite . + the article is structured as follows . in section 2 we recall the definition of the time - derivative , of linear material laws and some basic facts on maximal monotone relations on hilbert spaces . moreover , we recall the notion of so - called abstract boundary data spaces ( see @xcite ) , which will enable us to formulate evolutionary equations with nonlinear boundary condition as an inclusion of the form . in section 3 the solution theory for is presented . in the remaining part of the article we consider a certain type of partial differential equations with nonlinear boundary conditions . those problems occur for instance in frictional contact problems ( see e.g. @xcite ) in the field of elasticity , where the boundary condition is given by a differential inclusion . in subsection 4.1 we study an abstract nonlinear differential operator @xmath2 , which is given by the restriction of a linear operator to elements , which satisfy the nonlinear boundary condition . we show that under certain constraints on the relation , which occurs in the boundary condition , the operator @xmath2 gets maximal monotone and thus , the solution theory developed in section 3 applies . we illustrate our results in subsection 4.2 . there , in a first example we consider the wave equation with an impedance - type boundary condition . this problem was originally considered in @xcite and we show that it fits in our framework . the second example deals with the equations of visco - elasticity with a frictional boundary condition , which is modelled by a differential inclusion on the boundary . a similar kind of this problem was considered in @xcite for a cylindrical domain and antiplane shear deformations . again we show that the equations are covered by our abstract inclusion . + throughout every hilbert space is assumed to be complex . its inner product and its norm are denoted by @xmath9 and @xmath10 respectively , where the inner product is assumed to be linear with respect to the second and conjugate linear with respect to the first argument . moreover , for a hilbert space @xmath11 and a closed subspace @xmath12 we denote by @xmath13 the orthogonal projection onto @xmath14 is just the canonical embedding . ] then @xmath15 is the orthogonal projector on @xmath16 . in this subsection we recall , how to establish the time - derivative as a normal , boundedly invertible linear operator and we recall the notion of linear material laws . for the proofs and more details we refer to @xcite . let @xmath11 be a hilbert space . for @xmath17 we define @xmath18 as the space of measurable , @xmath11-valued functions on @xmath19 , which are square - integrable with respect to the exponentially weighted lebesgue - measure @xmath20 where by @xmath21 we denote the one - dimensional lebesgue measure . moreover , we set @xmath22 we define the operator @xmath23 as the closure of the mapping @xmath24 where by @xmath25 we denote the space of arbitrarily differentiable functions with compact support in @xmath26 indeed , the operator @xmath23 turns out to be normal and continuously invertible and its inverse is given by @xmath27 for each @xmath28 moreover @xmath29 , which gives @xmath30 . indeed , one can show that the operator norm of @xmath31 is equal to @xmath32 ( see e.g. @xcite ) . moreover , equation shows the causality be vector spaces and @xmath33 . @xmath34 is called _ causal _ ( cf . @xcite ) , _ _ if for each @xmath35 and @xmath36 the implication @xmath37}(m)\left(f - g\right)=0\right)\rightarrow\left(\chi_{(-\infty , a]}(m)\left(g(f)-g(g)\right)=0\right)\ ] ] holds . by @xmath38}(m)$ ] we denote the multiplication operator with the cut - off function @xmath38}$ ] , i.e. @xmath39}(m)f\right)(t)\coloneqq\chi_{(-\infty , a]}(t)f(t).$ ] ] of @xmath31 . + furthermore , we can give a spectral representation for @xmath23 in terms of the so - called _ fourier - laplace - transform _ @xmath40 which is given by @xmath41 where @xmath42 denotes the fourier - transform in @xmath43 given by the unitary extension of @xmath44 and @xmath45 is defined by @xmath46 clearly , @xmath47 is a unitary operator and so is @xmath48 using the well - known spectral representation of the weak derivative on @xmath43 via the fourier - transform @xmath42 , we get that @xmath49 where @xmath50 denotes the multiplication - by - the - argument operator with maximal domain , i.e. @xmath51 for all @xmath52 and @xmath53 . note that the operators @xmath23 and @xmath54 can be extended to @xmath55 in the canonical way . + next we consider a bounded strongly measurable mapping @xmath56 , where @xmath57 denotes the open ball in @xmath58 around the center @xmath59 with radius @xmath60 we define the bounded linear operator @xmath61 by setting @xmath62 the spectral representation of the derivative @xmath23 now leads to the definition of the linear and bounded operator @xmath63 if we assume additionally that @xmath64 is analytic , we call @xmath65 a _ linear material - law_. ] moreover , as @xmath66 is a function of @xmath31 , we obtain the translation invariance of @xmath67 i.e. for each @xmath68 we have @xmath69 , where by @xmath70 we denote the translation operator on @xmath18 given by @xmath71 for @xmath72 and @xmath73 we recall the notion of monotone and maximal monotone relations in hilbert spaces . moreover we state minty s famous theorem , which gives a characterization of maximal monotonicity , which will be frequently used in the forthcoming parts . additionally , we recall the definition of the yosida - approximation of a maximal monotone relation and state a well - known perturbation result , which will be the key argument for proving our solution theory for inclusions of the form . throughout let @xmath11 be a hilbert space and @xmath74 be a binary relation . the relation @xmath2 is called _ monotone , _ if for each @xmath75 the inequality @xmath76 holds . we call @xmath2 _ maximal monotone _ , if @xmath2 is monotone and there exists no proper monotone extension of @xmath2 , i.e. for each monotone relation @xmath77 with @xmath78 it follows that @xmath79 from this definition we can immediately derive the _ demi - closedness _ of maximal monotone relations , which in particular means that if @xmath80 and @xmath81 are sequences in @xmath11 such that @xmath82 and @xmath83 and @xmath84 for each @xmath85 then @xmath86 + before we state minty s theorem , we introduce a linear structure on the set of binary relations . let @xmath87 be two hilbert spaces , @xmath88 and @xmath89 then we define the relation @xmath90 by @xmath91 moreover , for @xmath92 and @xmath93 we set @xmath94 & \coloneqq\left\ { y\in h_{1}\,|\,\exists x\in m:(x , y)\in b\right\ } , \\ { } [ n]b & \coloneqq\left\ { x\in h_{0}\,|\,\exists y\in n:(x , y)\in b\right\ } .\end{aligned}\ ] ] for a monotone relation we can characterize the maximal monotonicity by using minty s theorem . a proof can be found for instance in @xcite . [ thm : minty ] let @xmath2 be monotone . then the following statements are equivalent : a. @xmath2 is maximal monotone , b. there exists @xmath95 such that @xmath96=h,$ ] c. for each @xmath95 it holds @xmath96=h.$ ] [ rem : maximal monotone ] let @xmath2 be maximal monotone . a. due to the monotonicity of @xmath2 , the relation @xmath97 is right - unique for each @xmath95 , i.e. a mapping , whose domain equals @xmath11 according to . indeed , one can show that @xmath98 is lipschitz - continuous with a lipschitz - constant less than or equal to @xmath99 ( see e.g. ( * ? ? ? * theorem 1.3 ) , ( * ? ? ? * proposition 1.12 ) ) . moreover , we define @xmath100 for @xmath95 and obtain again a lipschitz - continuous mapping with a lipschitz - constant less than or equal to @xmath101 . furthermore , @xmath102 is monotone for each @xmath95 . the mappings @xmath103 are called the _ yosida - approximation _ of @xmath104 moreover , for each @xmath95 and @xmath105 we have @xmath106 b. for @xmath107a$ ] the set @xmath108 $ ] is convex and closed . hence , we can find the element in @xmath108 $ ] possessing minimal norm , and we denote this element by @xmath109 then one can show , that for each @xmath107a$ ] and @xmath95 the inequality @xmath110 holds ( see ( * ? ? ? * theorem 1.3 ) , ( * ? ? ? * proposition 1.12 ) ) . finally , we want to state two perturbation results for maximal monotone relations . if @xmath111 are two monotone relations , then clearly @xmath112 is also monotone . however , in general the sum of two maximal monotone relations is not maximal monotone again . so the question is : if @xmath2 and @xmath113 are maximal monotone , when is @xmath112 maximal monotone ? a positive answer can be given , if one of the relations is a lipschitz - continuous mapping defined on the whole space @xmath114 [ lem : yosida]let @xmath2 be maximal monotone and @xmath115 monotone and lipschitz - continuous . then @xmath112 is maximal monotone . the proof is based on an application of the contraction mapping theorem and can be found for instance in ( * ? ? ? * lemme 2.4 ) or in ( * ? ? ? * lemma 1.15 ) . if @xmath2 and @xmath113 are maximal monotone relations in @xmath116 we can use the previous lemma and the fact that @xmath117 is monotone and lipschitz - continuous for each @xmath95 to derive the following perturbation result . [ prop : pert]let @xmath111 be maximal monotone and @xmath118 for @xmath95 define @xmath119 if @xmath120 then @xmath121 converges to some @xmath105 as @xmath122 with @xmath123 in particular @xmath112 is maximal monotone . for a proof of this statement we refer to ( * ? ? ? * lemma 7.41 ) or ( * ? ? ? * lemma 1.17 ) . the last concept we need for our main results are so - called abstract boundary data spaces , introduced in @xcite . let @xmath87 be two hilbert spaces and @xmath124 be two densely defined closed linear operators . moreover , assume that @xmath125 and @xmath126 are _ formally skew - adjoint _ , i.e. @xmath127 , which also yields @xmath128 and @xmath126 as the closure of the gradient and the divergence on @xmath8 , defined on test - functions . then the adjoints @xmath34 and @xmath129 are nothing but the usual weak gradient and weak divergence in @xmath8 ( see subsection 4.2 ) . ] if we equip the domains of @xmath34 and @xmath129 with their respective graph - norms , we get that @xmath130 and @xmath131 are closed linear subspaces of @xmath132 and @xmath133 and hence , the projection theorem yields @xmath134 where the orthocomplements have to be taken with respect to the inner products induced by the graph norms of @xmath34 and @xmath129 , respectively . an easy computation yields @xmath135(1-dg),\\ \mathcal{bd}(d)\coloneqq\mathcal{d}(d_{c})^{\bot } & = [ \{0\}](1-gd).\end{aligned}\ ] ] the spaces @xmath136 and @xmath137 are called _ abstract boundary data spaces _ of @xmath34 and @xmath138 respectively . and @xmath126 will be defined as the closure of differential operators , defined on test functions , the elements of the domains of @xmath125 and @xmath126 can be interpreted as those elements in the domain of @xmath34 and @xmath129 with vanishing traces . for instance , if @xmath125 is the closure of the gradient defined on test - functions , then @xmath130 is the classical sobolev - space @xmath139 , i.e. the space of weakly differentiable functions in @xmath8 with vanishing traces . hence , the boundary values of a function in @xmath132 just depend on the boundary values of its projection onto @xmath140 clearly , @xmath141\subseteq\mathcal{bd}(d)$ ] and @xmath142\subseteq\mathcal{bd}(g).$ ] we define the operators @xmath143{c } \bullet \vspace{-1mm}\\ \rule{3pt}{0pt}\hspace{-1.5 mm } g \end{array}}:\mathcal{bd}(g ) & \to\mathcal{bd}(d),\\ { \begin{array}[b]{c } \bullet \vspace{-1mm}\\ \rule{3pt}{0pt}\hspace{-1.5 mm } d \end{array}}:\mathcal{bd}(d ) & \to\mathcal{bd}(g),\end{aligned}\ ] ] as the respective restrictions of @xmath34 and @xmath129 . as @xmath136 and @xmath137 are closed subspaces of @xmath132 and @xmath144 respectively , they inherit the hilbert space structure from these supersets . with respect to these topologies , the operators @xmath145{c } \bullet \vspace{-1mm}\\ \rule{3pt}{0pt}\hspace{-1.5 mm } g \end{array}}$ ] and @xmath145{c } \bullet \vspace{-1mm}\\ \rule{3pt}{0pt}\hspace{-1.5 mm } d \end{array}}$ ] enjoy a surprising property . the operators @xmath145{c } \bullet \vspace{-1mm}\\ \rule{3pt}{0pt}\hspace{-1.5 mm } g \end{array}}$ ] and @xmath145{c } \bullet \vspace{-1mm}\\ \rule{3pt}{0pt}\hspace{-1.5 mm } d \end{array}}$ ] are unitary . furthermore , @xmath146{c } \bullet \vspace{-1mm}\\ \rule{3pt}{0pt}\hspace{-1.5 mm } g \end{array}}\right)^{\ast}={\begin{array}[b]{c } \bullet \vspace{-1mm}\\ \rule{3pt}{0pt}\hspace{-1.5 mm } d \end{array}}$ ] and so @xmath146{c } \bullet \vspace{-1mm}\\ \rule{3pt}{0pt}\hspace{-1.5 mm } d \end{array}}\right)^{\ast}={\begin{array}[b]{c } \bullet \vspace{-1mm}\\ \rule{3pt}{0pt}\hspace{-1.5 mm } g \end{array}}.$ ] let @xmath11 be a hilbert space . in this section we study evolutionary inclusions of the form @xmath147 where @xmath56 is a linear material law for some @xmath17 and @xmath148 is an autonomous maximal monotone relation , i.e. @xmath2 is maximal monotone and for each @xmath68 and @xmath149 it follows that @xmath150 the function @xmath151 is a given source term and @xmath72 is the unknown . we address the question of well - posedness of this problem , i.e. we show the uniqueness , existence and continuous dependence of a solution @xmath4 on the given data @xmath152 more precisely , we show that @xmath153 is a lipschitz - continuous mapping , whose domain is the whole space @xmath154 the second property we consider , is causality of the _ solution operator @xmath155 _ this subsection is devoted to the well - posedness of differential inclusions of the form . throughout , let @xmath11 be a hilbert space , @xmath56 a bounded strongly measurable function for some @xmath17 and @xmath148 . we say that @xmath64 and @xmath2 satisfy the hypotheses ( h1 ) , ( h2 ) and ( h3 ) respectively , if 1 . there exists @xmath156 such that for all @xmath157 the inequality @xmath158 holds . @xmath2 is maximal monotone and _ autonomous _ , i.e. for every @xmath68 and @xmath149 we have @xmath150 3 . for all @xmath75 the estimate @xmath159 holds . [ rem : autonomous relations]@xmath160 a. a typical example for an autonomous maximal monotone relation is the extension of a maximal monotone relation @xmath161 given by @xmath162 if one assumes that @xmath163 then @xmath2 is maximal monotone ( cf . b. in case of @xmath2 being the canonical extension of a skew - selfadjoint operator on @xmath11 , the evolutionary problem was originally considered by picard in several works ( cf . @xcite ) . the more general case of an autonomous operator @xmath164 was treated in @xcite . c. if @xmath2 is the extension of a maximal monotone relation @xmath77 ( compare ( a ) ) and @xmath65 is of the particular form @xmath165 , the problem was addressed by the author in @xcite . indeed , in @xcite the inclusion was studied on the half line @xmath166 instead of @xmath167 which in particular implies that one can omit the additional assumption @xmath168 in ( a ) in order to obtain the maximal monotonicity of the extension @xmath104 first , we state the following simple , but useful observation . [ lem : core]let @xmath17 and @xmath169 . assume that @xmath170 . then @xmath171 is a core of @xmath172 and @xmath173 let @xmath174 i.e. @xmath175 for @xmath176 we define @xmath177 and we obtain that as @xmath178 as well as @xmath179 in particular , this yields that @xmath180 and hence , @xmath181 [ lem : maximal monoton material law ] assume that @xmath64 satisfies ( h1 ) . then @xmath182 is a maximal monotone operator in @xmath154 using the unitary equivalence of @xmath183 in @xmath18 and @xmath184@xmath185 in @xmath186 , where @xmath187 denotes the multiplication - by - the - argument operator ( see subsection 2.1 ) , it suffices to prove the maximal monotonicity of @xmath188 in @xmath189 the monotonicity follows immediately from ( h1 ) . furthermore , by ( h1 ) we get the strict monotonicity of @xmath190 , where for @xmath191 and almost every @xmath53 we have that @xmath192 the latter implies that the operator @xmath193 has a dense range , which gives , using the closedness and the continuous invertibility of the operator , that @xmath193 is onto . hence yields the assertion . [ prop : uniq ] let @xmath194 be monotone and let @xmath64 satisfy ( h1 ) . then for each @xmath195 the estimate @xmath196 holds . since the operator @xmath197 is monotone , according to , we obtain that @xmath198 is monotone . an easy argument , using the cauchy - schwarz - inequality , yields the assertion . especially yields the uniqueness and continuous dependence of a solution @xmath4 of on the given right hand side @xmath5 . the next proposition yields the existence of a solution of for every right hand side @xmath151 . the proof uses the perturbation result given in in order to show the existence of a solution for @xmath199 and follows the same strategy as the one in ( * ? ? ? * proposition 4.6 ) . [ prop : existence]let @xmath64 satisfy ( h1 ) and @xmath2 satisfy ( h2 ) . then for each @xmath199 there exists @xmath72 such that @xmath200 we replace @xmath2 by its yosida approximation @xmath102 for @xmath95 , which is a lipschitz - continuous monotone mapping on @xmath154 then , using and we find an element @xmath201 such that @xmath202 which is equivalent to @xmath203 we prove that @xmath204 is uniformly bounded . for that purpose let @xmath205 then , using that @xmath102 and @xmath206 commute with the operator @xmath207 given by @xmath208 for @xmath209 we obtain that @xmath210 since @xmath102 is monotone , yields @xmath211 and hence , using the mean - value inequality @xmath212 for each @xmath213 where @xmath214 denotes the support of @xmath5 . the latter gives , by choosing a weak - convergent subsequence of @xmath215 , that @xmath216 and that @xmath217 hence , using we obtain @xmath218 for each @xmath219 thus , applies and hence , we find an element @xmath72 satisfying @xmath220 which gives @xmath221 we summarize our findings of this subsection in the following theorem . [ thm : well_posedness ] let @xmath11 be a hilbert space , @xmath56 a bounded strongly measurable function for some @xmath17 satisfying ( h1 ) and @xmath148 a relation satisfying ( h2 ) . then for each @xmath151 there exists a unique @xmath72 such that @xmath222 moreover , @xmath223 is lipschitz - continuous with a lipschitz constant less than or equal to @xmath224 in this subsection we prove the causality of the solution operator @xmath223 of an inclusion of the form . first we give an equivalent condition for the causality of the operator @xmath225 , which will enable us to prove the causality for the case of non - vanishing @xmath104 the statement was already given in ( * ? ? ? * proposition 2.65 ) in a slightly more general version . however , for sake of completeness we present the proof . [ lem : eq_causality]let @xmath17 and @xmath56 be a strongly measurable bounded mapping satisfying the hypothesis ( h1 ) . then the following statements are equivalent : a. the operator @xmath226 is causal , b. for every @xmath227 we have that @xmath228 assume that @xmath225 is causal and let @xmath229 we set @xmath230 and obtain @xmath231}(m)\left(\partial_{0,\nu}m\left(\partial_{0,\nu}^{-1}\right)\right)^{-1}v\right.\right\rangle _ { h_{\nu,0}(\mathbb{r};h)}\\ & = { \operatorname{\mathfrak{re}}}\left\langle v\left|\chi_{(-\infty,0]}(m)\left(\partial_{0,\nu}m\left(\partial_{0,\nu}^{-1}\right)\right)^{-1}\left(\chi_{(-\infty,0]}(m)v\right)\right.\right\rangle _ { h_{\nu,0}(\mathbb{r};h)}\\ & = { \operatorname{\mathfrak{re}}}\left\langle \chi_{(-\infty,0]}(m)v\left|\left(\partial_{0,\nu}m\left(\partial_{0,\nu}^{-1}\right)\right)^{-1}\left(\chi_{(-\infty,0]}(m)v\right)\right.\right\rangle _ { h_{\nu,0}(\mathbb{r};h)}.\end{aligned}\ ] ] using we get that @xmath232}(m)v\left|\left(\partial_{0,\nu}m\left(\partial_{0,\nu}^{-1}\right)\right)^{-1}\left(\chi_{(-\infty,0]}(m)v\right)\right.\right\rangle _ { h_{\nu,0}(\mathbb{r};h)}\\ & \geq c\left|\left(\partial_{0,\nu}m\left(\partial_{0,\nu}^{-1}\right)\right)^{-1}\left(\chi_{(-\infty,0]}(m)v\right)\right|_{h_{\nu,0}(\mathbb{r};h)}^{2}\\ & \geq c\left|\chi_{(-\infty,0]}(m)\left(\partial_{0,\nu}m\left(\partial_{0,\nu}^{-1}\right)\right)^{-1}\left(\chi_{(-\infty,0]}(m)v\right)\right|_{h_{\nu,0}(\mathbb{r};h)}^{2}\\ & = c\left|\chi_{(-\infty,0]}(m)\left(\partial_{0,\nu}m\left(\partial_{0,\nu}^{-1}\right)\right)^{-1}v\right|_{h_{\nu,0}(\mathbb{r};h)}^{2}\\ & = c\intop_{-\infty}^{0}|u(t)|^{2}e^{-2\nu t}\mbox { d}t.\end{aligned}\ ] ] assume now that ( ii ) holds . using that @xmath183 is translation invariant , i.e. it commutes with the translation operator @xmath70 for each @xmath233 the asserted inequality implies @xmath234 for each @xmath235 due to the linearity of @xmath225 it suffices to prove that @xmath38}(m)f=0 $ ] implies @xmath38}(m)\left(\partial_{0,\nu}m\left(\partial_{0,\nu}^{-1}\right)\right)^{-1}f=0 $ ] for each @xmath236 so let @xmath151 with @xmath38}(m)f=0 $ ] for some @xmath35 and define @xmath237 then gives @xmath238 which implies @xmath38}(m)u=0.$ ] if @xmath56 is a linear material law and satisfies the hypothesis ( h1 ) , then gives that holds for every @xmath227 and every @xmath239 since the assumed analyticity for @xmath64 yields the causality of @xmath225 ( see ( * ? ? ? * lemma 2.5 ) ) . now we are able to show the causality of the solution operator associated to the evolutionary inclusion . [ prop : causality ] let @xmath11 be a hilbert space , @xmath17 and @xmath56 a linear material law , satisfying hypothesis ( h1 ) . furthermore let @xmath148 be a binary relation satisfying ( h2 ) and ( h3 ) . then , the solution operator @xmath240 which exists according to , is causal . first note that due to the translation invariance of @xmath2 and @xmath206 it suffices to check that for @xmath241 satisfying @xmath242}(m)(f - g)=0 $ ] we have that @xmath243}(m)\left(\left(\overline{\partial_{0,\nu}m\left(\partial_{0,\nu}^{-1}\right)+a}\right)^{-1}(f)-\left(\overline{\partial_{0,\nu}m\left(\partial_{0,\nu}^{-1}\right)+a}\right)^{-1}(g)\right)=0.\ ] ] so let @xmath241 with @xmath242}(m)(f - g)=0.$ ] we choose sequences @xmath244 and @xmath245 in @xmath246 $ ] ( see ) such that @xmath247 and @xmath248 in @xmath18 as @xmath249 for @xmath250 we define @xmath251 and due to the continuity of @xmath223 we get that @xmath252 as well as @xmath253 in @xmath154 we estimate , using and ( h3 ) @xmath254 for each @xmath255 passing to the limit as @xmath256 gives @xmath257 which yields the causality . in this subsection we study a class of evolutionary equations , which covers evolutionary problems with nonlinear boundary conditions , modelled by an inclusion on the boundary . those equations occur for instance in the study of problems in elasticity with frictional boundary conditions , where the behaviour on the boundary is described by variational inequalities ( cf . @xcite ) or by suitable differential inclusions ( cf . @xcite ) . moreover , since we consider operators acting in time and space , also problems with boundary conditions given by ordinary differential equations , delay equations or inclusions can be treated within our framework . + throughout let @xmath258 and @xmath259 be hilbert spaces and @xmath260 and @xmath261 two densely defined closed linear operators . moreover assume that @xmath125 and @xmath126 are formally skew - adjoint , i.e. @xmath262 we do not distinguish notationally between the operators @xmath263 and @xmath129 and their canonical extensions to @xmath264 and @xmath265 respectively . recall from subsection 2.3 that @xmath135(1-dg),\\ \mathcal{bd}(d)\coloneqq\mathcal{d}(d_{c})^{\bot } & = [ \{0\}](1-gd),\end{aligned}\ ] ] where the orthocomplements are taken with respect to the inner products in @xmath132 and @xmath144 respectively . we consider the following operator @xmath266 given by @xmath267{c } \bullet \vspace{-1mm}\\ \rule{3pt}{0pt}\hspace{-1.5 mm } d \end{array}}\pi_{\mathcal{bd}(d)}v\right)\in h\right.\right\ } , \label{eq : a}\\ a\left(\begin{array}{c } u\\ v \end{array}\right ) & \coloneqq\left(\begin{array}{cc } 0 & d\\ g & 0 \end{array}\right)\left(\begin{array}{c } u\\ v \end{array}\right).\nonumber \end{aligned}\ ] ] here @xmath268 is a binary relation . we will show that under appropriate assumptions on @xmath269 the operator @xmath2 satisfies the hypotheses ( h2 ) and ( h3 ) and thus and are applicable . the first subsection is devoted to the proof of the following theorem . [ thm : a_hypotheses]if @xmath270 satisfies the hypotheses ( h2 ) or ( h3 ) , then so does @xmath104 in subsection 4.2 we will illustrate the applicability of the abstract class of evolutionary inclusions of the form , where the relation @xmath2 is an operator of the form given in . we begin to show the monotonicity of @xmath2 under the assumption that @xmath270 is monotone . indeed , we will show an equality , which will imply , despite the monotonicity of @xmath2 , that @xmath2 satisfies ( h3 ) if @xmath270 does . [ lem : monotone ] let @xmath271 then we compute @xmath272{c } \bullet \vspace{-1mm}\\ \rule{3pt}{0pt}\hspace{-1.5 mm } d \end{array}}\pi_{\mathcal{bd}(d)}\left(v(t)-y(t)\right)\right.\right\rangle _ { \mathcal{bd}(g)}e^{-2\nu t}\mbox { d}t.\tag*{\qedhere}\end{aligned}\ ] ] [ lem : closed ] let @xmath273 be a sequence in @xmath274 such that @xmath275 as well as @xmath276 and @xmath277 in @xmath264 and @xmath278 respectively . due to the closedness of @xmath34 and @xmath129 we obtain that @xmath279 and @xmath280 note that this means that @xmath281 and @xmath83 in @xmath282 and @xmath283 , respectively . the latter implies @xmath145{c } \bullet \vspace{-1mm}\\ \rule{3pt}{0pt}\hspace{-1.5 mm } d \end{array}}\pi_{\mathcal{bd}(d)}v_{n}\to{\begin{array}[b]{c } \bullet \vspace{-1mm}\\ \rule{3pt}{0pt}\hspace{-1.5 mm } d \end{array}}\pi_{\mathcal{bd}(d)}v$ ] and @xmath284 in @xmath285 since @xmath286{c } \bullet \vspace{-1mm}\\ \rule{3pt}{0pt}\hspace{-1.5 mm } d \end{array}}\pi_{\mathcal{bd}(d)}v_{n}\right)\in h$ ] for each @xmath85 the closedness of @xmath270 yields @xmath287{c } \bullet \vspace{-1mm}\\ \rule{3pt}{0pt}\hspace{-1.5 mm } d \end{array}}\pi_{\mathcal{bd}(d)}v\right)\in h.$ ] this shows @xmath288 and hence , @xmath2 is closed . now we are able to state the main result of this subsection . [ prop : a_max_mon]let @xmath270 be maximal monotone . then @xmath2 is maximal monotone , too . since every maximal monotone relation is closed ( in fact it is even demi - closed , see subsection 2.2 ) , the operator @xmath2 is closed according to . moreover , by @xmath2 is monotone . thus , by , it suffices to prove that @xmath289 has a dense range . to this end let @xmath290 and @xmath291 we define and @xmath292 are boundedly invertible , since @xmath293 and @xmath294 are positive selfadjoint operators.]as well aswe show that @xmath288 and that @xmath295 first , we compute @xmath296 and @xmath297 moreover , we get that @xmath298{c } \bullet \vspace{-1mm}\\ \rule{3pt}{0pt}\hspace{-1.5 mm } d \end{array}}\pi_{\mathcal{bd}(d)}g_{c}{\widetilde}{u}\right)+{\widetilde}{u}+d\left(-\pi_{\mathcal{bd}(d)}^{\ast}{\begin{array}[b]{c } \bullet \vspace{-1mm}\\ \rule{3pt}{0pt}\hspace{-1.5 mm } g \end{array}}(1+h)^{-1}\left(-{\begin{array}[b]{c } \bullet \vspace{-1mm}\\ \rule{3pt}{0pt}\hspace{-1.5 mm } d \end{array}}\pi_{\mathcal{bd}(d)}g_{c}{\widetilde}{u}\right)+{\widetilde}{v}\right)\\ & = \pi_{\mathcal{bd}(g)}^{\ast}(1+h)^{-1}\left(-{\begin{array}[b]{c } \bullet \vspace{-1mm}\\ \rule{3pt}{0pt}\hspace{-1.5 mm } d \end{array}}\pi_{\mathcal{bd}(d)}g_{c}{\widetilde}{u}\right)+{\widetilde}{u}-\pi_{\mathcal{bd}(g)}^{\ast}{\begin{array}[b]{c } \bullet \vspace{-1mm}\\ \rule{3pt}{0pt}\hspace{-1.5 mm } d \end{array}}{\begin{array}[b]{c } \bullet \vspace{-1mm}\\ \rule{3pt}{0pt}\hspace{-1.5 mm } g \end{array}}(1+h)^{-1}\left(-{\begin{array}[b]{c } \bullet \vspace{-1mm}\\ \rule{3pt}{0pt}\hspace{-1.5 mm } d \end{array}}\pi_{\mathcal{bd}(d)}g_{c}{\widetilde}{u}\right)+d{\widetilde}{v}\\ & = { \widetilde}{u}+d{\widetilde}{v}=f,\end{aligned}\ ] ] as well as @xmath299{c } \bullet \vspace{-1mm}\\ \rule{3pt}{0pt}\hspace{-1.5 mm } g \end{array}}(1+h)^{-1}\left(-{\begin{array}[b]{c } \bullet \vspace{-1mm}\\ \rule{3pt}{0pt}\hspace{-1.5 mm } d \end{array}}\pi_{\mathcal{bd}(d)}g_{c}{\widetilde}{u}\right)+{\widetilde}{v}+g\left(\pi_{\mathcal{bd}(g)}^{\ast}(1+h)^{-1}\left(-{\begin{array}[b]{c } \bullet \vspace{-1mm}\\ \rule{3pt}{0pt}\hspace{-1.5 mm } d \end{array}}\pi_{\mathcal{bd}(d)}g_{c}{\widetilde}{u}\right)+{\widetilde}{u}\right)\\ & = { \widetilde}{v}+g_{c}{\widetilde}{u}=g,\end{aligned}\ ] ] which shows that @xmath300 hence , the only thing which is left to show is @xmath287{c } \bullet \vspace{-1mm}\\ \rule{3pt}{0pt}\hspace{-1.5 mm } d \end{array}}\pi_{\mathcal{bd}(d)}v\right)\in h.$ ] since @xmath301 takes values in @xmath130 we get that @xmath302{c } \bullet \vspace{-1mm}\\ \rule{3pt}{0pt}\hspace{-1.5 mm } d \end{array}}\pi_{\mathcal{bd}(d)}g_{c}{\widetilde}{u}\right).\label{eq : fixpoint}\ ] ] using that @xmath303 we obtain @xmath304 thus , we compute @xmath305{c } \bullet \vspace{-1mm}\\ \rule{3pt}{0pt}\hspace{-1.5 mm } g \end{array}}\pi_{\mathcal{bd}(g)}u\\ & = -{\begin{array}[b]{c } \bullet \vspace{-1mm}\\ \rule{3pt}{0pt}\hspace{-1.5 mm } g \end{array}}(1+h)^{-1}\left(-{\begin{array}[b]{c } \bullet \vspace{-1mm}\\ \rule{3pt}{0pt}\hspace{-1.5 mm } d \end{array}}\pi_{\mathcal{bd}(d)}g_{c}{\widetilde}{u}\right)+\pi_{\mathcal{bd}(d)}{\widetilde}{v}+{\begin{array}[b]{c } \bullet \vspace{-1mm}\\ \rule{3pt}{0pt}\hspace{-1.5 mm } g \end{array}}(1+h)^{-1}\left(-{\begin{array}[b]{c } \bullet \vspace{-1mm}\\ \rule{3pt}{0pt}\hspace{-1.5 mm } d \end{array}}\pi_{\mathcal{bd}(d)}g_{c}{\widetilde}{u}\right)\\ & = \pi_{\mathcal{bd}(d)}{\widetilde}{v}=-\pi_{\mathcal{bd}(d)}g_{c}{\widetilde}{u},\end{aligned}\ ] ] which gives @xmath306{c } \bullet \vspace{-1mm}\\ \rule{3pt}{0pt}\hspace{-1.5 mm } d \end{array}}\pi_{\mathcal{bd}(d)}v+\pi_{\mathcal{bd}(g)}u=-{\begin{array}[b]{c } \bullet \vspace{-1mm}\\ \rule{3pt}{0pt}\hspace{-1.5 mm } d \end{array}}\pi_{\mathcal{bd}(d)}g_{c}{\widetilde}{u}.\ ] ] together with the latter yields @xmath307{c } \bullet \vspace{-1mm}\\ \rule{3pt}{0pt}\hspace{-1.5 mm } d \end{array}}\pi_{\mathcal{bd}(d)}v+\pi_{\mathcal{bd}(g)}u\right),\ ] ] which is equivalent to @xmath308{c } \bullet \vspace{-1mm}\\ \rule{3pt}{0pt}\hspace{-1.5 mm } d \end{array}}\pi_{\mathcal{bd}(d)}v)\in h.$ ] summarizing we have shown that if @xmath270 satisfies the hypothesis ( h3 ) , then so does @xmath2 , which follows by . moreover , it follows that if @xmath270 satisfies ( h2 ) , then @xmath2 satisfies ( h2 ) as well . indeed , the maximal monotonicity was shown in while the statement that @xmath2 is autonomous if @xmath270 is autonomous follows directly by the definition of @xmath104 hence , we have proved . in this subsection we apply our findings of the sections 3 and 4.1 to two examples . the first problem is linear and deals with the model of acoustic waves with an impedance - type boundary condition ( see @xcite ) . the second one , describing a frictional contact problem for a visco - elastic medium , was also treated by migorski et al . ( see @xcite ) for an antiplane model , using a variational approach . throughout let @xmath309 . we define the following differential operators , which will play the role of @xmath34 and @xmath129 . we define @xmath310 as the closure of the operator @xmath311 and likewise @xmath312 as the closure of @xmath313 the operators @xmath310 and @xmath312 are formally skew - adjoint and as above we set @xmath314 and @xmath315 note that an element @xmath4 lying in the domain of @xmath310 satisfies the abstract dirichlet - boundary condition `` @xmath316 '' on @xmath317 likewise , an element @xmath318 satisfies the neumann - boundary condition `` @xmath319 '' on @xmath320 in case of a smooth boundary @xmath321 where @xmath322 denotes the unit outer normal vector - field on @xmath320 . the corresponding boundary data spaces are given by @xmath323(1-{\operatorname{div}}{\operatorname{grad}}),$ ] the space of 1-harmonic functions on @xmath324 , and @xmath325(1-{\operatorname{grad}}{\operatorname{div}}).$ ] following @xcite we consider the following boundary value problem : @xmath326 here @xmath327 is a linear material law for some @xmath328 , having the power series representation @xmath329 where the @xmath330 s are @xmath331-vector fields , and @xmath332 denotes the corresponding multiplication operator . moreover we assume that @xmath333 such that @xmath334 gets bounded . we choose @xmath335 using the product rule @xmath336 for @xmath337 we can establish @xmath338 as a bounded operator @xmath339 moreover , one can show that @xmath340\subseteq h_{\nu,0}\left(\mathbb{r};\mathcal{d}({\operatorname{div}}_{c})\right).$ ] we define @xmath341 by @xmath342{c } \bullet \vspace{-1mm}\\ \rule{3pt}{0pt}\hspace{-1.5 mm } { \operatorname{div}}\end{array}}\pi_{\mathcal{bd}({\operatorname{div}})}a(\partial_{0,\nu}^{-1})\pi_{\mathcal{bd}({\operatorname{grad}})}^{\ast}$ ] with maximal domain , which is a densely defined closed linear operator . consider the operator @xmath343 where @xmath344{c } \bullet \vspace{-1mm}\\ \rule{3pt}{0pt}\hspace{-1.5 mm } { \operatorname{div}}\end{array}}\pi_{\mathcal{bd}({\operatorname{div}})}q\right.\right\ } .\ ] ] note that , by definition , the operator @xmath270 is autonomous in the sense of ( h2 ) . the evolutionary equation for the boundary value problem , is then given by @xmath345 indeed , if @xmath346 is a solution of then we formally get that @xmath347 , which gives @xmath348 moreover , the first line of gives @xmath349 hence , the formula for @xmath350 yields @xmath351 which gives . moreover , by the domain of @xmath2 we get that @xmath143{c } \bullet \vspace{-1mm}\\ \rule{3pt}{0pt}\hspace{-1.5 mm } { \operatorname{div}}\end{array}}\pi_{\mathcal{bd}({\operatorname{div}})}q & = h\left(\pi_{\mathcal{bd}({\operatorname{grad}})}u\right)\\ & = \partial_{0,\nu}{\begin{array}[b]{c } \bullet \vspace{-1mm}\\ \rule{3pt}{0pt}\hspace{-1.5 mm } { \operatorname{div}}\end{array}}\pi_{\mathcal{bd}({\operatorname{div}})}a(\partial_{0,\nu}^{-1})\pi_{\mathcal{bd}({\operatorname{grad}})}^{\ast}\pi_{\mathcal{bd}({\operatorname{grad}})}u\\ & = \partial_{0,\nu}{\begin{array}[b]{c } \bullet \vspace{-1mm}\\ \rule{3pt}{0pt}\hspace{-1.5 mm } { \operatorname{div}}\end{array}}\pi_{\mathcal{bd}({\operatorname{div}})}a(\partial_{0,\nu}^{-1})u,\end{aligned}\ ] ] where in the third equality we have used that @xmath340\subseteq h_{\nu,0}\left(\mathbb{r};\mathcal{d}({\operatorname{div}}_{c})\right).$ ] since @xmath352 the latter equation gives @xmath353 which is the appropriate formulation for the boundary condition within our framework . according to our findings in subsection 4.1 it suffices to check , whether @xmath270 satisfies the hypotheses ( h2 ) and ( h3 ) . in @xcite we find an additional constraint on @xmath338 , namely @xmath354 for all @xmath355 . this condition implies the hypothesis ( h3 ) . indeed , for @xmath356 we estimate @xmath357 since @xmath358 is a core of @xmath270 , according to and since @xmath270 commutes with the operators @xmath70 for each @xmath233 this yields the monotonicity of @xmath359 for showing the maximal monotonicity of @xmath269 we determine its adjoint . using we get that @xmath360{c } \bullet \vspace{-1mm}\\ \rule{3pt}{0pt}\hspace{-1.5 mm } { \operatorname{grad}}\end{array}},\ ] ] and we obtain that @xmath361 is a core of @xmath362 hence , @xmath363 is monotone , since for each @xmath364 we estimate @xmath365 we summarize our findings in the following theorem . let @xmath366 be of the form and let @xmath367 satisfy and . then the evolutionary equation is well - posed and the corresponding solution operator is causal . since @xmath270 and @xmath363 are monotone , we get the maximal monotonicity of @xmath269 since @xmath270 is closed . thus , @xmath270 satisfies the hypotheses ( h2 ) and ( h3 ) . and now give the well - posedness of and shows the causality of the solution operator . this subsection is devoted to the study of the following system : @xmath368 on a domain @xmath309 completed by a nonlinear boundary condition of the form @xmath369 where @xmath370 and @xmath322 denotes the outward unit normal vector - field on @xmath317 here @xmath371 denoting the displacement - field of the body @xmath324 , and @xmath372 standing for the stress tensor , are the unknowns . equation is the well - known equation of elasticity , describing the deformation of an elastic body @xmath324 with a density distribution @xmath373 , which is assumed to be real - valued and strictly positive . the operator @xmath374 denotes the divergence for symmetric - matrix - valued functions and the function @xmath375 is an external force . the constitutive relation , linking the stress @xmath6 and the strain @xmath376 , where by @xmath377 we denote the symmetrized jacobian of a vector - valued function , is known as the kelvin - voigt model in visco - elasticity . here @xmath378 and @xmath129 denote the elasticity and the viscosity tensor , respectively . the boundary condition is motivated by so - called frictional boundary conditions , where @xmath379 is usually a subgradient of a convex , lower semicontinuous function ( due to rockafellar and tyrell @xcite ) or a generalized gradient of a locally lipschitz - continuous function ( due to clarke @xcite ) . for typical examples of frictional boundary conditions in elasticity we refer to ( * ? ? ? * ff . ) and to ( * ? ? ? * section 5 ) . + a similar problem was treated in @xcite for antiplane shear deformations , where the domain @xmath324 was assumed to be a cylinder in @xmath380 . in this case the displacement can be expressed by a scalar - valued function and hence , reduces to an equation similar to . the boundary condition was assumed to hold on a part of the boundary , while on the other parts of @xmath320 dirichlet- and neumann - boundary conditions were prescribed . the relation @xmath379 was assumed to be a generalized gradient of a locally lipschitz - continuous function and the existence of a solution was proved , using a variational approach . the uniqueness of a solution was shown under the additional constraint that @xmath379 is quasi - monotone ( i.e. there exists some constant @xmath156 such that @xmath381 is monotone ) . here , for simplicity we assume that holds on the whole boundary and that @xmath379 satisfies a suitable monotonicity constraint . however , since our relation is allowed to act in time and space , it covers a broader class of possible boundary conditions . + we begin by introducing the differential operators @xmath377 and @xmath374 . let @xmath309 be open and denote by @xmath382 the space of @xmath383-matrix - valued @xmath384-functions equipped with the frobenius inner product given by @xmath385 moreover , we set @xmath386 which is a closed subspace of @xmath382 and hence , a hilbert space . we define the operator @xmath310 as the closure of @xmath387 and @xmath312 as the closure of @xmath388 furthermore , we introduce the operators @xmath389 and @xmath390 as the closures of @xmath391 and of @xmath392 respectively and of @xmath393 satisfy a generalized boundary condition of the form @xmath394 on @xmath320 and @xmath395 on @xmath320 for @xmath396 and @xmath397 , respectively , where @xmath322 denotes the outward normal vector - field on @xmath317 ] . an easy computation shows that @xmath310 and @xmath312 as well as @xmath389 and @xmath390 are formally skew - adjoint and we define @xmath399 , @xmath400 as well as @xmath401 and @xmath402 throughout we may assume that @xmath403 i.e. korn s inequality holds ( for sufficient conditions for korn s inequality we refer to @xcite and the references therein ) . we rewrite the equations and into a system of the form . we assume that @xmath404 is a selfadjoint , strictly positive definite operator and hence , we may rewrite as follows @xmath405 which gives @xmath406 if we choose @xmath17 large enough , such that @xmath407 , the latter equation can be written as @xmath408 thus , the equations and can be replaced by the system @xmath409 where @xmath410 hence , the material law @xmath65 is given by @xmath411 according to the assumptions on @xmath412 and @xmath138 this material law satisfies the hypothesis ( h1 ) if we choose @xmath17 large enough ( this corresponds to the so - called @xmath413-analytic case in @xcite ) . + we now have to formulate the boundary condition in our framework . since we do not want to impose regularity assumptions on the boundary @xmath321 we have to detour the space @xmath414 following ( * ? ? we define the following substitute for @xmath414 we assume that @xmath415 with @xmath416 to the whole domain @xmath324 . ] we define the operator @xmath417 by @xmath418 and assume that @xmath419{c } \bullet \vspace{-1mm}\\ \rule{3pt}{0pt}\hspace{-1.5 mm } { \operatorname{div}}\end{array}}\nu+\nu^{\ast}{\begin{array}[b]{c } \bullet \vspace{-1mm}\\ \rule{3pt}{0pt}\hspace{-1.5 mm } { \operatorname{grad}}\end{array}}\right)f\right|f\right\rangle _ { \mathcal{bd}({\operatorname{grad}})}>0\quad(f\in\mathcal{bd}({\operatorname{grad}})\setminus\{0\}).\ ] ] this gives that @xmath420{c } \bullet \vspace{-1mm}\\ \rule{3pt}{0pt}\hspace{-1.5 mm } { \operatorname{grad}}\end{array } } g\right.\right\rangle _ { \mathcal{bd}({\operatorname{div}})}+\left\langle \left.{\begin{array}[b]{c } \bullet \vspace{-1mm}\\ \rule{3pt}{0pt}\hspace{-1.5 mm } { \operatorname{grad}}\end{array } } f\right|\nu g\right\rangle _ { \mathcal{bd}({\operatorname{div}})}\right)$ ] defines an inner product on @xmath421 and we denote the completion of @xmath421 with respect to this inner product by @xmath422 moreover , we denote the embedding of @xmath421 into @xmath423 by @xmath424 . [ rem : u]a formal calculation ( see ( * ? ? ? * remark 5.2 ) ) yields that in the case of a smooth boundary we have that @xmath425 we define the mapping @xmath426 then , as it was pointed out in ( * ? ? ? * remark 5.4 ) , the term @xmath145{c } \bullet \vspace{-1mm}\\ \rule{3pt}{0pt}\hspace{-1.5 mm } { \operatorname{grad}}\end{array } } j^{\ast}u$ ] can be identified with @xmath4 on @xmath317 this construction allows us to compare dirichlet- and neumann - type boundary values . using this framework , we assume that @xmath379 is a binary relation on @xmath427 then the boundary condition can be formulated as @xmath428{c } \bullet \vspace{-1mm}\\ \rule{3pt}{0pt}\hspace{-1.5 mm } { \operatorname{grad}}\end{array } } j^{\ast}gj,\ ] ] or equivalently as @xmath429{c } \bullet \vspace{-1mm}\\ \rule{3pt}{0pt}\hspace{-1.5 mm } { \operatorname{div}}\end{array}}\pi_{\mathcal{bd}({\operatorname{div}})}(-t)\right)\in j^{\ast}gj.\ ] ] we define the operator @xmath2 by @xmath430{c } \bullet \vspace{-1mm}\\ \rule{3pt}{0pt}\hspace{-1.5 mm } { \operatorname{div}}\end{array}}\pi_{\mathcal{bd}({\operatorname{div}})}\sigma\right)\in j^{\ast}gj\right.\right\ } \nonumber \\ a\left(\begin{array}{c } v\\ \sigma \end{array}\right ) & \coloneqq\left(\begin{array}{cc } 0 & { \operatorname{div}}\\ { \operatorname{grad } } & 0 \end{array}\right)\left(\begin{array}{c } v\\ \sigma \end{array}\right),\label{eq : visco - elastic_a}\end{aligned}\ ] ] and consider the equation @xmath431 which is , according to our findings , an appropriate reformulation of the problem given by , and . hence , we are in the situation of subsection 4.1 and if we assume that @xmath432 satisfies the hypotheses ( h2 ) and ( h3 ) , applies , which gives the well - posedness and the causality of the problem . we summarize our findings in the following theorem . let @xmath433 be such that @xmath65 , given by , satisfies ( h1 ) and let @xmath434 be such that @xmath432 satisfies the hypotheses ( h2 ) and ( h3 ) . then the problem given by , where @xmath2 is defined by is well - posed and the corresponding solution operator is causal . in the last part of this subsection we will present a condition on @xmath379 , which implies the maximal monotonicity of @xmath435 + in @xcite the relation @xmath379 is defined as the canonical extension ( see ( a ) ) of a quasi - monotone relation @xmath436 on @xmath437 , which is given by @xmath438 where @xmath439 is a suitable function , which is locally lipschitz - continuous with respect to the second variable . moreover it is assumed that @xmath440\partial j(x,\cdot)=\mathbb{r}$ ] for almost every @xmath441 and that there exists @xmath156 such that for every @xmath442 and every @xmath443 satisfying @xmath444 the estimate @xmath445 holds . this gives that @xmath446{\widetilde}{g}=l_{2}(\partial\omega)$ ] and that @xmath436 is a _ bounded _ relation , i.e. @xmath447 $ ] is bounded for bounded @xmath64 . + a suitable realization of these assumptions in our framework is given as follows . we assume that @xmath448 is a maximal monotone bounded relation , satisfying @xmath449{\widetilde}{g}=u$ ] and for technical reasons @xmath450 in the next lemma we show that these conditions imply that @xmath451 is maximal monotone , too . this would give that @xmath452 satisfies the hypotheses ( h2 ) and ( h3 ) , where @xmath434 is the canonical extension of @xmath436 . let @xmath453 be two hilbert spaces , @xmath454 maximal monotone and bounded with @xmath455 and @xmath456t = h_{1}.$ ] moreover , let @xmath457 then @xmath458 is maximal monotone . the monotonicity of @xmath459 is clear . we will show that @xmath460=h_{2},$ ] which implies the maximal monotonicity according to . let @xmath461 we replace the relation @xmath6 by its yosida - approximation @xmath462 which is monotone and lipschitz - continuous with a lipschitz - constant less than or equal to @xmath463 hence , @xmath464 is monotone and lipschitz - continuous with a lipschitz - constant less than or equal to @xmath465 for @xmath466 the contraction mapping theorem yields the existence of an element @xmath467 satisfying @xmath468 hence , the mapping @xmath464 is maximal monotone and thus , for each @xmath95 there exists @xmath469 with @xmath470 the latter can be written as @xmath471 which implies that @xmath472 for each @xmath473 according to ( a ) . since @xmath6 is bounded , this gives that @xmath474 is bounded ( see ( b ) ) . we will prove that @xmath475 converges as @xmath21 tends to @xmath476 for @xmath477 we estimate @xmath478 for a suitable constant @xmath479 . this gives that @xmath121 converges to some @xmath480 as @xmath481 moreover , @xmath482 as @xmath481 furthermore , since @xmath474 is bounded there exists a weakly convergent subsequence @xmath483 with @xmath484 as @xmath256 and we denote its weak limit by @xmath485 then , since @xmath486 for each @xmath473 the demi - closedness of @xmath6 implies @xmath487 since @xmath488 we have that @xmath489 which gives @xmath490 or equivalently @xmath491 the author thanks marcus waurick for careful reading and fruitful discussions . r. picard . a class of evolutionary problems with an application to acoustic waves with impedance type boundary conditions . in spectral theory , mathematical system theory , evolution equations , differential and difference equations _ , volume 221 of _ operator theory : advances and applications _ , pages 533548 . springer basel , 2012 .

the penalized profile sampler for semiparametric inference is an extension of the profile sampler method obtained by profiling a penalized log - likelihood . the idea is to base inference on the posterior distribution obtained by multiplying a profiled penalized log - likelihood by a prior for the parametric component , where the profiling and penalization are applied to the nuisance parameter . because the prior is not applied to the full likelihood , the method is not strictly bayesian . a benefit of this approximately bayesian method is that it circumvents the need to put a prior on the possibly infinite - dimensional nuisance components of the model . we investigate the first and second order frequentist performance of the penalized profile sampler , and demonstrate that the accuracy of the procedure can be adjusted by the size of the assigned smoothing parameter . the theoretical validity of the procedure is illustrated for two examples : a partly linear model with normal error for current status data and a semiparametric logistic regression model . as far as we are aware , there are no other methods of inference in this context known to have second order frequentist validity .

math0701540

semiparametric models are statistical models indexed by both a finite dimensional parameter of interest @xmath0 and an infinite dimensional nuisance parameter @xmath1 . the profile likelihood is typically defined as @xmath2 where @xmath3 is the likelihood of the semiparametric model given @xmath4 observations and @xmath5 is the parameter space for @xmath1 . we also define @xmath6 the convergence rate of the nuisance parameter @xmath1 is the order of @xmath7 , where @xmath8 is some metric on @xmath1 , @xmath9 is any sequence satisfying @xmath10 , and @xmath11 is the true value of @xmath1 . typically , @xmath12 where @xmath13 is the euclidean norm and @xmath14 . of course , a smaller value of @xmath15 leads to a slower convergence rate of the nuisance parameter . for instance , the nuisance parameter in the cox proportional hazards model with right censored data , the cumulative hazard function , has the parametric rate , i.e. , @xmath16 . if current status data is applied to the cox model instead , then the convergence rate will be slower , with @xmath17 , due to the loss of information provided by this kind of data . the profile sampler is the procedure of sampling from the posterior of the profile likelihood in order to estimate and draw inference on the parametric component @xmath0 in a semiparametric model , where the profiling is done over the possibly infinite - dimensional nuisance parameter @xmath1 . @xcite show that the profile sampler gives a first order correct approximation to the maximum likelihood estimator @xmath18 and consistent estimation of the efficient fisher information for @xmath0 even when the nuisance parameter is not estimable at the @xmath19 rate . another bayesian procedure employed to do semiparametric estimation is considered in @xcite who study the marginal semiparametric posterior distribution for a parameter of interest . in particular , @xcite show that marginal semiparametric posterior distributions are asymptotically normal and centered at the corresponding maximum likelihood estimates or posterior means , with covariance matrix equal to the inverse of the fisher information . unfortunately , this fully bayesian method requires specification of a prior on @xmath1 , which is quite challenging since for some models there is no direct extension of the concept of a lebesgue dominating measure for the infinite - dimensional parameter set involved @xcite . the advantages of the profile sampler for estimating @xmath0 compared to other methods is discussed extensively in @xcite , @xcite and @xcite . in many semiparametric models involving a smooth nuisance parameter , it is often convenient and beneficial to perform estimation using penalization . one motivation for this is that , in the absence of any restrictions on the form of the function @xmath1 , maximum likelihood estimation for some semiparametric models leads to over - fitting . seminal applications of penalized maximum likelihood estimation include estimation of a probability density function in @xcite and nonparametric linear regression in @xcite . note that penalized likelihood is a special case of penalized quasi - likelihood studied in @xcite . under certain reasonable regularity conditions , penalized semiparametric log - likelihood estimation can yield fully efficient estimates for @xmath0 ( see , for example , @xcite ) . as far as we are aware , the only general procedure for inference for @xmath0 in this context known to be theoretically valid is a weighted bootstrap with bounded random weights ( see @xcite ) . it is even unclear whether the usual nonparametric bootstrap will work in this context when the nuisance parameter has a convergence rate @xmath20 . in contrast , @xcite and @xcite have shown that the profile sampler procedure without penalization can essentially yield second order frequentist valid inference for @xmath0 in semiparametric models , where the estimation accuracy is dependent on the convergence rate of the nuisance parameter . in other words , a faster convergence rate of the nuisance parameters can yield more precise frequentist inference for @xmath0 . these second order results are verified in @xcite and @xcite for several examples , including the cox model for both right censored and current status data , the proportional odds model , case - control studies with missing covariates , and the partly linear normal model . the convergence rates for these models range from the parametric to the cubic . the work in @xcite has shown clearly that the accuracy of the inference for @xmath0 based on the profile sampler method is intrinsically determined by the semiparametric model specifications through its entropy number . the purpose of this paper is to ask the somewhat natural question : does sampling from a profiled penalized log - likelihood ( which process we refer hereafter to as the penalized profile sampler ) yield first and even second order accurate frequentist inference ? the conclusion of this paper is that the answer is yes and , moreover , the accuracy of the inference depends in a fairly simple way on the size of the smoothing parameter . the unknown parameters in the semiparametric models we study in this paper includes @xmath0 , which we assume belongs to some compact set @xmath21 , and @xmath1 , which we assume to be a function in the sobolev class of functions supported on some compact set on the real line , whose @xmath22-th derivative exists and is absolutely continuous with @xmath23 , where @xmath24 here @xmath22 is a fixed , positive integer and @xmath25 is the @xmath26-th derivative of @xmath1 with respect to @xmath27 . obviously @xmath28 is some measurement of complexity of @xmath1 . we denote @xmath29 as the sobolev function class with degree @xmath22 . the penalized log - likelihood in this context is : @xmath30 where @xmath31 , @xmath32 is the log - likelihood of the single observation @xmath33 , and @xmath34 is a smoothing parameter , possibly dependent on data . in practice , @xmath34 can be obtained by cross - validation @xcite or by inspecting the various curves for different values of @xmath34 . the penalized maximum likelihood estimators @xmath18 and @xmath35 depend on the choice of the smoothing parameter @xmath34 . consequently we use the notation @xmath36 and @xmath37 for the remainder of this paper to denote the estimators obtained from maximizing ( [ penlik ] ) . in particular , a larger smoothing parameter usually leads to a less rough penalized estimator of @xmath11 . for the purpose of establishing first order accuracy of inference for @xmath0 based on the penalized profile sampler , we assume that the bounds for the smoothing parameter are in the form below : @xmath38 the condition ( [ smooth ] ) is assumed to hold throughout this paper . one way to ensure ( [ smooth ] ) in practice is simply to set @xmath39 . or we can just choose @xmath40 which is independent of @xmath22 . it turns out that the upper bound guarantees that @xmath36 is @xmath19-consistent , while the lower bound controls the penalized nuisance parameter estimator convergence rate . another approach to controlling estimators is to use sieve estimates with assumptions on the derivatives ( see @xcite ) . we will not pursue this further here . the log - profile penalized likelihood is defined as follows : @xmath41 where @xmath42 is @xmath43 for fixed @xmath0 and @xmath34 . the penalized profile sampler is just the procedure of sampling from the posterior distribution of @xmath44 by assigning a prior on @xmath0 . by analyzing the corresponding mcmc chain from the frequentist s point of view , our paper obtains the following conclusions : * _ distribution approximation _ : the posterior distribution with respect to @xmath44 can be approximated by the normal distribution with mean the maximum penalized likelihood estimator of @xmath0 and variance the inverse of the efficient information matrix , with error @xmath45 ; * _ moment approximation _ : the maximum penalized likelihood estimator of @xmath0 can be approximated by the mean of the mcmc chain with error @xmath46 . the efficient information matrix can be approximated by the inverse of the variance of the mcmc chain with error @xmath45 ; * _ confidence interval approximation _ : an exact frequentist confidence interval of wald s type for @xmath0 can be estimated by the credible set obtained from the mcmc chain with error @xmath46 . obviously , given any smoothing parameter satisfying the upper bound in ( [ smooth ] ) , the penalized profile sampler can yield first order frequentist valid inference for @xmath0 , similar as to what was shown for the profile sampler in @xcite . moreover , the above conclusions are actually second order frequentist valid results , whose approximation accuracy is directly controlled by the smoothing parameter . note that the corresponding results for the usual ( non - penalized ) profile sampler with nuisance parameter convergence rate @xmath15 in @xcite are obtained by replacing in the above @xmath45 with @xmath47 and @xmath46 with @xmath48 , for all respective occur where @xmath15 is as defined in ( [ convr ] ) . our results are the first higher order frequentist inference results for penalized semiparametric estimation . the layout of the article is as follows . the next section , section 2 , introduces the two main examples we will be using for illustration : partly linear regression for current status data and semiparametric logistic regression . some background is given in section 3 , including the concept of a least favorable submodel as well as some notations and the main model assumptions . in section [ rate ] , some preliminary results are developed , including three rather different theorems concerning the convergence rates of the penalized nuisance parameters and the order of the estimated penalty term under different conditions . the corresponding rates for the two featured examples are also calculated in this section . the main results and implications are discussed in section 5 , and all remaining model assumptions are verified for the examples in section 6 . a brief discussion of future work is given in section 7 . we postpone all technical tools and proofs to the last section , section 8 . in this example , we study the partly linear regression model with normal residue error . the continuous outcome @xmath49 , conditional on the covariates @xmath50 , is modeled as @xmath51 where @xmath52 is an unknown smooth function , and @xmath53 with finite variance @xmath54 . for simplicity , we assume for the rest of the paper that @xmath55 . the theory we propose also works when @xmath56 is unknown , but the added complexity would detract from the main issues . we also assume that only the current status of response @xmath49 is observed at a random censoring time @xmath57 . in other words , we observe @xmath58 , where indicator @xmath59 . current status data may occur due to study design or measurement limitations . examples of such data arise in several fields , including demography , epidemiology and econometrics . for simplicity of exposition , @xmath0 is assumed to be one dimensional . under the model ( [ eg2lik ] ) and given that the joint distribution for @xmath60 does not involve parameters @xmath61 , the log - likelihood for a single observation at @xmath62 is @xmath63 where @xmath64 is the standard normal distribution . the parameter of interest , @xmath0 , is assumed to belong to some compact set in @xmath65 . the nuisance parameter is the function @xmath52 , which belongs to the sobolev function class of degree @xmath22 . we further make the following assumptions on this model . we assume that @xmath66 is independent given @xmath67 . the covariates @xmath67 are assumed to belong to some compact set , and the support for random censoring time @xmath68 is an interval @xmath69 $ ] , where @xmath70 . in addition , @xmath71 is strictly positive and @xmath72 . the first order asymptotic behaviors of the penalized log - likelihood estimates of a slightly more general version of this model have been extensively studied in @xcite . let @xmath73 , @xmath74 , be independent copies of @xmath75 , where @xmath49 is a dichotomous variable with conditional expectation @xmath76 . @xmath77 is the logistic distribution defined as @xmath78 . obviously the likelihood for a single observation is of the following form : @xmath79 this example is a special case of quasi - likelihood in partly linear models when the conditional variance of response @xmath49 is taken to have some quadratic form of the conditional mean of @xmath49 . in the absence of any restrictions on the form of the function @xmath1 , the maximum likelihood of this simple model often leads to over - fitting . hence @xcite propose maximizing instead the penalized likelihood of the form @xmath80 ; and @xcite studied the asymptotic properties of the maximum penalized likelihood estimators for @xmath0 and @xmath1 . for simplicity , we will restrict ourselves to the case where @xmath81 and @xmath82 have bounded support , say @xmath83^{2}$ ] . to ensure the identifiability of the parameters , we assume that @xmath84 is positive and that the support of @xmath85 contains at least @xmath22 distinct points in @xmath83 $ ] . another interesting potential example we may apply the penalized profile sampler method to is the classic proportional hazards model with current status data by penalizing the cumulative hazard function with its sobolev norm . there are two motivations for us to penalize the cumulative hazard function in the cox model . one is that the estimated step functions from the unpenalized estimation can not be used easily for other estimation or inference purposes . another issue with the unpenalized approach is that without making stronger continuity assumptions , we can not achieve uniform consistency even on a compact set @xcite . the asymptotic properties of the corresponding penalized m - estimators have been studied in @xcite . in this section , we present some necessary preliminary material concerning least favorable submodels , general notational conventions for the paper , and an enumeration of the main assumptions . in this subsection , we briefly review the concept of a least favorable submodel . a submodel @xmath86 is defined to be least favorable at @xmath87 if @xmath88 , given @xmath89 , where @xmath90 is the efficient score function for @xmath0 . the efficient score function for @xmath0 can be viewed as the projection of the score function for @xmath0 onto the tangent space of @xmath1 . the inverse of its variance is exactly the efficient information matrix @xmath91 . we abbreviate hereafter @xmath92 and @xmath93 with @xmath94 and @xmath95 , respectively . the `` direction '' along which @xmath96 approaches @xmath1 in the least favorable submodel is called the least favorable direction . an insightful review about least favorable submodels and efficient score functions can be found in chapter 3 of @xcite . by the above construction of the least favorable submodel , @xmath97 can be rewritten in the following form : @xmath98 where @xmath99 , @xmath100 is a general map from the neighborhood of @xmath0 into the parameter set for @xmath1 , with @xmath101 . the concrete forms of ( [ penlfs ] ) will depend on the situation . we present in this subsection some notation that will be used throughout the paper . the derivatives of the function @xmath102 are with respect to its first argument , @xmath103 . for the derivatives relative to the other two arguments @xmath0 and @xmath1 , we use the following shortened notation : @xmath104 indicates the first derivative of @xmath102 with respect to @xmath0 . similarly , @xmath105 denotes the derivative of @xmath106 with respect to @xmath0 . also , @xmath107 and @xmath108 indicate the maps @xmath109 and @xmath110 , respectively . for brevity , we denote @xmath111 , @xmath112 and @xmath113 , where @xmath114 , @xmath115 are the true values of @xmath0 and @xmath1 . of course , we can write @xmath116 as @xmath117 . @xmath13 and @xmath118 indicate the euclidean norm and @xmath119 norm , respectively . the notations @xmath120 and @xmath121 mean greater than , or smaller than , up to a universal constant . the symbols @xmath122 and @xmath123 are used for the empirical distribution and the empirical processes of the observations , respectively . we now make the following three classes of assumptions : rate assumptions ( r1 ) for the penalized nuisance parameter and the estimated penalty term ; smoothness assumptions ( s1-s2 ) and empirical processes assumptions ( e1 ) for @xmath102 and its related derivatives . * : assume : @xmath124 * : the maps @xmath125 have integrable envelope functions in @xmath126 in some neighborhood of @xmath127 , for @xmath128 . * : assume : @xmath129 for all @xmath1 in some neighborhood of @xmath11 . * : for all random sequences @xmath130 and @xmath131 , we have @xmath132 assumption r1 implicitly assumes that we have a metric or topology defined on the set of possible values of the nuisance parameter @xmath1 . the form of @xmath133 may vary for different situations and does not need to be specified in this subsection beyond the given conditions . ( [ penrate ] ) implies that @xmath134 is consistent for @xmath11 as @xmath135 in probability . additionally , from ( [ jrate ] ) we know that the smoothing parameter @xmath34 plays a role in determining the complexity degree of the estimated nuisance parameter . ( [ jrate ] ) implies that @xmath136 if the @xmath36 is asymptotically normal , which has been shown in ( [ penmle ] ) . note that @xmath137 , where @xmath138 for a fixed @xmath0 , based on the inequality that @xmath139 . clearly , the assumptions s1 and s2 are separately the smoothness conditions for the euclidean parameters @xmath140 and the infinite dimensional nuisance parameter @xmath1 . the boundedness of the frchet derivatives of the maps @xmath141 and @xmath142 ensures the validity of conditions ( [ smcon2 ] ) and ( [ smcon3 ] ) . based on the discussions in section 2 of @xcite , under the given regularity conditions , it suffices to show ( [ nobias3 ] ) if the map @xmath143 is frchet differentiable and the map @xmath144 is second order frchet differentiable . condition ( [ smcon1 ] ) is concerned with the asymptotic equicontinuity of the empirical process measure of @xmath145 with @xmath1 ranging around the neighborhood of @xmath11 . it suffices to show ( [ eg2em1 ] ) and ( [ eg2em2 ] ) if @xmath146 and @xmath147 , provided @xmath148 and @xmath149 are square integrable . thus we will be able to use technical tools t2 and t6 given in the appendix to show ( [ smcon1])([eg2em2 ] ) . for the verification of ( [ eg2em3 ] ) , we need to make use of a glivenko - cantelli theorem for classes of functions that change with @xmath4 which is a modification of theorem 2.4.3 in @xcite and is explained in the appendix . in principle , assumptions s1 , s2 and e1 on the functions of the least favorable submodel directly imply the following empirical no - bias conditions : @xmath150 the derivations of ( [ emno1 ] ) and ( [ emno2 ] ) are simply based on the regular taylor expansions around the true values . the detailed arguments can be found in the proof of lemmas 1 and 2 in @xcite . the two empirical no - bias conditions ensure that the penalized profile likelihood behaves like a penalized likelihood in the parametric model asymptotically and therefore yields a second order asymptotic expansion of the penalized profile log - likelihood . in the previous section , we have imposed two assumptions about the convergence rates of the estimated nuisance parameter and the order of the estimated penalty term , i.e. ( [ penrate ] ) and ( [ jrate ] ) . to compute the convergence rates , we present three different theorems below which require different sets of conditions . these theorems can be viewed as extension of general results on m - estimators to penalized m - estimators , and are therefore of independent interest . we first state the classical definitions for the covering number ( entropy number ) and bracketing number ( bracketing entropy number ) for a class of functions . _ definition : _ let @xmath151 be a subset of a ( pseudo- ) metric space @xmath152 of real - valued functions . the @xmath153-covering number @xmath154 of @xmath151 is the smallest @xmath155 for which there exist functions @xmath156 in @xmath157 , such that for each @xmath158 , @xmath159 for some @xmath160 . the @xmath153-bracketing number @xmath161 is the smallest @xmath155 for which there exist pairs of functions @xmath162\}_{j=1}^{n}\subset\mathcal{l}$ ] , with @xmath163 , @xmath164 , such that for each @xmath158 there is a @xmath160 such that @xmath165 . the @xmath153-entropy number ( @xmath153-bracketing entropy number ) is defined as @xmath166 ( @xmath167 ) . before we present the first theorem , define @xmath168 for a known constant @xmath169 : [ ratethm1 ] assume conditions ( [ entrocon ] ) , ( [ eg1bou ] ) , ( [ egsm3 ] ) and ( [ egsm1 ] ) below hold for every @xmath170 and @xmath171 : @xmath172 then we have @xmath173 for @xmath174 satisfying @xmath175 . condition ( [ entrocon ] ) determines the order of the increments of the empirical processes indexed by @xmath176 . a detailed discussion about how to compute the increments of the empirical processes can be found in chapter 5 of @xcite . condition ( [ eg1bou ] ) is equivalent to the condition that @xmath177 is bounded away from zero uniformly in @xmath178 for @xmath87 ranging over @xmath179 . given that the distance function @xmath180 in ( [ egsm3 ] ) is just @xmath181 , ( [ egsm3 ] ) trivially holds provided that condition ( [ eg1bou ] ) holds . for the verification of ( [ egsm1 ] ) , we can do an analysis as follows . the natural taylor expansions of the criterion function @xmath182 around the maximum point @xmath183 implies that @xmath184 , and ( [ disine ] ) implies that @xmath185 given condition ( [ eg1bou ] ) . we now apply theorem [ ratethm1 ] to derive the related convergence rates in the partly linear model in corollary [ eg1rate ] . however , we need to strengthen our previous assumptions to require the existence of a known @xmath186 such that @xmath187 , where @xmath188 and that the density for the joint distribution @xmath189 is strictly positive and finite . the additional assumptions here guarantee condition ( [ eg1bou ] ) . the following theorem [ nuirate ] and theorem [ ratethm3 ] can also be employed to derive the convergence rate of the non - penalized estimated nuisance parameter by setting @xmath34 to zero . however , we would need to assume that @xmath190 for some known @xmath191 when applying these theorems . thus we can argue that the the penalized method enables a relaxation of the assumptions needed for the nuisance parameter . [ eg1rate ] under the above set - up for the partly linear normal model with current status data , we have , for @xmath192 , @xmath193 moreover , if we also assume that @xmath194 for some known @xmath191 , then @xmath195 provided condition ( [ smooth ] ) holds . corollary [ eg1rate ] implies that the convergence rate of the estimated nuisance parameter is slower than that of the the regular nuisance parameter by comparing ( [ eg1pratre ] ) and ( [ eg10ratre ] ) . this result is not surprising since the slower rate is the trade off for the smoother nuisance parameter estimator . however the advantage of the penalized profile sampler is that we can control the convergence rate by assigning the smoothing parameter with different rates . corollary [ eg1rate ] also indicates that @xmath196 and @xmath197 . note that the convergence rate of the maximum penalized likelihood estimator , @xmath198 , is deemed as the optimal rate in @xcite . similar remarks also hold for corollary [ eg2rate ] below . the boundedness condition ( [ eg1bou ] ) appears hard to achieve in some examples . hence we propose theorem [ nuirate ] below to relax this condition by choosing the criterion function @xmath199 $ ] . obviously , @xmath200 is trivially bounded away from zero . it is also bounded above for @xmath87 around the their true values if @xmath201 is bounded away from zero uniformly in @xmath178 and @xmath177 is bounded above . the first condition is satisfied if the map @xmath202 is continuous around @xmath203 and @xmath204 is uniformly bounded away from zero . the second condition is trivially satisfied in the semiparametric logistic regression model by the given form of the density . the boundedness of @xmath200 thus permits the application of lemma [ lerate ] below which is used to verify condition ( [ ratea2 ] ) in the following theorem : [ nuirate ] assume for any given @xmath170 , @xmath205 satisfies @xmath206 for given measurable functions @xmath207 . assume conditions ( [ ratea1 ] ) and ( [ ratea2 ] ) below hold for every @xmath170 , every @xmath171 and every @xmath208 : @xmath209 suppose that ( [ ratea2 ] ) is valid for functions @xmath210 such that @xmath211 is decreasing for some @xmath212 and sets @xmath179 such that @xmath213 . then @xmath214 for any sequence of positive numbers @xmath215 such that @xmath216 for every @xmath4 . lemma [ lerate ] below is presented to verify the modulus condition for the continuity of the empirical process in ( [ ratea2 ] ) . let @xmath217 and write @xmath218 [ lerate ] suppose the functions @xmath219 are uniformly bounded for @xmath87 ranging over a neighborhood of @xmath183 and that @xmath220 then condition ( [ ratea2 ] ) is satisfied for any functions @xmath210 such that @xmath221 consequently , in the conclusion of the above theorem we may use @xmath222 rather than @xmath223 . theorem [ eg2rate ] and lemma [ lerate ] are theorem 3.2 and lemma 3.3 in @xcite , respectively . we can apply theorem [ nuirate ] to the penalized semiparametric logistic regression model by including @xmath224 in @xmath0 , i.e. @xmath225 . this is accomplished in the following corollary . note that we assume that the uniform norm and sobolev norm of @xmath1 are bounded above with known upper bounds when deriving ( [ eg1ratre ] ) of the corollary , but this assumption is not needed for ( [ eg2pratre ] ) and ( [ eg2j ] ) . [ eg2rate ] under the above set - up for the semiparametric logistic regression model , we have for @xmath226 satisfying condition ( [ smooth ] ) and any @xmath227 that @xmath228 if we also assume that @xmath229 for some known @xmath191 , then @xmath230 corollary [ eg1rate ] and [ eg2rate ] imply that @xmath136 and @xmath231 , respectively . thus the maximum likelihood estimators of the nuisance parameters in the two examples of this paper are consistent in the uniform norm , i.e. @xmath232 and @xmath233 , since the sequences @xmath35 and @xmath234 consist of smooth functions defined on a compact set with asymptotically bounded first - order derivatives . the preceding two theorems imply that the convergence rate of the penalized estimated nuisance parameter is affected by the assigned smoothness parameter . however , the next theorem shows that , under different conditions , the above phenomena may not hold . let @xmath235 where @xmath236 for @xmath237 and @xmath238 is the appropriate score operator for the model . note that @xmath239 for sufficiently small @xmath103 . obviously @xmath240 and @xmath241 . we assume that the maps @xmath242 and @xmath243 are uniformly bounded such that @xmath244 and @xmath245 can be viewed as maps from the parameters set @xmath246 into @xmath247 . further we require the following regularity conditions : for some @xmath248 , @xmath249 [ ratethm3 ] suppose that @xmath250 is frchet differentiable at @xmath183 with derivative @xmath251 such that the map @xmath252 is invertible with an inverse that is continuous on its range . furthermore , we assume that ( [ con1th3 ] ) and ( [ con2th3 ] ) hold . then @xmath253 for @xmath254 and @xmath255 in probability . the preceding theorem is a variation of theorems used in @xcite and @xcite , among others , to prove the asymptotic normality of the maximum likelihood estimator @xmath256 . if we can show that @xmath257 by some other means , then ( [ thm3rate ] ) implies that @xmath258 . this indicates that the smoothing effect of the penalized method does not occur , which may be due to some very smooth non - penalized estimated nuisance parameter . the high degree of the smoothness of the non - penalized estimated nuisance parameter can be deduced from its fast convergence rate which equals the parametric rate in this instance . in this section we first present second order asymptotic expansion of the log - profile penalized likelihood which prepare us for deriving the main results about the higher order structure of the penalized profile sampler . the assumptions in section 3 and condition ( [ smooth ] ) are assumed throughout . [ penexp ] given @xmath259 , we have @xmath260 where @xmath261 , provided the efficient information @xmath95 is positive definite . the results in theorem [ penexp ] are useful in there own right for inference about @xmath0 . ( [ penmle ] ) is a second higher order frequentist result in penalized semiparametric estimation regarding the asymptotic linearity of the maximum penalized likelihood estimator of @xmath0 . we now state the main results on the penalized posterior profile distribution . a preliminary result , theorem [ pdist ] with corollary [ pmome ] below , shows that the penalized posterior profile distribution is asymptotically close enough to the distribution of a normal random variable with mean @xmath36 and variance @xmath262 with second order accuracy , which is controlled by the smoothing parameter . similar conclusions also hold for the penalized posterior moments . another main result , theorem [ pquant ] , shows that the penalized posterior profile log - likelihood can be used to achieve second order accurate frequentist inference for @xmath0 . let @xmath263 be the penalized posterior profile distribution of @xmath0 with respect to the prior @xmath264 . define @xmath265 [ pdist ] assume that @xmath266 for every random @xmath267 . if @xmath268 and @xmath269 has continuous and finite first order derivative in some neighborhood of @xmath203 , then we have , for any @xmath270 , @xmath271 where @xmath272 is the distribution of the @xmath273-dimensional standard normal random variable . [ pmome ] under the assumptions of theorem [ pdist ] , we have that if @xmath0 has finite second absolute moment , then @xmath274 where @xmath275 and @xmath276 are the penalized posterior profile mean and penalized posterior profile covariance matrix , respectively . we now present another second order asymptotic frequentist property of the penalized profile sampler in terms of quantiles . the @xmath277-th quantile of the penalized posterior profile distribution , @xmath278 , is defined as @xmath279 . without loss of generality , @xmath280 . we can also define @xmath281 , i.e. , @xmath282 . [ pquant ] under the assumptions of theorem [ pdist ] and assuming that @xmath283 has finite third moment with a nondegenerate distribution , then there exists a @xmath284 based on the data such that @xmath285 and @xmath286 for each choice of @xmath287 . theorem [ pquant ] ensures that there exists a unique @xmath277-th quantile for @xmath0 up to @xmath46 in the frequentist set - up for each fixed @xmath278 . note that @xmath278 is not unique if the dimension of @xmath0 is larger than one . theorem [ pdist ] , corollary [ pmome ] and theorem [ pquant ] above show that the penalized profile sampler generates second order asymptotic frequentist valid results in terms of distributions , moments and quantiles . moreover , the second order accuracy of this procedure is controlled by the smoothing parameter . another interpretation for the role of @xmath34 in the penalized profile sampler is that we can view @xmath34 as the prior on @xmath288 , or on @xmath1 to some extent . to see this , we can write @xmath289 in the following form : @xmath290\end{aligned}\ ] ] this idea can be traced back to @xcite . in other words , the prior on @xmath288 is a normal distribution with mean zero and variance @xmath291 . hence it is natural to expect @xmath34 has some effect on the convergence rate of @xmath1 . other possible priors on the functional parameter include dirichlet and gaussian processes which are more commonly used in nonparametric bayesian methodology . we now illustrate verification of the assumptions in section 3.3 with the two example that were introduced in section 2 . thus this section is a continuation of the earlier examples . we will concentrate on the estimation of the regression coefficient @xmath0 , considering the infinite dimensional parameter @xmath292 as a nuisance parameter . the score function of @xmath0 , @xmath293 , is given as follows : @xmath294 where @xmath295 @xmath296 , and @xmath297 is the density of a standard normal random variable . the least favorable direction at the true parameter value is : @xmath298 where @xmath299 is the expectation relative to the true parameters . the derivation of @xmath293 and @xmath300 is given in @xcite . thus , the least favorable submodel can be constructed as follows : @xmath301 where @xmath302 . by differentiating ( [ eg2lfs ] ) with respect to @xmath103 or @xmath0 , we can obtain the maps assessed in assumption s1 , @xmath303 . the concrete forms of these maps are given in @xcite which considers a more rigid model with a known upper bound on the @xmath304 norm of the @xmath22th derivative . the rate assumptions ( [ penrate ] ) and ( [ jrate ] ) have been verified previously in corollary [ eg1rate ] . the remaining assumptions are verified in the following two lemmas : [ eg2le0 ] under the above set - up for the partly linear normal model with current status data , assumptions s1 , s2 and e1 are satisfied . [ eg2le ] under the above set - up for the partly linear normal model with current status data , condition ( [ asyuni ] ) is satisfied . in the semiparametric logistic regression model , we can obtain the score function for @xmath0 and @xmath1 by similar analysis performed in the first example , i.e. @xmath305 and @xmath306 for @xmath307 . and the least favorable direction at the true parameter is given in @xcite : @xmath308}{p_{0}[\dot{f}(\theta_{0 } w+\eta_{0}(z))|z = z]},\end{aligned}\ ] ] where @xmath309 . the above assumptions plus the requirement that @xmath310 ensures the identifiability of the parameters . thus the least favorable submodel can be written as : @xmath311 where @xmath312 . by differentiating @xmath102 with respect to @xmath103 or @xmath0 , we obtain , @xmath313 where @xmath314 is the second derivative of the function @xmath315 . the rate assumptions have been shown in corollary [ eg2rate ] . the remaining assumptions are verified in the following two lemmas : [ eg1le0 ] under the above set - up for the semiparametric logistic regression model , assumptions s1 , s2 and e1 are satisfied . [ eg1le ] under the above set - up for the semiparametric logistic regression model , condition ( [ asyuni ] ) is satisfied . our paper evaluates the penalized profile sampler method from the frequentist view and discusses the effect of the smoothing parameter on estimation accuracy . one potential problem of interest is how to select a proper smoothing parameter in applications . a formal study about the higher order comparisons between the profile sampler procedure and fully bayesian procedure @xcite , which assign priors to both the finite dimensional parameter and the infinite dimensional nuisance parameter , is also interesting . we expect that the involvement of a suitable prior on the infinite dimensional parameter would at least not decrease the estimation accuracy of the parameter of interest . another worthwhile avenue of research is to develop analogs of the profile sampler and penalized profile sampler to likelihood estimation under model misspecification and to general m - estimation . some first order results for this setting in the case where the nuisance parameter may not be root-@xmath4 consistent have been developed for a weighted bootstrap procedure in @xcite . we first present some technical tools about the entropy calculations and increments of empirical processes which will be employed in the proofs that follow . let @xmath324 be a class of functions satisfying @xmath325 for every @xmath326 and @xmath103 and some fixed function @xmath327 . then , for any norm @xmath13 , @xmath328}(2\epsilon\|f\|,\mathcal{f},\|\cdot\|)\leq n(\epsilon , t , d).\end{aligned}\ ] ] t5 . let @xmath319 be a class of measurable functions @xmath330 on a product of a finite set and an arbitrary measurable space @xmath331 . let @xmath332 be a probability measure on @xmath333 and let @xmath334 be its marginal on @xmath335 . for every @xmath336 , let @xmath337 be the set of functions @xmath338 as @xmath52 ranges over @xmath319 . if every class @xmath337 is @xmath332-donsker with @xmath339 for every @xmath273 , then @xmath319 is @xmath332-donsker . let @xmath319 be a uniformly bounded class of measurable functions such that for some measurable @xmath340 , @xmath341 . moreover , assume that @xmath342 for some @xmath343 and @xmath344 and for all @xmath208 . then @xmath345}}\right]=o_{p}(n^{-1/2}).\end{aligned}\ ] ] t7 . for a probability measure @xmath332 , let @xmath346 be a class of measurable functions @xmath347 , and let @xmath348 denote a class of nondecreasing functions @xmath349 $ ] that are measurable for every probability measure . then , @xmath350 t8 . let @xmath319 and @xmath351 be classes of measurable functions . then for any probability measure @xmath352 and any @xmath353 , @xmath354 and , provided @xmath319 and @xmath351 are bounded by 1 , @xmath355 [ techdis ] the proof of t1 is found in @xcite . t1 implies that the sobolev class of functions with known bounded sobolev norm is @xmath332-donsker . t2 and t3 are separately lemma 3.4.2 and theorem 2.7.11 in @xcite . ( [ disine ] ) in t4 relates the kullback - leibler divergence and hellinger distance . its proof depends on the inequality that @xmath356 for every @xmath357 . t5 is lemma 9.2 in @xcite . t6 is a result presented on page 79 of @xcite and is a special case of lemma 5.13 on the same page , the proof of which can be found in pages 7980 . t7 and t8 are separately lemma 15.2 and 9.24 in @xcite . _ proof of theorem [ ratethm1 ] : _ the definition of @xmath134 implies that @xmath358 note that by t6 and assumption ( [ entrocon ] ) , we have @xmath359 by assumption ( [ egsm1 ] ) , we have @xmath360 combining with the above , we can deduce that @xmath361 where @xmath362 , @xmath363 and @xmath364 . the above inequality follows from assumption ( [ egsm3 ] ) . combining all of the above inequalities , we can deduce that @xmath365 where @xmath366 and @xmath367 . the equation ( [ ratee2 ] ) implies that @xmath368 . inserting @xmath368 into ( [ ratee1 ] ) , we can know that @xmath369 , which implies @xmath370 has the desired order . this completes the whole proof . @xmath371 _ proof of corollary [ eg1rate ] : _ conditions ( [ eg1bou])([egsm1 ] ) can be verified easily in this example based on the arguments in theorem [ ratethm1 ] because @xmath372 has finite second moment , and @xmath373 is bounded away from zero and infinity uniformly for @xmath61 ranging over the whole parameter space . note that @xmath374 by taylor expansion . then by the assumption that @xmath375 is positive definite , we know that @xmath376 implies @xmath377 . thus we only need to show that the @xmath378-bracketing entropy number of the function class @xmath379 defined below is of order @xmath380 to complete the proof of ( [ eg1pratre])([eg1j ] ) : @xmath381 for some constant @xmath382 . note that @xmath383 can be rewritten as : @xmath384 where @xmath385 and @xmath386 , where @xmath387 and where we know @xmath388 by t1 . we next calculate the @xmath378-bracketing entropy number with @xmath119 norm for the class of functions @xmath389 . by some analysis we know that @xmath390 is strictly decreasing in @xmath391 for @xmath392 , and @xmath393 because @xmath394 is bounded uniformly over @xmath392 . in addition , we know that @xmath395 because the function @xmath396 has bounded derivative for @xmath397 uniformly over @xmath392 . the above two inequalities imply that the @xmath378-bracketing number with uniform norm is of order @xmath398 for @xmath399 $ ] and is 1 for @xmath400 . thus we know @xmath401 . by applying a similar analysis to @xmath402 , we obtain that @xmath403 . combining this with t7 and t8 , we deduce that @xmath404 . this completes the proof of ( [ eg1pratre])([eg1j ] ) . for the proof of ( [ eg10ratre ] ) , we apply arguments similar to those used in the proof of theorem [ ratethm1 ] but after setting @xmath34 , @xmath405 and @xmath406 to zero in ( [ tech2 ] ) . then we obtain the following equality : @xmath407 . by treating @xmath408 and @xmath409 differently in the above equality , we obtain ( [ eg10ratre]).@xmath371 _ proof of corollary [ eg2rate ] : _ lemma 7.1 in @xcite establishes that @xmath410 after choosing @xmath411 in theorem [ nuirate ] . note that the map @xmath412 is uniformly bounded away from zero at @xmath413 and continuous around a neighborhood of @xmath203 . hence @xmath414 is well defined . moreover , @xmath415 by the inequality that @xmath416 . ( [ coro21 ] ) now directly implies ( [ eg2j ] ) . for the proof of ( [ eg2pratre ] ) , we need to consider the conclusion of lemma 7.4 ( i ) , which states that @xmath417 thus we have proved ( [ eg2pratre ] ) . for ( [ eg1ratre ] ) , we just replace the @xmath414 with @xmath418 in the proof of lemma 7.1 in @xcite . thus we can show that @xmath419 . by combining lemma [ lerate ] and ( [ coro22 ] ) , we know that @xmath420 , for @xmath215 satisfying @xmath421 . note that @xmath222 is as defined in ( [ thm2k ] ) . by similar analysis as used in the proof of lemma 7.1 in @xcite and the strengthened assumption on @xmath1 , we then find that @xmath422 , which leads to the desired convergence rate given in ( [ eg1ratre ] ) . @xmath371 _ proof of theorem [ ratethm3 ] . _ note that @xmath423 the last two equalities in the above follow from assumptions ( [ con1th3 ] ) and ( [ con2th3 ] ) . the frchet differentiability of @xmath424 at @xmath183 establishes that @xmath425 combining the above two sets of equations , we have , by the linearity of @xmath426 , established that @xmath427 now by the invertibility of @xmath428 , we can deduce that @xmath429 . @xmath371 _ proof of theorem [ penexp ] . _ we first show ( [ penmle ] ) , and then we need to state one lemma before proceeding to the proof of ( [ penexpf ] ) . for the proof of ( [ penmle ] ) , note that @xmath430 combining the third order taylor expansion of @xmath431 around @xmath203 , where @xmath432 and @xmath433 , with conditions ( [ emno1 ] ) and ( [ emno2 ] ) , the first term in the right - hand - side of the above displayed equality equals @xmath434 . by the inequality @xmath435 and assumption ( [ jrate ] ) , the second term in the right - hand - side of the above equality is equal to @xmath436 . combining everything , we obtain the following : @xmath437 the right - hand - side of ( [ mleint ] ) is of the order @xmath438 , where @xmath439 represents @xmath440 . however , its left - hand - side is trivially @xmath441 . considering the fact that @xmath442 , we can deduce that @xmath443 . inserting this into the previous display completes the proof of ( [ penmle ] ) . we next prove ( [ penexpf ] ) . note that @xmath443 . hence the order of the remainder terms in ( [ emno1 ] ) and ( [ emno2 ] ) becomes @xmath444 and @xmath445 , respectively . expression ( [ lnplexpps ] ) in lemma [ le4 ] below implies that @xmath446 the difference between ( [ difint ] ) and ( [ lnplexpps ] ) generates @xmath447 ( [ penexpf ] ) is now immediately obtained after considering ( [ penmle ] ) . @xmath371 _ proof of theorem [ pdist ] . _ suppose that @xmath448 is the penalized posterior profile distribution of @xmath449 with respect to the prior @xmath264 , where the vector @xmath450 is defined as @xmath451 . the parameter set for @xmath450 is @xmath452 . @xmath448 can be expressed as : @xmath453\cap\xi_{n}}\rho(\hat{\theta}_{\lambda_{n } } + \tilde{i}_{0}^{-\frac{1}{2}}\varrho_{n } ) \frac{pl_{\lambda_{n}}(\hat{\theta}_{\lambda_{n}}+\tilde{i}_{0}^{-\frac{1}{2}}\varrho_{n } ) } { pl_{\lambda_{n}}(\hat{\theta}_{\lambda_{n}})}d\varrho_{n } } { \int_{\varrho_{n}\in\xi_{n}}\rho(\hat{\theta}_{\lambda_{n}}+\tilde{i}_{0}^{-\frac{1}{2}}\varrho_{n } ) \frac{pl_{\lambda_{n}}(\hat{\theta}_{\lambda_{n}}+\tilde{i}_{0}^{-\frac{1}{2}}\varrho_{n } ) } { pl_{\lambda_{n}}(\hat{\theta}_{\lambda_{n}})}d\varrho_{n}}.\label{pdistab}\end{aligned}\ ] ] note that @xmath454 in the above is the short notation for @xmath455 . to prove theorem [ pdist ] , we first partition the parameter set @xmath452 as @xmath456 . by choosing the proper order of @xmath457 , we find the posterior mass in the first partition region is of arbitrarily small order , as verified in lemma [ pdist].1 immediately below , and the mass inside the second partition region can be approximated by a stochastic polynomial in powers of @xmath458 with error of order dependent on the smoothing parameter , as verified in lemma [ pdist].2 below . this basic technique applies to both the denominator and the numerator , yielding the quotient series , which gives the desired result . _ proof : _ fix @xmath463 . we then have @xmath464 where @xmath465 . then by lemma 3.2 in @xcite , @xmath466 for any fixed @xmath463 . this implies that there exists a positive decreasing sequence @xmath459 with @xmath460 such that ( [ postout ] ) holds . @xmath371 _ proof : _ the posterior mass over the region @xmath468 is bounded by @xmath469 by ( [ penexpf ] ) , we obtain @xmath470d\varrho_{n}.\end{aligned}\ ] ] obviously the order of @xmath471 depends on that of @xmath472 for @xmath34 satisfying ( [ smooth ] ) and @xmath473 . in order to analyze its order , we partition the set @xmath474 with the set @xmath475 , i.e. @xmath476 and @xmath477 . for the set @xmath478 , we have @xmath479 . for the set @xmath480 , we have @xmath481 . we can take @xmath482 for some @xmath317 such that @xmath460 and @xmath459 . then @xmath483 . combining with the above , we know that @xmath484 . by similar analysis , we can also show that @xmath485 has the same order . this completes the proof of lemma [ pdist].2 . @xmath371 we next start the formal proof of theorem [ pdist ] . by considering both lemma [ pdist].1 and lemma [ pdist].2 , we know the denominator of ( [ pdistab ] ) equals @xmath486d\varrho_{n}+o_{p}(\lambda_{n}^{2}).\end{aligned}\ ] ] the first term in the above display equals @xmath487 where @xmath488 . the above equality follows from the inequality that @xmath489 for any @xmath357 . consolidating the above analyses , we deduce that the denominator of ( [ pdistab ] ) equals @xmath490 . the same analysis also applies to the numerator , thus completing the whole proof . @xmath371 _ proof of corollary [ pmome ] : _ we only show ( [ pmon1 ] ) in what follows . ( [ pmon2 ] ) can be verified similarly . showing ( [ pmon1 ] ) is equivalent to establishing @xmath491 . note that @xmath492 can be written as : @xmath493 by analysis similar to that applied in the proof of theorem [ pdist ] , we know the denominator in the above display is @xmath494 and the numerator is a random vector of order @xmath495 . this yields the conclusion.@xmath371 _ proof of theorem [ pquant ] . _ note that ( [ ppostexp2 ] ) implies @xmath496 , for any @xmath497 , where @xmath498 . note also that the @xmath277-th quantile of a @xmath273 dimensional standard normal distribution , @xmath499 , is not unique if @xmath500 . the classical edgeworth expansion implies that @xmath501 , where @xmath502 , for @xmath497 . note that @xmath503 is uniquely determined for each fixed @xmath499 since @xmath504 has at least one absolutely continuous component . let @xmath505 . then @xmath506 . combining with ( [ penmle ] ) , we obtain @xmath507 . the uniqueness of @xmath284 up to order @xmath508 follows from that of @xmath503 for each chosen @xmath499.@xmath371 _ proof of lemma [ eg2le0 ] . _ assumptions s1 and s2 are verified in lemma 5 of @xcite . for the verifications of the assumption e1 , we first show the asymptotic equicontinuity condition ( [ smcon1 ] ) . without loss of generality , we assume that @xmath34 is bounded below by a multiple of @xmath509 and bounded above by @xmath510 in view of ( [ smooth ] ) . thus @xmath511 where ( [ eg1pratre ] ) implies the equality in the above expression . by ( [ eg1j ] ) , we know that @xmath512 and @xmath513 is bounded by some constant , since @xmath292 . we then define the set @xmath514 as follows : @xmath515 for some @xmath317 . obviously the function @xmath516 on a set of probability arbitrarily close to one , as @xmath517 . if we can show @xmath518 by t2 , then assumption ( [ smcon1 ] ) is verified . note that @xmath519 depends on @xmath52 in a lipschitz manner . consequently we can bound @xmath520 by the product of some constant and @xmath521 in view of t3 . @xmath522 is defined as @xmath523 where @xmath524 . by @xcite , we know that @xmath525 note that @xmath526 and @xmath527 in t2 . thus by calculation we know that @xmath528 . then by t2 we can show that @xmath518 . we next show ( [ eg2em3 ] ) . it suffices to verify that the sequence of classes of functions @xmath529 is @xmath332-glivenko - cantelli , where @xmath530 , for every random sequence @xmath531 and @xmath135 in probability . a glivenko - cantelli theorem for classes of functions that change with @xmath4 is needed . by revising theorem 2.4.3 in @xcite with minor notational changes , we obtain the following suitable extension of the uniform entropy glivenko - cantelli theorem : let @xmath532 be suitably measurable classes of functions with uniformly integrable functions and @xmath533 for any @xmath208 . then @xmath534 in probability for every @xmath208 . we then apply this revised theorem to the set @xmath532 of functions @xmath535 with @xmath103 and @xmath0 ranging over a neighborhood of @xmath203 and @xmath536 bounded by a constant . by the form of @xmath535 , the entropy number for @xmath529 is equal to that of @xmath537 by arguments similar to those used in lemma 7.2 of @xcite , we know that @xmath538 . moreover , the @xmath539 are uniformly bounded since @xmath292 . considering the fact that the probability that @xmath529 is contained in @xmath539 tends to 1 , we have completed the proof of ( [ eg2em3 ] ) . for the proof of ( [ eg2em1 ] ) , we only need to show that @xmath540 since @xmath541 is uniformly bounded in @xmath178 . note that we only need to show ( [ eg2em1 ] ) holds for @xmath542 based on the arguments in lemma [ pdist].2 . we next show that @xmath543 . by the rate assumptions r1 , we have @xmath544 we next define @xmath545 as follows : @xmath546 obviously the function @xmath547 on a set of probability arbitrarily close to one , as @xmath517 . if we can show @xmath548 by t2 , then the proof of ( [ eg2em1 ] ) is completed . accordingly , note that @xmath549 depends on @xmath61 in a lipschitz manner . consequently we can bound @xmath550 by the product of some constant and @xmath551 in view of t3 . @xmath552 is defined as @xmath553 where @xmath554 . by @xcite , we know that @xmath555 then by analysis similar to that used in the proof of ( [ smcon1 ] ) , we can show that @xmath548 in view of t2 . this completes the proof of ( [ eg2em1 ] ) . for the proof of ( [ eg2em2 ] ) , it suffices to show that @xmath556 for @xmath542 and for @xmath557 between @xmath558 and @xmath203 , in view of lemma [ pdist].2 . then we can show that @xmath559 by similar analysis as used in the proof of ( [ eg2em1]).@xmath371 _ proof of lemma [ eg2le ] . _ by the assumption that @xmath560 , we have @xmath561 . thus the following inequality holds : @xmath562 -n^{-1}\lambda_{n}^{2}[j^{2}(\hat{f}_{\tilde{\theta}_{n},\lambda_{n } } ) -j^{2}(\hat{f}_{\theta_{0},\lambda_{n}})]\geq o_{p}(1)\end{aligned}\ ] ] by considering assumption ( [ jrate ] ) , the above inequality simplifies to @xmath563\geq o_{p}(1),\end{aligned}\ ] ] where @xmath564 . by arguments similar to those used in lemma [ eg2le0].2 and by t5 , we know @xmath565 belongs to some @xmath332-donsker class . combining the above conclusion and the inequality @xmath566 for some @xmath567 and any @xmath357 , we can show that @xmath568\geq o_{p}(1).\label{leineq}\end{aligned}\ ] ] the remainder of the proof follows the proof of lemma 6 in @xcite.@xmath371 _ proof of lemma [ eg1le0 ] . _ the maps ( [ mapcon ] ) are uniformly bounded since @xmath315 , @xmath569 and @xmath314 are all uniformly bounded in @xmath570 . this completes the verifications of @xmath571 . note that @xmath82 are in @xmath83^{2}$ ] and @xmath572 is intrinsically bounded over @xmath83 $ ] . hence we can show that the frchet derivatives of @xmath141 and @xmath142 for any @xmath573 are bounded operators , from which we can deduce that @xmath574 is bounded by the product of some integrable function and @xmath575 . this ensures ( [ smcon2 ] ) and ( [ smcon3 ] ) . for ( [ nobias3 ] ) , @xmath576 can be written as @xmath577 since @xmath578 . note that @xmath579 . this implies that @xmath580 . however , by the common taylor expansion , we have @xmath581 . this proves ( [ nobias3 ] ) . we next verify assumption e1 . for the asymptotic equicontinuity condition ( [ smcon1 ] ) , we first apply analysis similar to that used in the proof of lemma [ eg2le0 ] to obtain @xmath582 by lemma 7.1 in @xcite , we know that @xmath583 and @xmath584 is bounded in probability by a multiple of @xmath585 . now we construct the set @xmath586 as follows : @xmath587 clearly , the probability that the function @xmath588 approaches 1 as @xmath517 . we next show that @xmath589 by t2 . note that @xmath145 depends on @xmath1 in a lipschitz manner . consequently , we can bound @xmath590 by the product of some constant and @xmath521 in view of t3 , where @xmath522 is as defined in the proof of lemma [ eg2le0 ] . by similar calculations as those performed in lemma [ eg2le0 ] , we can obtain @xmath591 . thus @xmath589 , and ( [ smcon1 ] ) follows . next we define @xmath592 . similar arguments as those used in the proof of lemma [ eg2le0 ] can be directly applied to the verification of ( [ eg2em3 ] ) in this second model . by the form of @xmath593 , the entropy number for @xmath594 is bounded above by that of @xmath595 . similarly , we know @xmath596 . moreover , the @xmath597 are uniformly bounded . this completes the proof for ( [ eg2em3 ] ) . the proof of ( [ eg2em1 ] ) and ( [ eg2em2 ] ) follows arguments quite similar to those used in the proof of lemma [ eg2le0 ] . in other words , we can show that @xmath598 and @xmath599 . this concludes the proof.@xmath371 _ proof . _ @xmath602 is bounded above and below by @xmath603 and @xmath604 respectively . by the third order taylor expansion of @xmath605 around @xmath203 , for @xmath606 and @xmath607 , and the above empirical no - bias conditions ( [ emno1 ] ) and ( [ emno2 ] ) , we can find that the order of the difference between @xmath608 and @xmath609 is @xmath610 . by the inequality @xmath611 , we know that @xmath612 provided assumptions ( [ smooth ] ) and ( [ jrate ] ) hold . similar analysis also applies to the lower bound . this proves ( [ lnplexpps]).@xmath371

the shard intersection order is a new lattice structure on a finite coxeter group which encodes the geometry of the reflection arrangement and the lattice theory of the weak order . in the case where is the symmetric group , we characterize shard intersections as certain pre - orders which we call permutation pre - orders . we use this combinatorial characterization to determine properties of the shard intersection order . in particular , we give an el - labeling .

1103.1910

_ shards _ were introduced in @xcite as a way to understand lattice congruences of the weak order on a finite coxeter group . they are defined in terms of the geometry of the associated simplicial hyperplane arrangement . the collection @xmath1 of arbitrary intersections of shards studied in @xcite , forms a lattice under reverse containment . this lattice is called the _ shard intersection order_. surprisingly , @xmath1 was found to be in bijection with the elements of the finite coxeter group @xmath0 , and thus the shard intersection order defines a new lattice structure on @xmath0 . this lattice is graded and contains the @xmath0-noncrossing partition lattice nc(@xmath0 ) as a sublattice . indeed , for any coxeter element , the subposet induced by @xmath2-sortable elements @xcite is a sublattice isomorphic to nc(@xmath0 ) . a formula for calculating the mbius numbers of lower intervals was given in @xcite , but overall the structure of the shard intersection order is not yet well - understood . + in this paper we consider the most classical coxeter group , the symmetric group , whose associated hyperplane arrangement is the braid arrangement . in section [ sectshard ] , we give necessary background information on hyperplane arrangements and the general construction of shards . we then specifically describe shards and shard intersections in the symmetric group . throughout the paper , no prior knowledge of coxeter groups will be assumed . following a suggestion from aguiar @xcite , in section [ sectpermpre ] we characterize shard intersections of type @xmath3 by realizing them combinatorially as certain pre - orders , which we call _ permutation pre - orders_. in section [ sectintord ] , we realize the shard intersection order as an order on the permutation pre - orders and use this realization to determine properties of the order , including an el - labeling . finally , in section [ sectnon ] we characterize _ noncrossing pre - orders _ , which correspond to @xmath2-sortable permutations . in this section we begin by defining a central hyperplane arrangement and the construction of shards within it . we then focus on the symmetric group , starting with necessary background and concluding with an explicit description of the correspondence between shard intersections and permutations . a _ linear hyperplane _ in a real vector space @xmath4 is a codimension-1 linear subspace of @xmath4 . a _ central hyperplane arrangement _ @xmath5 in @xmath4 is a finite collection of linear hyperplanes . the _ regions _ of @xmath5 are the closures of the connected components of @xmath6 . each region is a closed convex polyhedral cone whose dimension equals dim(@xmath4 ) . + fix a _ base region _ @xmath7 in the set of regions . we define a partial order on the set of regions called the _ poset of regions_. in the poset of regions , @xmath8 is below @xmath9 if and only if the set of hyperplanes separating @xmath8 from @xmath7 is contained in the set of hyperplanes separating @xmath9 from @xmath7 . the unique minimal element in the poset is @xmath7 and the unique maximal element is @xmath10 , the region antipodal to @xmath7 . a region @xmath9 covers @xmath8 if and only if @xmath9 and @xmath8 share a facet - defining hyperplane which separates @xmath9 from @xmath7 but does not separate @xmath8 from @xmath7 . ( more information on the poset of regions can be found in @xcite . ) + a region is _ simplicial _ if the normal vectors to its facet - defining hyperplanes form a linearly independent set . a central hyperplane arrangement is _ simplicial _ if each of its regions is simplicial . rank - two subarrangement _ @xmath11 of @xmath5 is a hyperplane arrangement consisting of all of the hyperplanes of @xmath5 which contain some subspace of codimension-@xmath12 , provided @xmath13 . in this subarrangement there exists a unique region @xmath14 containing @xmath7 , and the two facet - defining hyperplanes of @xmath14 are called the _ basic hyperplanes _ of @xmath15 . + _ shards _ are defined by imposing a cutting relation on the hyperplanes in @xmath5 . for each nonbasic hyperplane @xmath16 in a rank - two subarrangement @xmath11 , cut @xmath16 into connected components by removing the subspace @xmath17 from @xmath16 . a hyperplane @xmath16 is cut in every rank - two subarrangement in which it is non - basic , so it may be cut many times . cutting every hyperplane @xmath16 in this way , we obtain a set of connected components , the closures of which are the _ shards _ of @xmath5 . the set of _ intersections of shards _ of an arrangement @xmath5 is denoted @xmath18 or simply @xmath1 when @xmath5 and the choice of @xmath7 are clear . the empty intersection of shards is the entire space @xmath4 . + a shard @xmath19 is a _ lower shard _ of a region @xmath9 if it is contained in a facet - defining hyperplane of @xmath9 which separates @xmath9 from a region covered by @xmath9 . one of the primary results regarding shard intersections is a bijection ( * ? ? ? * proposition 4.7 ) between regions of a simplicial hyperplane arrangement and intersections of shards . the bijection sends a region @xmath9 to the intersection of the lower shards of @xmath9 . the shard intersections form a lattice under reverse containment , which induces , via the bijection , a partial order on the regions . called the _ shard intersection order _ , this partial order is different from the poset of regions and will be discussed further in section [ sectintord ] . + now that we have considered the construction of shards in a general setting , let us turn to the symmetric group @xmath20 . we begin with some background on @xmath20 and its coxeter arrangement . throughout this paper , permutations @xmath21 will be written in one - line notation as @xmath22 with @xmath23 . an _ inversion _ of @xmath24 is a pair @xmath25 such that @xmath26 and @xmath27 . a _ descent _ of @xmath24 is a pair @xmath28 such that @xmath29 . a _ descending run _ of @xmath24 is a maximal descending sequence @xmath30 . in this context , maximal implies that either @xmath31 or @xmath32 and either @xmath33 or @xmath34 . for example , the permutation @xmath35 has descending runs @xmath36 and @xmath37 . + the coxeter arrangement of the symmetric group consists of the hyperplanes @xmath38 for @xmath39 . the permutations in @xmath20 are in bijection with the regions of @xmath40 as follows : the permutation @xmath21 corresponds to the region @xmath41 . we will choose the base region to be the one corresponding to @xmath42 . + there are @xmath43 shards in each hyperplane @xmath44 . each shard is obtained by choosing @xmath45 for each @xmath46 with @xmath47 , and defining @xmath19 to be the cone @xmath48 . an intersection of shards can be represented similarly as a set of vector equalities and inequalities , by taking the union of the equalities and inequalities defining the shards being intersected . @xmath49 the coxeter arrangement @xmath50 consists of six hyperplanes through the origin in @xmath51 . these planes , intersected with the unit sphere in @xmath51 , define an arrangement of six great circles on the sphere . a stereographic projection yields an arrangement of six circles in the plane . this arrangement of circles is shown in figure [ fig : s4shards ] . the three largest circles are the hyperplanes @xmath52 on the top left , @xmath53 on the bottom , and @xmath54 on the top right . the two medium circles are the hyperplanes @xmath55 on the left and @xmath56 on the right . the smallest circle is the hyperplane @xmath57 . regions of @xmath50 appear as curve - sided triangles . the base region , corresponding to @xmath58 , is shaded in gray . the shards are closed 2-dimensional cones , thus they appear as full circles or as circular arcs in the figure . to clarify the picture , where shards intersect , certain shards are offset slightly from the intersection to indicate that they do not continue through the intersection . the four shards contained in the hyperplane @xmath57 are marked by arrows in the figure . the top shard in @xmath57 is defined by @xmath59 . the left shard in @xmath57 is defined by @xmath60 . the bottom shard in @xmath57 is defined by @xmath61 . the right shard in @xmath57 is defined by @xmath62 . given a permutation @xmath24 and the corresponding region @xmath9 , the descents of @xmath24 correspond to the hyperplanes containing the lower shards of @xmath9 . for example , @xmath63 and @xmath64 are descents in @xmath65 and the hyperplanes containing the shards below @xmath9 are @xmath55 and @xmath54 . the cone formed by the intersection of the lower shards of @xmath9 satisfies @xmath66 for each descent @xmath67 of @xmath24 . now we need to determine which shards from these hyperplanes are the lower shards of @xmath9 . the shard below @xmath9 contained in @xmath44 is the shard on the same side as @xmath9 of each hyperplane cutting @xmath44 . thus , for each @xmath46 with @xmath47 , the cone satisfies @xmath68 if and only if @xmath69 is an inversion of @xmath24 , and the cone satisfies @xmath70 if and only if @xmath69 is not an inversion of @xmath24 . continuing the example , the shard below @xmath9 contained in @xmath55 will satisfy @xmath71 , since @xmath72 is not an inversion in @xmath65 . the following proposition summarizes this explicit description for the shard intersection associated to a given permutation . [ permtoshardint ] for a permutation @xmath24 , the corresponding shard intersection is the cone consisting of points @xmath73 satisfying the following conditions for each descent @xmath67 of @xmath24 . 1 . @xmath74 2 . @xmath68 if and only if @xmath47 and @xmath69 is an inversion of @xmath24 3 . @xmath70 if and only if @xmath47 and @xmath69 is not an inversion of @xmath24 . we conclude this section with a lemma relating inversions of a permutation to intersections between descending runs . the notation @xmath75 $ ] stands for the set of integers @xmath76 with @xmath77 for @xmath78 . we will say that two descending runs @xmath79 and @xmath80 _ overlap _ if @xmath81\cap[e_n , e_1]\neq\emptyset$ ] . this lemma will be used in section [ sectpermpre ] . [ inversionint ] suppose @xmath82 is an inversion in @xmath24 such that @xmath83 and @xmath84 are in distinct descending runs @xmath85 and @xmath86 respectively . if @xmath87\cap[j_t , j_1]=\emptyset$ ] , then there exists a chain of descending runs @xmath88 between @xmath89 and @xmath90 such that @xmath87\cap d_1\neq\emptyset , d_n\cap[j_t , j_1]\neq\emptyset$ ] , and @xmath91 and @xmath92 overlap for @xmath93 . let @xmath94 as in the statement of the lemma . since @xmath87 \cap [ j_t , j_1 ] = \emptyset$ ] and @xmath95 , we know that @xmath96 for all @xmath97 $ ] and @xmath98 $ ] . we claim there must be a descending run @xmath99 between @xmath89 and @xmath90 such that @xmath100 and @xmath101 . we know @xmath102 and @xmath89 and @xmath90 are both part of different descending runs . directly to the right of @xmath89 is the descending run @xmath103 with @xmath104 since it is a descending run distinct from @xmath105 . suppose @xmath103 does not satisfy the claim . then @xmath106 for all @xmath107 $ ] . directly to the right of @xmath108 is the descending run @xmath109 with @xmath110 since it is a distinct descending run . this implies @xmath111 . suppose @xmath112 does not satisfy the claim . then @xmath113 for all @xmath114 $ ] . we can continue considering adjacent descending runs until we reach the descending run @xmath115 which is directly to the left of @xmath90 . again @xmath116 and if @xmath115 does not satisfy the claim , then @xmath117 for all @xmath118 $ ] . from this we obtain @xmath119 for all @xmath118 $ ] and in particular @xmath120 . this implies that @xmath115 and @xmath121 are not separate descending runs , which is a contradiction . therefore , at least one descending run @xmath122 between @xmath89 and @xmath90 must satisfy the claim . thus @xmath87\cap[d_p , d_1]\neq\emptyset$ ] . let @xmath123 be @xmath124 . also , since @xmath100 , we know that @xmath125 . then , by induction on the number of descending runs between @xmath89 and @xmath90 , there exists a chain of descending runs @xmath88 between @xmath89 and @xmath90 such that @xmath87\cap d_1\neq\emptyset , d_n\cap[j_t , j_1]\neq\emptyset$ ] , and @xmath91 and @xmath92 overlap for @xmath93 . this section begins with background and notation concerning pre - orders , leading to an injection from shard intersections to pre - orders . next , the bijection @xmath126 from permutations to pre - orders is defined and its image is characterized . we finish the section with the description of the inverse of @xmath126 . + a pre - order is a reflexive , transitive , binary relation . a pre - order @xmath127 on @xmath128 $ ] defines an equivalence relation on @xmath128 $ ] by setting @xmath129 if and only if @xmath130 . we will call the classes of this equivalence relation _ blocks_. we simultaneously think of @xmath127 as a pre - order on @xmath128 $ ] and as a partial order on its blocks . notationally , when considering a specific pre - order @xmath127 , a block will be labeled as @xmath131}$ ] where @xmath83 is the minimal element in the block and @xmath84 is the maximal element in the block . for example , a singleton block containing the element @xmath132 $ ] will be labeled as @xmath133}$ ] . @xmath134 is defined to be the block in @xmath127 containing the element @xmath135 $ ] . we will say blocks @xmath131}$ ] and @xmath136}$ ] _ overlap _ if the intervals @xmath137 $ ] and @xmath138 $ ] have a non - empty intersection . a block @xmath131}$ ] is covered by a block @xmath136}$ ] if @xmath139}}\prec{b_{[k , l]}}$ ] and there does not exist a block @xmath140}$ ] such that @xmath131}\prec b_{[s , t ] } \prec b_{[k , l]}$ ] . to denote that block @xmath141}}$ ] covers block @xmath139}}$ ] we will write @xmath139}}{\smash{\raisebox{0ex}{\setlength{\tabcolsep}{-1.1pt}\begin{tabular}{@{}cc@{}}\:$\prec$&$\cdot$\:\end{tabular}}}}{b_{[k , l]}}$ ] . an element @xmath83 is covered by an element @xmath84 in @xmath127 if @xmath142 and for any @xmath143 such that @xmath144 , either @xmath145 or @xmath146 . + we now describe an injection from the set @xmath1 of shard intersections in @xmath40 to the set @xmath147 of pre - orders on @xmath128 $ ] . the pre - order @xmath148 corresponding to a shard intersection @xmath149 is found by only considering the indices in the equalities and inequalities defining @xmath150 . thus for @xmath151 $ ] , @xmath152 in @xmath127 if and only if the inequality @xmath153 holds in @xmath150 . in particular , if @xmath66 in @xmath150 then @xmath154 in @xmath127 . @xmath155 the shard intersection @xmath156 in @xmath157 corresponds to the pre - order in figure [ shardint ] which has blocks @xmath158}=\{3\},\ b_{[1,4]}=\{1,4\},\ b_{[2,2]}=\{2\},\ b_{[5,5]}=\{5\}$ ] and @xmath159}=\{6,7\}$ ] . @xmath150 is the intersection of the shards @xmath59 and @xmath160 in @xmath161 . let @xmath162 be the map that takes a permutation @xmath21 to a pre - order @xmath163 as follows : each descending run @xmath164 in @xmath24 is a block @xmath165}=\{\pi_m,\pi_{m+1},\ldots,\pi_{m+t}\}$ ] in @xmath127 . for distinct blocks @xmath166}$ ] and @xmath167}$ ] which overlap , @xmath167 } \succeq b_{[\pi_{i+j},\pi_i]}$ ] in @xmath127 if and only if @xmath168 occurs to the right of @xmath169 in @xmath24 , i.e. @xmath170 . the transitive closure of these relations defines the pre - order @xmath127 . @xmath171 for @xmath172 the descending runs form the following blocks : @xmath173}=\{2\ } , b_{[1,6]}=\{1,3,6\ } , b_{[4,4]}=\{4\ } , b_{[5,7]}=\{5,7\ } , b_{[8,8]}=\{8\}$ ] . the pre - order @xmath174 is shown in figure [ exofmu ] . [ muiscomp ] @xmath126 is the composition of the bijection between permutations and shard intersections with the bijection between shard intersections and pre - orders . all that needs to be shown is that the relations between elements established by @xmath126 correspond to the relations given in proposition [ permtoshardint ] . suppose @xmath67 is a descent in @xmath24 . then @xmath83 and @xmath84 are in the same descending run in @xmath24 and in @xmath174 they will be in the same block , implying that @xmath129 . suppose @xmath46 is an element of @xmath24 with @xmath47 . in @xmath174 the block @xmath175 must be related to the block @xmath176 since they overlap . if @xmath69 is an inversion in @xmath24 , then we have @xmath177 implying @xmath178 . if @xmath69 is not an inversion , then we have @xmath179 implying @xmath180 . this corresponds to the relations given in proposition [ permtoshardint ] . permutation pre - order _ as a pre - order @xmath127 on @xmath128 $ ] with the following two conditions : 1 . if any two blocks in @xmath127 overlap , they must be comparable in @xmath127 2 . all covering relationships in @xmath127 must be between blocks that overlap . let @xmath181 . + as noted previously , a single shard @xmath182 can be represented as @xmath183 . thus @xmath19 corresponds to a pre - order @xmath127 with one block @xmath131}$ ] of size two and @xmath184 singleton blocks : @xmath185}$ ] for @xmath186\setminus\{i , j\}$ ] . each block @xmath187}$ ] for @xmath47 will be comparable to @xmath131}$ ] in @xmath127 and all other blocks will be incomparable to it . all covering relationships in @xmath127 will involve overlapping blocks . therefore @xmath188 . define @xmath189 to be the set of permutation pre - orders corresponding to single shards . [ permtopreorder ] @xmath126 is a bijection from permutations in @xmath20 to pre - orders in @xmath190 . by proposition [ muiscomp ] , @xmath126 is a bijection onto its image . all that remains to be shown is that @xmath190 is the image of @xmath126 . it is clear from the definition of @xmath126 that the image of @xmath126 is contained in @xmath190 . thus we will prove that @xmath190 is contained in the image of @xmath126 by showing that every @xmath191 is an intersection of shards . + let @xmath191 . it is sufficient to find a set of pre - orders in @xmath192 such that the transitive closure of the unions of these pre - orders is @xmath193 . for each @xmath194}$ ] in @xmath193 consider each @xmath195}$ ] with @xmath196 . define @xmath197 to be the pre - order on @xmath128 $ ] with one two element block @xmath198}$ ] and the remaining elements in singleton blocks @xmath185}$ ] for @xmath135\setminus \{i , s\}$ ] with the following order relations : + ( 1 ) for @xmath199 : \begin{cases } \text{if}\ i\preceq v\ \text{in}\ \omega , \text{then}\ i\preceq v\ \text{in}\ \omega_{is } \\ \text{if}\ i\succeq v\ \text{in}\ \omega , \text{then}\ i\succeq v\ \text{in}\ \omega_{is } \\ \text{if}\ i\equiv v\ \text{in}\ \omega , \text{then}\ i\preceq v\ \text{in}\ \omega_{is } \end{cases}$ ] + ( ( p1 ) rules out the possibility that @xmath83 and @xmath200 are incomparable . ) + ( 2 ) for @xmath201 $ ] : @xmath200 is incomparable in @xmath197 to @xmath202 for all @xmath203\setminus \{v\}$ ] . + clearly , @xmath204 . taking the transitive closure of unions of @xmath197 for each pair @xmath205 we obtain all the blocks of @xmath193 . ( p2 ) guarantees that all of the covering relationships of @xmath193 will be obtained as well . thus , taking the transitive closure of unions of the set of pre - orders @xmath197 ( for each pair @xmath205 described above ) gives @xmath193 . @xmath206 [ cols= " < , < , < " , ] figure [ fig : covers ] shows a permutation pre - order labeled @xmath193 and the eight permutation pre - orders labeled _ ( a ) _ through _ ( h ) _ which cover it . notice that if we combine blocks @xmath207}}$ ] and @xmath208}}$ ] in @xmath193 , blocks @xmath209}}$ ] and @xmath210}}$ ] overlap in the permutation pre - order which covers @xmath193 . by ( p2 ) this means that @xmath209}}$ ] and @xmath210}}$ ] must be related . since @xmath210}}$ ] is unrelated to blocks @xmath207}}$ ] and @xmath208}}$ ] in @xmath193 there is no restriction on the direction of this relation , so @xmath210}}$ ] can be either greater than or less than @xmath209}}$ ] . thus we obtain the two permutation pre - orders _ ( d ) _ and _ ( e)_. each of these pre - orders arise by intersecting @xmath193 with a permutation pre - order corresponding to a single shard in @xmath192 with @xmath211 and @xmath37 together in one block . + in @xmath193 any two blocks can be combined to form a permutation pre - order which covers @xmath193 . on the other hand , in the permutation pre - order labeled _ ( f ) _ , blocks @xmath207}}$ ] and @xmath210}}$ ] can not be combined to form a cover since this would force all numbers to be in one block and the rank would go up by two instead of one . the main result of this paper is an edge - lexicographic or el - labeling @xcite of @xmath212 . a sequence @xmath213 is _ lexicographically smaller _ than a sequence @xmath214 if there exists a @xmath215 $ ] such that @xmath216 for all @xmath26 but @xmath217 . define a labeling of the poset to be an _ el - labeling _ if the edges of the poset are labeled with positive integers such that the following two conditions hold . 1 . each interval @xmath218 in the poset has a unique maximal chain @xmath219 whose edge label sequence , @xmath220 is weakly increasing in value , and 2 . this weakly increasing edge label sequence is lexicographically smaller than the label sequences of all other maximal chains from @xmath193 to @xmath221 . given a permutation pre - order @xmath193 , number the descending runs in @xmath222 from @xmath223 to @xmath224 , proceeding from left to right . define this number to be the _ placement _ of the descending run @xmath225 in @xmath222 or the _ placement _ of its corresponding block @xmath226}}$ ] in @xmath193 . we will denote the placement of a block @xmath7 in @xmath193 as @xmath227 . the total order on the blocks of @xmath193 defined by the placements is equivalent to the total order defined by @xmath228 . also , if @xmath139}}\prec{b_{[k , l]}}$ ] in @xmath193 , then @xmath229}})<pl_{\omega}({b_{[k , l]}})$ ] since the descending run corresponding to @xmath139}}$ ] is to the left of the descending run corresponding to @xmath141}}$ ] in @xmath222 . + for @xmath212 we will describe a labeling @xmath230 . let @xmath231 the maximum of the placements of the two blocks from @xmath232 that were combined to form @xmath233 . we will prove the following theorem . [ ellabel ] @xmath230 is an el - labeling . @xmath234 the unique increasing maximal chain and the thirteen strictly decreasing maximal chains in the shard intersection order on @xmath235 with respect to the labeling @xmath230 are shown in figure [ exofellabel ] . the proof of theorem [ ellabel ] requires a proposition and several lemmas concerning relations in @xmath212 . given a permutation pre - order @xmath193 and the interval @xmath218 in @xmath212 , we will denote by @xmath236 the set of pairs of blocks in @xmath193 that could be combined to form a permutation pre - order in @xmath218 which covers @xmath193 . a pair of blocks is in @xmath236 if and only if the pair is in the same block in @xmath221 and is either incomparable or related by a cover in @xmath193 . a pair of blocks in @xmath193 will be called _ combinable _ if the pair is in @xmath236 . [ superprop ] let @xmath193 and @xmath221 be permutation pre - orders with @xmath237 . suppose @xmath238 and @xmath239 are the blocks in @xmath193 that are combined to form a block @xmath7 in @xmath221 , with @xmath240 . let @xmath241 be in @xmath193 . 1 . if @xmath242 , then @xmath241 is comparable to @xmath7 in @xmath221 . [ btwn2 ] if @xmath241 is incomparable to @xmath238 and @xmath239 in @xmath193 and comparable to @xmath7 in @xmath221 , then @xmath242 . [ labelorder ] if @xmath243 , then @xmath244 . if @xmath245 , then @xmath246 . if @xmath247 , then @xmath248 . [ comb ] let @xmath14 be a block in @xmath221 such that @xmath249 . if blocks @xmath7 and @xmath14 are combinable in @xmath221 , then @xmath14 is combinable with @xmath238 in @xmath193 . 5 . [ staysmall ] if @xmath250 , then @xmath14 , the block containing @xmath241 in @xmath221 , has @xmath251 . let @xmath252}}$ ] and @xmath253}}$ ] . then @xmath254}}$ ] for @xmath255 and @xmath256 . + _ proof of ( 1 ) _ : there are two cases to consider . suppose , in the first case , that @xmath139}}$ ] and @xmath141}}$ ] form a cover in @xmath193 . arguing as in case 2a of lemma [ linear ] , any block which is incomparable to both @xmath139}}$ ] and @xmath141}}$ ] must be to the left of both @xmath139}}$ ] and @xmath141}}$ ] or to the right of both @xmath139}}$ ] and @xmath141}}$ ] in @xmath222 and thus does not have a placement between @xmath257 and @xmath2 . therefore , all blocks @xmath241 with @xmath242 must be comparable to either @xmath139}}$ ] or @xmath141}}$ ] in @xmath193 . since all relations from @xmath193 must also be in @xmath221 , the blocks @xmath241 must be comparable to @xmath258}}$ ] in @xmath221 . + in the second case we suppose that @xmath139}}$ ] and @xmath141}}$ ] are incomparable in @xmath193 . again , any block @xmath241 with @xmath242 which is comparable to either @xmath139}}$ ] or @xmath141}}$ ] in @xmath193 must be comparable to @xmath258}}$ ] in @xmath221 . thus , all that is left to consider is a block @xmath259}}$ ] with @xmath260}})<c$ ] which is incomparable to both @xmath139}}$ ] and @xmath141}}$ ] in @xmath193 . since the three blocks @xmath139 } } , { b_{[k , l]}}$ ] and @xmath259}}$ ] are incomparable in @xmath193 , we know that @xmath137 , [ k , l]$ ] and @xmath261 $ ] are pairwise disjoint and the three blocks are ordered in @xmath222 based on the positions of @xmath137 , [ k , l]$ ] and @xmath261 $ ] in @xmath262 . given that the placements of the three blocks correspond to their ordering in @xmath222 and @xmath229}})<pl_{\omega}({b_{[p , q ] } } ) < pl_{\omega}({b_{[k , l]}})$ ] , we can conclude that @xmath263 . combining @xmath139}}$ ] and @xmath141}}$ ] forms the block @xmath264}}={b_{[s , t]}}$ ] , and we know that @xmath261\subset[i , l]=[s , t]$ ] . therefore by ( p1 ) , @xmath259}}$ ] must be related to @xmath258}}$ ] in @xmath221 . + _ proof of ( 2 ) _ : suppose for the purpose of contradiction , a block @xmath265}}$ ] with @xmath266}})<a$ ] is incomparable to both @xmath139}}$ ] and @xmath141}}$ ] in @xmath193 and is comparable to @xmath258}}$ ] in @xmath221 . since @xmath228 orders incomparable blocks based on their intervals , @xmath267 and @xmath268 . thus , @xmath269 and @xmath270\cap[s , t]=\emptyset$ ] . this implies that @xmath265}}$ ] and @xmath258}}$ ] do not cover one another in @xmath221 and therefore must be related by a chain of covers : @xmath265}}{\smash{\raisebox{0ex}{\setlength{\tabcolsep}{-1.1pt}\begin{tabular}{@{}cc@{}}\:$\prec$&$\cdot$\:\end{tabular}}}}{b_{[x_1,x_2]}}{\smash{\raisebox{0ex}{\setlength{\tabcolsep}{-1.1pt}\begin{tabular}{@{}cc@{}}\:$\prec$&$\cdot$\:\end{tabular}}}}{b_{[x_3,x_4 ] } } { \smash{\raisebox{0ex}{\setlength{\tabcolsep}{-1.1pt}\begin{tabular}{@{}cc@{}}\:$\prec$&$\cdot$\:\end{tabular}}}}\cdots { \smash{\raisebox{0ex}{\setlength{\tabcolsep}{-1.1pt}\begin{tabular}{@{}cc@{}}\:$\prec$&$\cdot$\:\end{tabular}}}}{b_{[x_{d-1},x_d ] } } { \smash{\raisebox{0ex}{\setlength{\tabcolsep}{-1.1pt}\begin{tabular}{@{}cc@{}}\:$\prec$&$\cdot$\:\end{tabular}}}}{b_{[s , t]}}$ ] . by ( p2 ) , we have @xmath270\cap[x_1,x_2]\neq \emptyset , [ x_1,x_2]\cap[x_3,x_4]\neq\emptyset , \ldots , [ x_{d-1},x_d]\cap[s , t]\neq\emptyset$ ] . thus , some interval @xmath271 $ ] for @xmath272 must overlap either @xmath137 $ ] or @xmath138 $ ] . the block @xmath273}}$ ] corresponding to this interval must be less than either @xmath139}}$ ] or @xmath141}}$ ] in @xmath193 ( whichever it overlaps ) because if it were greater than either of them it would be greater than @xmath258}}$ ] in @xmath221 and thus would not be part of the chain of covers . since all blocks except for @xmath139}}$ ] and @xmath141}}$ ] are the same in both @xmath193 and @xmath221 , we have that @xmath265}}$ ] is related to either @xmath139}}$ ] or @xmath141}}$ ] by a chain of covers in @xmath193 . this contradicts our assumption that @xmath265}}$ ] is incomparable to both @xmath139}}$ ] and @xmath141}}$ ] in @xmath193 and thus @xmath266}})\nless a$ ] . similarly , if @xmath266}})>c$ ] we see that @xmath274 and there must exist a chain of covers between @xmath258}}$ ] and @xmath265}}$ ] such that one block in the chain has an interval that overlaps either @xmath137 $ ] or @xmath138 $ ] . reaching the same contradiction we conclude that @xmath266}})\ngtr c$ ] and thus @xmath275}})<c$ ] . + _ proof of ( 3 ) _ : let @xmath276}}$ ] be a block in @xmath193 with @xmath277}})=x < a$ ] . we first show that @xmath278}})<pl_{\omega'}({b_{[s , t]}})$ ] . suppose that @xmath276}}$ ] is incomparable to both @xmath139}}$ ] and @xmath141}}$ ] . since @xmath279 and @xmath280 , @xmath281 and the descending run corresponding to @xmath258}}$ ] must be to the right of the descending run corresponding to @xmath276}}$ ] in @xmath282 . now suppose that @xmath276}}$ ] is comparable to @xmath139}}$ ] or @xmath141}}$ ] in @xmath193 . then either @xmath276}}\prec{b_{[i , j]}}$ ] or @xmath276}}\prec{b_{[k , l]}}$ ] , since @xmath279 and @xmath280 . all relations from @xmath193 must also hold in @xmath221 so @xmath276}}\prec{b_{[s , t]}}$ ] in @xmath221 . thus the descending run corresponding to the new block @xmath258}}$ ] will always be to the right of the descending run corresponding to @xmath276}}$ ] in @xmath282 . + we next show that @xmath277}})=pl_{\omega'}({b_{[u , v]}})$ ] . let @xmath265}}$ ] be any block in @xmath193 other than @xmath139 } } , { b_{[k , l]}}$ ] and @xmath276}}$ ] . if @xmath276}}$ ] is incomparable to @xmath265}}$ ] in both @xmath193 and @xmath221 then the descending runs corresponding to @xmath276}}$ ] and @xmath265}}$ ] must be ordered the same in both @xmath222 and @xmath282 , since @xmath228 orders them based on the positions of @xmath283 $ ] and @xmath270 $ ] in @xmath262 . if @xmath276}}$ ] is comparable to @xmath265}}$ ] in @xmath193 , then again the descending runs corresponding to each of them must be ordered the same in both @xmath222 and @xmath282 , since relations from @xmath193 are preserved in @xmath221 . if @xmath276}}$ ] is incomparable to @xmath265}}$ ] in @xmath193 and comparable to it in @xmath221 , then we claim that @xmath284 and @xmath276}}\prec{b_{[f , g]}}$ ] in @xmath221 . two blocks which are comparable in @xmath221 and incomparable in @xmath193 can only occur when both blocks are comparable through the new block @xmath258}}$ ] in @xmath221 . since we already showed that @xmath276}}$ ] must be less than @xmath258}}$ ] in @xmath221 if they are comparable , we must have @xmath276}}\prec{b_{[s , t]}}\prec{b_{[f , g]}}$ ] in @xmath221 . in order for the relation @xmath258}}\prec{b_{[f , g]}}$ ] to exist in @xmath221 , the descending run corresponding to @xmath265}}$ ] must be to the right of the descending run corresponding to @xmath139}}$ ] in @xmath222 . since @xmath276}}$ ] and @xmath265}}$ ] are incomparable in @xmath193 , this implies that @xmath284 . thus the claim is true . therefore , all descending runs to the right of the descending run corresponding to @xmath276}}$ ] in @xmath222 will remain to the right in @xmath282 and all descending runs to the left in @xmath222 will remain to the left in @xmath282 . thus @xmath276}}$ ] will retain the placement @xmath202 in @xmath221 , i.e. @xmath277}})=pl_{\omega'}({b_{[u , v]}})$ ] . + now let @xmath276}}$ ] be a block in @xmath193 with @xmath277}})>c$ ] . by similar arguments , all descending runs to the right of the descending run corresponding to @xmath276}}$ ] in @xmath222 will remain to the right in @xmath282 and all descending runs to the left in @xmath222 will remain to the left in @xmath282 . since the descending runs corresponding to @xmath139}}$ ] and @xmath141}}$ ] have been combined to form a single descending run , there is one less descending run to the left of the descending run corresponding to @xmath276}}$ ] . thus @xmath278}})=pl_{\omega}({b_{[u , v]}})-1 $ ] . we can now also conclude that if @xmath245 , then @xmath246 . + _ proof of ( 4 ) _ : suppose , for the purpose of contradiction , @xmath238 and @xmath14 are not combinable in @xmath193 . then @xmath238 and @xmath14 are comparable by a chain of covers in @xmath193 , i.e. either @xmath285 or @xmath286 for @xmath287 . since @xmath238 and @xmath239 are combinable in @xmath193 they must either be incomparable or related by a cover in @xmath193 . if @xmath238 and @xmath239 are incomparable in @xmath193 then we would have either @xmath288 or @xmath289 in @xmath221 . thus @xmath7 and @xmath14 would not be combinable in @xmath221 , which is a contradiction . therefore @xmath238 and @xmath239 are related by a cover in @xmath193 . if @xmath238 and @xmath239 are related by a cover then @xmath290 and @xmath291 in the chain of covers since @xmath7 and @xmath14 are combinable in @xmath221 . thus we have either @xmath292 or @xmath293 in @xmath193 . @xmath293 is not possible because @xmath294 . @xmath292 implies by part [ labelorder ] of this proposition that @xmath295 , which is a contradiction . therefore @xmath238 and @xmath14 are combinable in @xmath221 . + _ proof of ( 5 ) _ : there are two cases to consider . in the first case suppose that @xmath241 is @xmath238 ( it can not be @xmath239 due to its placement ) . then @xmath296 . by part [ labelorder ] of this proposition , @xmath297 . in the second case suppose that @xmath241 is neither @xmath238 nor @xmath239 . then @xmath298 . if @xmath299 , then by part [ labelorder ] of this proposition , @xmath300 . if @xmath301 , then again by part [ labelorder ] of this proposition , @xmath302 . let @xmath7 and @xmath14 be blocks in @xmath303 with @xmath304 and @xmath305 . then @xmath7 and @xmath14 are combinable in @xmath193 . if @xmath7 and @xmath14 are incomparable in @xmath193 then they are combinable and we are done . if @xmath7 and @xmath14 are comparable in @xmath193 suppose that one block does not cover the other . then there exists a block @xmath76 such that @xmath306 in @xmath193 . this implies that @xmath307 , which is a contradiction . thus @xmath7 and @xmath14 are related by a cover and are therefore combinable . [ coverlemma ] suppose @xmath193 and @xmath221 are permutation pre - orders with @xmath308 . let @xmath238 and @xmath239 be combinable blocks in @xmath193 such that they are contained in the same block in @xmath221 . then there exists a permutation pre - order @xmath309 with @xmath310 such that @xmath238 and @xmath239 are in the same block in @xmath309 . label the elements of @xmath238 as @xmath311 with @xmath312 for @xmath313 and the elements of @xmath239 as @xmath314 with @xmath315 for @xmath316 . without loss of generality suppose that @xmath317 . since blocks @xmath238 and @xmath239 are in the same block in @xmath221 we know that @xmath318 in @xmath221 for @xmath319 , j\in[n]$ ] . define @xmath320 to be the permutation pre - order on @xmath128 $ ] with one two element block @xmath321}}$ ] and the remaining elements in singleton blocks @xmath322}}$ ] for @xmath186\setminus\{a_1,b_1\}$ ] with the following order relations : + ( 1 ) for @xmath323 : \begin{cases } \text{if}\ a_1\preceq v\ \text{in}\ \omega ' , \text{then}\ a_1\preceq v\ \text{in}\ \omega^ * \\ \text{if}\ a_1\succeq v\ \text{in}\ \omega ' , \text{then}\ a_1\succeq v\ \text{in}\ \omega^ * \\ \text{if}\ a_1\equiv v\ \text{in}\ \omega ' , \text{then}\ a_1\preceq v\ \text{in}\ \omega^ * \end{cases}$ ] + ( ( p1 ) rules out the possibility that @xmath324 and @xmath200 are incomparable . ) + ( 2 ) for @xmath325 $ ] : @xmath200 is incomparable in @xmath320 to @xmath202 for all @xmath203\setminus \{v\}$ ] . + clearly , @xmath326 . taking the transitive closure of the union of @xmath193 with @xmath320 gives the permutation pre - order @xmath309 . [ min ] given @xmath193 and an interval @xmath327 , there exists a unique pair of blocks @xmath238 and @xmath239 whose larger placement is minimal among all pairs in @xmath236 . furthermore , there exists a unique cover of @xmath193 in which @xmath238 and @xmath239 are combined . suppose for the purpose of contradiction there are two pairs of blocks with minimal larger placement in @xmath193 . thus we have placements @xmath328 and @xmath2 such that @xmath301 and the blocks with placements @xmath257 and @xmath2 and the blocks with placements @xmath329 and @xmath2 are combinable in @xmath193 . for convenience we will denote the blocks by their respective placements . if blocks @xmath257 and @xmath329 are incomparable in @xmath193 or block @xmath329 covers block @xmath257 in @xmath193 , then they are combinable in @xmath193 . thus the pair @xmath330 is in @xmath236 , contradicting the minimality of @xmath2 . block @xmath257 can not be greater than block @xmath329 in @xmath193 due to their placements , so the only case left to consider is when block @xmath329 is greater than block @xmath257 but does not cover it . in this case there exists a chain of covers from @xmath257 to @xmath329 such that all of the blocks in the chain have placements less than @xmath329 , i.e. @xmath331 where @xmath332 for all @xmath333 $ ] . therefore @xmath334 is in @xmath236 , contradicting the minimality of @xmath2 . + we have established that there exists a unique pair of blocks @xmath238 and @xmath239 whose larger placement @xmath2 is minimal among all pairs in @xmath236 . we know by lemma [ coverlemma ] that there exists at least one cover of @xmath193 in which @xmath238 and @xmath239 are combined . suppose , again for the purpose of contradiction , there is more than one cover that can be obtained by combining @xmath238 and @xmath239 in @xmath193 . the only way for this to occur is if a block is incomparable to both @xmath238 and @xmath239 in @xmath193 and comparable to the combined block in the covers . thus we have the following situation . @xmath238 and @xmath239 are combined to form block @xmath7 in @xmath335 and @xmath336 for @xmath337 and @xmath338 . block @xmath14 is incomparable to both @xmath238 and @xmath239 in @xmath193 . in @xmath335 , @xmath339 and in @xmath336 , @xmath340 implying that @xmath14 and @xmath7 are in the same block in @xmath221 . this means that @xmath14 is also in the same block as @xmath238 and @xmath239 in @xmath221 . by proposition [ superprop ] part [ btwn2 ] we know that @xmath341 such that @xmath301 . since @xmath14 and @xmath238 are incomparable in @xmath193 they are a pair in @xmath236 , which contradicts the minimality of @xmath2 . therefore , there exists a unique cover of @xmath193 in which @xmath238 and @xmath239 are combined . we now have the necessary tools to prove theorem [ ellabel ] . we need to establish that @xmath230 satisfies conditions ( el1 ) and ( el2 ) . let @xmath218 be an interval in @xmath212 . denote by @xmath342 the maximal chain @xmath219 where each @xmath233 is obtained by combining a pair of blocks from @xmath343 such that the larger of the two placements is minimal among all pairs . by lemma [ min ] , this maximal chain is unique and thus does not share its label sequence with any other chain . + now we will show that the edge label sequence for @xmath342 is weakly increasing . suppose , for the purpose of contradiction , there is a descent in the edge label sequence , i.e. @xmath344 and @xmath345 for @xmath346 . let @xmath7 and @xmath14 with @xmath347 be the blocks in @xmath348 that were combined to form @xmath349 . we claim that the pair of blocks combined in @xmath350 did not have minimal larger placement . + _ proof of claim : _ there are three cases to consider . in the first case we suppose that @xmath7 and @xmath14 are the same blocks in @xmath350 and @xmath348 . since blocks @xmath7 and @xmath14 are not the blocks that were combined in @xmath350 to form @xmath348 and they both have placements less than @xmath2 in @xmath348 , proposition [ superprop ] part [ labelorder ] tells us that @xmath351 and @xmath352 . the only way in which @xmath7 and @xmath14 are not combinable in @xmath350 but are combinable in @xmath348 is if @xmath353 in @xmath350 and either @xmath7 and @xmath241 or @xmath241 and @xmath14 are combined to form @xmath348 . since we are assuming that @xmath7 and @xmath14 are the same blocks in @xmath350 and @xmath348 , this can not occur . this implies that @xmath7 and @xmath14 are combinable in @xmath350 . thus @xmath7 and @xmath14 are a combinable pair of blocks in @xmath350 with larger placement less than @xmath2 . in the second case , suppose that @xmath7 is the new block in @xmath348 formed by combining blocks @xmath238 and @xmath239 in @xmath350 where @xmath354 . then by proposition [ superprop ] part [ comb ] , @xmath14 and @xmath238 are a combinable pair in @xmath350 with larger placement less than @xmath2 . in the third case , suppose that @xmath14 is the new block in @xmath348 formed by combining blocks @xmath355 and @xmath356 where @xmath357 . then again by proposition [ superprop ] part [ comb ] , @xmath7 and @xmath355 are a combinable pair in @xmath350 with larger placement less than @xmath2 . in each case , we obtain a pair of blocks in @xmath350 such that the larger placement of the pair is less than @xmath2 . therefore , the claim is true and we have reached a contradiction . thus , there can not be a descent in the edge label sequence for @xmath342 , meaning @xmath342 is weakly increasing . + all that remains to be shown for ( el1 ) is that @xmath342 is the only weakly increasing maximal chain . as established above , @xmath342 does not share its label sequence with any other chain . suppose we follow some other maximal chain @xmath358 from @xmath193 to @xmath221 . at some step these chains must differ , meaning that in @xmath342 we combine two blocks from @xmath350 to form @xmath348 such that the larger placement of the pair of blocks in @xmath350 is minimal among pairs in @xmath359 , whereas in @xmath358 we combine two blocks from @xmath350 to form @xmath360 such that the larger placement of the pair of blocks in @xmath350 is not minimal among pairs in @xmath359 . thus we have the following situation . let @xmath238 and @xmath239 be blocks of @xmath350 with @xmath361 that were combined to form @xmath348 . let @xmath362 and @xmath363 be blocks of @xmath350 with @xmath364 , for @xmath365 , that were combined to form @xmath360 . since @xmath238 and @xmath239 are a pair in @xmath359 , they must be in the same block in @xmath221 . thus , at some later step in @xmath358 , a block containing @xmath238 must be combined with a block containing @xmath239 . by proposition [ superprop ] part [ staysmall ] , if we continue to pick pairs of blocks that cause the label sequence of @xmath358 to increase , the placements of the blocks containing @xmath238 and @xmath239 in @xmath366 for @xmath367 will always be smaller than the edge label leading to @xmath366 assigned by @xmath230 ( because in @xmath360 @xmath238 and @xmath239 have edge labels less than @xmath368 ) . eventually we are forced to combine the blocks containing @xmath238 and @xmath239 ( or a pair with smaller larger placement ) creating a descent in the edge label sequence for @xmath358 . thus , all other maximal chains have at least one descent . therefore , @xmath342 satisfies ( el1 ) . + since we chose the smallest possible label at each step in @xmath342 , its label sequence is lexicographically smaller than the label sequences of all other maximal chains . therefore , @xmath342 satisfies ( el2 ) . since @xmath212 is a graded poset , theorem [ ellabel ] implies that it is el - shellable and hence shellable ( * ? ? ? * theorem 2.3 ) . thus the mbius number of an interval in @xmath212 , @xmath369 , is equal to @xmath370 times the number of strictly decreasing maximal chains from @xmath193 to @xmath221 . to count the strictly decreasing maximal chains from @xmath193 to @xmath221 , we need to determine the number of ways in which the combinable block with the highest placement can be combined with blocks with lower placements at each step in the chain . at each step in a maximal chain there may be multiple ways to go up by a cover and obtain the maximum possible label on that edge . for this reason , counting strictly decreasing maximal chains is not completely straightforward , and at this time we do not know how to count strictly decreasing maximal chains for general intervals . interestingly , the mbius number of the entire lattice has a simple description @xcite . it is the number of indecomposable permutations , or equivalently , the number of permutations with no global descents . see sequence a003319 of @xcite for details and references . in this final section , we consider a subset of @xmath20 , the @xmath2-sortable permutations , and describe the corresponding subset of the permutation pre - orders . + a coxeter group @xmath0 is generated by a set @xmath371 of _ simple generators_. for @xmath372 the simple generators are @xmath373 where @xmath374 for @xmath375 . a _ coxeter element _ @xmath2 is the product of the simple generators in any order . for a chosen coxeter element @xmath376 , we can define a _ barring _ on the numbers @xmath377 as follows : 1 . if @xmath378 is before @xmath379 in @xmath2 then @xmath83 is lower - barred , and we denote this by @xmath380 2 . if @xmath378 is after @xmath379 in @xmath2 then @xmath83 is upper - barred , and we denote this by @xmath381 . a coxeter element in @xmath20 can be written as an @xmath382-cycle using the barring by placing @xmath223 at the 12 oclock position on a circle followed clockwise by the lower - barred numbers in ascending numerical order , the number @xmath382 , and then the upper - barred numbers in descending numerical order . @xmath383 [ barringex ] for @xmath384 the lower - barred numbers are @xmath385 and the upper - barred numbers are @xmath386 . this gives us the cycle shown in figure [ fig : cycle ] . for a permutation @xmath21 and @xmath387 we will say that @xmath24 _ contains the pattern @xmath202 _ if there are integers @xmath388 such that for all @xmath389 we have @xmath390 if and only if @xmath391 . otherwise we will say that @xmath24 _ avoids the pattern @xmath202_. in this paper we will be concerned with patterns involving barred numbers , specifically the patterns @xmath392 and @xmath393 . for @xmath21 with chosen coxeter element @xmath2 , we say that @xmath24 _ contains the pattern _ @xmath392 if it contains an instance of the pattern @xmath394 in which the number representing the @xmath12 in @xmath24 is upper - barred . similarly , we say @xmath24 _ contains the pattern _ @xmath393 if it contains an instance of the pattern @xmath395 in which the number representing the @xmath12 in @xmath24 is lower - barred . if these conditions do not hold , we again say that @xmath24 _ avoids the given pattern_. as a continuation of example [ barringex ] , suppose @xmath384 . then @xmath396 contains four instances of the pattern @xmath393 : @xmath397 , @xmath398 , @xmath399 , and @xmath400 , and avoids the pattern @xmath392 . there is a general definition of @xmath2-sortable elements of a coxeter group which can be found in @xcite . for this paper , the following characterization of @xmath2-sortable elements in @xmath20 is sufficient . given a coxeter element @xmath2 , a permutation @xmath21 is _ @xmath2-sortable _ if and only if @xmath24 avoids the patterns @xmath392 and @xmath393 ( * ? ? ? * lemmas 4.1 and 4.8 ) . + a _ @xmath2-noncrossing partition _ is a partition of @xmath128 $ ] into blocks such that when the blocks are drawn on the cycle @xmath2 , the convex hulls of the blocks do not overlap . @xmath401 continuing example [ barringex ] , a @xmath2-noncrossing partition of the cycle @xmath2 is shown in figure [ fig : noncross ] . define a _ noncrossing pre - order _ to be a pre - order @xmath191 on @xmath128 $ ] with respect to a coxeter element @xmath2 with the following two conditions : 1 . blocks in @xmath193 form a @xmath2-noncrossing partition of the cycle @xmath2 , and 2 . suppose blocks @xmath402}$ ] and @xmath403}$ ] intersect in @xmath193 , so that , without loss of generality , there exists a @xmath404}$ ] for @xmath405 $ ] such that @xmath406 . if @xmath407 is upper - barred with respect to @xmath2 then @xmath402}\prec b_{[j_t , j_1]}$ ] in @xmath193 and if @xmath407 is lower - barred with respect to @xmath2 then @xmath402 } \succ b_{[j_t , j_1]}$ ] in @xmath193 . let @xmath408 , and let @xmath409 be the image of the @xmath2-sortable permutations under @xmath126 . @xmath410 for any @xmath2-noncrossing partition there is a unique partial order on the blocks which defines a noncrossing pre - order . this defines a bijection from @xmath411 to @xmath2-noncrossing partitions . in ( * theorem 6.1 ) it was shown that there exists a bijection between @xmath2-noncrossing partitions and @xmath2-sortable permutations . thus @xmath412 , and it is enough to show that @xmath413 . let @xmath414 and let @xmath415 . suppose @xmath105 and @xmath121 are distinct descending runs in @xmath24 corresponding respectively to blocks @xmath402}$ ] and @xmath403}$ ] in @xmath193 such that @xmath402}$ ] and @xmath403}$ ] overlap . then without loss of generality there exists a @xmath404}$ ] such that @xmath406 for @xmath416 $ ] . suppose @xmath407 is upper - barred with respect to @xmath2 . then if @xmath402}\succ b_{[j_t , j_1]}$ ] in @xmath193 , @xmath417 and @xmath418 forms a @xmath392 pattern , which is a contradiction of @xmath24 being @xmath2-sortable . thus @xmath402}\prec b_{[j_t , j_1]}$ ] in @xmath193 . suppose @xmath407 is lower - barred with respect to @xmath2 . then if @xmath402}\prec b_{[j_t , j_1]}$ ] , @xmath419 and @xmath420 forms a @xmath393 pattern which is a contradiction of @xmath24 being @xmath2-sortable . thus @xmath402}\succ b_{[j_t , j_1]}$ ] in @xmath193 . the choice @xmath421 is the case where every number is lower - barred and the choice @xmath422 is the case where every number is upper - barred . in these specific cases , if two distinct blocks @xmath131}$ ] and @xmath136}$ ] overlap in @xmath193 , then either @xmath137\subset[k , l]$ ] or @xmath138\subset[i , j]$ ] . thus we have the following corollaries . let @xmath421 and let @xmath423 . then @xmath131}\preceq b_{[k , l]}$ ] in @xmath193 if and only if @xmath137\subset[k , l]$ ] . let @xmath422 and let @xmath423 . then @xmath131}\preceq b_{[k , l]}$ ] in @xmath193 if and only if @xmath138\subset[i , j]$ ] . the author thanks her advisor dr . nathan reading for presenting her with the idea to explore the shard intersection order on permutations and for insight and direction along the way . 9 m. aguiar , personal communication , 2009 . a. bjrner , _ shellable and cohen - macaulay partially ordered sets . _ trans . * 260 * ( 1980 ) , no . 1 , 159 - 183 . a. bjrner and m. wachs , _ on lexicographically shellable posets . math . soc . * 277 * ( 1983 ) , no . 1 , 323 - 341 . a. bjrner , p. edleman , g. ziegler , _ hyperplane arrangements with a lattice of regions . _ discrete comput . geom . * 5 * ( 1990 ) , no . 3 , 263 - 288 . p. edelman , _ a partial order on the regions of @xmath424 dissected by hyperplanes . * 283 * ( 1984 ) , no . 2 , 617 - 631 . n. reading , _ clusters , coxeter - sortable elements and noncrossing partitions . _ * 359 * ( 2007 ) , no . 12 , 5931 - 5958 . n. reading , _ lattice and order properties of the poset of regions in a hyperplane arrangement . _ algebra universalis * 50 * ( 2003 ) , no . 2 , 179 - 205 . n. reading , _ noncrossing partitions and the shard intersection order . _ preprint , 2009 ( arxiv:0909.3288 ) . n. j. a. sloane , _ the on - line encyclopedia of integer sequences . _ published electronically at http://www.research.att.com/~njas/sequences[`www.research.att.com/\simnjas/sequences ` ] .

as experimental platforms for quantum information processing continue to mature , characterization of the quality of unitary gates that can be applied to their quantum bits ( qubits ) becomes essential . eventually , the quality must be sufficiently high to support arbitrarily long quantum computations . randomized benchmarking already provides a platform - independent method for assessing the quality of one - qubit rotations . here we describe an extension of this method to multi - qubit gates . we provide a platform - independent protocol for evaluating the performance of experimental clifford unitaries , which form the basis of fault - tolerant quantum computing . we implemented the benchmarking protocol with trapped - ion two - qubit phase gates and one - qubit gates and found an error per random two - qubit clifford unitary of , thus setting the first benchmark for such unitaries . by implementing a second set of sequences with an extra two - qubit phase gate at each step , we extracted an error per phase gate of . we conducted these experiments with movable , sympathetically cooled ions in a multi - zone paul trap a system that can in principle be scaled to larger numbers of ions .

1203.3733

quantum information processing ( qip ) has the potential to solve difficult problems in many - body quantum mechanics and mathematics that lack efficient algorithms on classical computers . however , achieving useful qip will require precise control of many qubits ( two - level quantum systems ) and the ability to execute quantum gates ( operations that manipulate the quantum states of the qubits ) with low error per gate . here , the error per gate ( epg ) is @xmath2 , where @xmath3 is the ( average ) gate fidelity defined as the uniform average over pure input states of @xmath4 , where @xmath5 is the ( typically mixed ) output state and @xmath6 is the intended output state ( see ref . @xcite ) . a convincing demonstration of the potential for practical fault - tolerant qip should include verification of consistent epgs below a threshold of @xmath7 @xcite . so far , there has been substantial experimental progress on the basic techniques needed for qip , including the manipulation of small numbers of qubits and the implementation of the basic quantum gates that are needed to perform useful algorithms @xcite . the main challenges for qip experiments that remain are to scale up to larger numbers of qubits and to decrease the epg below the fault - tolerant threshold . therefore , it is desirable to efficiently characterize or benchmark the performance of multi - qubit qip experiments so as to extract the epg of specific gates and enable comparison between different quantum computing platforms . with these goals in mind , we give a benchmarking protocol for arbitrary numbers of qubits and show the results from an experimental implementation for two qubits . the protocol builds on previous work that used randomized sequences of clifford gates to measure the epg of one - qubit gates , first implemented in refs . @xcite . compared to techniques such as process tomography @xcite , randomized benchmarking offers several key advantages for characterizing epgs of quantum gates . for example , while process tomography offers more complete information about the performance of a gate , it does not scale efficiently with the number of qubits in the system , it can not readily measure epgs below the error probabilities of state preparation and readout , and it does not verify performance of a gate in arbitrary computational contexts . in contrast , randomized benchmarking can determine epgs with a number of measurements that scales polynomially with the number of qubits @xcite . because randomized benchmarking measures an exponential decay of fidelity as a function of the number of gates in the sequences , errors in state preparation and readout do not limit the minimum epg that one can measure . also , randomized benchmarking involves gates in the context of long sequences of operations and therefore establishes an epg that takes into consideration a computational context similar to that expected in the implementation of lengthy qip algorithms . because of these advantages , randomized benchmarking following the protocols of refs . @xcite has been used to measure one - qubit gate errors in a range of systems including trapped ions @xcite , superconducting qubits @xcite , liquid - state nmr @xcite , and neutral atoms in an optical lattice @xcite . recently , randomized benchmarking was used to measure an epg of @xmath8 for one - qubit operations with a trapped ion @xcite . a number of previous works have described properties of two - qubit gates with various measurement techniques . with trapped ions , the fidelity for creating a bell state has been measured at @xmath9 @xcite , @xmath10 @xcite , @xmath11 @xcite , @xmath12 @xcite and @xmath13 @xcite . process tomography was used to look at single and repeated applications of a two - qubit entangling gate @xcite . the average fidelities were found to be @xmath14 for one and @xmath15 for two gates in ref . two - qubit gates have also been studied in other quantum computing platforms including superconducting and photonic qubits ( see ref . @xcite and citations therein ) , with measured fidelities ranging from @xmath16 to @xmath17 , disregarding photon loss for photonic qubits . in a liquid - state nmr system , a randomized benchmarking technique was used to study the errors of sequences of randomized gates on three nuclear spins @xcite and found epgs of @xmath18 . the gates in this experiment were randomly chosen in a platform - dependent way from a special - purpose probability distribution where the probability of a two - qubit gate ( the cnot ) was @xmath19 . however , gate sets vary by platform , and other experiments may choose different probability distributions , for example to improve randomization . as a result the error probabilities from ref . @xcite may be difficult to compare to those obtained in future experiments . the multi - qubit protocol we describe first establishes a platform - independent error per operation ( epo ) for clifford unitaries by applying random sequences of clifford unitaries of varying lengths . here , a clifford unitary is any operator in the clifford group defined below . it then determines the epg of individual gates of our choice by inserting them into these sequences . the individual gates to be characterized may depend on the platform . of particular interest are implementations of one of the standard universal two - qubit gates such as the controlled - not ( cnot ) , phase gate or square - root of swap . the basic principles of the protocol are similar to the theoretical randomized clifford - based benchmarking sequences described and analyzed in ref . however , like the standard protocol used in one - qubit benchmarking experiments so far , the last gate of the sequence does not strictly reverse the effects of the previous ones , thus enabling the protocol to detect certain large errors that can otherwise masquerade as no errors . in addition , we discuss the practical aspects of choosing the random clifford unitaries in the sequence and extend the protocol to enable characterization of specific gates , thus enabling diagnostics that were previously unavailable in randomized benchmarking . the proposed protocol is flexible without affecting the ability to compare results from unrelated implementations . we note that the theoretical relationship between the protocol s epgs and epos and the detailed physical noise parameters is not known in general @xcite . however , we suggest that subject to simple consistency checks , the protocol s reported epgs and epos are nevertheless useful quantities for comparison and reflect computationally relevant error behavior . for our demonstration with trapped ions , we take advantage of a multi - zone ion trap @xcite . the universal two - qubit gate chosen here is a phase gate , @xmath20 , implemented via a mlmer - srensen gate @xcite and acting as the diagonal matrix with diagonal @xmath21 $ ] in the basis labeled by @xmath22,@xmath23,@xmath24,@xmath25 , where @xmath26 and @xmath27 represent the two eigenstates of each qubit , with @xmath28 , etc . qubit addressing is implemented by separation of the ions into different wells , and long sequences of gates are supported by sympathetic cooling techniques as required for the approach to scalable ion - trap quantum computing described in refs . the experiment extends the technology demonstrated in @xcite by using longer sequences of gates and a different implementation of the phase gate @xcite to act directly on a magnetic - field - insensitive transition in @xmath29be@xmath30 . from sequences of up to seven clifford unitaries , each requiring an average of @xmath31 phase gates , we deduced an epo of @xmath32 for the clifford unitaries and an epg of @xmath33 for the phase gates . although we implemented relatively long sequences , the experiment does not yet demonstrate stationary behavior because ion loss prevented routine implementation of longer sequences . there are also indications that the errors increased with sequence length by two to three standard deviations , with the epos ranging from @xmath34 to @xmath35 and the epgs from @xmath36 to @xmath37 as the sequences lengthened . our epo sets the first benchmark for random two - qubit clifford unitaries . the epg shows no improvement over the gates used in @xcite , but applies to gates used in computationally relevant contexts in longer sequences . the paper is structured as follows : we first describe the protocol and its main features for two qubits . we then discuss the experimental implementation of the protocol and show the experimental results . the data - analysis methods are detailed next , followed by a discussion of necessary consistency checks and estimates of physical sources of error . we finally define the protocol for arbitrary numbers of qubits , and make recommendations for how to apply and compare it when qubit numbers vary . clifford unitaries are fundamental to most error - correcting procedures envisioned for quantum computing ( see , for example , ref . @xcite ) and thus serve as a foundation on which universal fault - tolerant quantum computing is built . as a result , a large fraction of the fundamental processes in proposed quantum computing architectures involve clifford unitaries . the three main features of the group of clifford unitaries that make it useful for our purposes are that its members have compact representations that can be efficiently converted to circuits of elementary quantum gates , outcomes of standard measurements of sequences of clifford unitaries can be efficiently predicted by classical computation , and the group is sufficiently rich that error operators can be perfectly depolarized . these features are explained in context below . for a system of @xmath38 qubits , clifford unitaries can be constructed by combining one - qubit @xmath39 rotations , defined as @xmath40 with @xmath41 , about the @xmath42 and @xmath43 axes , and two - qubit cnot gates . alternatively , the clifford unitaries are the members of the clifford group , which is defined as the set of unitaries @xmath44 with the property that for every pauli operator @xmath45 , @xmath46 is a signed product of pauli operators . we consider two gates or unitaries that differ only by a global phase to be identical . the randomized benchmarking protocol is an extension of `` clifford twirling '' @xcite . in the simplest instance of clifford twirling , an arbitrary quantum process @xmath47 is sandwiched between a random clifford unitary @xmath48 picked from a uniform distribution and its inverse @xmath49 . alternatively , we can think of clifford twirling as averaging the process @xmath50 over all elements @xmath51 in the set @xmath52 , of clifford unitaries . the key property of clifford twirling is that this new process behaves like one that uniformly depolarizes with some probability . in other words , a single parameter , the probability that a pure input state is mapped to an orthogonal state , characterizes the new , average process @xcite . when @xmath47 is a noisy implementation of the identity gate , such as a long self - reversing sequence of gates , we use this parameter as the definition of the average error of @xmath47 . clifford twirling can be generalized to learn the average error of an arbitrary process @xmath47 intended to implement a specific clifford unitary @xmath44 : the inverting clifford unitary @xmath49 that is applied after the process is modified to an implementation of the unitary @xmath53 . with this modification , the net effect is @xmath44 if there are no errors in @xmath47 . because clifford unitaries form a group , @xmath54 is also clifford . the implementation of @xmath54 should not rely on @xmath47 to provide @xmath44 , and it is better not to decompose it into a composition of three processes according to the given expression . this can be satisfied by first evaluating the unitary operator @xmath55 as an element of the clifford group and then implementing it by an efficient procedures for translating clifford unitaries into quantum circuits @xcite . if @xmath47 contains errors , then the net process applies the unitary @xmath44 followed by a uniformly depolarizing error whose parameter defines the average error of @xmath47 . randomized benchmarking extends the idea of clifford twirling from the simple three - step sequence described above to randomized sequences of clifford unitaries with errors . these sequences consist of steps where each step implements a randomly - chosen clifford unitary and may have errors . each step in the sequence simultaneously acts as a process that undergoes twirling and contributes to the twirling of errors in the other steps . under optimistic assumptions described later , each step effectively behaves as an ideal unitary followed by a depolarizing process . the first goal is to establish the average error per step , which can then be reported as the epo for clifford unitaries . the method for implementing the clifford unitaries making up the steps in the randomized sequences is up to the experimenter . here we describe our approach . to improve stability of the twirling process and take advantage of the typically lower errors of one - qubit gates , each step is composed of a clifford unitary preceded by a pauli unitary , where the two parts are chosen so that together they implement a uniformly random clifford unitary . the choice of pauli unitary on each qubit is random and independent from the choice of clifford unitary ; each pauli unitary involves applying either no pulse or a major - axis @xmath56-pulse . there are eight possible such pulses acting as @xmath57 , where @xmath58 is a pauli matrix or the identity : @xmath59 , @xmath60 . the sign in the exponent affects only the global phase and results in two choices for each possible matrix in the exponent . we keep the sign because in many cases including ours , the change in sign can involve a physically different device setting , such as the phase in a pulse generator that determines the orientation of the fields that mediate the pulse . each qubit s @xmath56 pulse is chosen uniformly at random from the above eight pulses . because of this pauli randomization procedure , it suffices to choose the unitary in the second part of the step uniformly at random from the clifford group modulo the group of pauli products . for this we can take advantage of the fact that the group of pauli products is a normal subgroup of the clifford group , and the quotient group ( of clifford unitaries modulo pauli products ) has a representation in terms of binary symplectic matrices @xmath61 of dimension @xmath62 such that @xmath63 modulo @xmath64 , where @xmath65 is a @xmath66 block matrix with @xmath67 blocks whose diagonal blocks are zero and whose off - diagonal blocks are the identity , see , for example ref . @xcite . the terminology is based on ref . @xcite . for two qubits , there are @xmath68 such matrices @xmath61 . uniformly and randomly choosing from among these matrices is computationally straightforward and efficient . determining an implementation of the clifford unitary described by such a matrix in terms of the elementary gates available in a particular experiment is more challenging . there are efficient algorithms that translate an arbitrary symplectic binary matrix into order of @xmath69 elementary one- and two - qubit gates @xcite , each of which can then be mapped into experimentally available operations . however , there is strong motivation to obtain shorter implementations , as this is a sure way to improve the measured epo . while it is unlikely that optimal implementations can be readily obtained for arbitrary numbers of qubits , we used the following strategy for two qubits optimized for our demonstration : by exhaustively listing compact circuits of one - qubit clifford gates and phase gates @xmath20 , we determined for each of the @xmath68 symplectic binary matrices a circuit with the minimum number of phase gates implementing the corresponding clifford unitary ( modulo a pauli product ) . on average , @xmath31 phase gates were required . these circuits were then translated into appropriate actions in our ion - trap platform . ( 6,1.1)(0,0 ) ( 0,.5 ) ( 0,0)[l]@xmath70 ( 1.85,0.5 ) ( 0,0)[c]@xmath71 ( 2.5,0.5 ) ( 0,0)[l ] matrix @xmath72 is generated . in general , the size of such matrices is twice the number @xmath38 of qubits . they efficiently encode a clifford unitary whose size is @xmath73 in general . the second step is to convert @xmath72 into a sequence of elementary gates that enacts the corresponding clifford unitary and is suitable for implementation in the ion trap platform . we minimize the number of @xmath74 gates in such sequences . the sequence found for the example @xmath72 shown is given on the right . , title="fig:",height=96 ] ( 3.45,0.45)@xmath20 ( 4.63,0.45)@xmath20 ( 2.7,0.17)@xmath75 ( 3.88,0.72)@xmath75 ( 5.08,0.72)@xmath75 ( 3.9,0.17)@xmath76 ( 5.08,0.17)@xmath76 given the method for generating the random unitaries for one step , a benchmarking experiment is configured by first deciding on a set of lengths @xmath77 that determine the numbers of steps in sequences to be generated . the epo is determined by fitting an exponential decay to fidelities ( @xmath78 , where @xmath79 is the error probability ) measured for each sequence length . the choice of lengths therefore contributes to how well the epo can be extracted . in particular , there should be enough lengths for stable curve fitting , and lengths much greater than the inverse epo contribute little additional information . for each length @xmath80 , many sequences of @xmath80 random steps are produced . at the end of each such sequence , a randomized measurement step is added . this step consists of a pauli randomization followed by a clifford unitary that inverts the @xmath80 preceding clifford steps . the final clifford unitary is chosen independently of the pauli randomization . this ensures that in the absence of errors , the final state is again in the computational basis but randomized . which basis element it should be in can be computed by use of standard efficient methods for simulating sequences of clifford unitaries @xcite . the sequence can then be experimentally implemented after preparation of each qubit in the @xmath81 eigenstate of @xmath82 and followed by a measurement in the @xmath82 basis of each qubit . one should implement sufficiently many runs of each sequence to have good signal - to - noise on the inferred probabilities of getting a correct or incorrect answer in the measurement for this particular sequence . the process of generating and implementing random sequences at each length is repeated in order to ensure randomization of the unitaries and their associated implementation errors . for our two - qubit benchmarking demonstration , we used the set of lengths @xmath83 and generated between @xmath84 and @xmath85 random sequences of each length . the variation in numbers of sequences is explained below . we implemented @xmath86 runs for each sequence to determine its probability of error of experiments , the record for one run is missing . thus , for experiments with nominally @xmath86 runs , occasionally only @xmath87 runs were recorded . ] . the experimental runs yield an average probability of error @xmath88 for each length @xmath80 , where the average is over the sequences of this length and their runs . to analyze @xmath88 we start by making the simple assumption that each step s error behaves as a completely depolarizing channel ( see , for example , ref . @xcite , pg . 378 ) characterized by error probability @xmath89 independent of its gates or position in the sequence . similarly , we assume an overall error probability @xmath90 for state preparation , the last inverting gate and its pauli randomization , and measurement . then the mean of @xmath88 with respect to repetitions of the experiment satisfies @xmath91 for two qubits . the case of more than two qubits is discussed in sec . [ sec : benchn ] . note that @xmath92 is the depolarization probability if the error probability is @xmath93 . the probability of not depolarizing the state in a sequence is the product of the probabilities of not depolarizing in each step . to derive the equation , note that in terms of @xmath94 , the probability of not depolarizing is @xmath95 . assuming that the experimental observations are consistent with the simple exponential behavior suggested by this formula , we use it as the defining formula for the epo @xmath89 of a random clifford unitary , regardless of the actual behavior of errors . in particular , in the context of these benchmarks , we associate the epo with the decay parameter of the error probabilities @xmath88 rather than a particular exact parameter of the underlying physical errors . this supports the platform - independent use of randomized benchmarking . if the simple depolarizing assumption does not hold , then @xmath94 may exhibit non - exponential and transient behaviors ; see the discussion below . however , the twirling effected by the randomization is intended to induce behavior that matches the one implied by this assumption . to isolate the epg of the phase gate @xmath20 ( or any other gate ) we generate a second set of sequences by inserting @xmath20 after each random clifford unitary . the final inverting clifford unitary is chosen in the same way as before , taking into account the effect of the additional @xmath20 gates to ensure that the final state is a predictable computational basis state in the absence of errors . under the same idealizing assumptions that yield eq . , the average probability of error @xmath96 measured for the implementation of this experiment satisfies eq . [ eq1 ] , but with a different value of @xmath89 due to the additional operation in each step . in an ideal experiment @xmath90 should be the same , but the model must take into consideration that it might have changed , for example due to experimental drifts . explicitly , @xmath97 where @xmath98 is the probability of error of a step consisting of a random clifford gate and @xmath20 . in this context , the assumptions on the error behavior of @xmath99 could be relaxed from simple depolarization . we can isolate the epg @xmath100 of @xmath20 by solving the identity @xmath101 which gives @xmath102 it is helpful to run randomized benchmarks on subsets of the available qubits so that results can be compared to other experimental platforms that have different numbers of available computational qubits and for investigating differences in behavior that depend on ( for example ) geometrical relationships between qubits . if possible , these benchmarks should be run in parallel on disjoint subsets . for these reasons , we checked the performance of the one - qubit gates in parallel on the two ion qubits . because of the pre - existing benchmarks , we did not implement the above protocol for each qubit , but used a one - qubit benchmarking protocol similar to that of ref . briefly , the length of a sequence is the number of steps that consist of a pauli gate ( @xmath56-pulse ) followed by a clifford gate ( @xmath103-pulse ) on each qubit . each step can be thought of as implementing a random computational gate . the gate sequence is followed by a pauli gate and clifford gate chosen to yield a predictable measurement outcome in the @xmath104 basis for each qubit . the pauli gates are chosen with equal probability to be rotations about the @xmath105 or @xmath106 axis or the identity . the clifford gates are chosen with equal probability from the following five options : @xmath107 @xmath108 or the identity . when many subsequent gates are composed together , this distribution of clifford gates demonstrates favorable convergence to a uniformly random clifford unitary in comparison with the distribution in ref . the introduction of identity gates into the clifford gate step reduces the average expected number of @xmath109-pulses in that step from @xmath110 to @xmath111 . we perform the benchmarking demonstration with the ion - trap system described in @xcite using updated techniques . this system includes most of the features of the scalable quantum computing architecture of @xcite . we trap four ions in a six - zone linear paul trap : two @xmath29be@xmath30 ions that serve as the qubits , and two @xmath112mg@xmath30 ions that are used for sympathetically recooling the qubit ions during the sequences . the ions form a linear chain along the axis of the trap , which is the axis of weakest confinement . the two - qubit phase gates are performed with all four ions in the same trap zone , in the order @xmath29be@xmath113mg@xmath113mg@xmath114be@xmath30 ( fig . [ trapfig]a bottom left ) . individual addressing of the ions for one - qubit rotations is achieved by separating the ions into two trap zones @xmath115 apart with a single @xmath29be@xmath30-@xmath112mg@xmath30 pair in each zone ( fig . [ trapfig]a below electrodes ) . be@xmath113mg@xmath30 pair is trapped in each zone ( depicted directly under the electrodes ) . to perform the entangling gate , all four ions are combined in trap zone a ( bottom left ) . the beam waists are approximately @xmath116 in the vertical direction and @xmath117 along the trap axial direction , which is large compared to the extent of the two - ion and four - ion crystals ( @xmath118 and @xmath119 ) . ( b ) laser beam configurations and frequencies used for different operations . ( i ) two co - propagating beams induce raman carrier transitions in either trap zone used for single - qubit gates . an acousto - optic deflector is used to direct the co - carrier to either trap zone . ( ii ) the 90@xmath120 beam is directed to trap zone a at 90@xmath120 with respect to the co - carrier beam paths such that the wave - vector difference is along the trap axis . these beams induce carrier transitions used as part of the phase gate @xmath20 . ( iii ) three beams induce the mlmer - srensen gate used as part of @xmath20 . beams with different frequencies depicted as slightly displaced arrows are actually overlapped in the experiment . details and frequency definitions are provided in the text . ] the qubit states are the @xmath121 and @xmath122 hyperfine states of @xmath29be@xmath30 , where @xmath3 and @xmath123 are the total angular momentum quantum numbers . the energy difference between these states is first - order insensitive to magnetic - field fluctuations at the applied field of @xmath124 @xcite . at the beginning of each experiment we prepare the @xmath29be@xmath30 ions in the @xmath125 state . at the end of each experiment , we detect the qubit states by transfering the @xmath126 and @xmath127 states to the @xmath128 and @xmath129 states , respectively and then apply a @xmath130-polarized laser beam that is directed to trap zone a ( fig . [ trapfig ] ) and resonant with the s@xmath131 @xmath128 @xmath132 p@xmath133 @xmath134 cycling transition . the presence ( absence ) of ion fluorescence observed with a photomultiplier tube indicates the @xmath26 ( @xmath27 ) state . for a single @xmath29be@xmath30 ion , the average number of photons collected in @xmath135 is typically @xmath136 for the @xmath128 state and @xmath31 for the @xmath129 ( limited by stray light ) . this allows us to analyze each detection individually with a threshold detection level of around @xmath137 counts . to measure both qubits we first detect the state of the left qubit while the other is held in trap zone b ( fig . [ trapfig ] ) . then we optically pump the left qubit to the @xmath128 state and transfer it into the `` dark '' @xmath129 state . finally , we bring both qubit ions into trap zone a and apply the same procedure to detect the state of the right qubit . one - qubit rotations about a vector in the @xmath138 plane are implemented with the `` co - carrier '' laser beams ( fig . [ trapfig ] ) that cause stimulated - raman @xmath139 transitions on the @xmath29be@xmath30 ions after they are separated and held in different trap zones @xcite . specifically , the carrier transitions perform @xmath140 , where @xmath141 , and @xmath142 depends on the phase difference of the laser beams at the position of the ion(s ) . for co - carrier transitions ( on - resonance qubit transitions ) , two co - propagating laser beams have frequencies @xmath143 and @xmath144 ( see fig . [ trapfig]b ( i ) ) , where @xmath143 is the principal laser frequency at @xmath145 thz , which is approximately @xmath146 ghz below the @xmath147 to @xmath148 transition frequency and @xmath149 ghz is the qubit transition frequency . the co - carrier laser beam position along the trap axial direction is controlled with an acousto - optic deflector that allows the beam to address ions in either trap zone . the pulse duration for a single - qubit rotation by @xmath56 and @xmath103 , using the co - carrier beam configuration shown in fig . [ trapfig]b ( i ) , was approximately @xmath150 and @xmath151 @xmath152s , respectively . one - qubit @xmath82 gates ( @xmath153 ) are implemented in software by shifting the rf phase of all future rotations for that qubit by @xmath154 . identity gates are implemented with a wait time equal to @xmath155 . one laser beam propagating along the co - carrier beam path and a pair of laser beams propagating along the 90@xmath156 beam path ( fig . [ trapfig ] ) are used to implement phase gates @xmath20 . in contrast to the experiments in @xcite , which implemented two - qubit @xmath20 phase gates directly on hyperfine states , we use one - qubit rotations and a mlmer - srensen ( ms ) gate @xcite to implement @xmath20 @xcite . in the previous experiments , implementation of the phase gate required that the qubit ions states be transferred from the qubit manifold , where the qubit frequency is first - order independent of magnetic - field fluctuations , to other hyperfine states @xcite . however , the ms gate can be performed directly on the qubit states . to implement @xmath20 we surround a ms gate pulse , @xmath157 , with carrier @xmath158-pulses on both ions by use of two laser beams as shown in fig . [ trapfig]b ( ii ) and ( iii ) . the resulting three pulse sequence is @xmath159 , where we use @xmath160 and where the overall phase factor after @xmath20 has no physical consequence in this setting . the advantage of using @xmath20 as our elementary two - qubit gate rather than the ms gate is that this three - pulse sequence is insensitive to slow changes in the optical path - length difference of the non - copropagating beams , which cause @xmath142 to change @xcite . the duration of the @xmath161 pulse is @xmath162 and the duration of each carrier @xmath158 pulse using the beam configuration shown in fig . [ trapfig]b ( ii ) , is approximately @xmath163 . due to wait periods between pulses that are necessary to stabilize the feedback loops that control the laser pulse amplitudes and phases , the three pulse sequence requires @xmath164 to complete . before performing each @xmath20 gate we sympathetically laser - cool the four - ion crystal , first using doppler and then raman sideband cooling of the @xmath112mg@xmath30 ions @xcite . this ensures that each time we implement @xmath20 , the motional modes along the axial direction are cooled to near the ground state . the cooling light interacts only with @xmath112mg@xmath30 and thus preserves the qubit state coherences . in more detail , the ms gate requires the simultaneous application of detuned blue and red sidebands . to achieve this , we overlap three laser beams with different frequencies in trap zone a ( fig . [ trapfig]b ( iii ) ) . one laser beam propagates along the co - carrier beam path with frequency @xmath143 . the other two beams co - propagate in the 90@xmath120 beam path at frequencies of @xmath165 , where @xmath166 is the frequency of a motional mode and @xmath167 is a detuning . the two laser beams are derived from a single beam that is split and passed through different double - pass acousto - optic deflectors such that they end up with a frequency difference of @xmath168 . the split beams are then recombined on a @xmath169-@xmath169 beam splitter with one port directed to the ions and the other going to a photo - detector that is used to measure and stabilize the phase of the beat note , as required to realize @xmath161 @xcite . to implement @xmath20 , we simultaneously address the two highest - frequency axial motional modes for the four - ion crystal at @xmath170 and @xmath171 @xcite . the detuning @xmath172 must be chosen such that the detuning from one mode is an integer multiple of the detuning from the other in order to fully disentangle both motional states from the qubit states at the end of the gate . experimentally , we found a detuning of @xmath173 above @xmath166 was optimal given our laser beam intensities , which implies that the ms gate was implemented with one phase - space loop on the @xmath166 mode and four loops on the @xmath174 mode @xcite . in fig . [ msgate ] we plot the observed fraction of both ions in the @xmath26 state ( red squares ) , both ions in the @xmath27 state ( blue circles ) , and one ion in each state ( green triangles ) as a function of the duration of the red and blue sideband pulses applied to an initial state of @xmath175 . the ms gate is completed in approximately @xmath162 ( @xmath176 ) . state for varying durations and observe the frequency with which we find both ions in the @xmath26 state ( red squares ) , both ions in the @xmath27 state ( blue circles ) , or one ion in each state ( green triangles ) . from these curves , we can determine the gate time ( @xmath177 ) for the ms gate . here it is approximately @xmath178 @xmath152s , at which point the qubit states are entangled and ideally in the state @xmath179 , where @xmath142 depends on the phases of laser beams at the ions position and can vary from experiment to experiment ( see text ) . the solid lines show the theoretical results for an ideal gate . to perform a phase gate we surround the ms - gate pulse with two @xmath103-pulses by use of the laser beams as depicted in fig . [ trapfig]b . the points and their error bars were determined by photon - count histogram fitting from @xmath180 runs . ] for the clifford and phase - gate benchmarks , we generated random sequences of lengths @xmath181 . the respective number of sequences implemented was @xmath182 for the clifford benchmark , and @xmath183 for the phase - gate benchmark , in order of sequence length . each time we performed a sequence for the clifford benchmark we then immediately performed the corresponding phase - gate benchmark sequence . each sequence was implemented @xmath86 times . from the measurement outcomes , we determined the fraction of measurements that matched the prediction . the data shown in fig . [ rbdata ] was obtained in four successive sets of experiments on the same day . during and between the sets we periodically recalibrated the magnetic field and the laser frequencies needed for sympathetic cooling of @xmath112 mg but not the other pulse parameters . within each set , the sequences were randomized with respect to length . however , the first ( second , third , fourth ) set involved sequences of lengths @xmath110 to @xmath184 ( to @xmath185 , @xmath186 , @xmath187 , respectively ) . in particular , sequences of length @xmath187 were run only in the last set of experiments , which is why there are fewer sequences of length @xmath187 contributing to the data . during the experimental runs for these sets , we observed one ion - loss event . we did not implement longer sequences because ion - loss events became a problem for lengths greater than @xmath187 . in our implementation , a clifford unitary took @xmath188 on average . for the sequences with an extra @xmath20 gate inserted after each step , @xmath189 per step was typical . the most time - consuming elements of the sequence implementations were the sympathetic recooling of the ions after each recombination of the ions into trap zone a , followed by the separation and recombination processes . each sequence began with approximately @xmath190 for state preparation and laser lock stabilization . thus , a sequence of length @xmath187 with an extra @xmath20 inserted at each step lasted approximately @xmath191 . longer sequences resulted in an accelerated rate of ion loss events ( on the order of a loss event per minute ) , which can likely be attributed to a decreased probability of recovery from background gas collisions that can occur at any point during the sequences . before running each sequence , two warmup sequences with @xmath86 experiments each were run to make sure the experiment was in a steady state ; the results of these experiments were not recorded . switching from one sequence to the next required @xmath192 to @xmath193 of computer time to reprogram the control hardware . in total , all of the clifford benchmarks , including the sequences used to benchmark @xmath20 , were completed in approximately @xmath110 hour and @xmath194 minutes , which also includes the time durations needed for periodic recalibrations of the magnetic field as well as the time period to reload a set of ions following the only ion - loss event . the parallel one - qubit benchmark whose results are shown in fig . [ rbdatasq ] was executed in one set , after all of the two - qubit benchmarks and following a recalibration of the one - qubit gates . the number of sequences implemented was @xmath195 for sequence lengths of @xmath196 , respectively . we ran each sequence @xmath86 times , as before . in order to approximately replicate the conditions of the experiment for the two - qubit benchmark , in each step , the ions were recombined into a single trap zone , recooled and then held for approximately the same duration required to execute @xmath20 before being separated again for the next sequence step . the red data points and curve in fig . [ rbdata ] show the results from the experimental clifford gate benchmark and their match to an exponential decay . the match gives a clifford unitary epo of @xmath197 . as a function of the sequence length @xmath80 . by fitting the data to the expression in eq . ( red line ) , we find an error per random clifford unitary @xmath197 . the preparation / measurement error , @xmath90 , is @xmath198 ( recall that measurement error includes the error for an additional inverting gate before detection ) . blue squares show the results for running random sequences with an additional @xmath20 inserted after each step . fitting this data to eq . yields an error of @xmath199 . in this case the preparation / measurement error , @xmath200 , is @xmath201 . the error bars in the plot represent the standard deviation of the mean of the sequences frequency of correct measurement outcome . error bars for inferred parameters are based on bootstrap resampling ; see the text . ] the blue data points and curve in fig . [ rbdata ] show the results from the @xmath20 benchmark . curve - fitting and solving the above equations for @xmath100 give an epg of @xmath202 . we determined the errors per step on each qubit independently with the parallel one - qubit benchmarks explained above . the results from the benchmarks are shown in fig . [ rbdatasq ] . the inferred one - qubit errors per step are @xmath203 and @xmath204 for the respective qubits . using the assumption that laser pulses dominate the error per step , these results can be compared to the protocol of ref . @xcite through multiplication by the ratio @xmath205 of @xmath109-pulses per step in the two protocols . where @xmath80 is the sequence length , @xmath90 is related to the state preparation and readout fidelities of the two qubits , and @xmath89 is the error per step in the sequence . we find the errors per step to be @xmath203 and @xmath204 , respectively . ] in the limit of very large numbers of sequences for each length , we can use a simple , nonlinear , weighted - least - squares fit of eq . to the fidelity curves as a function of length . the weights are determined by the standard error of the mean for the fidelity at each length . note that some non - linear least - squares fitting functions compute the error in the inferred parameters from the fitting error and ignore the scale of the errors implied by the weights given to the individual points . because we already have a good estimate of the standard errors of these means , a better estimate of the error in the inferred parameters can be obtained by direct propagation of errors , particularly when there are few points . for smaller numbers of sequences for each length , we must consider that we know little about the distribution of the fidelities for different random sequences of a given length . this distribution is affected not only by the differences in actual pulses applied , but also by factors such as the amount of coherence in error ( see below ) . however , the estimate of a given sequence s fidelity from the @xmath86 experimental runs is binomially distributed . thus we used a partially parametric bootstrap @xcite procedure to determine a standard error for the parameters inferred by fitting . let @xmath206 be the number of different sequences of length @xmath80 used in the experiment . denote the experimentally measured fidelity of the @xmath207th sequence of length @xmath80 as @xmath208 , which is the fraction of times the correct result was obtained during the @xmath86 runs of the @xmath207th sequence . we generated artificial data for each bootstrap resample as follows . for each length @xmath80 , we constructed @xmath209 by letting @xmath210 be a random element of the sequence of fidelities @xmath211 , picked independently ( with replacement , that is , the same element can be picked multiple times ) for different @xmath212 . for each @xmath210 , we generated a random @xmath213 according to a binomial distribution , so that @xmath213 was the fraction of @xmath110 s in @xmath86 random instances of a @xmath214 variable where the probability of @xmath110 was @xmath210 . we then averaged the @xmath213 for each length @xmath80 and computed the fit in the same way as it was computed for the real data to obtain the inferred parameters for this resample . this resampling procedure was repeated @xmath215 times , yielding @xmath215 resampled values for each of the parameters . the standard errors in the reported parameters are determined as the square - root of the variances of the corresponding resampled parameters . the values of the parameters are still the ones from the fit of the original experimental data , as this is the less biased estimate and requires no bootstrapping . although the relationship between epos and physical errors in gates is not known in general , specific benchmarking protocols provide well - defined epos that can be compared across platforms . however , for an implementation of the benchmark to be convincing , there are several assumed or expected properties that can be checked . these include the following : we can determine whether or not the fidelity curves are consistent with a simple exponential as a function of sequence length , and if not , analyze the deviations . given that we know the implementations of the clifford unitaries , we can compare the epo for a clifford unitary with that inferred from the epg for a phase gate and the one - qubit benchmark . first we consider the exponential fits shown in fig . [ rbdata ] . the @xmath216 values ( four degrees of freedom ) for the two curves are @xmath217 and @xmath218 , respectively . the higher one corresponds to a @xmath219-value of @xmath220 , approximately the conventional boundary for significance . there is other evidence that the exponential model may not be a good fit . first , the two preparation / measurement errors are expected to be the same , but the fits seem to suggest otherwise , although the statistical significance is not strong . second , both sets of data seem to dip below the fit near the end . together these observations suggest an increased epo for later clifford unitaries . indeed , dropping the first points from the analysis suggests higher epos . for example , the fits for the last four and three sequence lengths have epos of @xmath221 and @xmath222 , respectively . the corresponding epgs are @xmath37 and @xmath223 . note that the second values are from a two - parameter fit to three points , which reduces their significance . the fits to the first four sequence lengths give an epo of @xmath34 and an epg of @xmath224 . the higher epos for longer sequences could be related to the fact that unlike the shorter ones , they were run only in the last or last two sets of experiments , without recalibrating pulse parameters except as noted above . we attempted to confirm this hypothesis by analyzing the results for sequence lengths one to three separately for each set . the results of this analysis are consistent with a drop but have insufficient signal - to - noise to be conclusive . in principle , we would like the platform to have the property that errors reach stationary behavior soon after state preparation , and the benchmark s reported epo should reflect the stationary error . as noted above , we were not able to implement long enough or sufficiently many sequences to clearly observe stationary behavior , or to determine the extent to which the behavior is nonstationary . the epo and epg reported in the abstract are determined from all six lengths tested given the `` early '' and `` late '' values above , we believe that they are a good representation of mid - length behavior of gate errors . our inability to consistently run sequences of length longer than six prevents any claims of stationary behavior . other issues with the exponential fits that can arise include the possibility that the curves are a mixture of exponentials , as would be expected if the epos change slowly compared to the time required to run a sequence . in this case , the apparent epos would tend to decrease with increasing sequence length , as the higher - epo runs affect the loss at shorter lengths , but tail behavior is dominated by the slowest decay . given the observations of the previous paragraph and the available statistics , we can not usefully test for this possibility . now we consider consistency between the measured epos and epgs . we estimate the epo that we should have measured given the epgs obtained from the one - qubit and the phase gate benchmarks . for this estimate , we count the complexity of sequences of one - qubit pulses in terms of the number of effective @xmath103-pulses applied . this counts only pulses around the @xmath225 or @xmath226 axes , taking into consideration that @xmath227-axis pulses and identity gates are essentially error - free . the @xmath56-pulses are counted as two @xmath103-pulses . coherent error addition is neglected . the one - qubit benchmark s steps each have an average of @xmath228 effective @xmath103-pulses per qubit . if we use @xmath229 as a representative error probability per step from the one - qubit benchmark ( fig . [ rbdatasq ] ) , we obtain @xmath230 as the linearized error probability per one - qubit @xmath103-pulse . the factor of @xmath231 converts the average probability of error for one qubit to that for two qubits under the assumptions that the other qubit has no error and any pauli error is twirled to a depolarizing error . for the purpose of this calculation , we take the error probability per phase gate to be @xmath232 . each step in the clifford benchmark has an average of @xmath233 effective @xmath103 pulses and @xmath31 phase gates . the linearized error probability for a step can therefore be estimated as @xmath234 , with a standard error of about @xmath235 , if we add the statistical errors in quadrature . this linear approximation is expected to give a pessimistic estimate , but in this case , the nonlinear correction is smaller than the error in the estimate . while our estimate gives a value below the measured epo , the difference appears not to be statistically significant . we emphasize that the above strategy for estimating the epo from epgs neglects coherent error addition , which tends to increase the error , and internal error cancellation that could arise from the way pulses are combined within a step . we consider known sources of errors and estimate their contribution to the epg of @xmath20 . spontaneous emission is a fundamental source of error for transitions driven by stimulated - raman transitions ; here the laser beams are tuned approximately @xmath236 below the @xmath148 state @xcite . we simulate that for our laser parameters , this should contribute an error probability of @xmath237 to a one - qubit @xmath56-pulse and @xmath238 to the phase gate . ( recall that the phase gate consists of an ms gate surrounded by @xmath158-pulses . ) errors can also arise from imperfect calibrations and slow drifts of the gate parameters . these parameters include beam intensities , frequencies , phases , and pulse durations . these types of errors are coherent in the sense that for any given run , each implemented gate still causes a unitary change in state , but not exactly the intended one . to determine whether such errors contribute significantly to the measured epos , we consider the variation in fidelities for different sequences of a given length . coherent error contributions typically result in a variation that is larger than that expected from a simple statistical analysis @xcite . for our experiment and in the absence of coherent errors , we attribute the largest sources of variation in the fidelities of fig . [ rbdata ] to the varying number of @xmath20 gates and single - qubit rotations needed to implement each step s random clifford unitary , and to the binomial statistics for the fidelities inferred from the @xmath86 runs of each sequence . [ alldata1 ] compares the actual variation and the variation predicted from the statistics of the number of phase gates per step and the binomial statistics . we did not include the variation in numbers of one - qubit gates due to their significantly smaller error . the gate statistics and binomial statistics are independent , so their contributions were added in quadrature . the predicted variation is generally somewhat less than the measured one but does not indicate coherence of the errors in a given sequence because our simple model does not account for all incoherent effects . the main sources of coherent errors are due to drifts in beam intensity , relative laser field phases for the beams implementing the phase gate , and stark - shifted frequencies . these drifts result in errors in pulse time , and phase and frequency calibration , each of which are estimated to contribute approximately equally to the epg . we estimate that their total contribution to the phase gate epg is less than @xmath239 . . the total numbers of sequences shown at sequence lengths @xmath181 are @xmath240 , respectively . the average error is the percentage of times at least one of the qubits was not found in the expected state in @xmath86 experiments . the red error bars to the left of the data at each length show the expected standard deviation if the error variation is due to variation in the number of phase gates needed to implement the random clifford unitaries used in the sequences and the binomial statistics for the @xmath86 runs for each sequence . the black error bars show the standard deviation of the set of fidelities measured for the corresponding length . the solid line is the fit to eq . [ eq1 ] . ] there are a number of lesser sources of error to consider . in addition to the slow intensity drifts included above , there are also fluctuations in intensities that can be slow compared to sequence duration but are too fast to be calibrated out . for example , these can arise from fluctuations in laser power or from noise in the position of the laser beams with respect to the ions , due to vibration and air movement . such fluctuations in intensity lead to loss of visibility in rabi flopping curves @xcite . from such curves , we determined that the rabi rate on the carrier transition with the non - copropagating beams fluctuates by @xmath241 from experiment to experiment . this results in a contribution of @xmath242 @xcite to the phase gate epg . due to the finite lambe - dicke parameter for the ions , fluctuations in ion motional energy can cause errors in the ms gate @xcite . as in ref . @xcite , we estimate that each motional mode is cooled to an average excitation of at most @xmath243 quanta before the implementation of each phase gate . this leads to errors in the ms gate of @xmath244 due to the finite excitation of the modes directly involved in the gate and @xmath245 due to the fluctuating debye - waller factors of the other modes combined @xcite . intrinsic background heating for the ions results in motional decoherence in the ms gate while the spins and motion of the ions are entangled . we measure a heating rate for a single @xmath29be@xmath30 ion to be @xmath246 to @xmath247 quanta per millisecond in the common axial mode at a confinement frequency of @xmath248 . however , the motional modes used for the ms gate have only a small component of the center - of - mass motion . conditions here are essentially the same as those of a previous experiment @xcite , and imply a contribution of less than @xmath249 to the phase gate s error . to perform the phase gate , the optical path lengths of the two non - copropagating beams should ideally remain constant throughout the gate s three - pulse sequence , whose duration is @xmath164 . measurements of the optical beat - note between the non - copropagating raman beams give a linewidth of order @xmath250 ; from this we estimate that optical - path - length fluctuations result in a phase - gate error of the order of @xmath249 , assuming the relative phase fluctuations at the beat - note detector are the same as those experienced by the ions . the qubit coherence time and thus the benchmark error probabilities are affected by fluctuations in the magnetic field and its gradients , which cause differential frequency shifts in the qubits . experimentally we determine a qubit coherence time by measuring the decay of the contrast in a two - pulse ramsey experiment as a function of the duration between pulses . we find a coherence time of @xmath251 . due to the resolution limit of our frequency synthesizer , which serves as a clock to keep track of the qubit phases during the randomized benchmarking experiments , we are systematically detuned from the qubit frequency by @xmath252 . the error due to this frequency offset should be negligible , given the typical time required for a clifford unitary of a few milliseconds . we measure a magnetic field difference between trap zones a and b of @xmath253 , which leads to a systematic frequency difference between the qubits of a few millihertz depending on the exact value of the magnetic - field . as an independent check on phase - gate fidelity , we measured the state fidelity for a bell state created by use of the phase gate @xmath20 , as was done in refs . such measurements were performed before and after the randomized benchmarking data was taken . before the benchmark we determined a bell state fidelity of @xmath254 . after the benchmark we obtained @xmath255 . these fidelities include errors due to imperfect state initialization , detection , and three carrier @xmath103-pulses using the co - propagating beams that are needed to prepare and analyze the state . a measurement of the state fidelity where the bell state was prepared using only the ms gate and analyzed with a single non - copropagating carrier @xmath103-pulse , thereby removing one non - copropagating carrier pulse and three co - propagating carrier pulses from the measurement , gave @xmath256 . the fidelities are consistent with the epg determined by the benchmark . errors for one - qubit gates implemented with copropagating beams are likely dominated by changes in the rabi rate @xcite . an indication of whether or not long - term drifts may have affected the two - qubit benchmark can be obtained by comparing two one - qubit benchmarks . the first was run immediately after the two - qubit benchmark , without recalibrating . the second followed recalibration of the relevant pulses and is the one that we reported above . the first one found epgs @xmath257 and @xmath258 for the two qubits , respectively , suggesting that at least the second qubit s gates may have been in need of recalibration by the end of the two - qubit benchmarks . error can also be caused by loss of ions due to background gas collisions . we checked for loss of ions after each sequence , and if an ion - loss event was detected , we removed the previous sequence from the data set . we observed a significant increase in the rate of ion loss events for sequences involving more than @xmath259 ion separation / recombination processes one such process is needed for each phase gate performed . this limited the maximum length of the sequences used in the randomized benchmarking . the reason for the increase in the rate of ion loss is not understood . in summary , the errors discussed in the previous paragraphs amount to a phase gate epg of about @xmath260 ( linearized , incoherent error addition ) to be compared to the benchmark - determined epg of @xmath261 . the epg estimate of @xmath260 includes @xmath238 for spontaneous emission , @xmath239 for calibration imperfections , @xmath262 for intensity fluctuations , @xmath263 for ion motion , @xmath264 for motional heating , and @xmath237 for optical path length fluctuations . the fact that our measured error is greater than our estimated error based on known physical sources suggests that our model of errors for the phase gate is incomplete . clifford benchmarks as defined above can serve as a platform - independent strategy for comparing the quality of quantum operations in a computational context . for @xmath38 qubits , the expressions for the epos and epgs of eq . and are generalized as follows : @xmath265 where @xmath266 is a gate being characterized by insertion after each step . these equations can be established under idealizing assumptions in the same way as eqs . [ eq1 ] and [ eq4 ] , after observing that for @xmath38 qubits , the probability that the sequence does not depolarize is @xmath267 . the comparison on the basis of epos and epgs obtained from the length - dependent loss of fidelity makes sense in the absence of significant deviations from the simple exponential - decay model . the thus - measured epgs can be meaningfully interpreted as true average gate errors in the idealized case where the errors are independent , depolarizing and stationary . in general , the connection to the actual error - behavior of elementary gates is not well understood @xcite . nevertheless , we believe that sufficiently small randomized - benchmark - determined epgs are a good indication of gate quality that can be compared to a general - purpose fault - tolerance threshold goal such as the often - mentioned @xmath7 . in the above , we have considered how to take into account the indications for non - exponential decay in our data . generally , if there is clear evidence of non - exponential decay , the behavior and range of observed epos and epgs and the extent to which stationary behavior was achieved need to be discussed . given sufficiently many sequence lengths , these ranges can be determined by considering different - length intervals . initial transients in error behavior may be expected even in the case where stationary behavior is achieved for longer sequences and can be analyzed separately . the translation of a given clifford unitary into a circuit of elementary gates suitable for a given platform is up to the experimenter , so some improvements are possible by greater efficiency of the translation rather than higher quality gates . we consider such `` software '' improvements to be potentially as useful as strictly `` hardware''-based ones . furthermore they are usually easier for others to implement . however , we believe that for small numbers of qubits , such circuit translations are already sufficiently close to optimal for software improvements of this sort to be self - limiting . the individual gate benchmarks implemented by inserting specific gates after the clifford unitaries can show directly how much the gate quality has improved , independent of how the clifford unitaries have been translated into circuits of elementary gates . when benchmarking @xmath38 qubits , we suggest that the benchmarks are applied to different subsets of the qubits so that comparable epos are obtained for @xmath268 qubits we recommend that such benchmarks be applied in parallel to disjoint subsets , if possible . this solves the problem of comparing new results to earlier ones involving platforms with fewer qubits . nevertheless , it would be helpful to have a way of comparing epos for the clifford benchmark that is independent of the number of qubits . one possibility is to divide the epo by @xmath269 , the average over clifford unitaries of the minimum number of controlled - not gates needed to implement them with a circuit consisting of controlled - not gates and arbitrary one - qubit clifford gates . for two qubits , this normalization factor is determined by @xmath270 , so the normalized clifford epo for our benchmark is @xmath271 . for three qubits we determined @xmath272 rounded at the last digit . since the number of clifford unitaries grows very rapidly with @xmath38 , it may be difficult to determine the normalization factor exactly for @xmath273 . it is known that @xmath269 scales as @xmath69 @xcite . while this complexity may seem relatively large at first , any viable platform must be able to implement circuits of this size , if not much larger . in particular , any successful demonstration of the clifford benchmark also establishes the ability to implement non - trivial circuits for algorithmic purposes . if the primary purpose of the experiment is to benchmark individual gates by inserting them after randomized unitaries , it is desirable to find ways to achieve sufficient randomization that are more efficient than random clifford unitaries . in particular , to exhibit an epg of a gate in this way , it suffices for the random unitaries to approximate a so - called unitary @xmath64-design as explained in @xcite . such approximations are possible with circuits involving a logarithmic multiple of @xmath38 two - qubit gates @xcite . how to best translate these theoretical ideas into a practical benchmark remains to be determined . for the clifford benchmark , the strategy for choosing the final unitary @xmath274 of a sequence given above is more constrained than necessary for ensuring that the measurement outcome is deterministic in the absence of error . this may result in less effective error depolarization . we suggest two alternatives that greatly reduce the constraints on @xmath274 . the first is to choose @xmath274 uniformly at random from all clifford unitaries that ensure that the final state in the absence of error is a logical state . this is equivalent to constructing @xmath274 as an implementation of the inverse of the previous unitaries followed by a random gate that can be decomposed into cnot and pauli product operators . an even more randomizing approach is to choose @xmath274 as suggested in ref . @xcite . in this case , @xmath274 is composed of a uniformly random clifford unitary followed by one - qubit clifford gates randomly chosen to ensure that a randomly chosen joint @xmath104-measurement is deterministic in the absence of errors . by a joint @xmath104-measurement we mean measurement of a product of @xmath82 operators on a subset of the qubits . the product s eigenvalues are @xmath275 and can be determined by multiplying the standard basis measurement outcomes of the qubits in the subset , where a qubit s @xmath276 and @xmath110 measurement outcomes are mapped to @xmath110 , @xmath81 , respectively . this strategy takes advantage of the often much better fidelity of one - qubit gates for the sequence - dependent part of the last unitary . a disadvantage is that instead of @xmath38 deterministic bit values , only one bit value is obtained in each run of the experiment . to fit the resulting data , we use eqs . [ eq1n ] and [ eq4n ] with @xmath277 in summary , we have described a protocol for randomized benchmarking of gates in a quantum information processor and implemented the protocol experimentally on two qubits to measure the error per operation of arbitrary two - qubit clifford unitaries . the protocol we propose is independent of the gate set that is experimentally implemented and so can provide an easily portable method for evaluating the performance of clifford unitaries on different physical platforms . furthermore , with this method it is straightforward to isolate the fidelity of a specific two - qubit gate . we have emphasized some of the consistency checks that can be performed to qualify the reported errors per operation or gate . looking ahead , this randomized benchmarking protocol should prove useful as different experimental implementations of quantum information processors aim to increase the number of qubits and work to decrease the errors towards what is required for fault - tolerance . this work was supported by nsa , iarpa , onr and the nist quantum information program . the authors would like to thank scott glancy and brian sawyer for careful readings of the manuscript . jpg is supported by nist through an nrc fellowship . this paper is a contribution by nist and not subject to u.s . copyright

we present a catalog of 34 diffuse features identified in x - ray images of the galactic center taken with the _ chandra _ x - ray observatory . several of the features have been discussed in the literature previously , including 7 that are associated with a complex of molecular clouds that exhibits fluorescent line emission , 4 that are superimposed on the supernova remnant sgr a east , 2 that are coincident with radio features that are thought to be the shell of another supernova remnant , and one that is thought to be a pulsar wind nebula only a few arcseconds in projection from . however , this leaves 20 features that have not been reported previously . based on the weakness of iron emission in their spectra , we propose that most of them are non - thermal . one long , narrow feature points toward , and so we propose that this feature is a jet of synchrotron - emitting particles ejected from the supermassive black hole . for the others , we show that their sizes ( 0.12 pc in length for=8 kpc ) , x - ray luminosities ( between and , 28 kev ) , and spectra ( power laws with ) are consistent with those of pulsar wind nebulae . based on the star formation rate at the galactic center , we expect that pulsars have formed in the last 300 kyr , and could be producing pulsar wind nebulae . only one of the 19 candidate pulsar wind nebulae is securely detected in an archival radio image of the galactic center ; the remainder have upper limits corresponding to . these radio limits do not strongly constrain their natures , which underscores the need for further multi - wavelength studies of this unprecedented sample of galactic x - ray emitting structures .

0707.1907

the galactic center is host to a high concentration of energetic processes , driven by the interplay between the energetic output of several young star clusters , radiation and outflows powered by accretion onto the supermassive black hole , and dense molecular clouds that have large turbulent velocities . the young stellar population naturally explains the presence of the mixed - morphology supernova remnant sgr a east @xcite , a shell - like radio supernova remnant that exhibits two regions of bright , non - thermal x - ray emission @xcite , and at least two x - ray features that resemble pulsar wind nebula ( e.g. , * ? ? ? * wang , lu , & gotthelf 2006 ) . these processes also produce astronomical phenomena unique to the galactic center . for instance , fluorescent iron emission is produced where molecular clouds are bombarded by hard x - rays from transient sources ( most likely ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ) . finally , numerous milligauss - strength bundles of magnetic fields illuminated by energetic electrons produce radio ( * ? ? ? * ; * ? ? ? * yusef - zadeh , hewitt , & cotton 2004 ) and sometimes x - ray emission ( wang , lu , & lang 2002 ; lu , wang , & lang 2003 ) ; the origin of these non - thermal filaments is under debate . the x - ray emission from the brightest of these features has been described by various authors in the references above . however , whereas at radio wavelengths comprehensive catalogs of hii regions and non - thermal features near the galactic center are available ( e.g. , * ? ? ? * ; * ? ? ? * ; * ? ? ? * ) , no similar catalog has been produced in the x - ray band . here , we remedy this by presenting a catalog of x - ray features identified in images produced from 1 msec of _ chandra _ observations of the central 20 pc of the galaxy . we focus on features that are clearly larger than the _ chandra _ point spread function , but that are smaller than @xmath81.5 ; i.e. , we do not include in our catalog the bulk of the emission from sgr a east @xcite , or the apparent outflow oriented perpendicular to the galactic plane and centered on the sgr a complex ( * ? ? ? * and in prep . ) . lccccc[htp ] 2000 oct 26 18:15:11 & 1561 & 35,705 & 266.41344 & @xmath929.0128 & 265 + 2001 jul 14 01:51:10 & 1561 & 13,504 & 266.41344 & @xmath929.0128 & 265 + 2001 jul 17 14:25:48 & 2284 & 10,625 & 266.40417 & 28.9409 & 284 + 2002 feb 19 14:27:32 & 2951 & 12,370 & 266.41867 & @xmath929.0033 & 91 + 2002 mar 23 12:25:04 & 2952 & 11,859 & 266.41897 & @xmath929.0034 & 88 + 2002 apr 19 10:39:01 & 2953 & 11,632 & 266.41923 & @xmath929.0034 & 85 + 2002 may 07 09:25:07 & 2954 & 12,455 & 266.41938 & @xmath929.0037 & 82 + 2002 may 22 22:59:15 & 2943 & 34,651 & 266.41991 & @xmath929.0041 & 76 + 2002 may 24 11:50:13 & 3663 & 37,959 & 266.41993 & @xmath929.0041 & 76 + 2002 may 25 15:16:03 & 3392 & 166,690 & 266.41992 & @xmath929.0041 & 76 + 2002 may 28 05:34:44 & 3393 & 158,026 & 266.41992 & @xmath929.0041 & 76 + 2002 jun 03 01:24:37 & 3665 & 89,928 & 266.41992 & @xmath929.0041 & 76 + 2003 jun 19 18:28:55 & 3549 & 24,791 & 266.42092 & 29.0105 & 347 + 2004 jul 05 22:33:11 & 4683 & 49,524 & 266.41605 & 29.0124 & 286 + 2004 jul 06 22:29:57 & 4684 & 49,527 & 266.41597 & 29.0124 & 285 + 2004 aug 28 12:03:59 & 5630 & 5,106 & 266.41477 & 29.0121 & 271 + 2005 feb 27 06:26:04 & 6113 & 4,855 & 266.41870 & 29.0035 & 91 + 2005 jul 24 19:58:27 & 5950 & 48,533 & 266.41520 & 29.0122 & 277 + 2005 jul 27 19:08:16 & 5951 & 44,586 & 266.41514 & 29.0122 & 276 + 2005 jul 29 19:51:11 & 5952 & 43,125 & 266.41509 & 29.0122 & 275 + 2005 jul 30 19:51:11 & 5953 & 45,360 & 266.41508 & 29.0122 & 275 + 2005 aug 01 19:54:13 & 5954 & 18,069 & 266.41505 & 29.0122 & 275 the _ chandra _ x - ray observatory has observed the inner @xmath810 of the galaxy with the advanced ccd imaging spectrometer imaging array ( acis - i ; * ? ? ? * ) for a total of 1 msec between 2000 and 2006 ( table [ tab : obs ] ; * ? ? ? * ; * ? ? ? the observations were processed using ciao version 3.3.0 and the calibration database version 3.2.2 . we processed the event lists for each observation by correcting the pulse heights of the events for position - dependent charge - transfer inefficiency , excluding events that did not pass the standard asca grade filters and _ chandra _ x - ray center ( cxc ) good - time filters , and removing 5.7 ks of exposure during which the background rate flared to @xmath10 above the mean level . we then refined the default estimate of the detector positions of each photon received by applying the single - event repositioning algorithm of @xcite . finally , we applied a correction to the absolute astrometry of each pointing in two steps . first , for the deepest observation , we aligned the positions of 23 foreground x - ray sources detected at energies below 2 kev within 5of the aim - point to sources in the 2mass catalog @xcite . second , we registered the remaining observations to the deepest one by comparing the locations of x - ray sources in each image . the resulting astrometric frame should be accurate to 02 . a composite image of the field is displayed with 2resolution in figure [ fig : image ] . we searched for diffuse features using two independently - developed wavelet - based algorithms . for the first , we used the standard ciao tool wavdetect @xcite to search for sources , and identified candidate diffuse features from the resulting list by searching the raw images ( no images were smoothed for this work ) by eye for sources with sizes larger than that of the point spread function . we ran our search using nine composite images , composed of three energy bands ( the full 0.58.0 kev band , the 0.52.0 kev band to increase our sensitivity to foreground sources , and the 48 kev band to increase our sensitivity to highly absorbed sources ) and three spatial resolutions ( primarily for computational efficiency : the full 05 resolution for the inner 1024@xmath111024 pixels , 1resolution covering the inner 2048@xmath112048 detector pixels , and 2 resolution covering the entire image ) . for the purposes of source detection only , we removed events that had been flagged as possible cosmic - ray afterglows . we employed the default `` mexican hat '' wavelet , and used a sensitivity threshold that corresponded to a @xmath12 chance of detecting a spurious source per psf element if the local background is spatially uniform . we expect roughly one spurious source in our field . we used wavelet scales that increased by a factor of @xmath13 : 14 pixels for the 05 image , 18 pixels for the 1 image , and 116 pixels for the 2 image . the final source list was the union of the lists from the three images taken at different pixel scales and energies . sources were identified as duplicates if the they fell within the average radius of the 90% encircled - energy contour for the point spread function ( psf ) for 4.5 kev photons at the position of the source . when duplicate sources were found in the list , we retained the positions from the list derived from the highest - resolution images , either the finest pixel scale or the lowest energy band . among the list of 2933 sources identified with wavdetect , we selected 73 candidate diffuse features . for the second search , we used the wavelet reconstruction algorithm of @xcite . the algorithm searched for features in the 0.58.0 kev band image that were significant at the @xmath143@xmath15 level . the results of the decomposition were used to create an image of the field , and then a multi - scale vision model was used to extract the structural parameters of both point - like and extended sources . we defined as extended those sources that either ( 1 ) were clearly anisotropic , for which the ratio of the major to minor axes were greater than 1.8 , or ( 2 ) appeared extended , in that their major and minor dimensions were at least 3@xmath15 larger than the mean values at that offset from . in total , the algorithm tabulated 2709 objects , of which we identified 91 as anisotropic and 80 as unusually large , for a total of 171 extended sources . we then compared the lists of candidate sources , and found that only 36 objects were identified using both wavelet analyses . all of the sources that were only in one of the two lists were faint , and visual inspection suggested that many were likely to be close alignments of point sources . we examined this hypothesis by extracting the average spectra of these faint features ( the technique is described in 2.2 ) . we found that the average spectra could be described as a @xmath16 power law absorbed by a column density of @xmath17 @xmath18 , along with line emission at 6.7 kev from he - like fe with an equivalent width of 220 ev . this spectrum is similar to those of faint point - like sources in @xcite , which supports our hypothesis that these are not truly extended features . at the same time , at least one feature previously identified as extended is not identified as such by either algorithm cxogc 174545.5285829 , the candidate neutron star suggested to be associated with sgr a east by @xcite . our difficulty in automatically identifying this source as diffuse probably stems from the faintness of the extended portion of its emission . therefore , we caution that our algorithm for identifying extended features is certainly ( 1 ) not complete , in the sense that it could miss extended features , and ( 2 ) contains spurious features that are chance alignments of point sources . we have mitigated the second problem by examining only the 34 sources that are found through both wavelet - based algorithms . the remaining 172 objects will be flagged as being `` possibly extended '' in an updated catalog of the point - like sources that we are preparing . in figures [ fig : image][fig : sub_med ] , we display images with the 34 diffuse features marked . in figure [ fig : image ] , we display the entire field , but for readability we only mark the locations of features with offsets of at least 6 from , and plot boxes around portions of the field closer to the aim point that contain diffuse features . these sub - regions are displayed in figures [ fig : sub_small ] and [ fig : sub_med ] . the clustering of sources in the center of the image results mostly from the fact that the _ chandra _ psf is smallest near the aim point , which provides us the ability to identify @xmath81 features as extended . in order to compute the number of x - ray events received from each source , we drew by hand polygonal or elliptical regions enclosing each of the candidate diffuse features . the regions are plotted in figures [ fig : image][fig : sub_med ] . we estimated the background counts in each region using surrounding annular regions that contained @xmath81000 total counts , after excluding both the diffuse features and circles enclosing 92% of photons from each point source ( regions excluding larger fractions of the psf would cover the inner part of the image , leaving no events suitable for estimating a local background ) . we then computed approximate photon fluxes by dividing the net counts received in each of four energy bands 0.52.0 , 2.03.3 , 3.34.7 , and 4.78.0 kev by the mean effective area derived using the ciao tool mkacisarf . these energy bands were chosen to contain roughly equal numbers of counts , and so that the effective area in each energy range was roughly constant @xcite . in table [ tab : sources ] we report the mean location , net counts , and estimated photon fluxes for each source . to estimate the intrinsic slopes of the spectra and the absorption toward each source , we computed two hardness ratios . following @xcite , the ratios were defined as the fractional difference between two energy bands , @xmath19 , where @xmath20 is the number of counts in the hard band , and @xmath21 is the number of counts in the soft . the soft color was calculated using @xmath20 as the 2.03.3 kev band and @xmath21 as the 0.52.0 kev band , and was most sensitive to the absorption toward each source . the hard color was calculated using @xmath20 as the 4.78.0 kev band , and @xmath21 as the 3.34.7 kev band , and provided a good measure of the intrinsic hardness of a source @xcite . we list the hardness ratios in table [ tab : sources ] , and plot them as a function of the 0.58.0 kev photon fluxes in figure [ fig : hit ] . we also plot the hardness ratios and fluxes that we would expect for a variety of spectral shapes and luminosities and a single absorption column of @xmath22 @xmath23 of gas and dust , using the thick black lines . we find that all but one of diffuse sources have soft colors @xmath24@xmath25 , which implies that they are absorbed by @xmath26 @xmath18 of interstellar medium . this suggests that these sources lie near or beyond the galactic center . one source , g359.9770.076 , has a soft color of only @xmath27 , which implies that it is less absorbed , and probably lies within a few kpc of earth . the luminosities of these sources , assuming they lie at @xmath0@xmath288 kpc , are between @xmath1 and @xmath2 , and in general their hard colors imply that their spectra are consistent with either a power law with photon index @xmath29 , or a @xmath30@xmath243 kev thermal plasma . for comparison , in figure [ fig : hit ] we have also plotted contours denoting the locations of the point - like sources in this diagram @xcite . we find that the diffuse sources are slightly softer , with median hard color of 0.1 , compared to 0.2 for the point sources . we then modeled the spectra of the brightest sources in more detail . spectra were created by binning the source and background events over their pulse - heights . the background - subtracted spectra were then grouped such that each bin had a signal - to - noise ratio of at least 5 . we found that because of their extended nature and the high background in the image , sources required @xmath8400 net counts in order to provide 4 independent spectral bins , which is required for us to constrain independently the interstellar absorption and a two - parameter continuum model . for those sources with enough signal , the observed spectra were compared with models of an absorbed power law using the chi - squared minimization implemented in xspec version 12.2.0 @xcite . the efficiency and energy resolution of the detector were characterized using the above effective - area function and the the energy response computed using mkacisrmf . representative spectra of four sources are displayed in figure [ fig : spec ] . most of the sources could be adequately described by these simple absorbed continuum models , but this was largely a result of the low - signal to noise . the spectral parameters are consistent with those that would be derived from figure [ fig : hit ] . however , the models did not adequately describe several of the sources with @xmath311000 net counts , and visual inspection revealed that these exhibited strong emission from low - ionization fe at 6.4 kev ( see also * ? ? ? * ) . therefore , for all sources for which the spectra contained at least 5 independent spectral bins , we attempted to model the spectra as the sum of a power law plus gaussian line emission at 6.4 kev , absorbed by the interstellar medium . if the addition of a line to the model resulted in a decrease in reduced chi - squared , we tested the significance of the added line using markov chain monte carlo simulations of an @xmath32 test , as described in arabadjis , bautz , & arabadjis ( 2004 ) and @xcite . the @xmath32 value is given by @xmath33 where @xmath34 and @xmath35 are the values of chi - squared and the number of degrees of freedom for the model without the line , and @xmath36 and @xmath37 are the same values for the model with a line . we simulated a set of 1001000 spectra with parameters that were consistent with the measured parameters of the source ( the markov chain ) , and computed a theoretical distribution of the @xmath32 value under the assumption that no line was present . if the observed value of @xmath32 exceeded the theoretical distribution under the null hypothesis in all of 1001000 trials , we considered a line to be detected with @xmath2499%99.9% confidence . in table [ tab : spec ] , we list the parameters of the best - fit absorbed power law for each source we modeled , and either the measured value of the equivalent width of the 6.4 kev iron line or an upper limit to that value . we performed a similar search for the 6.7 kev line from the he - like 21 transition of fe , and only found one source with such a line , g359.9420.045 ( source 2 in table [ tab : sources ] ) . the upper limits to the equivalent widths of the 6.7 kev lines from the other sources were generally 25% higher than those on the 6.4 kev lines , for the cases in which neither line was detected . lcccccccc[htp ] 1 & [email protected] & @xmath38 & @xmath39 & @xmath40 & @xmath4160 & 187.4/175 & 33 & 66 + 2 & [email protected] & @xmath42 & @xmath43 & @xmath44 & @xmath41220 & 102.6/57 & 9 & 16 + 3 & [email protected] & @xmath45 & @xmath46 & @xmath47 & @xmath41470 & 4.3/6 & 3 & 2 + 4 & [email protected] & @xmath48 & @xmath49 & @xmath50 & @xmath41140 & 10.2/12 & 4 & 4 + 6 & [email protected] & @xmath51 & @xmath52 & @xmath53 & & 1.0/1 & 1 & 1 + 7 & [email protected] & @xmath54 & @xmath55 & @xmath56 & @xmath57 & 6.7/9 & 2 & 3 + 8 & [email protected] & @xmath58 & @xmath59 & @xmath60 & @xmath41180 & 6.6/5 & 2 & 2 + 9 & [email protected] & @xmath61 & @xmath62 & @xmath63 & @xmath412060 & 5.6/5 & 2 & 3 + 10 & [email protected] & @xmath64 & @xmath65 & @xmath66 & @xmath4170 & 43.0/53 & 11 & 17 + 11 & [email protected] & @xmath67 & @xmath68 & @xmath69 & & 6.4/3 & 2 & 3 + 13 & [email protected] & @xmath70 & @xmath71 & @xmath72 & & 6.6/2 & 2 & 2 + 14 & [email protected] & @xmath73 & @xmath74 & @xmath75 & @xmath4175 & 6.5/12 & 3 & 6 + 15 & [email protected] & @xmath76 & @xmath77 & @xmath78 & @xmath41130 & 32.8/22 & 6 & 10 + 17 & [email protected] & @xmath79 & @xmath80 & @xmath81 & @xmath411300 & 5.0/5 & 1 & 3 + 20 & [email protected] & 2.2 & 1.1 & @xmath82 & & 4.7/3 & 1 & 1 + 22 & [email protected] & @xmath83 & @xmath84 & @xmath85 & @xmath41110 & 5.9/11 & 2 & 6 + 23 & [email protected] & @xmath86 & @xmath87 & @xmath88 & & 4.4/3 & 1 & 3 + 24 & [email protected] & @xmath89 & @xmath90 & @xmath91 & & 0.1/1 & 1 & 2 + 25 & [email protected] & @xmath92 & @xmath93 & @xmath94 & @xmath4130 & 86.9/95 & 26 & 70 + 26 & [email protected] & @xmath95 & @xmath96 & @xmath97 & @xmath98 & 11.6/14 & 8 & 11 + 28 & [email protected] & @xmath99 & @xmath100 & @xmath101 & @xmath102 & 15.4/11 & 7 & 9 + 29 & [email protected] & @xmath103 & @xmath104 & @xmath105 & @xmath41110 & 26.6/27 & 8 & 10 + 31 & [email protected] & @xmath106 & @xmath107 & @xmath108 & @xmath109 & 28.8/28 & 11 & 13 + 32 & g0.062 + 0.010 & @xmath110 & @xmath111 & @xmath112 & @xmath113 & 22.8/11 & 6 & 9 + 33 & [email protected] & @xmath114 & @xmath115 & @xmath116 & @xmath117 & 135.4/50 & 48 & 100 finally , for the 20 faintest features , we extracted their average spectrum ( fig . [ fig : ave ] ) . the background was computed from an elliptical region enclosing all of the sources , but excluding the sgr a complex . the effective area function was computed using mkwarf , and the response function using mkacisrmf . we found that the spectrum could be modeled as a @xmath118 power law absorbed by @xmath119 @xmath18 , along with a 6.7 kev iron line with an equivalent width of 150@xmath120 ev ( @xmath121 . the spectrum of the emission from these diffuse features is sightly steeper than that of the faint point sources ( with @xmath122 ) , and the he - like fe emission is weaker ( 400@xmath12360 ev for point sources * ? ? ? * ) . this suggests that some , but not all , of these faint features are produced by chance alignments of point sources . alternatively , the presence of the he - like fe line could be explained if a diffuse thermal plasma contributes some of the flux . we therefore also modeled the emission as a non - equilibrium - ionization plasma , and found a best - fit temperature of @xmath124 kev , metal abundances of @xmath125 solar , an ionization time scale @xmath126 s @xmath127 , and an absorption column of @xmath128 @xmath18 . we examine whether it is reasonable to presume these diffuse features are composed of hot plasma in 3 . several diffuse x - ray features elsewhere in the galactic center region have been associated with radio sources @xcite . therefore , we have examined an archival image of the galactic center taken at 6 cm with the very large array @xcite in order to search for counterparts to the x - ray features . the image is displayed in figure [ fig : radiob ] , with the regions from figures 12 overlaid . the beam size was 34 by 29 . to compute fluxes for each feature contained within the radio image , we integrated the source flux over each feature and the background flux from an annular region around the feature , and subtracted the two . the background region had a minimum radius of 5 for the most compact features , and was 50% bigger than the radius of the source region for the larger features . as an estimate of the uncertainty in the flux measurement , we computed the standard deviation of the flux from the source region . the fluxes and standard deviations are listed in table [ tab : sources ] . using the image in figure [ fig : radiob ] , we obtained upper limits to the radio fluxes for the majority of the sources . the limits ranged from a few mjy at an offset of @xmath311 from , to tens of mjy for objects located within the bright radio emission of the sgr a complex . only three sources have radio fluxes measured above the 5@xmath15 level : g359.9500.043 ( source 4 in tab [ tab : sources ] ) , g359.8990.065 ( source 23 in tab . [ tab : sources ] , and source f in @xcite ) , and g359.9590.027 ( source 14 in tab . [ tab : sources ] ) . of these , we believe that the association between the radio and x - ray emission is physical only for the latter two sources . g359.9500.043 ( source 4 ) lies on the `` minispiral , '' a feature with prominent hydrogen recombination emission in the optical and radio , and that is therefore not hot enough to be an x - ray source on its own ( @xmath129 ev ; * ? ? ? * ) . two regions ( g0.021 - 0.051 and g0.0320.056 ) with offsets @xmath244.5 that fell within the image contained negative artifacts left by the cleaning algorithm , so we could not compute fluxes for them . better measurements and limits would require a dedicated re - analysis of the archival vla observations , which is beyond the scope of this paper . we can roughly divide the diffuse features in the galactic center into two groups , based on their spectral properties and sizes . three of the four features with areas larger than 100 arcsec@xmath130 exhibit strong fluorescent iron emission with equivalent widths @xmath24500 ev ( the exception is g359.889 - 0.081 , which we will return to shortly ) . these features are discussed extensively in @xcite and @xcite , and are produced because hard x - rays ( @xmath247 kev ) from a transient source are being scattered by molecular gas , and causing iron to fluoresce . these features all lie in the northeast of the field . in fact , this entire quadrant of the image exhibits unusually strong diffuse fe k-@xmath131 emission ( equivalent width 570 ev ; * ? ? ? * ) . the features that we identify in this paper are simply the brightest manifestations of this pervasive scattered and fluorescent emission , and they probably trace regions of molecular gas with the highest optical depth . it is notable , however , that similar features are not seen toward other clouds with similarly high optical depth , particularly those with velocities of @xmath13220 km s@xmath133 and @xmath13240 km s@xmath133 just to the east of ( e.g. , gsten , walmsley , & pauls 1981 ; oka 1998 ) . the clouds that do not exhibit fluorescent emission are slightly in the foreground of the field , as they are seen to cast a shadow in the number of x - ray sources along their lines of sight @xcite . the clouds without associated x - ray emission must not intersect the light - front of the transient that illuminated the fluorescent features . the features with surface area @xmath41100 arcsec@xmath134 either exhibit no evidence for iron emission at 6.4 kev , or are too faint for such emission to be detected . based on their spectra , the x - ray emission from these features must either originate from a hot thermal plasma ( @xmath135 kev ; fig . [ fig : hit ] ) , or from non - thermal processes such as synchrotron or inverse - compton scattering . as we describe below , previous work on individual features suggests that they have a variety of origins . this is the first paper to provide a comprehensive compilation of their properties . for brevity , we will refer to the features without obvious fluorescent iron emission as `` streaks . '' below , we speculate about their natures . the possibility that some of the features are thermal is suggested by the detection of the 6.7 kev line from he - like iron in the combined spectra of the faint diffuse features ( fig . [ fig : ave ] ) . however , this hypothesis requires that some mechanism must be found to confine the plasma . for a characteristic temperature of @xmath87 kev , and a luminosity of @xmath136 ( 28 kev ) , the emission measure is @xmath137 @xmath127 , where @xmath138 is the electron density , @xmath139 is the hydrogen density , and @xmath140 is the volume of the emitting region . the sizes of these features are typically @xmath141 arcsec@xmath130 , which for @xmath0@xmath288 kpc corresponds to projected areas of @xmath142 pc@xmath130 . if we assume that the volume is given by @xmath143 , then the electron density of the plasma is @xmath144 @xmath127 , where @xmath145 is normalized to 50 arcsec@xmath130 , and @xmath146 to @xmath147 ( 28 kev ) . therefore , the pressure is @xmath148 erg @xmath127 . this pressure is @xmath8300 times larger than that of the diffuse , @xmath149 kev plasma in the galactic center ( @xmath150 erg @xmath127 ) , and 50 times larger than the putative 8 kev plasma there ( @xmath151 erg @xmath127 ; * ? ? ? magnetic fields would have to be strong to confine this plasma , @xmath152 mg . these fields strengths are thought to be present within radio - emitting filaments toward the galactic center @xcite , but the average field at the galactic center may be significantly lower ( e.g. , * ? ? ? thermal features could also be produced by shocks with velocities of @xmath4@xmath153=1200 km s@xmath133 ( where @xmath154 is the proton mass and @xmath155=0.5 for a plasma of electrons and protons ) . however , there are no obvious candidates for the sources of the energy powering these putative shocks ; they are not arranged as part of a shell as would be expected if they originate from a supernova ( and the lack of metal lines would be puzzling ) , and the known massive stars in the region are mostly located within 30 of sgr a * , which is several parsecs in projection from these features ( e.g. , * ? ? ? * ; * ? ? ? . given the problems with assuming that these features are thermal sources , we follow other authors and suggest that the majority of these features are non - thermal . a few of the x - ray streaks appear to be non - thermal particles accelerated in shocks where supernova remnants impact the interstellar medium . it has been noted by several authors that g359.8890.081 and g359.8990.065 ( sources 25 and 23 in table [ tab : sources ] ) lie on part of a shell - like feature seen in radio maps @xcite . these features are coincident with the brightest parts of the radio - emitting shell , which are referred to as e and f by ( * ? ? ? * and fig . [ fig : radiob ] ) . we also find four filamentary features associated with the sgr a east supernova remnant , g359.9620.062 , g359.9650.056 , g359.9640.053 , and g359.9560.052 ( sources 6 , 10 , 11 , and 13 in table [ tab : sources ] ) . of these features associated with supernova remnants , we can put strict upper limits to the equivalent width of iron emission for g359.956 - 0.052 and g359.8890.081 , @xmath4170 and @xmath4130 ev , respectively , which strongly suggests that they are non - thermal . the x - rays are probably synchrotron emission from tev electrons @xcite , which could occur in parts of remnants where the shock velocity is high and the ambient density is low , so that particles are efficiently accelerated ( e.g. , * ? ? ? supernova shocks therefore can explain 5 of the 26 streaks . at least one non - thermal radio filament has been identified with x - ray emission in the central 100 pc of the galaxy , but outside of our deep images of the central 20 pc : g 359.54 + 0.18 @xcite . this confirmed non - thermal filament produces a flux density of 150 mjy at 6 cm over the region that also emits x - rays . assuming an @xmath156 spectrum , the estimated radio luminosity of g 359.54 + 0.18 between @xmath157 hz and @xmath158 hz is @xmath159 , compared to an x - ray luminosity of @xmath160 ( 0.210 kev * ? ? ? * ) . in our image , however , only one diffuse feature that is not associated with a supernova remnant has a firm radio counterpart , g359.9590.027 ( source 14 in table [ tab : sources ] ) . its x - ray luminosity is @xmath161 ( 28 kev ) . in fig . [ fig : radiob ] @xcite , the feature has a flux density of @xmath162 mjy , which for the same assumed radio spectrum as above implies @xmath163 . a second nearby feature with a similar size , spectrum , and even orientation , g359.9700.008 ( figure [ fig : sub_small ] , panel 5 ; source 22 in table [ tab : sources ] ) , also has a marginal radio counterpart with a flux of @xmath164 mjy . for this source we find @xmath165 ( 28 kev ) and @xmath166 . therefore , a couple of the streaks could be counterparts to non - thermal radio filaments , although the relative amount of radio and x - ray flux appears highly variable . the final conventional explanation for the x - ray streaks is that they could be pulsar wind nebulae . two candidate pulsar wind nebulae in this field have been discussed in the literature . @xcite have demonstrated that the spectral and morphological properties of g359.9450.044 resemble a pulsar wind nebula , and have suggested that it is the source of the tev emission that originates near @xcite . this source is included in our catalog . @xcite have identified the point - source cxogc j174545.5285829 as another candidate pulsar based on its featureless non - thermal spectrum , its lack of long - term variability , and the detection of a faint extended tail that may represent a bow shock formed by the wind nebula . this object is not in our catalog , because the faint nebulosity was not identified by our wavelet algorithms ( see [ sec : id ] ) . from this new work , one bright feature has a morphology that makes it an especially attractive candidate to be a pulsar wind nebulae : g 0.0320.056 ( source 29 in table [ tab : sources ] ) , which has a bright head producing about half its flux , and a long tail that curves to the northwest . pulsar wind nebulae are formed when a wind of particles that is powered by the rotational energy of the neutron star encounters the ism . the resulting shock accelerates the particles further , causing them to radiate synchrotron emission ( see * ? ? ? * for a review ) . these wind nebulae have diverse morphologies , which depend upon how the pulsar wind interacts with the surrounding supernova remnant , how far the pulsar travels during the lifetime of the electrons in the wind nebula , and whether the pulsar is moving supersonically with respect to the ism . in the galactic center , few of the pulsars are likely to be moving highly supersonically . most of the region is probably suffused with hot @xmath167 kev plasma with densities of @xmath168 @xmath127 , which implies that the sound speed of the ism is @xmath8500 km s@xmath133 @xcite . in contrast , the mean three - dimensional velocities of known radio pulsars are @xmath8400 km s@xmath133 ( e.g. , * ? ? ? * ; * ? ? ? * ) , and only 2 of 169 pulsars in @xcite have measured transverse velocities @xmath24800 km s@xmath133 , which , assuming the velocities are isotropic , would translate to expected three - dimensional velocities @xmath241000 km s@xmath133 . pulsars will only be highly supersonic ( mach numbers @xmath242 ) in regions of the ism that are dense and cool , such as near molecular clouds . therefore , we would not expect bow - shock nebulae to be common ( although see * ? ? ? nonetheless , the complex morphologies we observe could be explained by interactions between wind nebulae and unseen supernova remnants ( e.g. , * ? ? ? * ) , or if the proper motion of the pulsar moves the particle acceleration region faster than the electrons can cool ( e.g. , sn g371.1 - 1.1 in * ? ? ? the characteristic sizes of known pulsar nebulae are between @xmath169 and @xmath170 cm ( cheng , taam , & wang 2004 ) , which at the galactic center ( @xmath0=8 kpc ) would have angular sizes of 0880 . this bounds the sizes of the features we observe . moreover , the x - ray luminosities of the diffuse features in the galactic center are consistent with those of known pulsar wind nebulae . galactic pulsar wind nebulae have x - ray luminosities that are as large as @xmath4@xmath171 ( 210 kev ) ; typically , @xmath4@xmath172 of the spin - down energy @xmath173 is emitted as x - rays ( e.g. , * ? ? ? * ; * ? ? ? * ; * ? ? ? the diffuse features in our survey all have luminosities of @xmath174@xmath2 , which would imply spin - down luminosities of @xmath175 . there are @xmath886 known pulsars with @xmath173 in this range . most of these energetic pulsars have ages @xmath6@xmath176 years , although 7 of @xmath4100 known recycled millisecond pulsars have @xmath173 this high @xcite . many galactic pulsar wind nebulae are luminous at radio wavelengths , and yet we find that most of our x - ray sources do not have radio counterparts . the limits to their flux densities at 6 cm are @xmath63 mjy ( tab . [ tab : sources ] and fig . [ fig : radiob ] ) . if we assume a standard pulsar wind spectrum with logarithmic slope @xmath177 between @xmath157 and @xmath158 hz , our typical limit of 3 mjy corresponds to @xmath178 , or between @xmath179 and @xmath172 of the x - ray luminosity . this would translate to @xmath180 . for comparison , the youngest , most energetic galactic pulsars are detected with @xmath181 . however , for @xmath182 a few pulsars have @xmath183 @xcite : of @xmath830 pulsars that have been reasonably well - studied at x - ray @xcite and radio @xcite wavelengths , five stand out as having wind nebulae with @xmath184 and @xmath185 : psr b104658 @xcite , psr b182313 @xcite , psr b0355 + 54 @xcite , psr b1706 - 44 @xcite , and possibly psr j11056107 @xcite . the sources in our sample that are either bright in x - rays or that have strict limits to their radio fluxes would have to be analogous to these radio - faint pulsars . given that the lack of radio detections can be reconciled plausibly with the known properties of pulsars , we next examine whether we should expect enough energetic pulsars in the galactic center to account for the number of diffuse features that we observe . @xcite and @xcite have considered whether recycled millisecond pulsars in the galactic center could contribute to these x - ray features . based on the inferred population of galactic millisecond pulsars @xcite , they estimate that @xmath4200 could be present in our survey region . if , based on the catalog of @xcite , @xmath410% of the millisecond pulsars have @xmath186 , they also could contribute @xmath420 objects to the diffuse features at the galactic center . these pulsars will generally be traveling rather slow ( @xmath187@xmath4130 km s@xmath133 * ? ? ? * ) , relative to the sound speed of the @xmath80.8 kev plasma that takes up most of the volume of the ism at the galactic center @xcite , but they could form extended nebulae if they encounter regions of the ism with relatively low sound velocities ( @[email protected] kev ) . the filling factor of this cooler ism is unknown , but if it is @xmath3110% , a handful of millisecond pulsars could produce extended diffuse features . most of the known energetic pulsars with wind nebulas are young , and still interacting with their supernova remnants . we concentrate on these as the best candidates for the diffuse x - ray features . @xcite modeled the infrared color - magnitude diagrams of stars in several fields toward the galactic center , and , assuming an initial mass function ( imf ) of the form @xmath188 for stars with masses between 0.1 @xmath189 and 120@xmath189 , concluded that the star formation rate in the inner 30 pc of the galaxy is 0.02 @xmath189yr@xmath133 . if we assume that radio pulsars form from stars with initial masses between 8 and 30 @xmath189 ( e.g. , * ? ? ? * ; * ? ? ? * but see figer 2005 ; muno 2006 ) , then they would form at a rate of @xmath190 yr@xmath133 . however , the supernovae that produce pulsars provide them with kicks , so the loss of pulsars as they move out of the galactic center will also limit the number of pulsar wind nebulae observable there ( e.g. , * ? ? ? yr old located within 0.02 pc of . these have higher escape velocities and would be retained , but they would be faint and confused with and so would not be detected in our x - ray survey . ] there is still debate as to the three - dimensional velocity distribution of pulsars ( e.g. , * ? ? ? * ; * ? ? ? * ; * ? ? ? * ) , so we have adopted a simplified approach to estimate the number of pulsars remaining within the 8 ( 20 pc ) radial bound of our image , based upon the observed tangential velocities for 169 radio pulsars that were tabulated by @xcite . we assume that a pulsar would remain in our image if the product of its tangential velocity @xmath187 and its lifetime @xmath191 is less than @xmath0=20 pc . using the data in @xcite , we can compute the fraction of known pulsars in that sample that have traveled a projected distance @xmath0@xmath4120 pc at a given age @xmath191 , which we call @xmath192 . if pulsars are born at a constant rate @xmath193=@xmath190 yr@xmath133 and are bright pulsar wind nebulae for a time @xmath194 , then the number of nebulae in our image is roughly @xmath195 for @xmath194=100 kyr we find that only pulsars with @xmath187@xmath24200 km s@xmath133 may have time to escape , and that out of 20 pulsars born , roughly 14 would remain within our image . if @xmath194=300 kyr , pulsars with @xmath187>70 km s@xmath133 may escape , so that 29 out of 60 pulsars would remain . in this estimate , we have not considered that some pulsars will be gravitationally bound to the galactic center . if we use the enclosed mass in @xcite , the escape velocity from a radius of 10 pc is only @xmath196 km s@xmath133 . therefore , if @xmath194 is significantly longer than 100 kyr , then the fraction of pulsars retained in the galactic center would be larger than we estimate ( 66% of pulsars have tangential velocities smaller than 200 km s@xmath133 , although many of these will have larger three - dimensional velocities ) . for comparison , if we consider candidate pulsar wind nebulae to be those features that ( 1 ) exhibit no evidence for iron emission , and ( 2 ) are not obviously associated with supernova remnants already detected in the radio , then there are also @xmath420 of them in our image . this is consistent with the numbers we expect if pulsars produce wind nebulae for 100300 kyr after their birth . finally , we note that one of the x - ray features may be a jet produced by , g359.9440.052 ( source 3 in table [ tab : sources ] ) . we highlight this source in particular in figure [ fig : jetonly ] . the object is about 15 long , and is unresolved along its short axis . our hypothesis that it is a jet is motivated by the fact that it position angle points within 2@xmath197 of . from a visual inspection of the rest of our image , we find only one candidate feature that is similarly long and thin ( located at @xmath198= 266.37688 , 29.04517 ) , but this feature is not picked up by the automated wavelet search based on the work of @xcite because it is half as bright as g359.9440.052 . the candidate jet is found by both of our wavelet algorithms , so we believe this feature is unlikely to be an artifact produced by a chance alignment of point sources , or by poisson variations in the diffuse background . the feature has a flat spectrum of slope @xmath199 and an absorption column within 1@xmath15 of the average toward the galactic center , @xmath200 @xmath18 . its luminosity is @xmath201 ( 28 kev ) . the candidate jet is not detected in the radio , with an upper limit of @xmath4145 mjy at 6 cm ( 5 ghz ) . the radio upper limit is not constraining because the feature lies along the line - of - sight toward bright , diffuse radio emission . the relative strength of the radio and x - ray emission from the synchrotron - emitting jets of quasars can be quantified using the logarithmic slope @xmath202 , where @xmath203 and @xmath204 and the x - ray and radio flux densities , and @xmath205 and @xmath206 are the frequencies at which the flux densities are computed . from the values in table [ tab : spec ] , @xmath207 njy at @xmath208 hz ( 1 kev ) . therefore , we can only constrain @xmath209 . for comparison , the quasars in the sample of @xcite have @xmath210 between 0.9 and 1.1 . therefore , one would need radio observations with a sensitivity of @xmath411 mjy to make a useful comparison between the putative jet feature near and quasar jets . further observations are warranted to determine whether g359.9440.052 is a jet of synchrotron - emitting particles that is flowing away from . there are several processes that appear to produce diffuse x - ray features in the galactic center : x - rays scatter off of gas in molecular clouds , and cause them to fluoresce ; supernova shocks produce x - ray emitting , tev electrons when they propagate through low - density ism ; and non - thermal radio filaments produce x - rays through a mechanism that is still not understood . however , we propose that the majority of the diffuse x - ray features are pulsar wind nebulae . a significant number of pulsars are expected at the galactic center , but because of the high dispersion measure along the line of sight , only two have so far been reported there in the radio @xcite . x - ray identifications of pulsar wind nebulae bypass the problem of the dispersion measure , and in addition are sensitive to pulsars that are not beamed toward our line of sight . therefore , the identification of @xmath820 x - ray features that are candidate pulsar wind nebulae is consistent with a basic prediction about the number of massive stars that have recently ended their lives in the galactic center . however , at the moment we lack identificaations of most of these features at other wavelengths , so further multi - wavelength studies are required to confirm our hypotheses . we are grateful to b. gaensler for informative conversations about pulsar wind nebulae . mpm was supported by the national aeronautics and space administration through chandra award number 7910613 issued by the chandra x - ray observatory center , which is operated by the smithsonian astrophysical observatory for and on behalf of the national aeronautics space administration under contract nas8 - 03060 . 0 aharonian , f. 2004 , , 425 , l13 arabadjis , j. s. , bautz , m. w. , & arabadjis , g. 2004 , , 617 , 303 arnaud , k.a . , 1996 , astronomical data analysis software and systems v , eds . jacoby g. and barnes j. , p17 , asp conf . series volume 101 arzoumanian z. , chernoff d. f. , & cordes j. m. , 2002 , apj , 568 , 289 baganoff , f. k. 2003 , , 591 , 891 becker , w. & trmper , j. 1997 , , 326 , 682 bertin , e. & arnouts , s. 1996 , , 117 , 393 brarzier , k. t. s. & johnston , s. 1999 , , 305 , 671 bykov , a. m. 2002 , , 390 , 327 cheng , k. s. , taam , r. e. , & wang , w. 2004 , , 617 , 480 cheng , k. s. , taam , r. e. , & wang , w. 2006 , , 641 , 427 cordes , j. m. & lazio , t. j. w. 1997 , , 475 , 557 faucher - gigure , c .- a . & kaspi , v. m. 2006 , , 643 , 332 figer , d. f. , rich , r. m. , kim , s. s. , morris , m. , & serabyn , e. 2004 , , 610 , 317 frail , d. a. , & scharringhausen , b. r. 1997 , , 480 , 364 freeman , p. e. , kashyap , v. , rosner , r. , & lamb , d. q. 2002 , , 138 , 185 gaensler , b. m. & slane , p. o. 2006 , , 44 , 17 gaensler , b. m. , stappers , b. w. , frail , d. a. , moffat , d. a. , johnston , s. , & chatterjee , s. 2000 , , 318 , 58 gaensler , b. m. , schulz , n. s. , kaspi , v. m. , pivovaroff , m. j. , & becker , w. e. 2003 , , 588 , 441 genzel , r. , thatte , n. , krabbe , a. , kroker , h. , & tacconi - garman , l. e. 1996 , , 472 , 153 gonzales , m. e. , kaspi , v. m. , pivovaroff , m. j. & gaensler , b. m. 2006 , , 652 , 569 gotthelf , e. v. , halpern , j. p. , & dodson , r. 2002 , , 567 , l125 gotthelf , e. v. & kaspi , v. .m . 1998 , , 497 , 429 gsten , r. , walmsley , c. m. , & pauls , t. 1981 , , 103 , 197 heger , a. , fryer , c. l. , woosley , s. e. , langer , n. , & hartmann , d. h. 2003 , , 591 , 288 ho , p. t. p. , jackson , j. m. , barret , a. h. , & armstrong , t. j. 1985 , , 288 , 575 hobbs , g. , lorimer , d. r. , lyne , a. g. , & kramer , m. 2005 , , 360 , 974 johnston , s. , kramer , m. , lorimer , d. r. , lyne , a. g. , mclaughlin , m. , klein , b. , & manchester , r. n. 2006 , , 373 , l6 krabbe , a. 1995 , , 447 , l95 kroupa , p. 2002 , science , 295 , 82 koyama , k. , maeda , y. , sonobe , t. , takeshima , t. , tanaka , y. , & yamauchi , s. 1996 , , 48 , 249 larosa , r. n. , brogan , c. l. , shore , s. m. , lazio , t. j. w. , kassim , n. e. , & nord , m. e. 2005 , , 626 , l23 li , j. , kastner , j. h. , prigozhin , g. y. , schulz , n. s. , feigelson , e. d. , & getman , k. v. 2004 , , 610 , 1204 longair , m. s. 1994 , _ high energy astrophysics _ , volume 2 , 2nd ed . , cambridge university press lu , f. j. , wang , q. d. , & lang , c. c. 2003 , , 126 , 319 lyne , a. g. 1998 , , 295 , 743 maeda , y. 2002 , , 570 , 671 manchester , r. n. , hobbs , g. b. , teoh , a. , & hobbs , m. 2005 , , 129 , 1993 marshall , h. m. 2005 , , 156 , 13 mcgowan , k. e. , vestrand , w. t. , kennea , j. a. , zane , s. , cropper , m. , & crdova , f. a. , 2006 , , 647 , 1300 morris , m. 2003 , ans , 324 , 167a morris , m. & serabyn , e. 1996 , , 34 , 645 muno , m. p. 2003 , , 589 , 225 muno , m. p. 2004a , , 613 , 326 muno , m. p. 2004b , , 613 , 1179 muno , m. p. , baganoff , f. k. , brandt , w. n. , park , s. & morris , m. r. 2007 , , 656 , l69 murakami , h. , koyama , k. , & maeda , y. 2001 , , 558 , 687 murakami , h. , koyama , k. , tsujimoto , m. , maeda , y. , & sakano , m. 2001b , , 550 , 297 nord , m. e. , lazio , t. j. w. , kassim , n. e. , hyman , s. d. , larosa , t. n. , brogan , c. l. , & duric , n. 2004 , , 128 , 1646 oka , t. , hasegawa , t. , hayashi , m. , handa , t. , & sakamoto , s. 1998 , , 493 , 730 park , s. , muno , m. p. , baganoff , f. k. , maeda , y. , morris , m. , howard , c. , bautz , m. w. , & garmire , g. p. 2004 , , 603 , 548 park , s. 2005 , , 631 , 964 paumard , t. 2006 , , 643 , 1011 pfahl , e. & loeb , a. 2004 , , 615 , 253 revnivtsev , m. g. 2004 , , 425 , l49 revnivtsev , m. , vikhlinin , a. , & sazonov , s. 2007 , submitted to , astro - ph/0611952 romani , r. w. , ng , c. y. , dodson , r. , & brisker , w. 2005 , , 631 , 480 rybicki , g. , & lightman , a. _ radiative processes in astrophysics _ , 1979 , john wiley & sons , inc . sakano , m. , warwick , r. s. , decourchelle , a. , & predehl , p. 2003 , , 340 , 747 sakano , m. , warwick , r. s. , decourchelle , a. , & predehl , p. 2004 , , 350 , 129 scoville , n. z. , stolovy , s. r. , reike , m. , christopher , m. , & yusef - zadeh , f. 2003 , , 594 , 294 skrutskie , m. f. 2006 , , 131 , 1163 starck , j .- l . , bijauoi , a. , valtchanov , i. , & murtagh , f. 2000 , , 147 , 139 sunyaev , r. a. , markevitch , m. , & pavlinsky , m. 1993 , , 407 , 606 valinia , a. , tatischeff , v. , arnaud , k. , ebisawa , k. , & ramaty , r. 2000 , , 543 , 733 wang , q. d. , lu , f. j. , & gotthelf , e. v. 2006 , , 367 , 937 wang , q. d. , lu , f. j. , & lang , c. c. 2002b , , 581 , 1148 wang , w. , jiang , z. j. , & cheng , k. s. 2005 , , 358 , 263 weisskopf , m. c. , brinkman , b. , canizares , c. , garmire , g. , murray , s. , & van speybroeck , l. p. 2002 , , 114 , 1 vink , j. , bleeker , j. , van der heyden , k. , bykov , a. , bamba , a. & yamazaki , r. 2006 , submitted to , astro - ph/0607307 yusef - zadeh , f. & morris , m. 1987 , , 320 , 545 yusef - zadeh , f. , wardle , m. , muno , m. , law , c. , & pound , m. 2005 , adspr , 35 , 1074 yusef - zadeh , f. , hewitt , j. w. , & cotton , w. 2004 , , 155 , 421 clccccccccccc 1 & g359.945 - 0.044 & @xmath211 & @xmath212 & 22 & 0.1 & 927 & @xmath213 & @xmath4199.0 & @xmath214 & @xmath215 & 419.6 & @xmath41100 + 2 & g359.942 - 0.045 & @xmath216 & @xmath217 & 28 & 0.1 & 927 & @xmath218 & 10.8 & @xmath219 & @xmath220 & 132.3 & @xmath4130 + 3 & g359.944 - 0.052 & @xmath221 & @xmath222 & 20 & 0.3 & 927 & @xmath223 & 11.1 & @xmath224 & @xmath225 & 29.2 & @xmath4145 + 4 & g359.950 - 0.043 & @xmath226 & @xmath227 & 55 & 0.4 & 927 & @xmath228 & 8.9 & @xmath229 & @xmath230 & 50.6 & 210@xmath12330 + 5 & g359.933 - 0.039 & @xmath231 & @xmath232 & 15 & 0.8 & 927 & @xmath233 & 3.4 & @xmath234 & @xmath235 & 8.1 & @xmath413 + 6 & g359.956 - 0.052 & @xmath236 & @xmath237 & 10 & 0.8 & 927 & @xmath238 & 3.1 & @xmath239 & @xmath240 & 12.7 & @xmath419 + 7 & g359.933 - 0.037 & @xmath241 & @xmath242 & 13 & 0.9 & 927 & @xmath243 & 17.2 & @xmath244 & @xmath245 & 22.8 & @xmath415 + 8 & g359.941 - 0.029 & @xmath246 & @xmath247 & 17 & 1.0 & 927 & @xmath248 & 11.7 & @xmath249 & @xmath250 & 18.7 & @xmath412 + 9 & g359.925 - 0.051 & @xmath251 & @xmath252 & 19 & 1.2 & 927 & @xmath253 & 7.5 & @xmath254 & @xmath255 & 18.1 & @xmath412 + 10 & g359.964 - 0.053 & @xmath256 & @xmath257 & 76 & 1.2 & 927 & @xmath258 & 41.9 & @xmath259 & @xmath260 & 151.1 & 11@xmath1233 + 11 & g359.965 - 0.056 & @xmath261 & @xmath262 & 29 & 1.4 & 927 & @xmath263 & 6.8 & @xmath264 & @xmath265 & 22.4 & @xmath412 + 12 & g359.921 - 0.052 & @xmath266 & @xmath267 & 12 & 1.4 & 927 & @xmath268 & 3.6 & @xmath269 & @xmath270 & 7.5 & @xmath411 + 13 & g359.962 - 0.062 & @xmath271 & @xmath272 & 26 & 1.4 & 927 & @xmath273 & 1.1 & @xmath274 & @xmath275 & 20.3 & 2@xmath1231 + 14 & g359.959 - 0.027 & @xmath276 & @xmath277 & 34 & 1.4 & 927 & @xmath278 & 32.7 & @xmath264 & @xmath279 & 33.6 & 20@xmath1233 + 15 & g359.971 - 0.038 & @xmath280 & @xmath281 & 148 & 1.7 & 927 & @xmath282 & 19.8 & @xmath283 & @xmath284 & 71.4 & @xmath4110 + 16 & g359.969 - 0.033 & @xmath285 & @xmath286 & 17 & 1.7 & 927 & @xmath287 & 5.6 & @xmath288 & @xmath289 & 9.0 & @xmath411 + 17 & g359.921 - 0.030 & @xmath290 & @xmath291 & 30 & 1.7 & 927 & @xmath292 & 12.7 & @xmath293 & @xmath294 & 15.7 & @xmath411 + 18 & g359.915 - 0.061 & @xmath295 & @xmath296 & 22 & 2.0 & 927 & @xmath297 & 15.2 & @xmath298 & @xmath299 & 10.0 & [email protected] + 19 & g359.983 - 0.040 & @xmath300 & @xmath301 & 35 & 2.4 & 927 & @xmath302 & 0.5 & @xmath303 & @xmath304 & 5.4 & @xmath411 + 20 & g359.904 - 0.047 & @xmath305 & @xmath306 & 32 & 2.4 & 927 & @xmath307 & 6.7 & @xmath308 & @xmath309 & 13.2 & @xmath410.2 + 21 & g359.977 - 0.076 & @xmath310 & @xmath311 & 26 & 2.7 & 927 & @xmath312 & 1.6 & @xmath313 & @xmath314 & 6.5 & @xmath414 + 22 & g359.970 - 0.008 & @xmath315 & @xmath316 & 30 & 2.7 & 927 & @xmath317 & 32.6 & @xmath318 & @xmath319 & 28.1 & [email protected] + 23 & g359.899 - 0.065 & @xmath320 & @xmath321 & 30 & 2.9 & 927 & @xmath322 & 10.7 & @xmath323 & @xmath324 & 11.0 & 6@xmath1231 + 24 & g359.897 - 0.023 & @xmath325 & @xmath326 & 42 & 3.2 & 927 & @xmath327 & 2.3 & @xmath328 & @xmath329 & 9.2 & @xmath410.4 + 25 & g359.889 - 0.081 & @xmath330 & @xmath331 & 432 & 3.9 & 927 & @xmath332 & @xmath41 - 99.0 & @xmath333 & @xmath334 & 281.9 & + 26 & g0.014 - 0.054 & @xmath335 & @xmath336 & 260 & 4.2 & 927 & @xmath337 & 28.4 & @xmath338 & @xmath339 & 78.1 & @xmath410.6 + 27 & g0.008 - 0.015 & @xmath340 & @xmath341 & 51 & 4.3 & 927 & @xmath342 & 3.1 & @xmath343 & @xmath96 & 9.2 & [email protected] + 28 & g0.021 - 0.051 & @xmath344 & @xmath345 & 190 & 4.6 & 927 & @xmath346 & 27.8 & @xmath347 & @xmath348 & 69.2 & + 29 & g0.032 - 0.056 & @xmath349 & @xmath350 & 429 & 5.3 & 927 & @xmath351 & 30.8 & @xmath352 & @xmath353 & 100.9 & + 30 & g0.029 - 0.080 & @xmath354 & @xmath355 & 838 & 5.5 & 927 & @xmath356 & 18.3 & @xmath357 & @xmath358 & 73.0 & + 31 & g0.039 - 0.077 & @xmath359 & @xmath360 & 333 & 6.0 & 927 & @xmath361 & 33.5 & @xmath362 & @xmath363 & 118.4 & + 32 & g0.062 + 0.010 & @xmath364 & @xmath365 & 1187 & 7.8 & 927 & @xmath366 & 14.7 & @xmath367 & @xmath368 & 75.9 & + 33 & g0.097 - 0.131 & @xmath369 & @xmath370 & 4181 & 10.5 & 563 & @xmath371 & @xmath41 - 99.0 & @xmath372 & @xmath373 & 529.0 & + 34 & g0.116 - 0.111 & @xmath374 & @xmath375 & 2257 & 11.0 & 574 & @xmath376 & 7.3 & @xmath377 & @xmath378 & 121.6 &

we show that every internal biequivalence in a tricategory is part of a biadjoint biequivalence . we give two applications of this result , one for transporting monoidal structures and one for equipping a monoidal bicategory with invertible objects with a coherent choice of those inverses .

1102.0979

it is common in mathematics to regard two objects @xmath1 and @xmath2 as being the same if there is an isomorphism between them , that is a pair of maps @xmath3 and @xmath4 such that @xmath5 and @xmath6 . more importantly , any specific isomorphism @xmath3 gives an explicit means for transporting information about @xmath1 to information about @xmath2 . the choice of ambient category plays a very important role in this process , especially when we examine two categories with the same objects but different morphisms . for example , first take the category of cw - complexes and continuous maps . in this category , an isomorphism between @xmath1 and @xmath2 gives a formula for transporting cell structures from @xmath1 to @xmath2 . if we now take the category of cw - complexes and homotopy classes of continuous maps , then isomorphisms are now homotopy equivalences . an isomorphism then no longer gives a recipe for transporting cell structures , but instead only gives a recipe for transporting homotopical information like homotopy and homology groups . an important property of an isomorphism @xmath3 is that it uniquely determines the inverse @xmath4 by the formulas @xmath5 and @xmath6 ; the proof is exactly the same as that showing that any group element has a unique inverse . thus we have two concepts which are _ a priori _ different : * the property that @xmath3 is an isomorphism , and * the structure consisting of a pair @xmath7 of morphisms @xmath3 , @xmath4 such that @xmath5 and @xmath6 . we have then uncovered that a morphism @xmath8 has the property of being an isomorphism if and only if there is a pair @xmath7 with the isomorphism structure , and that moreover the pair @xmath7 is uniquely determined by @xmath8 alone . this is an example of a structure ( being part of an isomorphism pair ) that is determined in a unique way by a property ( the existence of an inverse ) . moving up to the case in which the objects of study are now the 0-cells of some 2-category ( or more generally , some bicategory ) , there are now more possible notions of sameness . while we can ask if two 0-cells of a 2-category are isomorphic , it is much more common to ask if they are equivalent . the canonical example of a 2-category is that of categories , functors , and natural transformations , and in this 2-category we see from experience that equivalence is the natural notion of sameness . a functor @xmath9 is often defined to be an equivalence if it is essentially surjective , full , and faithful . it is then shown that a functor @xmath10 is an equivalence if and only if there exists a functor @xmath11 such that the composites @xmath12 are naturally isomorphic to the identity functors @xmath13 , respectively . this definition identifies a property of a functor @xmath10 that allows us to conclude that two categories @xmath1 and @xmath2 are , in some sense , the same . on the other hand , we can make the definition of the structure of adjoint equivalence @xmath14 which consists of a functor @xmath9 , a functor @xmath11 , and two natural isomorphisms @xmath15 such that the following two diagrams commute . @xmath16 it is clear that if @xmath14 is an adjoint equivalence , then @xmath10 is an equivalence . it is also well - known @xcite that every equivalence @xmath10 can be completed to an adjoint equivalence . now @xmath10 no longer determines @xmath17 uniquely , but instead only determines them up to a unique isomorphism preserving the adjoint equivalence structure . we refer the reader to the paper @xcite for a general discussion of this phenomenon in the setting of algebras for a 2-monad . the aim of this paper is to establish an analogous result in three - dimensional category theory . we define two different notions of sameness internal to a tricategory , one a property of a 1-cell and the other a structure involving 1- , 2- , and 3-cells satisfying certain axioms . these two notions are that of a 1-cell having the property `` is a biequivalence '' on the one hand , and the structure of a biadjoint biequivalence on the other hand . our main result is that , in any tricategory @xmath0 , every 1-cell which is a biequivalence is part of a biadjoint biequivalence . the proof of this result proceeds in three steps . first , we show that it is true for the special case when @xmath18 , the tricategory of bicategories , functors , pseudo - natural transformations , and modifications . second , we prove a result about transporting biadjoint biequivalences ; more precisely , we show that if a functor @xmath19 satisfies a kind of local embedding condition and @xmath0 has the property that every biequivalence is part of a biadjoint biequivalence , then @xmath20 has that property as well . finally , we prove that this property is inherited by functor tricategories from the target , so that if @xmath0 is a tricategory in which every biequivalence is part of a biadjoint biequivalence then the same holds for the functor tricategory @xmath21 for any @xmath20 . the main result then follows from these theorems and coherence for tricategories by considering the yoneda embedding . we also give two applications of this result . the first is a transport - of - structure result , showing how biequivalences @xmath9 between bicategories can be used to transport monoidal structures from @xmath1 to @xmath2 . this relies on choosing a weak inverse @xmath11 for the definition of the tensor product on @xmath2 , and then requires the rest of the biadjoint biequivalence structure in order to define the higher cells that are part of the definition of a monoidal bicategory and to check that they satisfy the necessary axioms . we also indicate how to prove similar results when the monoidal structure is replaced with a braided monoidal , sylleptic monoidal , or symmetric monoidal one . the second application of the main result is to an elucidation of those monoidal bicategories in which every object is weakly invertible , called picard 2-categories here . we show that every picard 2-category is monoidally biequivalent to one in which a coherent choice of inverses has been made . in fact , the result is much stronger in that we show that the forgetful functor from coherent picard 2-categories ( those with a choice of inverses ) to picard 2-categories ( those monoidal bicategories which merely have the property that every object is invertible ) is a triequivalence . we go on to improve this result by defining functors that preserve a given coherent structure up to equivalence , and show that every monoidal functor between picard 2-categories can be given the structure of such . the paper is organized as follows . the first section is a warm - up in which we give a proof of the fact that every equivalence in a bicategory is part of an adjoint equivalence . we do this to give the reader a taste of the strategy that will be used later in the tricategorical case so as to clearly indicate the crucial points . this material is well - known although i am unaware of a reference that presents this result in full detail using the argument we give below . the second section gives the definitions of biequivalence and biadjoint biequivalence that are at the heart of this paper . both of these we express in the completely general case , working in an arbitrary tricategory @xmath0 . the definition of biadjoint biequivalence has two forms , with and without the `` horizontal cusp '' axioms , and we discuss briefly why these two definitions are logically equivalent using the calculus of mates . section 3 gives a proof of our main result in the special case where @xmath22 . this proof is largely calculation , much as the proof that every equivalence in @xmath23 is part of an adjoint equivalence is done by straightforward calculation in section 1 . we use that a biequivalence in @xmath24 can be characterized in two different ways : as a functor having a weak inverse or as a functor which is biessentially surjective and a local equivalence . the proof largely consists of using this alternate characterization of biequivalences , results from section 1 , and the biadjoint biequivalence axioms to construct the other cells of the biadjoint biequivalence . this is done by constructing the components on objects first , and then building up the rest of the structure afterwords . the reader will note that the technical difficulties lie in two places : in constructing these cells once the components on objects are given , and in checking that our constructions satisfy all of the biadjoint biequivalence axioms since we only need a subset of them in order to define all the needed cells . the fourth section provides the proof of our main result , that every biequivalence in a tricategory is part of a biadjoint biequivalence . the proofs in this section rely very heavily on coherence for tricategories in the form `` every diagram in a free tricategory commutes '' to simplify the pasting diagrams required . we often leave the particulars of checking axioms to the reader as the diagrams are very large , but we state exactly which axioms are needed in each case . the bulk of the technical work goes into showing that the components we construct satisfy the axioms required to be transformations , modifications , or perturbations , while the biadjoint biequivalence axioms are immediate . sections 5 and 6 give the two applications mentioned above . both of these results should be thought of in the form `` a certain 3-dimensional monad has property p '' in each case . in the case of lifting monoidal structures , the monad would be the free monoidal bicategory monad on the tricategory @xmath24 , and the property p would be a 3-dimensional version of flexibility @xcite . in the case of picard 2-categories , the monad would be the free picard 2-category monad on the tricategory of monoidal bicategories , and the property p would be a kind of 3-dimensional idempotency @xcite . this is the proper way to view these results , although we do not pursue the details here because of the lack of groundwork on 3-dimensional monads on tricategories . the author would like to thank bruce bartlett for his interest in these results , and john baez and eugenia cheng for enlightening discussions . we begin by reviewing the relevant results for bicategories that we will later generalize to tricategories . we assume that the reader is familiar with bicategories and the coherence theorem for bicategories ( see @xcite for coherence for bicategories , or @xcite for a discussion of coherence , including functors , for the case of monoidal categories instead of bicategories ) . we begin with some basic definitions . let @xmath25 be a bicategory , and let @xmath26 and @xmath27 be 1-cells in @xmath25 . an _ adjunction _ @xmath28 consists of a 2-cell @xmath29 and a 2-cell @xmath30 such that the following two diagrams ( the triangle identities ) commute . @xmath31^{l_{g}^{-1 } } \ar[rrrrd]_{1_{g } } & 1_{x}g \ar[r]^{\eta * 1_{g } } & ( gf)g \ar[r]^{a } & g(fg ) \ar[r]^{1_{g}*\varepsilon } & g1_{y } \ar[d]^{r } \\ & & & & g}\ ] ] @xmath32^{r_{f}^{-1 } } \ar[rrrrd]_{1_{f } } & f 1_{x } \ar[r]^{1_{f}*\eta } & f(gf ) \ar[r]^{a^{-1 } } & ( fg)f \ar[r]^{\varepsilon * 1_{f } } & 1_{y}f \ar[d]^{l } \\ & & & & f}\ ] ] we then say that @xmath8 is left adjoint to @xmath33 , or that @xmath33 is right adjoint to @xmath8 . in the bicategory @xmath34 , the associativity and unit isomorphisms are all identities . in that case , this definition reduces to the usual definition of an adjunction between functors . an adjunction @xmath35 is an _ adjoint equivalence _ if @xmath36 and @xmath37 are invertible . in this case , we write @xmath38 . [ cat ] let @xmath39 and @xmath11 be functors , and let @xmath40 and @xmath41 be natural isomorphisms . then there is a unique adjoint equivalence @xmath42 in @xmath34 such that @xmath43 . let @xmath43 . the second triangle identity states that @xmath44 . by the invertibility of @xmath36 , this equation is the same as @xmath45 . the righthand side of this equation is well - defined and @xmath10 is full and faithful since it is an equivalence of categories , so we define @xmath46 to be the unique arrow such that @xmath47 . we must now check that @xmath37 is natural and that the first triangle identity holds . for naturality , we consider the square below . @xmath48^{f } \ar[d]_{\eta_{x } } & y \ar[d]^{\eta_{y } } \\ gfx \ar[r]_{gff } & gfy } \ ] ] applying @xmath10 to the diagram and using functoriality gives this square . @xmath49^{ff } \ar[d]_{f\eta_{x } } & fy \ar[d]^{f\eta_{y } } \\ fgfx \ar[r]_{fgff } & fgfy } \ ] ] by the definition of @xmath37 , this square is the naturality square of @xmath50 and thus must commute . by the faithfulness of @xmath10 , the original square commutes as well and so @xmath37 is natural . for the first triangle identity , we consider the composite @xmath51^{\eta_{gy } } & gfgy \ar[r]^{g\varepsilon_{y } } & gy . } \ ] ] applying @xmath10 to this yields @xmath52 , which is by definition @xmath53 . now the following square commutes by the naturality of @xmath36 . @xmath54^{\varepsilon_{fgy } } \ar[d]_{fg \varepsilon_{y } } & fgy \ar[d]^{\varepsilon_{y } } \\ fgy \ar[r]_{\varepsilon_{y } } & y } \ ] ] by the invertibility of @xmath55 , we get that @xmath56 . therefore @xmath57 . once again by the faithfulness of @xmath10 , @xmath58 and thus the first triangle identity is satisfied . [ remcat ] we could have just as easily constructed an adjoint equivalence with @xmath59 instead of @xmath43 . in general it is not possible to require both of these conditions , though , as the choice of either @xmath37 or @xmath60 fixes the other . it would be possible at this point to prove , by a series of calculations , an analogous result with @xmath61 replaced by any bicategory , but instead we choose a different approach that generalizes more easily to the case of tricategories . let @xmath25 be a bicategory , and let @xmath62 be a 1-cell in @xmath25 . then @xmath8 is an _ equivalence _ if there exists a @xmath63 such that @xmath64 and @xmath65 . [ functorbicats ] let @xmath66 be bicategories , and assume that every equivalence @xmath8 in @xmath67 is part of an adjoint equivalence @xmath38 . then every equivalence in the functor bicategory @xmath68 $ ] is part of an adjoint equivalence . let @xmath69 be an equivalence in the functor category . then each 1-cell @xmath70 is an equivalence in @xmath67 , thus we can produce adjoint equivalences @xmath71 in @xmath67 for every object @xmath72 in @xmath25 . we will now define a transformation @xmath73 using the @xmath74 constructed above as the components on objects . now given a morphism @xmath62 in @xmath25 , we must also produce an invertible 2-cell @xmath75 in @xmath67 , subject to the transformation axioms . we define @xmath76 by the requirement that it provides the equality of pasting diagrams given below . @xmath77 here , @xmath78 is the counit of the adjoint equivalence @xmath71 . this gives a well - defined @xmath76 as follows . since all the 2-cells in this pasting diagram are invertible , this equality determines @xmath79 . but @xmath80 is an equivalence 1-cell , hence the functor @xmath81 is an equivalence of categories , so that @xmath82 determines @xmath76 . in a similar fashion , we can also construct an invertible 2-cell @xmath83 with the same source and target as @xmath76 by requiring it provides the equality of pasting diagrams given below . @xmath84 after applying left and right unit isomorphisms , the pasting diagram below can be shown to be equal to both @xmath76 and @xmath83 using the triangle identities , so @xmath85 . @xmath86 this defines the components of @xmath87 on both objects and morphisms . now we check that these components satisfy the axioms for a tranformation . first , we show that @xmath76 is natural in @xmath8 . given a 2-cell @xmath88 in @xmath25 , we must show that @xmath89 this follows from the naturality of both @xmath90 and the coherence isomorphisms used in the definition of @xmath76 . second , we must show that @xmath91 is the composite @xmath92 where every isomorphism is given by a unique coherence isomorphism . to do this , we need only show that the composite above gives the equality of pasting diagrams we used to define @xmath91 . this is trivial using the unit axiom for the transformation @xmath93 and the fact that @xmath94 is the unique coherence isomorphism @xmath95 by the definition of @xmath91 . the third and final transformation axiom follows from a similar proof . assume that every equivalence @xmath8 in @xmath67 is part of an adjoint equivalence @xmath38 , and let @xmath96 be a functor which is locally an equivalence . then every equivalence @xmath97 in @xmath25 is part of an adjoint equivalence @xmath98 . if @xmath99 is an equivalence in @xmath25 , then there is an @xmath100 such that @xmath101 and @xmath102 . then @xmath103 is an equivalence in @xmath67 with @xmath104 as a pseudoinverse . by hypothesis , we can find an adjoint equivalence @xmath105 . by the uniqueness of pseudo - inverses , we must have @xmath106 , so we have an adjoint equivalence @xmath107 . since @xmath10 is locally full , this means we can find a 2-cell in @xmath25 which maps to the following composite . @xmath108 ( here @xmath109 denotes the unit of the adjoint equivalence @xmath107 . ) this 2-cell will be the unit of our adjoint equivalence , and the counit is constructed similarly ; the triangle identities follow from coherence for functors and the fact that @xmath10 is locally faithful . [ adjeqinbicats ] let @xmath25 be a bicategory , and let @xmath8 be an equivalence in @xmath25 . then @xmath8 is part of an adjoint equivalence @xmath38 . since every equivalence in @xmath61 is part of an adjoint equivalence , the same is true for @xmath110 $ ] . let @xmath111 $ ] be the yoneda embedding . the functor @xmath2 satisfies the hypotheses of the above lemma , hence there is an adjoint equivalence @xmath38 in @xmath25 . we have actually shown something stronger than the fact that every equivalence is part of an adjoint equivalence . we have actually shown that given any equivalence @xmath8 , a pseudo - inverse @xmath33 , and an isomorphism @xmath112 , there is a unique adjoint equivalence @xmath35 with @xmath93 as its counit . this is true in @xmath61 , hence in any functor bicategory into @xmath61 . therefore any two adjoint equivalences @xmath113 with the same counit @xmath114 will necessarily have the same unit after applying the yoneda embedding , therefore must have the same unit before applying @xmath2 since it is a local equivalence . this section will provide the definition of a biadjoint biequivalence in an arbitrary tricategory @xmath0 . this proceeds in two steps : first we define a biadjunction in a tricategory , and then equip it with extra structure to define a biadjoint biequivalence . there are two possible options for the definition of a biadjoint biequivalence . we provide the concise definition first ( omitting the `` horizontal cusp '' axioms ) , and then explain how it is equivalent to a definition with additional data and axioms . before giving the definition of a biadjunction in a tricategory @xmath0 , we note that our definition is merely the weakening of previous definitions of biadjunctions in * gray*-categories . this weakening is done in the most straightforward manner , and is done because the most natural and concise definition of a biadjunction in a * gray*-category uses the * gray*-category axioms implicitly . a different approach to these structures might , for instance , involve constructing the `` free living biadjunction '' - this would be a tricategory @xmath115 with the property that biadjunctions in an arbitrary tricategory @xmath0 would correspond to maps @xmath116 . this is the approach taken by lack in @xcite in the context of * gray*-categories in order to discuss the relationship between biadjunctions and pseudomonads . since our focus is on biadjoint biequivalences , and not the more general biadjunctions , we do not proceed in this fashion . let @xmath0 be a tricategory . then a _ biadjunction _ @xmath117 consists of * 1-cells @xmath118 , * 2-cells @xmath119 , and * invertible 3-cells @xmath120 below , @xmath121 such that the pasting diagrams in figures 1 and 2 are both the identity . @xmath122 @xmath123 in the presence of the simplifying assumption that the tricategory @xmath0 is actually a strict , cubical tricategory ( i.e. , a @xmath124-category ) , the axioms simplify to the equality of pasting diagrams below . @xmath125 @xmath126 see @xcite , @xcite , or @xcite for earlier definitions . let @xmath0 be a tricategory . then a _ biadjoint biequivalence _ @xmath127 consists of * a biadjunction @xmath128 and * adjoint equivalences @xmath129 in the respective hom - bicategories . it is also possible to give a longer version of a biadjoint biequivalence which includes extra data satisfying the so - called horizontal cusp axioms . such a definition is equivalent to the one given above by using the calculus of mates , as we explain below . the extra data needed to express the horizontal cusp axioms are a pair of invertible 3-cells @xmath130 . @xmath131 these additional 3-cells are then required to satisfy the horizontal cusp axioms , named for their relationship with certain `` braid movie moves '' between braided surfaces in @xmath132 . one such axiom , written categorically , is given below . @xmath133 it is then clear that this axiom merely says that @xmath134 is the mate of @xmath135 , and similarly for @xmath136 and @xmath137 . all of the horizontal cusp - type axioms can be expressed in this fashion . this section presents a computational proof that every biequivalence in the tricategory @xmath138 is part of a biadjoint biequivalence . the proof here will proceed much as the proof in @xmath61 did , by relying on an alternate description of biequivalences . thus we begin with a simple lemma . let @xmath139 be a functor between bicategories . then @xmath10 is biessentially surjective and locally an equivalence of categories if and only if there is a functor @xmath140 such that @xmath141 and @xmath142 in the respective functor bicategories . since @xmath10 is biessentially surjective , for every object @xmath143 in @xmath67 we can find an object @xmath144 in @xmath25 and an adjoint equivalence @xmath145 between @xmath146 and @xmath143 by theorem [ adjeqinbicats ] ; here we choose @xmath147 to have source @xmath146 and target @xmath143 . we choose such an adjoint equivalence for every @xmath143 , and define the functor @xmath148 on objects by @xmath149 . now @xmath150 is an equivalence of categories for every pair @xmath151 , and we choose an adjoint equivalence @xmath152 . we define the functor @xmath148 on hom - categories @xmath153 to be the composite @xmath154 we must now construct isomorphisms @xmath155 , @xmath156 and check the axioms for a functor . for the first of these , we compute that @xmath157 while @xmath158 . now note that the adjoint equivalence @xmath152 has a unit isomorphism @xmath159 , and when specialized to the case @xmath160 and then evaluated at @xmath161 yields @xmath162 since @xmath10 is a functor , we have an isomorphism @xmath163 which we can compose with the previous isomorphism to get @xmath164 which we denote by @xmath165 . writing @xmath166 for the unit of the adjoint equivalence @xmath167 , we obtain the isomorphism @xmath168 as the following composite . @xmath169 the isomorphism @xmath170 is obtained in a similar fashion , and the functor axioms for @xmath148 follow from those for @xmath10 and the adjoint equivalence axioms for both @xmath167 and @xmath171 . we must finally check that @xmath141 and @xmath142 in the relevant functor bicategories . for the first of these , note that @xmath172 by construction . we already have equivalence 1-cells @xmath173 that we now need to complete to a natural transformation . this requires that we give natural isomorphisms @xmath174 one for each @xmath97 , which we define to be the obvious whiskering of the counit isomorphism @xmath175 composed with the left unit isomorphism . the adjoint equivalence axioms for @xmath145 and coherence for bicategories imply all of the transformation axioms . this shows that @xmath141 , and we leave it to the reader to prove @xmath142 . [ biadjbieq1 ] let @xmath96 be a biequivalence between bicategories . then there is a biadjoint biequivalence @xmath176 . since @xmath10 is a biequivalence , choose a functor @xmath140 such that @xmath177 is equivalent to @xmath178 in the bicategory @xmath179 and @xmath180 is equivalent to @xmath181 in @xmath182 . taking any equivalence @xmath183 exhibiting this fact , we can construct an adjoint equivalence @xmath184 in @xmath179 by theorem [ adjeqinbicats ] . we will write @xmath185 and @xmath186 for the counit and unit of this adjoint equivalence , respectively . now we construct the adjoint equivalence @xmath187 ( between @xmath181 and @xmath180 ) and the invertible modification @xmath135 simultaneously . the component of @xmath87 at an object @xmath188 is a 1-cell @xmath189 . the component of @xmath135 at @xmath188 is an invertible 2-cell in @xmath67 @xmath190 since the associativity and unit 2-cells in @xmath138 have identities as their components . by coherence , such an invertible 2-cells determines and is determined by an invertible 2-cell @xmath191 . since @xmath93 is an equivalence , giving such an isomorphism is equivalent to giving an isomorphism @xmath192 . now @xmath10 is locally an equivalence of categories , and in particular essentially surjective , so there exists a morphism @xmath193 such that @xmath194 . for every object @xmath188 , choose such a @xmath195 and a specified isomorphism @xmath196 . for the component of @xmath87 at @xmath197 , consider the following composite . @xmath198 since @xmath10 is locally an equivalence , there is a unique isomorphism @xmath199 that maps to the composite above . it is then simple to check that @xmath87 is a transformation @xmath200 , and that it is an equivalence . this construction also immediately implies that @xmath201 is an invertible modification @xmath202 we then define the adjoint equivalence @xmath187 to be any adjoint equivalence containing @xmath87 . the 2-cell @xmath203 is defined to be the following composite . @xmath204 by coherence for bicategories , this determines the 2-cell @xmath205 uniquely . these 2-cells @xmath205 then give the data for an invertible modification @xmath135 since all of the cells used to construct the @xmath205 are either components of modifications or are appropriately natural . all that remains is to construct the invertible 3-cell @xmath137 and to check the two biadjunction axioms . before doing so , we remind the reader that , for 3-cells in a tricategory , @xmath206 denotes the composition along 2-cell boundaries , @xmath207 denotes composition along 1-cell boundaries , and @xmath208 denotes composition along 0-cell boundaries . now the second axiom determines the 3-cell @xmath209 . since the 2-cell @xmath87 is an equivalence , the functor @xmath210 is an equivalence of categories , and in particular the cell @xmath209 uniquely determines the cell @xmath211 . similarly , since the functor @xmath10 is a biequivalence , @xmath148 is also , so the functor @xmath212 is a biequivalence of bicategories ; thus @xmath213 uniquely determines the invertible modification @xmath137 . by construction , the second biadjoint biequivalence axiom is satisfied . now we show that this choice of @xmath137 satisfies the first biadjoint biequivalence axiom . first , note that , while @xmath138 is not a @xmath214-category , it does have a strictly associative and unital composition law for 1-cells in the following sense . the composite @xmath215 equals the composite @xmath216 , and similarly @xmath217 , but we still have associativity and unit equivalences for this composition law . these are the 1-cells in the biadjoint biequivalence axioms labeled @xmath218 , and they have components on objects given by identites and components on morphisms given by unique coherence 2-cells . for examples , the transformation @xmath219 has its component at an object @xmath72 the identity @xmath220 and its component at a 1-cell @xmath62 the unique coherence cell @xmath221 similarly , the modifications @xmath222 and @xmath223 in @xmath138 all have unique coherence 2-cells as their components . thus coherence for bicategories reduces the first biadjoint biequivalence to checking that the pasting @xmath224 is equal to the identity on @xmath225 . ( here we use the same convention that @xmath226 is derived from @xmath137 via unique coherence isomorphisms . ) from this point on , we mark our naturality isomorphisms with a subscript to indicate which transformation they are naturality isomorphisms for to avoid confusion , and we refer to all instances of the above pasting diagram as `` axiom 1 '' , perhaps with some descriptor to indicate which object @xmath72 is being used . first , note that axiom 1 is the identity if and only if it is the identity when @xmath72 is of the form @xmath227 for some @xmath228 in @xmath25 . indeed , consider the following pasting diagram . @xmath229 using the modification and transformation axioms , it is equal to the pasting below . @xmath230 thus the pasting axiom 1 for @xmath72 is the identity if and only if axiom 1 for @xmath231 is the identity , so taking @xmath232 proves the claim . from this point , we replace @xmath72 with @xmath227 . now axiom 1 for @xmath227 is the identity if and only if the following pasting diagram is the identity since @xmath233 is an equivalence 1-cell . @xmath234 applying @xmath10 to the second biadjoint biequivalence axiom and rewriting , we see that the above pasting diagram is the identity if and only if the following one is . @xmath235 recall now that @xmath87 was constructed together with an invertible modification @xmath201 with components @xmath236 such that the composite @xmath237 is the counit of the adjoint equivalence @xmath238 . the previous pasting diagram is then the identity if and only if we pre - compose it with @xmath239 and post - compose it with @xmath240 . using the naturality axiom for @xmath241 , the modification axiom for @xmath201 , and the equality relating @xmath201 to the counit of the adjoint equivalence @xmath184 , the pre- and post - composed pasting is the identity if and only if the one displayed below is the identity . @xmath242 here we have written @xmath243 for the counit isomorphism of the adjoint equivalence @xmath244 , and @xmath245 for the inverse of the counit . this diagram is the identity following from a general lemma on mates that we give below . let @xmath25 be a bicategory , and @xmath246 a functor . let @xmath247 , @xmath248 be part of an adjoint equivalence @xmath184 . then for any object @xmath249 in @xmath25 , the pasting diagram below is the identity . @xmath250 first , the above pasting diagram is the identity if and only if the one below is . @xmath251 here we have written @xmath252 to denote a unit isomorphism , and @xmath253 the inverse of a unit isomorphism . it is now a simple calculation using the adjoint equivalence axioms for transformations to show that the pasting below ( modulo unit isomorphisms to alter the 1-cell source and target , which we ignore for now but record the presence of for later ) @xmath254 is equal to the naturality square below . @xmath255 thus we have shown that the pasting diagram in the previous paragraph is equal ( modulo units ) to the pasting diagram displayed below . @xmath256 using the triangle identities and the naturality axioms for @xmath93 , this is now equal to the pasting below ( once again modulo units ) . @xmath257 by the naturality axioms for @xmath93 and coherence for functors , this is equal to a composite of left and right unit isomorphisms so the original diagram is a composite of coherence isomorphisms , hence is the identity by coherence for bicategories . this section will establish the general result that every biequivalence in a tricategory @xmath0 is part of a biadjoint biequivalence in @xmath0 . our proof will proceed largely as did the general case for equivalences in bicategories by first examining the case of functor tricategories and then using a yoneda embedding . since a yoneda embedding is only known for cubical tricategories rather than the general case , the proof for tricategories is slightly longer although essentially the same . we refer the reader to @xcite or @xcite for the relevant tricategorial results . recall that if @xmath20 is any tricategory and @xmath0 is a @xmath214-category , then there is a @xmath214-category @xmath258 with objects functors @xmath259 , 1-cells transformations , 2-cells modifications , and 3-cells perturbations . let @xmath20 be any tricategory and @xmath0 be a @xmath214-category . assume that every biequivalence 1-cell @xmath8 in @xmath0 is part of a biadjoint biequivalence @xmath127 . then every biequivalence 1-cell @xmath93 in @xmath258 is part of a biadjoint biequivalence @xmath260 . we have written the proof of this proposition out so that it should be obvious to the reader that it remains true when @xmath0 is merely a tricategory and not a @xmath214-category . by this we mean the following : the assumption that @xmath0 is a @xmath214-category is only present to use the results of @xcite to give a concrete construction of the tricategory @xmath258 . using coherence for tricategories , it is possible to construct a tricategory @xmath258 when @xmath0 is any tricategory , and then the proof below applies verbatim to the analogous proposition . let @xmath69 be a biequivalence in @xmath258 . then @xmath261 is a biequivalence in @xmath0 for every object @xmath249 , so we choose biadjoint biequivalences @xmath262 for every object @xmath249 of @xmath20 . to complete the proof , we must do the following : 1 . equip the components @xmath263 with the structure of a transformation ; 2 . equip the componentwise adjoint equivalences @xmath264 with the structure of adjoint equivalences in the hom - bicategories @xmath265 , @xmath266 ; and 3 . check that @xmath120 are perturbations . since equations between perturbations are checked componentwise , the fact we have biadjoint biequivalences @xmath267 will then imply that there is a global biadjoint biequivalence @xmath260 in @xmath258 . we begin by defining a transformation @xmath268 with components given by these @xmath263 . since we have already given the components on objects , there are three pieces of data left to define . the first is an adjoint equivalence @xmath269 in the bicategory @xmath270 . we write down the component @xmath76 of @xmath87 at an object @xmath8 and leave it to the reader to construct the rest of the adjoint equivalence in the obvious fashion . @xmath271 we have written this out as if if were a 2-cell an arbitrary tricategory , not necessarily a @xmath272-category . in the case that @xmath0 is @xmath272 , this cell is as below . @xmath273 we now must produce a pair of invertible modifications @xmath274 to complete the definition of the data for the transformation @xmath87 . the component of the modification @xmath275 at the object @xmath249 is given by the isomorphism shown below where the unmarked isomorphisms are unique by coherence and the two marked 3-cells are both appropriate mates . @xmath276 the component of the modification @xmath277 at the composable pair @xmath278 is given by the pasting below , once again following the same conventions . to conserve space , we omit the subscripts for the components of @xmath93 and @xmath87 given that they can be deduced from the other 1-cells in any given term . @xmath279 all of the unmarked isomorphisms are isomorphisms of the form @xmath280 arising from the functoriality of the horizontal composition @xmath281 the transformation axioms for @xmath87 then follow from those for @xmath93 , thus completing the construction of a weak inverse for @xmath93 . the next step is to construct adjoint equivalences @xmath282 we already have the adjoint equivalences @xmath264 on components in the tricategory @xmath0 , we need only lift these to adjoint equivalences in the hom - bicategories @xmath283 , respectively . thus we need to equip the collection @xmath284 with the structure of a trimodification . the 2-cell @xmath285 is the 2-cell in @xmath0 of the same name , so we must give the invertible 3-cell displayed below . @xmath286 the component @xmath287 is given by @xmath288 , so once again using coherence we write this as the identity and can define @xmath289 as shown below . @xmath290 the unmarked isomorphisms in this pasting diagram are ( up to coherence ) counits for adjoint equivalences for the two left squares , functoriality of the tensor for the top right square , and a counit together with functoriality of the tensor for the bottom right square . there are now two trimodification axioms to check , one for composition and one for units . both of these axioms follows from coherence , the definitions of @xmath277 and @xmath275 , and the definitions of @xmath291 and @xmath292 ; they do not require the biadjunction axioms for @xmath262 . to complete the proof that we have an adjoint equivalence @xmath293 in @xmath294 , we need only note that a modification @xmath295 in this bicategory is an equivalence if and only if each component @xmath296 is an equivalence in the appropriate hom - bicategory of @xmath0 . in this case , the modification @xmath297 is @xmath36 , and each component @xmath285 is an equivalence in @xmath298 by construction . we note in passing that the components of the specified pseudo - inverse for @xmath36 can be taken to be the 2-cells @xmath299 that appear in the pointwise adjoint equivalences @xmath300 the construction of the adjoint equivalence @xmath301 follows exactly the same pattern as for @xmath302 , so we omit most of the details here . in order to check that @xmath135 and @xmath137 are perturbations , we will need to know the components @xmath303 for each 1-cell @xmath304 . using coherence , we write @xmath305 as the pasting below . @xmath306 the isomorphisms are all obtained from units of adjunctions and functoriality of the tensor . finally , we must check that the cells @xmath307 constitute a pair of invertible perturbations . this involves checking a single axiom for each 1-cell @xmath304 which we leave to the reader , but it follows from coherence and the biadjunction axioms for both @xmath262 and @xmath308 . to prove our main result , we require two lemmas . let @xmath304 and @xmath309 be 1-cells in a tricategory @xmath0 . if both @xmath310 is equivalent to @xmath311 in @xmath312 and @xmath313 is equivalent to @xmath161 in @xmath314 , then @xmath315 is equivalent to @xmath33 in @xmath316 . in particular , if @xmath8 is a biequivalence in @xmath0 and both @xmath317 are 1-cells such that @xmath318 and @xmath319 for @xmath320 , then @xmath321 . this follows in the standard way by considering the 1-cell @xmath322 in @xmath316 . let @xmath323 be tricategories , and assume that every biequivalence @xmath8 in @xmath0 is part of a biadjoint biequivalence @xmath324 . assume that @xmath19 is 2-locally an equivalence , i.e. , every functor @xmath325 is an equivalence of categories . then every biequivalence @xmath315 in @xmath20 is part of a biadjoint biequivalence . this lemma follows in the same way that lemma [ functorbicats ] did ; the proof requires using the previous lemma in exactly the same way that lemma [ functorbicats ] required the uniqueness of weak inverses in a bicategory . we now present our main result . [ biadjbieqintricats ] let @xmath0 be a tricategory , and let @xmath8 be a biequivalence in @xmath0 . then there is a 1-cell @xmath33 in @xmath0 and a biadjoint biequivalence @xmath127 . first , recall that for every tricategory @xmath0 there is a cubical tricategory ( i.e. , a tricategory in which the hom - bicategories are 2-categories and the unit and composition functors are cubical ) @xmath326 and a triequivalence @xmath327 . since triequivalences satisfy the hypotheses of the lemma above , we are left proving the theorem in the case that @xmath0 is a cubical tricategory . if @xmath0 is cubical , then it has a yoneda embedding @xmath328 which also satisfies the hypotheses of the above lemma . thus if we show that every biequivalence in @xmath214 is part of a biadjoint biequivalence , we will have proven the theorem for arbitrary @xmath0 . but the inclusion @xmath329 satisfies the hypotheses of the lemma and we have already proven the claim directly for @xmath138 . this section will show how to lift monoidal structures on bicategories using the results of section 4 . we refer the reader to @xcite for the definitions of the morphisms and higher cells between monoidal bicategories in the case of @xmath272-monoids , and @xcite or @xcite for the definitions of higher cells between tricategories from which these are derived . from this point forward , most calculations will only be described as they are generally straightforward but the pastings used can be very large . let @xmath25 be a monoidal bicategory and let @xmath67 be any bicategory . if @xmath330 is a biequivalence from @xmath67 to the underlying bicategory of @xmath25 , then @xmath67 can be equipped with the structure of a monoidal bicategory and @xmath10 can be compatibly equipped with the structure of a monoidal functor . first choose a biadjoint biequivalence @xmath176 in @xmath138 . we now define a monoidal structure on @xmath67 as follows . the tensor @xmath331 is the composite @xmath332 the unit @xmath333 is the composite @xmath334 the associativity adjoint equivalence is given by the pasting below . here we have marked identity adjoint equivalences with equal signs . @xmath335 the left unit adjoint equivalence is given by the pasting below , following the same conventions as above . @xmath336 the right unit adjoint equivalence is constructed in an analogous fashion . in order to construct the remaining data , it is useful to compute the components of the above transformations . the component of @xmath337 at @xmath338 is given as the following composite which we write omitting all the identity components from adjoint equivalences marked with an equal sign . @xmath339 the component of @xmath340 at @xmath72 is given by the following composite . @xmath341 the component of @xmath342 at @xmath72 is given by the following composite . @xmath343 now we define the invertible modifications @xmath344 . we explicitly define the unit modifications @xmath345 , but only describe the construction of the modification @xmath346 due to the size of the pasting diagram . the unit modification @xmath347 is the pasting below ( composed with the unique coherence isomorphism @xmath348 ) , where all unmarked isomorphisms are naturality isomorphisms and @xmath349 is the mate of @xmath135 . @xmath350 the unit modification @xmath351 is the pasting below . @xmath352 for the unit modification @xmath353 , we give the following non - standard presentation . let @xmath354 denote the mate of @xmath355 with source @xmath356 and target @xmath97 . we thus define @xmath354 as the pasting below and leave it to the reader to construct the usual modification @xmath223 as necessary . @xmath357 we now describe the construction of the modification @xmath346 for the monoidal structure on @xmath67 . this invertible modification is constructed like the unit modifications by pasting together * naturality isomorphisms for @xmath358 , and the associator @xmath249 ; * a single counit coming from @xmath359 ; and * the cell @xmath360 . we leave it to the reader to construct the appropriate pasting from these cells . there are now three monoidal bicategory axioms to check . we leave these to the reader as the diagrams are large but the computations simple the associativity axiom follows by coherence , naturality , and the associativity axiom in @xmath25 , while both unit axioms follow by coherence , naturality axioms , the corresponding unit axioms in @xmath25 , and the biadjoint biequivalence axioms . now we show that @xmath10 can be equipped with the structure of a monoidal functor . the adjoint equivalence @xmath361 is @xmath362 , and the adjoint equivalence @xmath363 is @xmath364 . the invertible modification @xmath365 @xmath366 is given by the pasting diagram below in which all the cells are naturality isomorphisms or a counit for @xmath367 . @xmath368 the invertible modification @xmath369 @xmath370 is the pasting diagram shown below where both cells are mates of naturality isomorphisms . @xmath371 the invertible modification @xmath372 is defined similarly . there are now two axioms to check to show that these data give a monoidal functor between monoidal bicategories , and once again we leave these simple albeit long computations to the reader . the associativity axiom follows from the transformation axioms for @xmath93 while the unit axiom requires using the biadjoint biequivalence axioms . it would be possible at this point to prove that the functor @xmath148 chosen in the proof above can also be given the structure of a monoidal functor . for instance , the transformation @xmath373 is given by the transformation @xmath87 used in constructing the biadjoint biequivalence @xmath176 . we could go further , and even show that the entire biadjoint biequivalence @xmath374 can be lifted from @xmath138 to @xmath375 , showing that the forgetful functor @xmath376 is the tricategorical analogue of an isofibration . put another way , the free monoidal bicategory construction is an example of a `` flexible 3-monad . '' the proofs of all of these statements follow in exactly the same fashion as the construction of the monoidal structure on @xmath67 and @xmath10 given above . we could also lift braided monoidal , sylleptic monoidal , or symmetric monoidal structures along biequivalences in a similar fashion . for instance , if @xmath25 is braided with braiding @xmath377 then @xmath67 can be given a braided structure with braiding @xmath378 in these cases , as above , the entire biadjoint biequivalence could be lifted from @xmath138 to the relevant tricategory of monoidal bicategories of the kind considered . this section will present an application of the main result to the study of picard 2-categories . this is the analogue , for monoidal bicategories , of the result of baez and lauda that the 2-category of 2-groups ( or picard groupoids ) is 2-equivalent to the 2-category of coherent 2-groups @xcite . it should be noted that all of the results of this section remain true when we add braided , sylleptic , or symmetric structures . the reader should note that we call these picard 2-categories even though the underlying object is a mere bicategory . we have also not assumed that all the 1- and 2-cells are invertible , nor that the monoidal structure is braided , sylleptic , or symmetric . all of these additional features ( strictness , invertible higher cells , and symmetry ) can be added as desired to produce the notion of picard 2-category appropriate to a particular application . analogous results to those we present below can then be proven . for the next definition , recall that every monoidal bicategory @xmath1 gives rise to a tricategory @xmath382 with a single object @xmath207 and single hom - bicategory given by @xmath383 horizontal composition is then given by the tensor product , and all of the coherence constraints for the tricategory are given by those for the monoidal structure on @xmath1 . thus a biadjoint biequivalence @xmath384 between objects of a monoidal bicategory is defined to be a biadjoint biequivalence @xmath384 in @xmath382 where now @xmath385 are treated as 1-cells of the tricategory @xmath382 . the underlying monoidal bicategory functor @xmath391 factors ( as a strict functor between tricategories ) through the inclusion of @xmath380 into @xmath381 . @xmath392 the strict functor @xmath393 is a triequivalence . the first statement is clear , as the underlying monoidal bicategory of a coherent picard 2-category @xmath386 is obviously a picard 2-category , and all of the higher dimensional structure involved in the definitions of these two tricategories agrees . for the second statement , we must prove that @xmath394 is locally a biequivalence and triessentially surjective . now @xmath394 is the identity functor on hom - bicategories , so is locally a biequivalence . to show that @xmath394 is triessentially surjective , note that theorem [ biadjbieqintricats ] actually implies that @xmath394 is surjective on objects as we can always choose a biadjoint biequivalence @xmath384 for any object @xmath72 with the property that @xmath395 . this theorem produces the most basic kind of equivalence between the theory of picard 2-categories and its coherent version . for the rest of this paper , we will sketch an improvement to this equivalence by explaining how one might go about proving that not only can picard 2-categories be replaced by coherent ones , but also that monoidal functors can be replaced by ones that preserve the choice of inverses up to equivalence . * underlying bicategory the same as @xmath2 , * @xmath403 , * @xmath404 , * all adjoint equivalences given by the opposites of the appropriate adjoint equivalences for the monoidal structure on @xmath2 , and * all invertible 2-cell data the same as that for @xmath2 . we have already defined @xmath410 on objects , so now it is time to define it on 1- and 2-cells . to do this , we must fix notation . for an object @xmath72 , the biadjoint biequivalence @xmath388 consists of * an adjoint equivalence @xmath411 between the objects @xmath412 and @xmath413 , * an adjoint equivalence @xmath414 between the objects @xmath413 and @xmath415 , * and invertible 2-cells @xmath416 , satisfying the necessary axioms . next we must define structure constraints @xmath421 and check that these give a functor of bicategories . now @xmath422 in @xmath423 is defined to be the composite @xmath424 , so we in fact require an isomorphism of the form @xmath425 it is given by the pasting diagram below , in which we have written @xmath426 as @xmath427 and all the unmarked isomorphisms are functoriality of the tensor product . @xmath428 the isomorphism @xmath429 is ( modulo coherence ) @xmath430 . there are now three axioms to check , one for associativity and two for units . all three of these axioms follow by using the functoriality of the tensor product and then invoking coherence for monoidal bicategories ( i.e. , coherence for tricategories in the single - object case ) . the unit axioms then require the equation @xmath431 , while the associativity axiom only uses naturality . thus we have shown that inv is a functor of bicategories @xmath432 . now we turn to showing that it is monoidal . the first step in showing that inv is a monoidal funtor @xmath433 is to construct an adjoint equivalence @xmath434 on components , this gives adjoint equivalences @xmath435 in @xmath423 , hence adjoint equivalences in @xmath1 with the source and target reversed . we define the component @xmath436 below , and the rest of the adjoint equivalence will be defined in the obvious fashion . we retain the same convention as above for writing @xmath426 as @xmath427 . @xmath437 the second step in giving inv a monoidal structure is to construct an adjoint equivalence @xmath438 in @xmath423 . since @xmath439 , this is the obvious adjoint equivalence with left adjoint shown below . @xmath440 the third step in giving inv a monoidal structure is to define three invertible modifications @xmath441 . we leave it to the reader to write down the pasting diagrams as they are quite large , but we explain here which cells will be included in each . in each case , there will be a large number of coherence cells from the monoidal bicategory structure , most of which will arise from the functoriality of the tensor product . the other cells in each case are as follows : finally , there are two monoidal functor axioms to check . both follow from coherence for monoidal functors ( a special case of coherence for functors of tricategories ) and the biadjoint biequivalence axioms . one can go further to prove that the monoidal functor @xmath410 is actually a monoidal biequivalence . first we can equip @xmath423 with a canonical coherent picard 2-category structure using the one given on @xmath1 . then it is easy to show that @xmath449 is equivalent to @xmath72 in @xmath1 as both are weak inverses for @xmath426 , from which it follows that @xmath410 squares to the identity . * a monoidal functor @xmath9 , * an equivalence 1-cell @xmath451 for each object @xmath452 , and * a pair of invertible 2-cells @xmath453 for each object @xmath452 as displayed below . @xmath454 @xmath455 we could have structured this definition in a slightly different fashion in a variety of ways . first , the equivalence 1-cells @xmath457 could have been the components of a transformation @xmath143 as shown here . @xmath458 second , we could have asked that the invertible 2-cells @xmath453 could have been the components of a pair of invertible modifications . to express the axioms above as diagrams of modifications would have required that the @xmath459 be the components of transformations , which in turn would require that the functor @xmath410 be covariant instead of contravariant . thus we would have to restrict attention to those picard 2-categories in which every 1-cell is an equivalence ; in fact , we would need every 1-cell to come as part of a specified adjoint equivalence in order to prescribe @xmath410 as a covariant _ functor_. third , we could have required a third axiom about how @xmath460 interact with @xmath137 . this axiom follows from the first axiom by using the biadjoint biequivalence axioms . in addition , the pastings involved are larger than the one for the axiom above as they involve two different uses of @xmath143 instead of just one , so it requires additional naturality squares . the 1-cell @xmath463 is given by the following composite where the last arrow is given by a composite of coherence cells and is thus unique up to unique isomorphism by coherence for functors . @xmath464 it is immediate that @xmath457 is an equivalence 1-cell . we must now construct the invertible 2-cells @xmath453 and check the two axioms . the cell @xmath465 is a pasting of coherence cells from both @xmath2 and the functor @xmath10 , together with @xmath466 . the cell @xmath467 is constructed similarly out of coherence cells and @xmath468 . the two axioms are straightforward diagram chases . let @xmath1 be a picard 2-category . then @xmath1 has a coherent structure which is unique in the following sense : if @xmath386 and @xmath469 are two coherent structures on @xmath1 , then the identity functor on @xmath1 lifts to a coherent functor @xmath470 we leave it to the reader to define coherent transformations and modifications . a coherent transformation will involve additional data , while a coherent modification will only involve a new axiom . defined correctly , it is then possible to prove that the forgetful functor from the tricategory in which all cells are coherent to the tricategory @xmath380 is a triequivalence . this shows that a coherent structure on a given picard 2-category is unique in the strongest possible sense .

the purpose of this study is to carry out a thorough investigation of the changes in morphology of the red giant branch ( rgb ) of galactic globular clusters ( ggc ) as a function of metallicity , in the bands . to this aim , two key points are developed in the course of the analysis . * ( a ) * using our photometric database for galactic globular clusters ( the largest homogeneous data sample to date ; rosenberg et al . ) _ we measure a complete set of metallicity indices _ , based on the morphology and position of the red - giant branch . in particular , we provide here the first calibration of the , and indices in the plane . we show that our indices are internally consistent , and we calibrate each index in terms of metallicity , both on the zinn & west ( 1984 ) and the carretta & gratton ( 1997 ) scales . our new calibrations of the , , and indices are consistent with existing relations . * ( b ) * using a grid of selected rgb fiducial points , _ we define a function in the $ ] space which is able to reproduce the whole set of ggc giant branches in terms of a single parameter _ ( the metallicity ) . as a first test , we show that the function is able to predict the correct trend of our observed indices with metallicity . the usage of this function will improve the current determinations of metallicity and distances within the local group , since it allows to easily map coordinates into,m_{i } $ ] ones . to this aim the `` synthetic '' rgb distribution is generated both for the currently used lee et al . ( 1990 ) distance scale , and for the most recent results on the rr lyr distance scale .

astro-ph0001266

rrrrrrr & & & & + & & & & & & + 104 & 0.05 & 0.06 & -0.71 & -0.70 & -0.78 & 14.05 @xmath12 0.05 + 288 & 0.03 & 0.04 & -1.40 & -1.07 & -1.14 & 15.40 @xmath12 0.05 + 362 & 0.05 & 0.06 & -1.33 & -1.15 & -1.09 & 15.51 @xmath12 0.05 + 1261 & 0.01 & 0.01 & -1.32 & & -1.08 & 16.68 @xmath12 0.05 + 1851 & 0.02 & 0.03 & -1.23 & & -1.03 & 16.18 @xmath12 0.05 + 1904 & 0.01 & 0.01 & -1.67 & -1.37 & -1.37 & 16.15 @xmath12 0.05 + 3201 & 0.21 & 0.27 & -1.53 & -1.23 & -1.24 & 14.75 @xmath12 0.05 + 4590 & 0.04 & 0.05 & -2.11 & -1.99 & -2.00 & 15.75 @xmath12 0.10 + 4833 & 0.33 & 0.42 & -1.92 & -1.58 & -1.71 & 15.70 @xmath12 0.10 + 5272 & 0.01 & 0.01 & -1.66 & & -1.33 & 15.58 @xmath12 0.05 + 5466 & 0.00 & 0.00 & -2.22 & & -2.13 & 16.60 @xmath12 0.05 + 5897 & 0.08 & 0.10 & -1.93 & -1.59 & -1.73 & 16.30 @xmath12 0.10 + 5904 & 0.03 & 0.04 & -1.38 & -1.11 & -1.12 & 15.00 @xmath12 0.05 + 6093 & 0.18 & 0.23 & -1.75 & & -1.47 & 16.25 @xmath12 0.05 + 6171 & 0.33 & 0.42 & -1.09 & & -0.95 & 15.65 @xmath12 0.05 + 6205 & 0.02 & 0.03 & -1.63 & -1.39 & -1.33 & 14.95 @xmath12 0.10 + 6218 & 0.19 & 0.24 & -1.40 & & -1.14 & 14.70 @xmath12 0.10 + 6254 & 0.28 & 0.36 & -1.55 & -1.41 & -1.25 & 15.05 @xmath12 0.10 + 6341 & 0.02 & 0.03 & -2.24 & & -2.10 & 15.20 @xmath12 0.10 + 6352 & 0.21 & 0.27 & -0.50 & -0.64 & -0.70 & 15.25 @xmath12 0.05 + 6362 & 0.09 & 0.12 & -1.18 & -0.96 & -0.99 & 15.35 @xmath12 0.05 + 6397 & 0.18 & 0.23 & -1.94 & -1.82 & -1.76 & 12.95 @xmath12 0.10 + 6541 & 0.12 & 0.15 & -1.79 & & -1.53 & 15.40 @xmath12 0.10 + 6637 & 0.17 & 0.22 & -0.72 & & -0.78 & 15.95 @xmath12 0.05 + 6656 & 0.34 & 0.44 & -1.75 & & -1.41 & 14.25 @xmath12 0.10 + 6681 & 0.07 & 0.09 & -1.64 & & -1.35 & 15.70 @xmath12 0.05 + 6723 & 0.05 & 0.06 & -1.12 & & -0.96 & 15.45 @xmath12 0.05 + 6752 & 0.04 & 0.05 & -1.54 & -1.42 & -1.24 & 13.80 @xmath12 0.10 + 6779 & 0.20 & 0.26 & -1.94 & & -1.61 & 16.30 @xmath12 0.05 + 6809 & 0.07 & 0.09 & -1.80 & & -1.54 & 14.45 @xmath12 0.10 + 7078 & 0.09 & 0.12 & -2.13 & -2.12 & -2.02 & 15.90 @xmath12 0.05 + in very recent times , new determinations of galactic globular cluster ( ggc ) metallicities have provided us with new homogeneous @xmath13\textrm { } $ ] scales . in particular , carretta & gratton ( @xcite ; cg ) obtained metallicities from high resolution spectroscopy for 24 ggcs , with an internal uncertainty of 0.06 dex . for an even larger sample of 71 ggcs , metallicities have been obtained by rutledge et al . ( @xcite ; rhs97 ) based on spectroscopy of the caii infrared triplet . the equivalent widths of the caii triplet have been calibrated by rhs97 on both the cg scale and the older zinn & west ( @xcite ; zw ) scale . the compilation by rhs97 is by far the most homogeneous one which is currently available . in the same period , we have been building the largest homogeneous @xmath0 photometric sample of galactic globular clusters ( ggc ) based on ccd imaging carried out both with northern ( isaac newton group , ing ) and southern ( eso ) telescopes ( rosenberg et al . @xcite , @xcite ) . the main purpose of the project is to establish the relative age ranking of the clusters , based on the methods outlined in saviane et al . ( @xcite , @xcite ; srp97 , srp99 ) and buonanno et al . ( @xcite ; b98 ) . the results of this investigation are presented in rosenberg et al . ( @xcite ; rspa99 ) . here suffice it to say that for a set of clusters we obtained @xmath14 vs. @xmath15 color - magnitude diagrams ( cmd ) , which cover a magnitude range that goes from a few mags below the turnoff ( to ) up to the tip of the red giant branch ( rgb ) . at this point both a spectroscopic and photometric homogeneous databases are available : the purpose of this study is to exploit them to perform a thorough analysis of the morphology of the rgb as a function of the cluster s metallicity . as a first step , we want to obtain a new improved calibration of a few classical photometric metallicity indices . secondly , we want to provide to the community a self - consistent , * analytic , * family of giant branches , which can be used in the analysis of old stellar populations in external galaxies . photometric indices have been widely used in the past to estimate the mean metallicities of those stellar systems where direct determinations of their metal content are not feasible . in particular , they are used to obtain @xmath13\textrm { } $ ] values for the farthest globulars and for those resolved galaxies of the local group where a significant pop ii is present ( e.g. the dwarf spheroidal galaxies ) . the calibration of @xmath0 indices is particularly important , since with comparable exposure times , deeper and more accurate photometry can be obtained for the cool , low - mass stars in these broad bands than in @xmath16 . moreover , our huge cmd database allows a test of the new cg scale on a large basis : we are able to compare the relations obtained for both the old zw and new scale , and check which one allows to rank ggcs in the most accurate way . indeed , the most recent calibration of the @xmath0 indices ( carretta & bragaglia @xcite ) is based on just 8 clusters . a reliable metallicity ranking of ggc giant branches also allows studies that go beyond a simple determination of the _ mean _ metallicity of a stellar population . as an illustration , we may recall the recent investigation of the halo metallicity distribution function ( mdf ) of ngc 5128 ( harris et al . @xcite ) , which was based on the fiducial gc lines obtained by da costa & armandroff ( @xcite , hereafter da90 ) . these studies can be made more straightforward by providing a suitable analytic representation of the rgb family of ggcs . indeed , assuming that most of the ggcs share a common age ( e.g. rosenberg et al . @xcite ) , one expects that there should exist a `` universal '' function of @xmath17\ } $ ] able to map any @xmath18 $ ] coordinate pair into the corresponding metallicity ( provided that an independent estimate of the distance and extinction of the star are available ) . we will show here that such relatively simple mono - parametric function can actually be obtained , and that this progress is made possible thanks to the homogeneity of both our data set and analysis . in order to enforce a proper use of our calibrations , we must clearly state that , in principle , the present relations are valid only for rigorously old stellar populations ( i.e. for stars as old as the bulk of galactic globulars ) . at fixed abundance , giant branches are somewhat bluer for younger ages ( e.g. bertelli et al . moreover , in real stellar systems agb stars are also present on the blue side of the rgb ( cf . [ f : pars - partb ] ) . both effects must be taken into account when dealing with lg galaxies , since they could lead to systematic effects in both the mean abundances and the abundance distributions ( e.g. saviane et al . @xcite ) . the observational sample , on which this investigation is based , is presented in sect . [ s : sample ] . [ s : indices ] is devoted to the set of indices which are to be calibrated . they are defined in sect [ s : defindices ] . the reliability of our sample is tested in sect . [ s : checks ] , where we demonstrate that our methodology produces a set of well - correlated indices . in sect . [ s : newda90 ] we show that , once a distance scale is assumed for the ggcs , our whole set of rgbs can be approximated by a _ single _ analytic function , which depends on the metallicity alone . this finding allows a new and easier way to determine the distances and mean metallicities of the galaxies of the local group , extending the methods of da costa & armandroff ( @xcite ) , and lee et al . ( @xcite ) . the metallicity indices are calibrated in sect . [ s : calibrations ] , where analytic relations are provided both for the zw and for the cg scales . using these indices , we are able to test our analytic rgb family in sect [ s : testfits ] . our conclusions are in sect . [ s : conclusioni ] . clusters have been observed with the eso / dutch 0.9 m telescope at la silla , and at the rgo / jkt 1 m telescope in la palma . this database comprises @xmath19 of the ggc whose distance modulus is @xmath20 . the zero - point uncertainties of our calibrations are @xmath21 mag for each band . three clusters were observed both with the southern and the northern telescopes , thus providing a consistency check of the calibrations : no systematic differences were found , at the level of accuracy of the zero - points . a detailed description of the observations and reduction procedures will be given in forthcoming papers ( rosenberg et al . @xcite , @xcite ) presenting the single clusters . a subsample of this database was used for the present investigation . we retained those clusters whose cmd satisfied a few criteria : ( a ) the hb level could be well determined ; ( b ) the rgb was not heavily contaminated by foreground / background contamination ; and ( c ) the rgb was well defined up to the tip . this subsample largely overlaps that used for the age investigation , but a few clusters whose to position could not be measured , are nevertheless useful for the metallicity indices definition . conversely , in a few cases the lower rgb could be used for the color measurements , while the upper branch was too scarcely defined for a reliable definition of the fiducial line . two of the cmds that were used are shown in figs . [ f : pars - partea ] ( ngc 1851 ) and [ f : pars - partb ] ( ngc 104 ) , and they illustrate the good quality of the data . the dataset of clusters used in this paper is listed in table [ t : the - sample ] . from left to right , the columns contain the ngc number , the reddening both in @xmath22 and @xmath15 , the metallicity according to three different scales , and the apparent magnitude of the horizontal branch ( hb ) . the @xmath23 values were taken from the harris ( @xcite ) on - line table . the @xmath15 reddenings were obtained by assuming that @xmath24 ( dean et al . the values of the metallicity were taken from rhs97 : they represent the equivalent widths of the caii infrared triplet , calibrated either onto the zinn & west ( @xcite ) scale ( zw column ) or the carretta & gratton ( @xcite ) scale ( rhs97 column ) . moreover , the original carretta & gratton metallicities ( cg column ) are also given for the clusters comprised in their sample . the hb level was found in different ways for clusters of different metallicity . for the the metal rich and metal intermediate clusters , a magnitude distribution of the hb stars was obtained , and the mode of the distribution was taken . where the hb was too scarcely populated , a horizontal line was fitted through the data . the blue tail of the metal poorest clusters does not reach the horizontal part of the branch : in that case , a fiducial hb was fitted to the tail , and the magnitude of the horizontal part was taken as the reference level . the fiducial branch was defined by taking a cluster having a bimodal hb color distribution ( ngc 1851 , cf . [ f : pars - partea ] ) and then extending its hb both to the red and to the blue by `` appending '' clusters being more and more metal rich and metal poor , respectively . the details of this procedure , as well as the errors associated to the @xmath25 in table [ t : the - sample ] , are discussed in rspa99 . for ngc 1851 , @xmath26 was adopted ( dashed line in fig . [ f : pars - partea ] ) , and this value is just @xmath27 mag brighter than the value found by walker ( @xcite ) and saviane et al . ( @xcite ) . based on this observational sample , a set of metallicity indices were measured on the rgbs of the clusters . in the next section , the indices are defined and the measurement procedures are described . consistency checks are also performed . the metallicity indices calibrated in this study are represented and defined in fig . [ f : pars - partea ] and fig . [ f : pars - partb ] . the figures represent the cmd of ngc 1851 and ngc 104 in different color - magnitude planes , and the crosses mark the position of the rgb points used in the measurement of the indices . the left panel of fig . [ f : pars - partea ] shows the apparent colors and magnitudes for ngc 1851 : the inclined line helps to identify the first index , @xmath1 . this was defined , in the @xmath28 plane , by hartwick ( @xcite ) as the slope of the line connecting two points on the rgb : the first one at the level of the hb , and the second one 2.5 mag brighter . we use the same definition for the @xmath4 plane here ; however , in order to be able to use our metal richest clusters , we redefined @xmath1 by measuring the second rgb point mag brighter than the hb . since @xmath1 is measured on the apparent cmd , it is independent both from the reddening and the distance modulus . the right panel of the same figure , shows the apparent @xmath14 magnitude vs. the de - reddened @xmath29 color . in this panel , four other indices are identified , i.e. @xmath30 , @xmath2 , @xmath6 , and @xmath3 . the first one is the rgb color at the level of the hb , and the other three measure the magnitude difference between the hb and the rgb at a fixed color @xmath31 , 1.2 and 1.4 mag . the former index was originally defined by sandage & smith ( @xcite ) and the latter one by sandage & wallerstein ( @xcite ) , in the @xmath32 plane . the other two indices , @xmath2 and @xmath6 , are introduced later to measure the metal richest gcs . these indices require an independent color excess determination . finally , fig . [ f : pars - partb ] shows the cmd of ngc 104 ( 47 tuc ) in the absolute @xmath33 plane : the adopted distance modulus , @xmath34 , was obtained by correcting the apparent luminosity of the hb according to lee et al . ( @xcite ; cf . [ s : calibrations ] ) . by comparison , harris catalog reports @xmath35 . two other indices are represented in the figure : @xmath7 and @xmath8 . they are defined as the rgb color at a fixed absolute @xmath36 magnitude of @xmath37 ( da costa & armandroff @xcite ) or @xmath38 ( lee et al . the latter index was also discussed by armandroff et al . ( @xcite ) , and a calibration formula was given in caldwell et al . ( @xcite ) . this is based on the da90 clusters plus m5 and ngc 362 from lloyd evans ( @xcite ) . since these two indices are defined on the bright part of the rgb , they can be measured even for the farthest objects of the local group ( lg ) . due to the fast luminosity evolution of the stars on the upper rgb , this part of the branch was typically under - sampled by the early small - size ccds , so no wide application of these indices has been made for galactic globulars . however , this is of no concern for galaxy - size stellar systems . it will be shown in sect . [ s : calibrations ] that good accuracies can be obtained even for gcs , provided that the analytic function of eq . ( [ e : iperbole ] ) is used . lrrrrrrr & & & & + & & & & & & & + 104 & 0.99 & 4.13 & 0.78 & 1.27 & 1.87 & 1.94 & 1.57 + 288 & 0.95 & 6.39 & 1.25 & 1.75 & 2.36 & 1.51 & 1.35 + 362 & 0.90 & 7.28 & 1.67 & 2.09 & 2.57 & 1.45 & 1.28 + 1261 & 0.91 & 7.77 & 1.62 & 2.13 & 2.73 & 1.39 & 1.25 + 1851 & 0.97 & 7.41 & 1.23 & 1.82 & 2.55 & 1.45 & 1.31 + 1904 & 0.94 & 8.56 & 1.58 & 2.14 & 2.83 & 1.35 & 1.24 + 3201 & 0.99 & 8.72 & 1.19 & 1.91 & 2.71 & 1.39 & 1.27 + 4590 & 0.91 & 9.98 & 1.90 & 2.52 & 3.25 & 1.24 & 1.16 + 4833 & 0.92 & 9.25 & 1.80 & 2.36 & 3.12 & 1.28 & 1.19 + 5272 & 0.91 & 7.60 & 1.66 & 2.13 & 2.81 & 1.36 & 1.24 + 5466 & 0.91 & 9.85 & 1.93 & 2.50 & 3.18 & 1.24 & 1.16 + 5897 & 0.97 & 8.73 & 1.34 & 2.00 & 2.79 & 1.35 & 1.25 + 5904 & 0.93 & 6.91 & 1.41 & 1.91 & 2.55 & 1.44 & 1.30 + 6093 & 0.93 & 8.02 & 1.58 & 2.12 & 2.91 & 1.34 & 1.24 + 6171 & 1.07 & 5.66 & 0.31 & 1.09 & 1.93 & 1.67 & 1.49 + 6205 & 0.89 & 7.70 & 1.75 & 2.20 & 2.75 & 1.37 & 1.23 + 6218 & 0.95 & 7.09 & 1.34 & 1.88 & 2.51 & 1.46 & 1.31 + 6254 & 0.90 & 8.25 & 1.75 & 2.29 & 3.17 & 1.30 & 1.21 + 6341 & 0.88 & 9.92 & 2.15 & 2.69 & 3.40 & 1.21 & 1.13 + 6352 & 1.12 & 3.11 & -0.16 & 0.52 & 1.30 & 1.99 & 1.75 + 6362 & 0.93 & 5.84 & 1.31 & 1.76 & 2.32 & 1.55 & 1.37 + 6397 & 0.89 & 9.45 & 1.98 & 2.49 & 3.12 & 1.26 & 1.16 + 6541 & 1.01 & 8.59 & 1.03 & 1.77 & 2.67 & 1.39 & 1.29 + 6637 & 0.96 & 4.39 & 0.96 & 1.41 & 1.97 & 1.82 & 1.53 + 6656 & 0.86 & 10.32 & 2.27 & 2.69 & 2.96 & 1.24 & 1.12 + 6681 & 0.95 & 7.54 & 1.35 & 1.92 & 2.76 & 1.37 & 1.27 + 6723 & 1.01 & 6.02 & 0.76 & 1.38 & 2.18 & 1.55 & 1.41 + 6752 & 0.99 & 7.16 & 1.08 & 1.69 & 2.46 & 1.45 & 1.33 + 6779 & 0.94 & 8.74 & 1.60 & 2.18 & 2.94 & 1.32 & 1.22 + 6809 & 0.93 & 9.38 & 1.72 & 2.29 & 2.87 & 1.32 & 1.20 + 7078 & 0.88 & 9.82 & 2.10 & 2.62 & 3.27 & 1.23 & 1.14 + colors and magnitudes were measured on a fiducial rgb , which has been found by least - square fitting an analytic function to the observed branch . after some experimenting , it was found that the best solution is to use the following relation : @xmath39 where @xmath40 and @xmath41 represent the color and the magnitude , respectively . one can see from figs . [ f : pars - partea ] and [ f : pars - partb ] that the function is indeed able to represent the giant branch over the typical metallicity range of globular clusters . moreover , it is shown in sect . [ s : newda90 ] that , when the cmds are corrected for distance and reddening , the four coefficients can be parametrized as a function of [ fe / h ] , so that one is able to reproduce the rgb of each cluster , using just one parameter : the metallicity . at any rate , the indices were measured on the original loci , so that an independent check of the goodness of the generalized hyperbolae can be made , by comparison of the measured vs. predicted indices . all the indices values that have been measured are reported in table [ t : the - indices ] . in this table , the cluster ngc number is given in column 1 ; the following columns list , from left to right , @xmath5 , @xmath1 , @xmath2 , @xmath6 , @xmath3 , and finally the rgb color measured at @xmath42 and @xmath43 . the lee et al . ( 1990 ) distance scale was used to compute the last two indices ( cf . [ s : calibrations ] ) . before discussing the indices as metallicity indicators , we checked their internal consistency . we will show in sect . [ s : calibrations ] that the index @xmath1 is the most accurate one , as expected , since it does not require reddening and distance corrections . the rest of the indices are therefore plotted vs. @xmath1 in figs . [ f : intcheck2 ] and [ f : intcheck1 ] , and we expect that most of the scatter will be in the vertical direction . second order polynomials were fitted to the distributions , and the _ rms _ of the fit was computed for each index . in order to intercompare the different indices , a relative uncertainty has been computed by dividing the _ rms _ by the central value of each parameter ( this value is identified by a dotted line in each figure ) . in this way , the scatter of the metal index @xmath44 is @xmath45 , 0.02 , 0.04 , 0.06 , 0.12 , and 0.26 , for the indices @xmath46 , @xmath8 , @xmath5 , @xmath3 , @xmath6 , and @xmath2 , respectively . these values confirm the visual impression of the figures , that @xmath47 and @xmath48 are the lowest dispersion indices , followed by @xmath5 and @xmath3 . the indices will be calibrated in terms of [ fe / h ] in sect . [ s : calibrations ] ; however , before moving to this section , we want to present a new way to provide `` standard '' ggc branches in the @xmath10 plane , along the lines of the classical da costa & armandroff ( @xcite ) study . using this family of rgb branches , we are able to make predictions on the trend of the already defined indices with metallicity ; these trends can thus be compared to the observed ones , and therefore provide a further test of the reliability of our rgb family ( cf . sect [ s : testfits ] ) . lcccccc cluster & @xmath49 & @xmath50 & @xmath51 & [ fe / h]@xmath52 & [ fe / h]@xmath53 & [ fe / h]@xmath54 + ngc 104 & 14.05 & 0.050 & 0.064 & @xmath55 & @xmath56 & @xmath57 + ngc 5904 & 15.00 & 0.023 & 0.029 & @xmath58 & @xmath59 & @xmath60 + ngc 288 & 15.40 & 0.036 & 0.046 & @xmath61 & @xmath62 & @xmath63 + ngc 6205 & 14.95 & 0.000 & 0.000 & @xmath64 & @xmath65 & @xmath66 + ngc 5272 & 15.58 & 0.002 & 0.003 & @xmath67 & @xmath65 & + ngc 6341 & 15.20 & 0.010 & 0.013 & @xmath68 & @xmath69 & + rrrrrrrrrrrr & & & & & + @xmath70 & @xmath71 & @xmath70 & @xmath71 & @xmath70 & @xmath71 & @xmath70 & @xmath71 & @xmath70 & @xmath71 & @xmath70 & @xmath71 + 13.782 & 0.978 & 15.359 & 0.914 & 15.492 & 0.852 & 14.725 & 0.926 & 14.645 & 0.867 & 15.060 & 0.852 + 13.604 & 0.994 & 15.107 & 0.939 & 15.151 & 0.874 & 14.457 & 0.942 & 14.322 & 0.890 & 14.720 & 0.872 + 13.443 & 1.008 & 14.849 & 0.960 & 14.789 & 0.892 & 14.221 & 0.961 & 14.033 & 0.909 & 14.395 & 0.894 + 13.317 & 1.021 & 14.593 & 0.984 & 14.597 & 0.910 & 14.040 & 0.978 & 13.788 & 0.929 & 14.079 & 0.916 + 13.075 & 1.045 & 14.342 & 0.999 & 14.359 & 0.929 & 13.878 & 0.994 & 13.595 & 0.944 & 13.789 & 0.937 + 12.862 & 1.070 & 14.109 & 1.018 & 14.143 & 0.955 & 13.700 & 1.009 & 13.381 & 0.966 & 13.533 & 0.953 + 12.619 & 1.101 & 13.881 & 1.036 & 13.796 & 0.990 & 13.456 & 1.032 & 13.170 & 0.984 & 13.303 & 0.974 + 12.346 & 1.136 & 13.649 & 1.062 & 13.517 & 1.021 & 13.190 & 1.061 & 12.984 & 1.005 & 13.082 & 0.994 + 12.035 & 1.185 & 13.376 & 1.090 & 13.265 & 1.046 & 12.916 & 1.091 & 12.832 & 1.019 & 12.850 & 1.020 + 11.761 & 1.231 & 13.058 & 1.132 & 13.005 & 1.076 & 12.655 & 1.122 & 12.631 & 1.045 & 12.611 & 1.039 + 11.461 & 1.281 & 12.766 & 1.173 & 12.759 & 1.110 & 12.419 & 1.154 & 12.363 & 1.077 & 12.351 & 1.067 + 11.101 & 1.362 & 12.534 & 1.210 & 12.519 & 1.148 & 12.231 & 1.183 & 12.118 & 1.111 & 12.075 & 1.102 + 10.696 & 1.459 & 12.380 & 1.233 & 12.302 & 1.187 & 12.073 & 1.212 & 11.945 & 1.138 & 11.771 & 1.148 + 10.330 & 1.600 & 12.163 & 1.268 & 12.109 & 1.227 & 11.868 & 1.254 & 11.844 & 1.156 & 11.492 & 1.195 + 10.062 & 1.720 & 11.928 & 1.317 & 11.878 & 1.275 & 11.615 & 1.305 & 11.707 & 1.178 & 11.284 & 1.233 + 9.877 & 1.856 & 11.617 & 1.411 & 11.741 & 1.310 & 11.335 & 1.371 & 11.571 & 1.204 & 11.154 & 1.265 + 9.706 & 2.019 & 11.427 & 1.483 & 11.575 & 1.344 & 11.116 & 1.422 & 11.395 & 1.252 & 11.008 & 1.295 + 9.602 & 2.148 & & & 11.494 & 1.377 & 10.902 & 1.489 & 11.141 & 1.312 & 10.854 & 1.320 + 9.524 & 2.315 & & & 11.330 & 1.406 & 10.652 & 1.585 & 10.870 & 1.376 & 10.709 & 1.351 + 9.573 & 2.576 & & & 11.240 & 1.447 & 10.457 & 1.680 & 10.643 & 1.444 & & + 9.619 & 2.768 & & & 11.112 & 1.488 & 10.343 & 1.742 & 10.552 & 1.492 & & + & & & & 11.078 & 1.528 & & & & & & + & & & & 11.047 & 1.546 & & & & & & + da costa & armandroff ( @xcite ) presented in tabular form the fiducial ggc branches of 6 globulars , covering the metallicity range @xmath72\leq -0.71 $ ] . the rgbs were corrected to the absolute @xmath10 plane using the apparent @xmath14 magnitude of the hb , and adopting the lee et al . ( @xcite ) theoretical hb luminosity . since the da90 study , these branches have been widely used for stellar population studies in the local group . based on these rgbs , in particular , a method to determine both the distance and mean metallicity of an old stellar population was presented by lee et al . ( @xcite ) . both da90 and lee et al . ( 1993 ) provided a relation between the metallicity [ fe / h ] and the color of the rgb at a fixed absolute @xmath36 magnitude ( @xmath42 and @xmath43 , respectively ) , and recently a new relation for @xmath73 has also been obtained by caldwell et al . ( @xcite ) . once the distance of the population is known ( e.g. via the luminosity of the rgb tip ) , then an estimate of its _ mean _ metallicity can be obtained using one of the calibrations . it is assumed that the age of the population is comparable to that of the ggcs , and that the age spread is negligible compared to the metallicity spread ( rspa99 ) . in such case , one expects that any rgb star s position in the absolute cmd is determined just by its metallicity , and that a better statistical determination of the population s metal content would be obtained by converting the color of _ each _ star into a [ fe / h ] value . with this idea in mind , in the following sections we will show that this is indeed possible , at least for the bright / most sensitive part of the giant branch . we found that a relatively simple _ continuous _ function can be defined in the @xmath9 $ ] space , and that this function can be used to transform the rgb from the @xmath10 plane to the @xmath74},m_{i } $ ] plane . in order to obtain this function , we first selected a subsample of clusters with suitable characteristics , so that a reference rgb grid can be constructed . the fiducial branches for each cluster were then determined in an objective way , and they were corrected to the absolute @xmath75 plane . in this plane , the analytic function was fitted to the rgb grid . these operations are described in the following sections . the clusters that were used for the definition of the fiducial rgbs are listed in table [ t : fiducialgc ] , in order of increasing metallicity . the table reports the cluster name , and some of the parameters listed in table [ t : the - sample ] are repeated here for ease of use . the values of the reddening were in some cases changed by a few thousandth magnitudes ( i.e. well within the typical uncertainties on @xmath23 ) , to obtain a sequence of fiducial lines that move from bluer to redder colors as [ fe / h ] increases , and again the corresponding @xmath76 values were obtained assuming that @xmath24 ( dean et al . @xcite ) . indeed , due to the homogeneity of our sample , we expect that if a monotonic color / metallicity sequence is not obtained , then only the uncertainties on the extinction values must be taken into account . in order to single out these clusters from the total sample , some key characteristics were taken into account . in particular , we considered clusters whose rgbs are all well - defined by a statistically significant number of stars ; they have low reddening values ( @xmath77 ) ; and they cover a metallicity range that includes most of our ggcs ( @xmath78\leq -0.7 $ ] on the zw scale ) . the da90 fiducial clusters were ngc 104 , ngc 1851 , ngc 6752 , ngc 6397 , ngc 7078 and ngc 7089 ( m2 ) . ngc 104 is the only cluster in common with the previous study , and m2 is not present in our dataset . the other objects have been excluded from our fiducial sample since they have too large reddening values ( @xmath79 for ngc 6397 and ngc 7078 ) , or their rgbs are too scarcely populated in our cmds ( ngc 1851 and ngc 6752 ) . nevertheless , the calibrations that we obtain for the @xmath7 and @xmath8 are in fairly good agreement with those obtained by da90 ( for the small discrepancies at the high metallicity end , cf . [ s : da90indices1 ] and [ s : da90indices2 ] ) , and in particular with the recent caldwell et al . ( @xcite ) calibration for the @xmath73 index . the ridge lines of our fiducial rgbs were defined according to the following procedure . the rgb region was selected from the calibrated photometry , by excluding both hb and agb stars . all stars bluer than the color of the rr lyr gap were removed ; agb stars were also removed by tracing a reference straight line in the cmd , and by excluding all stars blue - side of this line . this operation was carried out in the @xmath80 plane , where the rgb curvature is less pronounced , and a straight line turns out to be adequate . the fiducial loci were then extracted from the selected rgb samples . the @xmath15 and @xmath36 vectors were sorted in magnitude , and bins were created containing a given number of stars . within each bin , the median color of the stars and the mean magnitude were used as estimators of the bin central color and brightness . the number of stars within the bins was exponentially increased going from brighter to fainter magnitudes . in this way , ( a ) one can use a small number of stars for the upper rgb , so that the color of the bin is not affected by the rgb slope , and ( b ) it is possible to take advantage of the better statistics of the rgb base . finally , the brightest two stars of the rgb were not binned , and were left as representatives of the top branch . after some experimenting , we found that a good rgb sampling can be obtained by taking for each bin a number of stars which is proportional to @xmath81 , where @xmath44 is an integer number . the resulting fiducial vectors were smoothed using an average filter with a box size of 3 . the rgb regions of the 6 clusters are shown in fig . [ f : zooms ] , together with the fiducial lines : it can be seen that in all cases the agbs are easily disentangled from the rgbs . the values of the fiducial points corresponding to the solid lines in fig . [ f : zooms ] , are listed in table [ t : fidtable ] . the fiducial branches defined in sect . [ s : fiducials ] were fitted with a parametrized family of hyperbolae . first , the rgbs were moved into the absolute @xmath10 plane . the distance modulus was computed from the apparent magnitude of the hb ( cf . table [ t : fiducialgc ] ) and by assuming the common law @xmath82+b $ ] ; in order to compare our results with those of da90 , @xmath83 and @xmath84 were used , but we also obtained the same fits using more recent values as in carretta et al . ( @xcite ) , i.e. @xmath85 and @xmath86 . the rgb was modeled with an hyperbola as in rosenberg et al . ( @xcite ) , but in this case the coefficients were taken as second order polynomials in [ fe / h ] . in other words , we parametrized the whole family of rgbs in the following way : @xmath87\ ] ] where @xmath88^{2}+k_{2}[{\rm fe / h}]+k_{3}\ ] ] @xmath89^{2}+k_{5}[{\rm fe / h}]+k_{6}\ ] ] @xmath90^{2}+k_{8}[{\rm fe / h}]+k_{9}\ ] ] @xmath91 the list of the parameters of the fits in magnitude is reported in table [ t : coeffs ] , together with the _ rms _ of the residuals around the fitting curves . the table shows that the parameter @xmath92 does not depend on the choice of the distance scale , as expected . even the other coefficients are little dependent on the distance scale , apart from @xmath93 . it is affected by the zero - point of the hb luminosity - metallicity relation , and indeed there is the expected @xmath94 mag difference going from the ldz to the c99 distance scale . one could question the choice of a constant @xmath92 , but after some training on the theoretical isochrones , we found that even allowing for a varying parameter , its value indeed scattered very little around some mean value . this empirical result is a good one , in the sense that it allows to apply a robust linear least - square fitting method for any choice of @xmath95 , and then to search for the best value of this constant by a simple _ rms _ minimization . we chose to fit the @xmath96\ } $ ] function , and not the @xmath97 ) $ ] function , since the latter one would be double - valued for the brightest part of the metal rich clusters rgbs . this choice implies that our fits are not well - constrained for the vertical part of the giant branch , i.e. for magnitudes fainter than @xmath98 . however , we show in the next section that our analytic function is good enough for the intended purpose , i.e. to obtain the [ fe / h ] of the rgb stars in far local group populations , and thus to analyze how they are distributed in metallicity . our synthetic rgb families are plotted in figs . [ f : fitszw ] and [ f : fitscg ] , for the ldz distance scale . in the former figure , the zw metallicity scale is used , while the cg scale is used in the latter one . the figures show that the chosen functional form represents a very good approximation to the true metallicity `` distribution '' of the rg branches . the _ rms _ values are smaller than the typical uncertainties in the distance moduli within the local group . we further stress the excellent consistency of the empirical fiducial branches for clusters of similar metallicity . we have two pairs of clusters whose metallicities differ by at most 0.03 dex ( depending on the scale ) : ngc 288 and ngc 5904 on the one side , and ngc 5272 and ngc 6205 on the other side . the figures show that the fiducial line of ngc 288 is similar to that of ngc 5904 , and the ngc 5272 fiducial resembles that of ngc 6205 , further demonstrating both the homogeneity of our photometry and the reliability of the procedure that is used in defining the cluster ridge lines . if the coefficients of the hyperbolae are taken as third order polynomials , the resulting fits are apparently better ( the _ rms _ is @xmath99 mag ) ; however , the trends of the metallicity indices show an unphysical behavior , which is a sign that further clusters , having metallicities not covered by the present set , would be needed in order to robustly constrain the analytic function . in the following section , the indices are calibrated in terms of metallicity , so that in sect . [ s : testfits ] they will be used to check the reliability of our generalized fits . .the coefficients that define the functions used to interpolate our rgbs ( see text ) ; the top header line identifies the two distance scales used , while the two metallicities are identified in the second line of the header[t : coeffs ] [ cols= " > , > , > , > , > " , ] cccccccc & d.sc . & metallicity & @xmath100 & @xmath101 & @xmath102 & _ rms _ & fit + & & cg & -0.03 & 0.23 & -1.19 & 0.13 & 2 + & & zw & -0.004 & -0.18 & 0.08 & 0.12 & 2 + & & zw & -0.24 & 0.28 & & 0.12 & 1 + @xmath8 & ldz & cg & 0.00487 & -0.0057 & & 0.13 & @xmath103 + & & zw & -2.12 & 8.81 & -9.75 & 0.13 & 2 + & c99 & cg & 0.0045 & -0.0053 & & 0.15 & @xmath103 + & & zw & -2.05 & 8.57 & -9.61 & 0.12 & 2 + @xmath7 & ldz & cg & 0.0068 & -0.0076 & & 0.15 & @xmath103 + & & zw & -3.34 & 12.37 & -11.91 & 0.14 & 2 + & c99 & cg & 0.0065 & -0.0073 & & 0.15 & @xmath103 + & & zw & -3.233 & 12.23 & -11.96 & 0.14 & 2 + @xmath3 & & cg & -0.34 & 0.93 & -1.37 & 0.16 & 2 + & & zw & -0.063 & -0.56 & 0.41 & 0.16 & 2 + & & zw & -0.87 & 0.77 & & 0.16 & 1 + @xmath6 & & cg & -0.36 & 0.55 & -0.97 & 0.19 & 2 + & & cg & -0.69 & 0.0007 & & 0.22 & 1 + & & zw & -0.13 & -0.38 & -0.28 & 0.20 & 2 + & & zw & -0.82 & 0.06 & & 0.20 & 1 + @xmath2 & & cg & -0.30 & 0.09 & -0.81 & 0.23 & 2 + & & cg & -0.59 & -0.52 & & 0.25 & 1 + & & zw & -0.13 & -0.42 & -0.68 & 0.25 & 2 + & & zw & -0.70 & -0.56 & & 0.25 & 1 + @xmath5 & & cg & 4.25 & -5.37 & & 0.32 & 1 + & & zw & 5.25 & -6.52 & & 0.33 & 1 + in order to obtain analytic relations between the indices and the actual metallicity , our photometric parameters were compared both with the zw and the cg values . a summary of the resulting equations is given in table [ t : rms ] . for each index ( first column ) both linear and quadratic fits were tried , of the form : @xmath104=\alpha \cdot index + \beta $ ] and @xmath104=\alpha \cdot index^{2}+\beta \cdot index + \gamma $ ] . the coefficients of the calibrating relation are given in the columns labelled @xmath105 , @xmath106 , and @xmath107 ; in column 7 , the _ rms _ of the residuals is also given . in the case of the @xmath7 and @xmath8 indices , neither the linear nor the quadratic fits give satisfactory results , when the cg scale is considered . instead , a good fit is obtained if a change of variables is performed , setting @xmath108 } $ ] , and linearly interpolating in the index ( i.e. setting @xmath109 ) . the column 8 of table [ t : rms ] identifies the kind of fitting function that is used for each parameter / metallicity combination : the symbols `` 1 '' , `` 2 '' and `` @xmath103 '' refer to the linear , quadratic , and linear in @xmath103 fits , respectively . relations on both the cg and zw metallicity scales are given , and column 3 flags the [ fe / h ] scale that is used . in order to measure the @xmath46 and @xmath8 indices ( cf . [ s : indices ] ) a distance scale must be adopted . the most straightforward way is to use the observed @xmath25 ( cf . table [ t : the - sample ] ) coupled with a suitable law for the hb absolute magnitude . it has become customary to parameterize this magnitude as @xmath110+b $ ] , although there is no consensus on the value of the two parameters @xmath111 and @xmath112 . the current calibrations of these two metallicity indices were obtained by da costa & armandroff ( @xcite ) and lee et al . ( @xcite ) , and they are based on the lee et al . ( @xcite ; ldz ) theoretical luminosities of the hb . ldz gave a relation @xmath113 + 0.82 $ ] valid for @xmath114 . as discussed in sect [ s : newda90 ] , since many current determinations of population ii distances within the local group are based on the lee et al . ( @xcite ) distance scale , and for the purpose of comparison with previous studies , we provide a calibration using the latter hb luminosity - metallicity relation . however , in the last ten years revisions of this relation have been discussed by many authors , so we also calibrated the two indices using @xmath115 + 0.90 $ ] ( carretta et al . @xcite ) , which is one of the most recent hb - based distance scales . we must stress that _ metallicities on the zw scale must be used in the @xmath116 vs. _ [ fe / h ] _ relation_. indeed , cg showed that their scale is not linearly correlated to that of zw , so not even the _ @xmath116 vs. _ [ fe / h ] relation will be linear : if one wishes to use the new scale , then _ the absolute magnitude of the hb must be re - calibrated _ in a more complicated way . the best calibrating relations are shown in figs . [ f : cals ] to [ f : caldv14 ] . in the following sections , for each index a few remarks on the accuracy of the calibrations and comparisons with past studies are given . on the cg scale , the second - order fit has a residual _ rms _ of 0.12 dex in [ fe / h ] . on the zw scale , the linear fit is obtained with a _ rms _ of 0.12 dex . this index can therefore be calibrated on both scales , with a comparable level of accuracy . a parabolic fit does not improve the relation on the zw scale , since the coefficient of the quadratic term is very small ( -0.004 ) and the _ rms _ is the same . these relations are shown in fig . [ f : cals ] as solid lines , where the upper panel is for the zw scale , and the lower panel for the cg scale ( this layout is reproduced in all the following figures ) . the cluster ngc 6656 ( m22 ) was excluded from the fits , and is plotted as an open circle in fig . [ f : cals ] . it is well - known that m22 is a cluster that shows a metallicity spread , and indeed it falls outside the general trend in most of the present calibrations . the first definition of the @xmath7 index was given in da costa & armandroff ( @xcite ) , where a calibration in terms of the zw scale was also given : @xmath11=-15.16 + 17.0\ , ( v - i)_{-3}-4.9\ , ( v - i)_{-3}^{2 } $ ] . the same index ( measured on the _ absolute _ rgbs corrected with the ldz hb luminosity - metallicity relation ) is plotted , in fig . [ f : calvi-3 ] , as a function of the metallicity on both scales , and the solid lines represent our calibrations . the top panel shows the quadratic relation on the zw scale , whose _ rms _ is 0.14 dex . the bottom panel of fig . [ f : calvi-3 ] shows the relation on the cg scale . in this case , a quadratic fit is not able to reproduce the trend of the observational data . a better result can be obtained by making a variable change , i.e. using the variable @xmath119 } $ ] ; in this case , a linear relation is found , and its _ rms _ is 0.15 dex . this measure of the residual scatter has been computed after transforming back to metallicity , so the reliability of the index can be compared to that of the other ones . again , the index can be calibrated on both scales with a comparable accuracy . the dashed curve in the upper panel of fig . [ f : calvi-3 ] shows the original relation obtained by da90 : there is a small discrepancy at the high - metallicity end , which can be explained by the different 47 tuc fiducial line that was adopted by da90 ( cf . below the discussion on @xmath8 ) . as already recalled , we checked the effect of adopting another distance scale , by repeating our measurements and fits , and adopting the c99 distance scale . for the zw metallicity scale , we obtain the quadratic relation whose coefficients are listed in table [ t : rms ] , and whose _ rms _ is 0.15 dex . the bottom panel of fig . [ f : calvi-3 ] shows the relation on the cg scale . again , a quadratic fit is not able to reproduce the trend of the observational data . making the already discussed variable substitution , the linear relation in @xmath103 has an _ rms _ of 0.16 dex , so the two metallicity scales yield almost comparable results . using the same `` standard '' gc branches of da90 , lee et al . ( @xcite ) defined a new index , @xmath8 , to be used for the farthest population ii objects . it was also calibrated in terms of the zw scale : @xmath11=-12.64 + 12.6\ , ( v - i)_{-3.5}-3.3\ , ( v - i)_{-3.5}^{2 } $ ] . a new calibration was also given recently in caldwell et al . ( @xcite ) : [ fe / h]@xmath121 , where @xmath122 $ ] . the index and our calibrations ( solid lines ) are plotted , in fig . [ f : calvi-3.5 ] , on both metallicity scales . again , the measurements were made in the absolute cmd , assuming the ldz distance scale . our quadratic calibration vs. the zw scale has a residual _ rms _ scatter of 0.13 dex , which is the same of the linear relation on the cg metallicity vs. @xmath103 . the lee et al . relation ( dashed line ) predicts slightly too larger metallicities on the zw scale , for @xmath104>-1 $ ] . this can also be interpreted as if the da90 47 tuc branch were @xmath123 mag bluer than ours . indeed , if one looks at fig . 5 of da90 , one can easily see that some weight is given to the brightest rgb star , which is brighter than the trend defined by the previous ones . the result is a steeper branch , which also justifies the da90 slightly bluer rgb fiducial . since our metal richest point is defined by two clusters , and since the two measured parameters agree very well , we are confident that our calibration is reliable . in any case , the discrepancy between the two scales is no larger than @xmath94 dex . it is also reassuring that the caldwell et al . ( @xcite ) relation ( pluses ) is closer to the present calibration , since the former is based on a larger set of clusters . this might be an indication that the lee et al . relation is actually inaccurate at the metal rich end , due to the small set of calibrating clusters . as before , we obtained a further calibration also using the c99 @xmath116 vs. [ fe / h ] relation ; the quadratic fit on the zw scale has a residual _ rms _ scatter of 0.13 dex , while the @xmath103 variable can be fitted with a straight line , with an _ rms _ of 0.14 dex . for any @xmath126 index , the quadratic relations vs. the zw metallicity do not improve the _ rms _ and they are not plotted in the figures . the coefficients are listed in table [ t : rms ] . the best metallicity estimates of the `` @xmath126 family '' are obtained with the @xmath3 index . the errors on @xmath11 $ ] are just slightly larger than the standard uncertainties of the spectroscopic determinations . the solid lines of fig . [ f : caldv14 ] show the calibrations that we obtain . the quadratic equation on the cg scale , and the linear one on the zw scale , are obtained with residual scatters of 0.16 dex . the rest of the indices in this family , and @xmath127 , lack the precision of the other abundance indicators . this is due to the fact that the error on any @xmath126 index is proportional to the uncertainty on the color of the rgb ( which depends on the reddening ) , times its local slope where the reference point is measured . since the rgb slope increases going away from the tip ( i.e. towards bluer colors ) , we expect that the scatter on the @xmath126 indices will also increase as the color of the reference point gets bluer . indeed , table [ t : rms ] shows that in most cases the _ rms _ uncertainties are @xmath128 dex for these indices . the residual scatter is largest for the @xmath127 index , which is the most affected by the uncertainties on the reddening . the @xmath6 and @xmath127 parameters have been earlier calibrated , on the cg scale , by carretta & bragaglia ( @xcite ) . using their quadratic relation for @xmath6 , and both their linear and quadratic relations for @xmath127 , the corresponding _ rms _ of the residuals in metallicity are 0.21 dex and @xmath129 dex , respectively . our new and the old calibrations are therefore compatible , within the ( albeit large ) uncertainties . a straightforward test of our new analytic rgbs can be made by generating the same metallicity indices that have been measured on the observed rgbs , and then checking the consistency of the predicted vs. measured quantities . to this aim , for a set of discrete [ fe / h ] values a @xmath130 vector was generated , and the combination of the two was used to compute the @xmath131 vector of the giant branch , using eqs . ( [ e : general]-[e : d ] ) . then for each branch the metallicity indices were measured as it was done for the clusters fiducials . in figs . [ f : cals ] to [ f : caldv14 ] , the predicted indices are identified by the small open squares ( spaced by 0.1 dex ) connected by a solid line . the best predictions are for those indices that rely on the brightest part of the rgb ( i.e. @xmath7 , @xmath8 and @xmath132 ) , while the computations are partially discrepant for those indices that rely on a point that is measured on the faint rgb . this is easily explained by the nature of our fit : since the best match is searched for along the ordinates ( for the reasons discussed in sect . [ s : newda90 ] ) , then it is better constrained in the upper part of the rgb , where its curvature becomes more sensitive to metallicity . we must also stress that the metal richest cluster in the reference grid is 47 tuc ( [ fe / h]@xmath133 on the zw scale ) , whereas ngc 6352 ( [ fe / h]@xmath134 on the same scale ) is the metal richest cluster for which metallicity indices have been measured . some of the discrepancies that are seen at the highest metallicities are therefore due to the lack of low - reddening clusters that can be used to extend the reference grid to the larger [ fe / h ] values . the mean differences between the predicted and fitted indices are , on the zw scale , around 0.03 dex for the @xmath7 and @xmath8 indices . they are around 0.08 dex for the @xmath6 , @xmath3 , and @xmath1 indices . they rise to @xmath94 and @xmath135 dex for the @xmath2 and @xmath5 indices . a similar trend is seen for the comparison on the cg scale . in this case , the mean differences are @xmath99 dex for @xmath7 , @xmath8 , and @xmath1 ; they are @xmath94 dex for @xmath6 and @xmath3 ; and they are 0.12 and 0.27 for the @xmath2 and @xmath5 indices . we can therefore conclude that , apart from the @xmath2 and @xmath5 indices , our mono - parametric rgb family gives a satisfactory reproduction of the actual changes of the rgb morphology and location , as a function of metallicity . it is then expected that , using this approach , one can exploit the brightest @xmath136 mags of the rgb to determine the mean metallicity , and even more important , the metallicity _ distribution _ of the old stellar population of any local group galaxy . in a forthcoming paper , we will demonstrate such possibility by re - analyzing our old photometric studies of the dwarf spheroidal galaxies tucana ( saviane et al . @xcite ) , phoenix ( held et al . @xcite ; martnez - delgado et al . @xcite ) , fornax ( saviane et al . @xcite ) , lgs 3 ( aparicio et al . @xcite ) , leo i ( gallart et al . @xcite ; held et al . @xcite ) and ngc 185 ( martnez - delgado et al . in this work , we have provided the first calibration of a few metallicity indices in the @xmath137 plane , namely the indices @xmath1 , @xmath2 and @xmath3 . calibrations on both the zinn & west ( 1984 ) and carretta & gratton ( 1997 ) scales have been obtained . the metallicity indices @xmath5 , @xmath6 , @xmath7 and @xmath8 have been also calibrated on both scales , and we have shown that our new relations are consistent with existing ones . in the case of the latter two indices , we have obtained the first calibration on the cg scale ; for both scales , we have also obtained the first calibration that takes into account new results on the rr lyr distances . the accuracy of the calibrations is generally better than 0.2 dex , regardless of the metallicity scale that is used . our results are an improvement over previous calibrations , since a new approach in the definition of the rgb is used , and since our formulae are based on the largest homogeneous photometric database of galactic globular clusters . the availability of such database also allowed us a progress towards the definition of a standard description of the rgb morphology and location . we were able to obtain a function in the @xmath9 $ ] space which is able to reproduce the whole set of ggc giant branches in terms of a single parameter ( the metallicity ) . we suggest that the usage of this function will improve the current determinations of metallicity and distances within the local group , extending the methods of lee et al . ( 1993 ) . we thank the referee , gary da costa , for helpful suggestions that improved the final presentation of the manuscript . i.s . acknowledges the financial support of italian and spanish foreign ministries , through an ` azioni integrate / acciones integradas ' grant . aparicio a. , gallart c. , bertelli g. , 1997 , aj 114 , 680 armandroff t.e . , da costa g.s . , caldwell n. , seitzer p. , 1993 , aj 106 , 986 bertelli g. , bressan a. , chiosi c. , fagotto f. , nasi e. , 1994 , a&as 106 , 275 buonanno r. , corsi c.e . , pulone l. , fusi pecci f. , bellazzini m. , 1998 , a&a 333 , 505 ( b98 ) caldwell n. , armandroff t.e . , da costa g.s . , seitzer p. , 1998 , aj 115 , 535 carretta e. , bragaglia a. , 1998 , a&a 329 , 937 carretta e. , gratton r. , 1997 , a&as 121 , 95 ( cg ) carretta e. , gratton r.g . , clementini g. , fusi pecci f. , 1999 , apj , in press ( c99 ) da costa g.s . , armandroff t.e . , 1990 , aj 100 , 162 ( da90 ) dean j.f . , warren p.r . , cousins a.w.j . , 1978 , mnras 183 , 569 gallart c. , freedman w. , aparicio a. , bertelli g. , chiosi c. , 1999 , aj , , 118 , 2245 harris g.l.h . , harris w.e . , poole g.b . , 1999 , aj 117 , 855 harris w.e , 1996 , aj 112 , 1487 hartwick f.d.a . , 1968 , apj 154 , 475 held e.v . , saviane i. , momany y. , 1999a , a&a 345 , 747 held e.v . , saviane i. , momany y. , carraro g. , 1999b , apj , in press lee m.g . , freedman w.l . , madore b.f . , 1993 , apj 417 , 553 lee y.w . , demarque p. , zinn r. , 1990 , apj 350 , 155 ( ldz ) lloyd evans t. , 1983 , s. afr . circ . 7 , 86 martnez - delgado d. , aparicio . a. , gallart c. , 1999a , aj , 118 , 2229 martnez - delgado d. , gallart c. , aparicio a. , 1999b , aj , 118 , 862 rosenberg a. , saviane i. , piotto g. , aparicio a. , 1999a , aj , 118 , 2306 ( rspa99 ) rosenberg a. , piotto g. , saviane i. , aparicio a. , 1999b , a&as , in press rosenberg a. , aparicio a. , saviane i. , piotto g. , 1999c , a&as , submitted rutledge a.g . , hesser j.e . , stetson p.b . , 1997 , pasp 109 , 907 ( rhs97 ) sandage a. , smith l.l . , 1966 , apj 144 , 886 sandage a. , wallerstein g. , 1960 , apj 131 , 598 saviane i. , held e.v . , bertelli g. , 1999a , a&a , in press saviane i. , held e.v . , piotto g. , 1996 , a&a 315 , 40 saviane i. , rosenberg a. , piotto g. , 1997 . in : rood , a.renzini ( eds . ) advances in stellar evolution , cambridge university press , cambridge , p. 65 ( srp97 ) saviane i. , piotto g. , fagotto f. , et al . , 1998 , a&a 333 , 479 saviane i. , rosenberg a. , piotto g. , 1999b . in : b. k. gibson , t. s. axelrod , m. e. putman ( eds . ) , `` the third stromlo symposium : the galactic halo '' ( srp99 ) walker a. , 1992 , pasp 104 , 1063 zinn r. , west m. , 1984 , apjs 55 , 45 ( zw )

we analyze neutral higgs boson decays into squark pairs in the minimal supersymmetric extension of the standard model and improve previous analyses . in particular the treatment of potentially large higher - order corrections originating from the soft susy breaking parameters , the trilinear higgs coupling to sbottoms , and , the higgsino mass parameter , is investigated . the remaining theoretical uncertainties including the susy - qcd corrections are analyzed quantitatively . -1.0 cm psi pr1101 neutral higgs boson decays to squark pairs reanalyzed e. accomando , g. chachamis , f. fugel , m. spira and m. walser _ university of southampton , theory group , southampton so17 1bj , united kingdom + paul scherrer institut , ch-5232 villigen psi , switzerland + institute for theoretical physics , eth zrich , ch8093 zrich , switzerland _

1103.4283

the search for higgs bosons plays one of the most important roles at high - energy collider experiments at the tevatron and the lhc . the higgs boson is the remnant of electroweak symmetry breaking in the scalar higgs sector of the standard model ( sm ) and its supersymmetric extensions . the minimal supersymmetric extension of the sm ( mssm ) requires the introduction of two higgs doublets in order to preserve supersymmetry @xcite . this leads to the existence of five elementary higgs particles , two cp - even ( @xmath6 ) , one cp - odd ( @xmath7 ) and two charged ( @xmath8 ) states . at lowest order ( lo ) all couplings and masses of the mssm higgs sector are described by two independent input parameters , which are usually chosen as @xmath9 , the ratio of the two vacuum expectation values @xmath10 , and the pseudoscalar higgs - boson mass @xmath11 . at lo , the light scalar higgs mass @xmath12 has to be smaller than the @xmath13-boson mass @xmath14 . including the one - loop and dominant two - loop corrections the upper bound is increased to @xmath15 gev @xcite . recent first three - loop results confirm this upper bound within less than 1 gev @xcite . the couplings of the various neutral higgs bosons to fermions and gauge bosons depend on the mixing angles @xmath16 and @xmath17 . normalized to the sm higgs couplings , they are listed in table [ tb : hcoup ] . .[tb : hcoup ] _ higgs couplings in the mssm to fermions and gauge bosons [ @xmath18 relative to sm couplings . the subscripts @xmath19 denote up- and down - type fermions . _ [ cols="<,^,^,^,^",options="header " , ] the soft supersymmetry - breaking parameters @xmath20 and @xmath21 will be introduced as @xmath22 parameters at an input scale @xmath23 which will in general be of the order of the susy scale . in order to resum large corrections for large values of @xmath24 we define the sbottom masses and couplings to the higgs bosons in terms of the effective bottom mass definition which has been used in ref . our effective bottom mass runs with five active flavours consistently , i.e. all heavier particles are decoupled , while the @xmath25 definition requires running with the contributions of all supersymmetric particles . the scale of the strong coupling @xmath26 in @xmath27 within the effective bottom mass is identified with the average mass of the sbottom and gluino states in order to account for the nnlo corrections to a large extent @xcite and is thus different from the scale @xmath28 . ] @xcite @xmath29 with @xmath30 , where @xmath31 denotes the running bottom mass in the @xmath22 scheme at the scale @xmath28 and @xmath32 are the sbottom and gluino pole masses respectively . the effective top mass is identified with the running @xmath22 top mass @xmath33 the use of these effective masses corresponds to the consistent inclusion of the resummed and rg - improved yukawa couplings of the higgs decays @xmath34 everywhere in the squark sectors , too . the stop / sbottom mass matrix at lo is then given by ( @xmath35 ) @xmath36 \\ \hat m_q(q_0 ) [ \bar{a}_q(q_0)-\mu r_q ] & \tilde{\overline{m}}_{\tilde q_r}^2(q_0 ) + \hat m_q^2(q_0 ) \end{array } \right ] \ , , \ ] ] where @xmath37 denotes the running trilinear @xmath22 coupling at the scale @xmath23 and @xmath38 . analogous to eqs . ( [ eq : dterms ] ) the @xmath39-terms @xmath40 have been absorbed in the parameters @xmath41 , @xmath42 the stop / sbottom mass matrix is modified by higher - order corrections in the diagonal and off - diagonal entries . while the corrections to the off - diagonal entries will be treated by the renormalization of the mixing angles , we compensate the radiative corrections to the diagonal matrix elements by shifts in the soft mass parameters @xmath43 , @xmath44 in order to arrive at simple expressions at nlo for the stop / sbottom masses . using the squark pole masses @xmath45 the tree - level definition @xmath46 of the mixing angle is given by @xmath47 } { m_{\tilde q_1}^2 - m_{\tilde q_2}^2 } \quad , \quad \cos 2\tilde\theta_q = \frac{\tilde m_{\tilde q_l}^2(q_0 ) - \tilde m_{\tilde q_r}^2(q_0 ) } { m_{\tilde q_1}^2 - m_{\tilde q_2}^2 } \ , , \label{eq : sqmix}\ ] ] where the shifted squark mass parameters @xmath48 have been used with the radiatively corrected squark pole masses . this definition will be used as the tree - level - like mixing angle with loop - corrected squark masses @xmath45 at nlo , too , but will only play the role of an auxiliary parameter in our analysis as will be explained later . at nlo the masses of the stop / sbottom eigenstates are given by @xmath49 ^ 2 + 4\hat m_q^2(q_0 ) [ \bar{a}_q(q_0 ) - \mu r_q]^2 } \right ] + \delta m_{\tilde q_{1/2}}^2 \nonumber \\ \delta m_{\tilde q_{1/2}}^2 & = & \sigma_{11/22}(m_{\tilde q_{1/2}}^2 ) + \delta \hat m_{\tilde q_{1/2}}^2 \label{eq : sqmass}\end{aligned}\ ] ] the diagonal parts @xmath50 of the stop / sbottom self - energies can be calculated from the diagrams in fig . [ fg : sqself ] , ( 130,40)(130,-10 ) ( 0,0)(100,0)5 ( 50,0)(10,20)(0)0.5 ( 130,0)(230,0)5 ( 180,0)(20,0,180)45 ( 260,0)(360,0)5 ( 290,0)(330,0 ) ( 310,0)(20,0,180 ) ( 310,0)(20,0,180)45 ( 390,0)(490,0)5 ( 440,20)(20,0,360)5 ( 440,0)2 ( 88,-3)@xmath51 ( 193,-3)@xmath52 ( 297,-3)@xmath52 ( 10,6)@xmath53 ( 70,6)@xmath54 ( 113,6)@xmath53 ( 143,-13)@xmath55 ( 143,25)@xmath56 ( 175,6)@xmath54 ( 218,6)@xmath53 ( 248,-12)@xmath57 ( 248,25)@xmath58 ( 280,6)@xmath54 ( 323,6)@xmath53 ( 351,35)@xmath55 ( 383,6)@xmath54 + @xmath59 + 4 m_{\tilde q_{1/2}}^2 b_0(m_{\tilde q_{1/2}}^2;0,m_{\tilde q_{1/2 } } ) \label{eq : sigii } \\ & + & \left . \!\!\!\!\ ! 2\!\left [ m_{{\tilde{g}}}^2 + \hat m^2_q(q_0 ) - m_{\tilde q_{1/2}}^2 \!\ ! \mp 2m_{{\tilde{g}}}\hat m_q(q_0 ) \sin 2\tilde\theta_q \right ] \ ! b_0[m_{\tilde q_{1/2}}^2;m_{{\tilde{g}}},\hat m_q(q_0 ) ] \right\ } \ , . \nonumber\end{aligned}\ ] ] the one - loop integrals are defined as @xcite @xmath60[(k+p)^2-m_2 ^ 2 ] } \nonumber \\ b_1(p^2;m_1,m_2 ) & = & \frac{1}{2p^2}\left\ { a_0(m_1)-a_0(m_2 ) - ( p^2+m_1 ^ 2-m_2 ^ 2 ) b_0(p^2;m_1,m_2 ) \right\ } \ , .\end{aligned}\ ] ] the parameter @xmath61 denotes the t hooft mass of dimensional regularization . the counter terms @xmath62 are given by @xmath63 \right\ } \nonumber \\ & = & - c_f\frac{\alpha_s}{\pi } \gamma(1+\epsilon ) ( 4\pi)^\epsilon \left\ { \frac{1}{\epsilon } + \log\frac{\bar \mu^2}{q_0 ^ 2 } \right\ } \left\{m^2_{{\tilde{g}}}\mp m_{{\tilde{g}}}\hat m_q(q_0 ) \sin 2\tilde \theta_q ) \right\ } \nonumber \\ & & + \frac{\delta \hat m_q}{\hat m_q(q_0 ) } \left\ { 2\hat m^2_q(q_0 ) \mp \frac{1}{2}(m_{\tilde q_2}^2 - m_{\tilde q_1}^2 ) \sin^2 2\tilde \theta_q \right\ } \ , , \label{eq : dmsq}\end{aligned}\ ] ] where the tree - level - like mixing angle @xmath46 of eq . ( [ eq : sqmix ] ) has been used and the parameters @xmath64 and @xmath65 have been renormalized in the @xmath22 scheme , @xmath66 the counter term of the effective quark mass @xmath67 for @xmath68 is given by expression used in ref . @xcite . ] @xmath69 + b_1[\hat m_q^2(q);m_{{\tilde{g}}},m_{\tilde q_2 } ] \right . \nonumber \\ & & \left . \hspace*{0.7 cm } + 2 m_{{\tilde{g}}}(\bar a_q-\mu r_q ) \frac{b_0[\hat m_q^2(q);m_{{\tilde{g}}},m_{\tilde q_1 } ] - b_0[\hat m_q^2(q);m_{{\tilde{g}}},m_{\tilde q_2}]}{m^2_{\tilde q_1}-m^2_{\tilde q_2}}\right\ } \ , , \label{eq : dmq}\end{aligned}\ ] ] where @xmath70 is a finite counter term required to restore the supersymmetric relations between the higgs boson couplings to top / bottom quarks and stops / sbottoms within dimensional regularization @xcite . the term @xmath71 denotes the correction @xmath27 of eq . ( [ eq : deltab ] ) for the bottom mass , while it vanishes for the top mass case , i.e. @xmath72 . the term @xmath27 in the counter term of the bottom mass cancels the leading term for large values of @xmath24 in the last line of eq . ( [ eq : dmq ] ) so that this counter term is free of large corrections . using the nlo corrected squark pole masses of eq . ( [ eq : sqmass ] ) and the tree - level - like mixing angle @xmath73 of eq . ( [ eq : sqmix ] ) , the shifted squared soft susy - breaking squark mass parameters @xmath74 can be obtained from the relations . an alternative and equivalent option would be to compensate the shift @xmath75 by a shifted trilinear coupling @xmath76 . moreover , it should be noted that the shifted parameters @xmath77 and @xmath78 are not equal to each other any more in accordance with the different radiative corrections to the stop and sbottom masses . ] , @xmath79 the tree - level definition of the mixing angle @xmath73 in eq . ( [ eq : sqmix ] ) corresponds to the following counter term at nlo , @xmath80 @xmath81 however , in order to avoid artificial singularities in physical observables for stop / sbottom masses @xmath82 close to each other , in our calculation the mixing angle of the squark fields has been renormalized via the anti - hermitian counter term with @xmath83 or @xmath84 @xcite which leads to a result free of the artificial singularities for stop / sbottom masses @xmath82 close to each other , too . another option free of these singularities is provided by taking the residue of eq . ( [ eq : sqmixct ] ) for equal masses for the mixing angle counter term @xcite . ] @xcite , @xmath85 where @xmath86 denotes the off - diagonal part of the stop / sbottom self - energy ( see fig . [ fg : sqself ] ) describing transitions from the first to the second mass eigenstate or _ vice versa _ , @xmath87 \right . \nonumber \\ & & \left . \hspace*{2 cm } + \frac{\sin 2\tilde\theta_q}{4 } \left [ a_0(m_{\tilde q_2 } ) - a_0(m_{\tilde q_1 } ) \right ] \right\ } \cos 2\tilde\theta_q \ , . \label{eq : sig12}\end{aligned}\ ] ] this implies a finite shift @xmath75 to the mixing angle @xmath73 of eq . ( [ eq : sqmix ] ) , @xmath88 which modifies the relations of eq . ( [ eq : mlr ] ) by replacing @xmath89 . the scale and scheme dependence of the input parameters determining the squark pole masses is explicitly compensated by the corrections @xmath90 to the tree - level relations of eq . ( [ eq : sqmass ] ) , while the scale dependence of the mixing angle is compensated by the finite shift @xmath91 in eq . ( [ eq : sqmix1 ] ) . it should be noted that the corrected mixing angle fulfills the relation @xmath92 at nlo consistently which will be necessary for the cancellation of the ultraviolet divergences of the nlo corrections to higgs decays into squarks . the scale of the strong coupling constants @xmath26 in eqs . ( [ eq : sigii ] , [ eq : dmsq ] , [ eq : daq ] , [ eq : dmq ] , [ eq : sig12 ] ) has been identified with the input scale @xmath23 . in order to obtain fully consistent input parameters ( [ eq : deltab ] , [ eq : sqmix ] , [ eq : sqmass ] , [ eq : mlr ] ) would have to be solved iteratively until the squark pole masses do not change anymore . performing this iteration explicitly , however , shows that for certain mssm scenarios convergence can not be reached . this happens in particular for scenarios where the parameters @xmath93 and @xmath94 are close to each other so that the mixing angle @xmath73 is driven towards @xmath95 and @xmath96 towards @xmath97 . this leads to a situation in which the squark pole masses do not change anymore , but the mixing angles @xmath98 do not correspond to the relations of eqs . ( [ eq : sqmix ] , [ eq : sqmix1 ] ) so that the iterated parameters are inconsistent . this situation can be avoided by stopping the iteration earlier before running into this inconsistent limit . since the iteration adds contributions beyond nlo to the input parameters and the squark masses , the iteration effects can be considered as arbitrary in a nlo analysis . we have chosen a procedure where no iteration is performed as described above . with the shifted parameters @xmath99 of eq . ( [ eq : mlr ] ) the squark pole masses are calculated by eq . ( [ eq : sqmass ] ) with vanishing @xmath100 terms and the mixing angles according to eqs . ( [ eq : sqmix ] , [ eq : sqmix1 ] ) , since the nlo corrections are absorbed in the shifted values for @xmath101 . however , for a consistent treatment of the @xmath71 terms of eq . ( [ eq : deltab ] ) iterations for the effective bottom mass @xmath102 are performed for the sbottom sector . this ensures that the effective bottom mass of the corrected mass matrix corresponds to the nlo sbottom pole masses consistently . the nlo neutral higgs couplings to squarks read in the current eigenstate basis as @xmath103 \label{eq : hsbsbcoup}\end{aligned}\ ] ] with the couplings @xmath104 listed in table [ tb : hsfcoup ] . note that we use a common renormalization scale @xmath105 for the effective quark mass @xmath106 and the trilinear coupling @xmath65 for simplicity . the corresponding couplings to the stop / sbottom mass eigenstates @xmath107 are obtained by the appropriate rotations according to eq . ( [ eq : couprot ] ) by the on - shell mixing angle @xmath108 of eq . ( [ eq : sqmix1 ] ) . the choice of the on - shell mixing angle @xmath108 for these rotations implies that this is also the relevant mixing angle involved in the physical processes . the on - shell mixing angle is uniquely related to the soft susy - breaking input parameters according to eqs . ( [ eq : sqmix],[eq : sqmix1 ] ) and thus in particular to the parameter @xmath76 . in our numerical analysis we include the running of the effective quark mass @xmath106 and the trilinear coupling @xmath65 up to the nll level in ( susy)qcd . note that due to the counter terms for the effective quark masses of eq . ( [ eq : dmq ] ) the heavy particles are decoupled from its running . we have included 5 ( 6 ) light flavours in the nll running of the effective bottom ( top ) masses . for the running of the trilinear couplings @xmath65 we have kept the contributions of all coloured particles including the top quark , the squarks and the gluino which is consistent with the counter term @xmath109 of eq . ( [ eq : daq ] ) . the running effective quark mass for @xmath110 light flavours is thus given by @xmath111}{c[\alpha_{s}(m_q)/\pi ] } \nonumber \\ c(x ) & = & \left(\frac{23}{6}\,x\right)^{\frac{12}{23 } } \ , \left[1+\frac{3731}{3174}\,x\right ] \qquad \mbox{for $ q = b \quad ( n_f=5)$ } \nonumber \\ c(x ) & = & \left(\frac{7}{2}\,x\right)^{\frac{4}{7 } } \ , \left[1+\frac{137}{98}\,x\right ] \qquad \quad \ ; \mbox{for $ q = t \quad ( n_f=6)$ } \ , , \label{eq : msbarev}\end{aligned}\ ] ] where @xmath112 denotes the corresponding quark pole mass . the nll evolution of the trilinear couplings @xmath65 with @xmath113 quark and squark flavours is determined as the solution of the corresponding two - loop renormalization group equations ( rges ) with respect to the strong coupling constant @xcite and can be expressed as scheme appropriately @xcite . in the @xmath25 scheme the coefficient @xmath114 is replaced by @xmath115 , the coefficient @xmath116 by @xmath117 and the @xmath22 gluino mass @xmath118 by the corresponding @xmath25 gluino mass in eq . ( [ eq : aqnll ] ) . ] @xmath119 \left[1+\frac{1}{6}\,\frac{\alpha_{s , susy}(q_0)}{\pi } \right ] \right . \nonumber \\ & & \left . \hspace*{3.5 cm } - \frac{16}{27}\,\frac{\alpha_{s , susy}(q_0)}{\pi } \left[\frac{\alpha_{s , susy}^2(\mu_r)}{\alpha_{s , susy}^2(q_0)}-1\right ] \right\ } \label{eq : aqnll}\end{aligned}\ ] ] with the running strong coupling @xmath120 including all quarks , squarks and gluinos , @xmath121 where @xmath122 is determined by the matching condition @xmath123 \right\}\ ] ] between the strong coupling @xmath120 including all quarks , squarks and gluinos and the low - energy qcd couplings @xmath26 with 5 active light flavours . the @xmath22 mass @xmath118 of the gluino can be obtained from the gluino pole mass @xmath124 , @xmath125 \right\ } \ , , \end{aligned}\ ] ] where @xmath126 . the sum is performed over all 3 ( s)quark generations . for the first two generations we neglected the quark masses in the squark mass matrices so that the corresponding squark pole masses are given by @xmath127 ^ 2}{2\tilde{\overline{m}}_{\tilde q_{l / r}}^2(q_0 ) } \log\left| \frac{m_{{{\tilde{g}}}}^2-\tilde{\overline{m}}_{\tilde q_{l / r}}^2(q_0)}{m_{{{\tilde{g}}}}^2 } \right| \right\}\end{aligned}\ ] ] ( 130,90)(0,0 ) ( 0,50)(50,50)5 ( 50,50)(100,100)5 ( 50,50)(100,0)5 ( -15,48)@xmath128 ( 105,98)@xmath55 ( 105,-2)@xmath129 + at leading order the mssm higgs boson decays into squark - antisquark pairs are described by the diagram shown in fig . [ fg : lodia ] . the corresponding partial decay width reads as @xmath130 ^ 2 \sqrt{\lambda_{ij}}\ ] ] with the two - body phase - space function @xmath131 while higgs boson decays into squarks of the first two generations are generally suppressed , higgs decays into sbottoms and stops develop sizable branching ratios , whenever they are kinematically possible @xcite . in particular scenarios their branching ratios can reach a level of 8090% @xcite . ( 130,90)(0,0 ) ( 0,50)(50,50)5 ( 50,50)(100,100)5 ( 50,50)(100,0)5 ( 50,50)(30,-45,45)35 ( -15,48)@xmath128 ( 105,98)@xmath55 ( 105,-2)@xmath129 ( 90,48)@xmath56 ( 130,120)(-20,0 ) ( 0,50)(50,50)5 ( 75,75)(100,100)5 ( 75,25)(100,0)5 ( 50,50)(75,75 ) ( 50,50)(30,-45,45 ) ( 50,50)(30,-45,45)35 ( 75,25)(50,50 ) ( -15,48)@xmath128 ( 105,98)@xmath55 ( 105,-2)@xmath129 ( 55,65)@xmath57 ( 55,30)@xmath57 ( 90,48)@xmath58 ( -35,48)@xmath52 ( 115,48)@xmath52 ( 130,120)(-40,0 ) ( 0,50)(20,50)5 ( 50,50)(100,100)5 ( 50,50)(100,0)5 ( 35,50)(15,0,360)5 ( -15,48)@xmath128 ( 105,98)@xmath55 ( 105,-2)@xmath129 ( 30,70)@xmath55 + ( 130,120)(0,0 ) ( 0,50)(50,50)5 ( 50,50)(100,100)5 ( 50,50)(100,0)5 ( 75,75)(8,15)(45)0.5 ( -15,48)@xmath128 ( 105,98)@xmath55 ( 105,-2)@xmath129 ( -40,48)@xmath52 ( 115,48)@xmath52 ( 130,120)(-20,0 ) ( 0,50)(50,50)5 ( 50,50)(100,100)5 ( 50,50)(100,0)5 ( 75,25)(8,15)(-45)0.5 ( -15,48)@xmath128 ( 105,98)@xmath55 ( 105,-2)@xmath129 + ( 130,100)(0,0 ) ( 0,50)(50,50)5 ( 50,50)(100,100)5 ( 50,50)(100,0)5 ( 65,65)(100,65)33 ( -15,48)@xmath128 ( 105,98)@xmath55 ( 105,63)@xmath56 ( 105,-2)@xmath129 ( 130,90)(-20,0 ) ( 0,50)(50,50)5 ( 50,50)(100,100)5 ( 50,50)(100,0)5 ( 65,35)(100,35)33 ( -15,48)@xmath128 ( 105,98)@xmath55 ( 105,33)@xmath56 ( 105,-2)@xmath129 + the susy qcd corrections at nlo have been calculated in refs . @xcite . however , the analytical results will be repeated in this section in our notation and for the investigation of calculational details once consistent input parameters at nlo are included in the theoretical analysis . the susy qcd corrections consist of one - loop virtual corrections , mediated by the feynman diagrams depicted in fig . [ fg : nlodiav ] , supplemented by the renormalization of the mass , coupling and wave function parameters involved , and real corrections due to gluon bremsstrahlung , see fig . [ fg : nlodiar ] . we have performed the loop and phase - space integration within dimensional regularization in @xmath132 dimensions so that ultraviolet and infrared singularities appear as poles in @xmath133 . the infrared singularities cancel after adding the real and virtual corrections . the ultraviolet poles disappear after adding the corresponding counter terms which will be discussed in detail in the following , since they have to correspond to the schemes in which the squark masses , mixing angles and couplings are defined . the wave - function counter terms @xmath134 are determined by normalizing the residues of the diagonalized squark propagators to unity . the explicit calculation of the diagrams of fig . [ fg : sqself ] leads to the expressions @xmath135 b'_0(m_{\tilde q_i}^2;m_{{\tilde{g}}},m_q ) \right\ } \ , , \end{aligned}\ ] ] where the derivative of the two - point function is defined as @xmath136 throughout our calculation in this section the quark mass @xmath112 is defined as @xmath137 of eq . ( [ eq : deltab ] ) for decays into sbottom pairs and as @xmath138 of eq . ( [ eq : tmass ] ) for decays into stops . the mixing angle is renormalized by the anti - hermitian counter term of eq . ( [ eq : sqmixct ] ) . the trilinear couplings @xmath139 ( @xmath140 ) are defined in the @xmath22 scheme at the scale @xmath105 which is identical to the running couplings in the @xmath25 definition at nlo . thus , the counter term of @xmath139 can be cast into the form @xmath141 the quark mass counter term @xmath142 is given by @xmath143 with the top and bottom mass counter terms @xmath144 of eq . ( [ eq : dmq ] ) with @xmath145 . the calculation of the diagrams of fig . [ fg : nlodiav ] for the virtual corrections and those of fig . [ fg : nlodiar ] for the real corrections leads to the final result after renormalization , @xmath146 + c^\phi_{ct } + c^\phi_{real } \nonumber \\ c^\phi_1 & = & b_0(m_{\tilde q_i}^2;0,m_{\tilde q_i } ) + b_0(m_{\tilde q_j}^2;0,m_{\tilde q_j } ) - b_0(m_\phi^2 ; m_{\tilde q_i } , m_{\tilde q_j } ) \nonumber \\ & & - 2(m_\phi^2-m_{\tilde q_i}^2-m_{\tilde q_j}^2 ) c_0(m_{\tilde q_i}^2,m_\phi^2,m_{\tilde q_j}^2;0,m_{\tilde q_i},m_{\tilde q_j } ) \nonumber \\ c^{h , h}_2 & = & \frac{g_q^{h , h}}{g_{\tilde q_i\tilde q_j}^{h , h } } \left\ { \delta_{ij } m_q^2 \left [ b_0(m_{\tilde q_i}^2;m_{{\tilde{g}}},m_q ) + b_0(m_{\tilde q_j}^2;m_{{\tilde{g}}},m_q ) + 2b_0(m_{h , h}^2;m_q , m_q ) \right . \nonumber \\ & & \left . + ( 2m_{{\tilde{g}}}^2 + 2m_q^2-m_{\tilde q_i}^2-m_{\tilde q_j}^2 ) c_0(m_{\tilde q_i}^2,m_{h , h}^2,m_{\tilde q_j}^2;m_{{\tilde{g } } } , m_q , m_q ) \right ] \right . \nonumber \\ & & + { \cal r}_{ij } m_{{\tilde{g}}}m_q \left [ ( m_{h , h}^2 - 4m_q^2 ) c_0(m_{\tilde q_i}^2,m_{h , h}^2,m_{\tilde q_j}^2;m_{{\tilde{g } } } , m_q , m_q ) \right . \nonumber \\ & & \left . -b_0(m_{\tilde q_i}^2;m_{{\tilde{g}}},m_q ) - b_0(m_{\tilde q_j}^2;m_{{\tilde{g}}},m_q ) \right ] \right\ } \nonumber \\ c^a_2 & = & \frac{g_q^a}{g_{\tilde q_i\tilde q_j}^a } \left\ { m_q^2 \sin 2\theta_q \left [ b_0(m_{\tilde q_i}^2;m_{{\tilde{g}}},m_q ) - b_0(m_{\tilde q_j}^2;m_{{\tilde{g}}},m_q ) \right . \right . \nonumber \\ & & \left . + ( m_{\tilde q_i}^2-m_{\tilde q_j}^2 ) c_0(m_{\tilde q_i}^2,m_a^2,m_{\tilde q_j}^2;m_{{\tilde{g } } } , m_q , m_q ) \right ] \right . \nonumber \\ & & \left . -(-1)^i m_{{\tilde{g}}}m_q \left [ m_a^2 c_0(m_{\tilde q_i}^2,m_a^2,m_{\tilde q_j}^2;m_{{\tilde{g } } } , m_q , m_q ) \right . \nonumber \\ & & \left . -b_0(m_{\tilde q_i}^2;m_{{\tilde{g}}},m_q ) - b_0(m_{\tilde q_j}^2;m_{{\tilde{g}}},m_q ) \right ] \right\ } \nonumber \\ c^{h , h}_3 & = & -\frac{{\cal s}_{ik } g^{h , h}_{\tilde q_k\tilde q_l } { \cal s}_{lj}}{g_{\tilde q_i\tilde q_j}^{h , h } } b_0(m_{h , h}^2;m_{\tilde q_k } , m_{\tilde q_l } ) \nonumber \\ c^a_3 & = & b_0(m_a^2;m_{\tilde q_i } , m_{\tilde q_j } ) \nonumber \\ \frac{\alpha_s}{\pi } c^{h , h}_{ct } & = & \delta z_{ii}+\delta z_{jj } + \frac{2}{g_{\tilde q_i\tilde q_j}^{h , h } } \left [ \frac{\partial g^{h , h}_{\tilde q_i\tilde q_j}}{\partial m_q } \delta m_q + \frac{\partial g^{h , h}_{\tilde q_i\tilde q_j}}{\partial a_q } \delta \bar a_q + { \cal t}_{ij } \delta\theta_q \right ] \nonumber \\ \frac{\alpha_s}{\pi } c^a_{ct } & = & \delta z_{ii}+\delta z_{jj } + \frac{2}{g_{\tilde q_i\tilde q_j}^a } \left [ \frac{\partial g^a_{\tilde q_i\tilde q_j}}{\partial m_q } \delta m_q + \frac{\partial g^a_{\tilde q_i\tilde q_j}}{\partial a_q } \delta \bar a_q \right ] \nonumber \\ c^\phi_{real } & = & c_f \gamma(1+\epsilon ) \left ( \frac{4\pi\bar\mu^2}{m_\phi^2}\right)^\epsilon \left\ { \frac{1-\rho_i-\rho_j}{2\beta } \left [ \frac{\log x_0}{\epsilon } + \log x_0 \log\frac{\rho_i\rho_j x_0}{\beta^4 } \right . . \nonumber \\ & & \left . \left . - \frac{1}{2 } \log^2 x_1 - \frac{1}{2 } \log^2 x_2 + 4 li_2\left(\frac{1-x_0}{-x_0}\right ) - 2 li_2(1-x_1 ) - 2 li_2(1-x_2 ) \right ] \right . \nonumber \\ & & \left . + \frac{1}{\epsilon } + \log\frac{\rho_i\rho_j}{\beta^4 } + 4 - \frac{1+\rho_i+\rho_j}{2\beta } \log x_0 + \frac{\rho_i\log x_1 + \rho_j\log x_2}{\beta}\right\ } \ , , \label{eq : cphi}\end{aligned}\ ] ] where in @xmath147 we used the abbreviations @xmath148 with the two - body phase - space function @xmath149 of eq . ( [ eq : lambda ] ) and @xmath150 with @xmath151 . the two mixing matrices @xmath152 and @xmath153 used in the coefficients @xmath154 and @xmath155 are given by @xmath156 the three - point function @xmath157 is defined as @xcite @xmath158[(k+p_1)^2-m_2 ^ 2][(k+p_{12})^2-m_3 ^ 2]}\ ] ] with @xmath159 . the finite remainders of the mixing angle renormalization at the external squark legs can be cast into the form of the matrix @xmath160 multiplied by the finite shift @xmath161 which includes the corresponding anti - hermitian counter term @xmath162 of the mixing angle as given in eq . ( [ eq : sqmixct ] ) . in eq . ( [ eq : deltath ] ) the singularity for @xmath163 is cancelled by the anti - hermitian counter term . finally the derivatives of the squark couplings @xmath164 to the higgs bosons are given by @xmath165 the results for the coefficients @xmath166 are ultraviolet and infrared finite after all individual contributions are added up since our renormalized mixing angles fulfill the relation @xmath167 consistently . the scale of @xmath26 in eq.([eq : cphi ] ) is identified with the renormalization scale everywhere apart from the @xmath27 term within the effective bottom mass of eq . ( [ eq : deltab ] ) . it should be noted that our final results do not contain any leading @xmath24-enhanced corrections of the @xmath27-type anymore due to our scheme choices of the effective masses and couplings at nlo . due to this property the problems of a naive renormalization program have been solved consistently leading to moderate radiative corrections to the higgs decay widths into squark - antisquark pairs as will be shown explicitly in the next section . by using power counting arguments along the lines of ref . @xcite it can be shown that our approach resums all leading terms of @xmath168 so that the residual ho corrections are free of these leading terms . in the current eigenstate basis there are three sources of terms proportional to @xmath24 : _ ( i ) _ the higgs couplings to squarks and quarks as in table [ tb : hcoup ] and eq . ( [ eq : hsbsbcoup ] ) which drop out in the relative corrections ; _ ( ii ) _ off - diagonal mass insertions in the sbottom propagators which are power - suppressed by @xmath169 due to the kln theorem and the @xmath169 structure of these insertions ; ( iii ) potential @xmath24-enhancements of the counter terms . the @xmath22 renormalization of @xmath0 as in eq . ( [ eq : daqp ] ) does not involve any @xmath24 enhanced corrections as can also be inferred from the beta functions of the rges for the trilinear couplings up to three - loop order @xcite . the @xmath24 enhancement of the bottom - mass counter term entering the corresponding bottom yukawa couplings involved in the higgs couplings to bottom quarks and the sbottom mass matrix is explicitly absorbed in the effective bottom mass of eq . ( [ eq : deltab ] ) which contains the corresponding threshold correction @xmath27 in resummed form @xcite . therefore the counter terms @xmath170 and @xmath171 are free of @xmath24 enhanced corrections at all orders up to terms which are suppressed by @xmath169 . close to the threshold @xmath172 the nlo susy qcd corrections develop coulomb singularities , @xmath173 which agrees with the usual sommerfeld rescattering correction factor @xcite . these singularities for @xmath174 can be regularized by taking into account the finite decay widths of the squarks . moreover , these coulomb factors can be resummed systematically . both effects are expected to be relevant close to threshold , but are beyond the scope of this paper . on the other hand far above the threshold @xmath175 the nlo corrections approach the asymptotic limit @xmath176 \right . \nonumber \\ & & \left . \hspace*{0.5 cm } + 2\frac{m_s}{g_{\tilde q_i\tilde q_j}^\phi } \frac{\partial g_{\tilde q_i\tilde q_j}^\phi}{\partial a_q } \left [ -\log\frac{\mu_r^2}{m_s^2 } + \frac{1}{4 } \log^2\frac{m_\phi^2}{m_s^2 } + \frac{\zeta_2}{2 } - 2 \right ] + \delta_\phi \right\ } \nonumber \\ \delta_{h , h } & = & \left [ \frac{m_q}{g_{\tilde q_i\tilde q_j}^\phi } \frac{\partial g_{\tilde q_i\tilde q_j}^\phi}{\partial m_q } - 1 \right ] \left [ \log\frac{m_\phi^2}{m_s^2 } - 2 \right ] + \frac{g^\phi_q m_q^2}{g_{\tilde q_i\tilde q_j}^{h , h } } \delta_{ij } \log\frac{m_\phi^2}{m_s^2 } \nonumber \\ & & + ~{\cal t}_{ij}~\frac{m_q}{m_s } \frac{\cos 2\theta_q}{g_{\tilde q_i\tilde q_j}^\phi } \left [ 1-\frac{1}{2}\log\frac{m_s^2}{m_q^2 } \right ] \nonumber \\ \delta_a & = & 0 \ , , \label{eq : largemh}\end{aligned}\ ] ] where we have identified all supersymmetric masses , @xmath177 , and neglected the quark mass against the other masses . the term @xmath178 is given by @xmath179 thus confirming the absence of large corrections of @xmath180 to the on - shell quark and squark mass as well as the mixing angle counter terms as in eq . ( [ eq : sqmixct1 ] ) and renormalizes the quark mass on - shell , large corrections of @xmath181 emerge for the partial higgs decay widths into squarks . ] . the asymptotic result far above the threshold develops a double logarithmic contribution in the second square bracket of eq . ( [ eq : largemh ] ) which originates from the second diagram of fig . [ fg : nlodiav ] in the sudakov limit . it can not be absorbed in an appropriate choice of the renormalization scale @xmath105 but will cancel against the corresponding double logarithm of gluino radiation in association with quark - squark pairs in the large higgs mass limit which has not been taken into account in this work . in the following we will identify @xmath105 with the corresponding higgs mass @xmath182 as the central scale choice . the numerical analysis of the neutral higgs boson decays into stop and sbottom pairs is performed for two mssm scenarios , one close to the one of ref . @xcite where we lifted the gluino and squark masses of the first two generations beyond the mass bounds from the lhc @xcite , and a second scenario with large susy - breaking masses and large mixing in the stop sector as representative cases : @xmath183 where @xmath184 and @xmath185 denote the squark and slepton mass parameters of the first two generations . the results of this work have been implemented in the program hdecay @xcite which calculates the mssm higgs decay widths and branching ratios including the relevant higher - order corrections @xcite . we use the rg - improved two - loop expressions for the higgs masses and couplings of ref . @xcite which yield predictions for the higgs boson masses that agree with the diagrammatic calculations of ref . @xcite within 34% in general . thus the leading one- and two - loop corrections have been included in the higgs masses and the effective mixing angle @xmath16 . consistency of our scheme with the scheme and scale choices of ref . @xcite requires the evolution of our sbottom parameters @xmath186 and @xmath187 to the scale @xmath188 and the stop parameters @xmath189 and @xmath190 to the scale @xmath191 since the scales @xmath23 and @xmath192 are of similar order of magnitude we neglect resummation effects so that the relations are given by and @xmath193 are not equal , because they are evaluated for different scales @xmath194 and @xmath195 . we have modified the calculation of ref . @xcite to account for these differences consistently . ] @xmath196 \right\ } \log\frac{q_b^2}{q_0 ^ 2 } \nonumber \\ \overline{m}^2_{\tilde b_r}(q_b ) & = & \overline{m}^2_{\tilde b_l}(q_0 ) + \left\ { - c_f \frac{\alpha_s(q_0)}{\pi } m_3 ^ 2(q_0 ) + \frac{\alpha_b}{4\pi } x_b \right\ } \log\frac{q_b^2}{q_0 ^ 2 } \nonumber \\ \bar a_t(q_t ) & = & \bar a_t(q_0 ) + \left\ { c_f \frac{\alpha_s(q_0)}{\pi } m_3(q_0 ) + \frac{\alpha_t}{4\pi } \bar a_t(q_0 ) + \frac{3}{2 } \frac{\alpha_b}{\pi } \bar a_b(q_0 ) \right\ } \log\frac{q_t^2}{q_0 ^ 2 } \nonumber \\ \overline{m}^2_{\tilde t_l}(q_t ) & = & \overline{m}^2_{\tilde t_l}(q_0 ) + \left\ { - c_f \frac{\alpha_s(q_0)}{\pi } m_3 ^ 2(q_0 ) + \frac{1}{4 } \left [ \frac{\alpha_t}{\pi } x_t + \frac{\alpha_b}{\pi } x_b \right ] \right\ } \log\frac{q_t^2}{q_0 ^ 2 } \nonumber \\ \overline{m}^2_{\tilde t_r}(q_t ) & = & \overline{m}^2_{\tilde t_l}(q_0 ) + \left\ { - c_f \frac{\alpha_s(q_0)}{\pi } m_3 ^ 2(q_0 ) + \frac{\alpha_t}{4\pi } x_t \right\ } \log\frac{q_t^2}{q_0 ^ 2 } \label{eq : transform}\end{aligned}\ ] ] with the abbreviations @xmath197 and the soft susy - breaking higgs mass parameters @xmath198 the top pole mass has been taken as @xmath199 gev , while the bottom quark pole mass has been chosen to be @xmath200 gev , which corresponds to a @xmath22 mass @xmath201 gev . the strong coupling constant has been normalized to @xmath202 which corresponds to the qcd scale @xmath203 mev for 5 active flavours . the related susy qcd scale @xmath122 of eq . ( [ eq : als_susy ] ) amounts to @xmath204 kev in scenario a and @xmath205 kev in scenario b. the @xmath22 masses involved in the scales @xmath206 can be derived as @xmath207 gev and @xmath208 gev while the @xmath22 gluino mass at the input scale @xmath23 amounts to @xmath209 tev in scenario a and @xmath210 tev in scenario b. the squark masses in the two scenarios amount in particular to @xmath211 where @xmath212 and @xmath213 denote the up- and down - type squark masses of the first two generations . in fig . [ fg : vienna_t ] we display the results for the partial decay widths of the heavy neutral higgs particles into stop pairs and the relative corrections in scenario a. since the stops are moderately heavy all decay modes are kinematically allowed for higgs masses above about 560 gev . the partial decay widths range at the few gev level and turn out to be similar for all higgs bosons above the corresponding thresholds except the scalar higgs decay widths into non - diagonal stop pairs which are smaller . apart from the coulomb singularities at threshold the susy qcd corrections are of moderate size in our scheme . ( 100,500)(0,0 ) ( 40.0,120.0 ) ( 40.0,-135.0 ) similar results have been obtained for the neutral higgs boson decay widths into sbottom pairs as shown in fig . [ fg : vienna_b ] . the moderate size of the full susy qcd corrections confirms the proper absorption of the @xmath27 terms as in eq . ( [ eq : deltab ] ) . if the trilinear coupling @xmath0 and the bottom quark mass @xmath214 would be renormalized in the on - shell scheme the susy qcd corrections would increase the lo results by more than an order of magnitude so that the result in the on - shell scheme becomes totally unreliable @xcite . the partial widths of the non - diagonal higgs decays @xmath215 are small , since the mixing factor @xmath216 is small and thus the contributions of the left - right couplings @xmath217 are suppressed , while the diagonal couplings @xmath218 and @xmath219 cancel each other to a large extent in the evaluation of the coupling @xmath220 of eq . ( [ eq : couprot ] ) . using the scheme of ref . @xcite for the mixing angle , i.e. defining the mixing angle counter terms as @xmath221 with @xmath222 or @xmath223 , the results agree with ours within less than 1% of eq . ( [ eq : sqmix ] ) develop sizeable differences to the results in our scheme for squark masses @xmath82 close to each other , where the partial decay widths @xmath224 turn out to be negative thus signalizing a basic problem with the tree - level - like mixing angle . ] . ( 100,500)(0,0 ) ( 40.0,120.0 ) ( 40.0,-135.0 ) ( 100,500)(0,0 ) ( 40.0,120.0 ) ( 40.0,-135.0 ) in order to obtain an estimate of the residual theoretical uncertainties we show the renormalization scale dependence of the partial decay widths into stop and sbottom pairs in fig . [ fg : scale_v ] . the scale dependences of the partial decay widths into stop and sbottom pairs are significantly reduced at nlo . varying the renormalization scale by a factor of 2 around the central scale @xmath225 the theoretical uncertainties can be estimated as 510% for decays into stop and sbottom pairs at nlo . [ fg : br_v ] displays the corresponding branching ratios of the dominant decay modes of the heavy scalar and the pseudoscalar higgs boson in scenario a. it is clearly visible that the decays into stop and sbottom pairs belong to the dominant decay modes for masses above about 300400 gev , i.e. where they are kinematically allowed , reaching branching ratios of up to about 50% in total . moreover , in scenario a the decay modes into @xmath226 s and neutralinos play a significant role , too . ( 100,500)(0,0 ) ( 40.0,120.0 ) ( 40.0,-135.0 ) in fig . [ fg : scen_b ] we display the final results for the scenario b. since the masses @xmath227 are close to or larger than 500 gev , the decay modes @xmath228 are kinematically forbidden for @xmath229 tev . the partial decay widths of decays into stop and sbottom pairs range at the few gev level as in scenario a once they are kinematically allowed . it can clearly be inferred from fig . [ fg : scen_b ] that the susy qcd corrections are of moderate size apart from the threshold regions where the coulomb singularities enhance the size of the corrections . it should be noted that the results with the sbottom mixing angle renormalized via the tree - level relation of eq . ( [ eq : sqmixct1 ] ) would lead to unphysical negative partial decay widths for heavy scalar higgs decays @xmath230 for higgs masses larger than 1 tev , where these decay channels open up , so that this scheme is strongly disfavoured . these negative contributions can be traced back to the uncancelled artificial singularity for @xmath231 in the finite remainders @xmath232 of eq . ( [ eq : deltath ] ) , if @xmath233 is replaced by the counter term @xmath234 of eq . ( [ eq : sqmixct1 ] ) @xcite . ( 100,500)(0,0 ) ( 40.0,120.0 ) ( 40.0,-135.0 ) in fig . [ fg : br_b ] we show the branching ratios of the heavy scalar and the pseudoscalar higgs boson in scenario b as functions of the corresponding higgs masses . as in scenario a the decay modes into stop and sbottom pairs play a significant role , once they are kinematically allowed . this starts to be the case for the heavy scalar higgs boson @xmath235 already for masses above about 400 gev , where the decay into light stop mass eigenstates opens up . the pseudoscalar higgs boson can only decay into mixed pairs of a light and heavy stop and sbottom eigenstates so that these decay modes open up for masses above 900 gev . the decays into charginos and neutralinos contribute significantly for larger higgs masses . ( 100,500)(0,0 ) ( 40.0,120.0 ) ( 40.0,-135.0 ) for smaller values of @xmath24 the decay modes @xmath236 and @xmath237 play a significant role , while the decays into sbottom pairs are usually suppressed . however , the neutral higgs boson decays into stop pairs still play a significant role and can even be the dominant heavy higgs boson decays in a large higgs mass range . qcd corrections to these decay modes are of similar size as in the scenarios with large values of @xmath24 analyzed in this work . in this work we have analyzed the nlo susy qcd corrections to heavy mssm higgs decays into stop and sbottom pairs and elaborated a consistent scheme and setup for the theoretical nlo results and the corresponding input parameters , i.e. the squark masses and their couplings to the higgs bosons . a reliable determination of these corrections requires a proper treatment of contributions which are enhanced by @xmath24 . the derivation of the input squark masses and couplings with nlo accuracy requires a modification of the tree - level relations among the stop and sbottom masses by higher - order corrections , if they should correspond to the physical pole masses . since the size of these modifications depends on the mssm parameters and pole masses themselves an iterative procedure should be applied . these iterations , however , do not converge for all scenarios so that we restricted the evaluation to a single iteration step . the final setup of the input parameters matches nlo accuracy consistently . a preferred choice of the scheme for the stop and sbottom mixing angles is the on - shell definition via the anti - hermitian counter term which removes spurious singularities for nearly mass degenerate cases which drive the off - diagonal decays @xmath238 to negative and thus unphysical values . the final nlo corrections are of moderate size apart from the expected coulomb singularities at the decay thresholds . the residual theoretical uncertainties at nlo have been estimated to be 510% . + * note added in proof . * during completion of this work ref . @xcite appeared which discusses several options for the renormalization of the stop and sbottom sector with nlo accuracy . their treatment of the stop and sbottom sectors is different from our approach in particular for the effective bottom mass and the squark mixing angles . + * acknowledgments . * + we are grateful to a. djouadi , m. mhlleitner , h. rzehak and p.m. zerwas for carefully reading our manuscript and useful comments . p. fayet , nucl . phys . * b90 * ( 1975 ) 104 , phys . * b64 * ( 1976 ) 159 and phys . * b69 * ( 1977 ) 489 . s. dimopoulos and h. georgi , nucl . b193 * ( 1981 ) 150 ; n. sakai , z. phys . * c11 * ( 1981 ) 153 ; k. inoue , a. kakuto , h. komatsu and s. takeshita , prog . * 67 * ( 1982 ) 1889 , prog . * 68 * ( 1982 ) 927 [ erratum - ibid . * 70 * ( 1983 ) 330 ] and prog . * 71 * ( 1984 ) 413 . see e.g. g. degrassi , s. heinemeyer , w. hollik , p. slavich and g. weiglein , eur . j. * c28 * ( 2003 ) 133 . s. p. martin , phys . * d75 * ( 2007 ) 055005 ; r. v. harlander , p. kant , l. mihaila and m. steinhauser , phys . * 100 * ( 2008 ) 191602 , erratum _ ibid . _ * 101 * ( 2008 ) 039901 ; p. kant , r. v. harlander , l. mihaila and m. steinhauser , arxiv:1005.5709 [ hep - ph ] . s. schael _ et al . _ [ aleph , delphi , l3 and opal collaborations ] , eur . phys . j. * c47 * ( 2006 ) 547 . m. s. carena _ et al . _ [ higgs working group collaboration ] , arxiv : hep - ph/0010338 . atlas collaboration , technical design report , cern lhcc 9914 ( may 1999 ) ; g. l. bayatian _ et al . _ [ cms collaboration ] , j. phys . * g34 * ( 2007 ) 995 . a. djouadi , j. kalinowski , p. ohmann and p. m. zerwas , z. phys . * c74 * ( 1997 ) 93 . a. bartl , h. eberl , k. hidaka , t. kon , w. majerotto and y. yamada , phys . * b402 * ( 1997 ) 303 ; a. arhrib , a. djouadi , w. hollik and c. jnger , phys . * d57 * ( 1998 ) 5860 . c. weber , k. kovarik , h. eberl and w. majerotto , nucl . * b776 * ( 2007 ) 138 . h. eberl , k. hidaka , s. kraml , w. majerotto and y. yamada , phys . * d62 * ( 2000 ) 055006 . s.p . martin and m.t . vaughn , phys . * b318 * ( 1993 ) 331 . l. j. hall , r. rattazzi and u. sarid , phys . * d50 * ( 1994 ) 7048 ; r. hempfling , phys . * d49 * ( 1994 ) 6168 ; m. carena , m. olechowski , s. pokorski and c. e. wagner , nucl . * b426 * ( 1994 ) 269 ; m. carena , d. garcia , u. nierste and c. e. wagner , nucl . * b577 * ( 2000 ) 88 . j. guasch , p. hfliger and m. spira , phys . rev . * d68 * ( 2003 ) 115001 . d. noth and m. spira , phys . * 101 * ( 2008 ) 181801 and arxiv:1001.1935 [ hep - ph ] ; l. mihaila and c. reisser , jhep * 1008 * ( 2010 ) 021 . a. crivellin , l. hofer , j. rosiek , jhep * 1107 * ( 2011 ) 017 . g. t hooft and m.j.g . veltman , nucl . * b153 * ( 1979 ) 365 ; g. passarino and m.j.g . veltman , nucl . * b160 * ( 1979 ) 151 . w. beenakker , r. hpker , t. plehn and p.m. zerwas , z. phys . * c75 * ( 1997 ) 349 - 356 . n. baro , f. boudjema , phys . d80 * ( 2009 ) 076010 . s. p. martin , m. t. vaughn , phys . * d50 * ( 1994 ) 2282 ; y. yamada , phys . rev . * d50 * ( 1994 ) 3537 ; i. jack and d. r. t. jones , phys . * b333 * ( 1994 ) 372 ; i. jack , d. r. t. jones , a. f. kord , phys . * b579 * ( 2004 ) 180 . m. spira , fortsch . * 46 * ( 1998 ) 203 ; a. djouadi , phys . * 459 * ( 2008 ) 1 . a. sommerfeld , _ atombau und spektrallinien _ , vieweg , braunschweig 1939 . v. khachatryan _ et al . _ [ cms collaboration ] , arxiv:1101.1628 [ hep - ex ] ; the atlas collaboration , arxiv:1102.2357 [ hep - ex ] . a. djouadi , j. kalinowski and m. spira , comput . commun . * 108 * ( 1998 ) 56 ; a. djouadi , j. kalinowski , m. mhlleitner and m. spira , in j. m. butterworth _ et al . _ , arxiv:1003.1643 [ hep - ph ] ; a. djouadi , m. m. mhlleitner and m. spira , acta phys . * b38 * ( 2007 ) 635 - 644 . carena , m. quiros and c.e.m . wagner , nucl . * b461 * ( 1996 ) 407 . manuel walser , diploma thesis , zrich , 2003 .

we report on searches for neutrino sources at energies above 200 gev in the northern sky of the galactic plane , using the data collected by the south pole neutrino telescope icecube and amanda . the galactic region considered in this work includes the local arm towards the cygnus region and our closest approach to the perseus arm . the searches are based on the data collected between 2007 and 2009 . during this time amanda was an integrated part of icecube , which was still under construction and operated with 22-strings ( 2007 - 8 ) and 40-strings ( 2008 - 9 ) of optical modules deployed in the ice . by combining the advantages of the larger icecube detector with the lower energy threshold of the more compact amanda detector , we obtain an improved sensitivity at energies below tev with respect to previous searches . the analyses presented here are : a scan for point sources within the galactic plane ; a search optimized for multiple and extended sources in the cygnus region , which might be below the sensitivity of the point source scan ; and studies of seven pre - selected neutrino source candidates . for one of them , cygnus x-3 , a time - dependent search for neutrino emission in coincidence with observed radio and x - ray flares has been performed . no evidence of a signal is found , and upper limits are reported for each of the searches . we investigate neutrino spectra proportional to e and e in order to cover the entire range of possible neutrino spectra . the steeply falling e neutrino spectrum can also be used to approximate neutrino energy spectra with energy cutoffs below 50 tev since these result in a similar energy distribution of events in the detector . for the region of the galactic plane visible in the northern sky , the 90% confidence level muon neutrino flux upper limits are in the range edn / de for point - like neutrino sources in the energy region [ 180.0 gev - 20.5 tev ] . these represent the most stringent upper limits for soft - spectra neutrino sources within the galaxy reported to date .

1210.3273

the icecube neutrino telescope at the south pole has successfully been completed in december 2010 . icecube is the most sensitive telescope to date to search for high - energy neutrino sources , whose existence is intimately related to the acceleration of hadrons and their interaction in the environment of their accelerator . the interaction of high - energy protons and nuclei with ambient matter or radiation leads to the generation of both gamma - rays and neutrinos of similar energy @xcite . however , it is difficult to infer the contribution of a possible hadronic component from the observed gamma - rays , since gamma - ray emission can also be produced by relativistic electrons via inverse compton scattering . moreover , the most energetic gamma - rays have a high probability to be absorbed on their way to earth , and the observed spectra may , after successive absorption and emission processes @xcite , not be the same as the primary spectra . the detection of a flux of high - energy neutrinos from astrophysical sources , even if challenging , can thus provide unique insights into the acceleration mechanisms and the origin of cosmic - rays . the icecube neutrino telescope has a full - sky field of view at any time and has thus the potential to observe neutrino point sources at any position in the sky albeit with different discovery potential depending on the source location and the neutrino energy spectrum of the source . for a source following an e@xmath1 spectrum in the energy range from 1 tev up to a few pev icecube can discover high - energy neutrino sources at the flux level of @xmath5 @xmath6 @xcite if the source location is known from other observations . assuming isotropic emission , this translates to source luminosities of @xmath7 for a source at a distance of 2 kpc . a search for neutrino point sources at any location in the sky with the icecube 40-strings detector has been presented in @xcite using e@xmath1 and flatter spectra for the optimization of the analysis . this work focusses on the more specific case of galactic neutrino sources and the energy spectra associated to them . among the most promising candidate sources of cosmic - rays in the galaxy are the remnants of supernovae ( both shell - type and pulsar wind nebulae ) , the jets of microquasars , and the collective winds of massive stars @xcite . due to the large amount of energy released in a supernova explosion ( @xmath8 erg ) , supernova remnants are prime candidates to be sources of galactic cosmic - rays . in microquasars , the kinetic energy carried by the jet accounts for at least @xmath9 , inferred from the observed non - thermal luminosities @xcite . the total energy injected into the interstellar medium by the winds of ob and wolf - rayet stars can be as high as @xmath10 , like in the case of the cygnus ob2 association @xcite ; and levels of @xmath11 can be achieved by a single young pulsar . what remains undetermined is the fraction of total energy per source that goes into cosmic - ray acceleration , as well as the probability for the interaction of cosmic - rays close to their source . the observation of the products of cosmic - ray interactions , i.e. gamma - rays and neutrinos , can shed light on these still unsolved problems . the highest energies ( e > 100 tev ) are only accessible by means of extensive air - shower ( eas ) arrays , in the case of gamma - rays , and @xmath12-volume neutrino detectors like icecube . results from milagro @xcite and argo - ybj @xcite , demonstrate that the gamma - ray emission is faint at very high energies @xcite . however , these gamma - ray observations do not impose constraints on neutrino production , due to internal and external absorption of gamma - rays at the energies considered . most of our knowledge of gamma - ray sources comes from cherenkov telescopes like h.e.s.s . , magic , and veritas , working in the energy range 100 gev - 100 tev . in the past few years , a large family of galactic accelerators have been observed to have the bulk of their gamma - ray emission at energies below 50 tev and/or to be softer than the @xmath13 spectrum that is generally expected from first - order fermi shock acceleration @xcite . many of these sources reside relatively nearby , and external gamma - ray absorption in the interstellar radiation field is not likely . if the detected gamma - rays are related to hadronic acceleration and are produced in transparent sources , the expected distribution of neutrino energies has the same spectral index and a lower energy cutoff than the gamma - ray spectrum @xcite . the modeling of cosmic - ray sources with diffusive shock acceleration also allows for the presence of spectra steeper or flatter than the generic e@xmath1 behavior , depending on the configuration of the shock @xcite . the influence of diffusion in the sources themselves may modify the spectra to produce primary spectra of e@xmath14 or steeper @xcite . in order to target soft - spectra sources , we have optimized the search here reported for a generic spectrum proportional to e@xmath2 . in this work , we use the 22- and 40-strings configurations of icecube ( ic22 , ic40 ) as well as the antarctic muon and neutrino detector array ( amanda ) ( ic22+a , ic40+a ) to enhance the sensitivity for soft - spectra sources , or sources with energy cut - offs . we have used amanda as an integrated low - energy extension of ic22 and ic40 and developed an analysis strategy that is optimized for a high retention of signal events below 10 tev . we have used the resulting low - energy optimized data samples to search for galactic neutrino emission above @xmath15200 gev . at these energies , icecube s field of view covers the range of galactic longitude @xmath16 , as illustrated in figure [ galaxy ] . the performed searches include a scan of the accessible part of the galactic plane , a dedicated analysis of the cygnus region , the search for neutrino emission from a pre - defined list of interesting astrophysical objects and an analysis that searches for time - dependent neutrino emission from cygnus x-3 in correlation with radio flares . the paper is organized as follows : section 2 describes the relevant technical aspects of icecube , of amanda and of its integration into icecube . section 3 reports about the analysis methods that have been applied and the respective astrophysical targets . section 4 explains the details and characteristics of the obtained data samples , and section 5 provides the results . ( i.e. lower left region).,scaledwidth=40.0% ] during the construction phase from 2004 to 2010 , the operational configuration of icecube increased year by year ( see figure [ ic_am ] ) to finally cover a volume of approximately one cubic kilometer . icecube , including its deepcore extension , is composed of 86 strings each holding 60 digital optical modules ( doms ) . each dom contains a 10-inch photomultiplier tube ( pmt ) and an on - board signal read - out and digitization system , all housed in a glass pressure vessel @xcite . 78 of the 86 strings in the array form a hexagonal grid with a typical distance of 125 m between neighboring strings . the vertical distance between doms on the same string is 17 m. the remaining 8 strings are part of the low - energy extension deepcore @xcite and are deployed in the deepest , clearest ice at the center of the detector with a smaller vertical and horizontal spacing between the doms of 7 m and 60 m , respectively . the doms detect cherenkov radiation emitted by secondary charged particles produced in interactions of high - energy neutrinos with nuclei in the ice or the bedrock below the ice . to enhance the detection of light from upward - going particles , the pmts point downwards . in order to avoid a deterioration of the analog pmt signal , the signal is digitized directly in the doms with a set of analog transient waveform digitizers ( atwds ) and a fast analog to digital converter ( fadc ) @xcite . the events that are used in this analysis are selected by a multiplicity trigger which requires at least 8 hit doms within a time window of @xmath17sec . the doms send their recorded signals to the surface and an event is constructed if the trigger conditions are met . in the detector configurations used in this work , only doms for which there is also a signal from one of the nearest two doms above or the nearest two doms below within 1 @xmath18sec ( so - called hard local coincidence ) are considered in the trigger and the event building to suppress noise contributions . an event contains all dom readouts associated with the trigger as well as all further readouts within @xmath19sec around the trigger time . after a construction phase from 1993 until 2000 , the completed amanda - ii detector took data as a standalone neutrino telescope from february 2000 until december 2006 . this configuration consisted of 677 optical modules ( oms ) on 19 strings . most of the optical modules were deployed at depths between 1500 and 2000 m whereas icecube extends down to 2450 m. for data analysis , a total of 526 oms have been used . the amanda strings follow a roughly cylindrical geometry , as is shown in figure [ ic_am ] . the typical distance between adjacent strings is around 40 m and the average vertical spacing between the modules is about 15 m . from february 2007 until april 2009 , amanda was operated as an integrated part of icecube . in many aspects , icecube is technologically more advanced than amanda , reflecting general progress as well as experience collected during the operation of amanda @xcite . in particular , the signal transfer from the optical modules to the surface is different . as mentioned above , icecube doms digitize the pmt signal directly in the ice . they also generate the hv for the pmts in the doms . in contrast , amanda oms produced analog signals that were sent to the amanda data acquisition system which was located in the martin a. pomerantz observatory ( mapo ) . the original data acquisition system ( daq ) could register the leading and trailing edge time of up to 8 pulses per om per event and only the total charge . the same cables were used to transfer the analog data and to provide the hv to the pmts . for amanda strings 11 to 19 , an additional connection via optical fibers was installed to transmit the pmt signals with a better time resolution . moreover , amanda string 18 @xcite was equipped with prototypes for the icecube doms , including the capability of on - board waveform digitization . this option however was used only in testing mode and was not included in the data acquisition schemes used for physics analysis . the amanda daq was upgraded starting from 2002 . flash adc modules called transient waveform recorders ( twrs ) were installed in the new daq in order to digitize the analog waveforms from the amanda oms at the surface . the upgraded data acquisition operated in parallel to the analog one until 2006 . from 2007 on , only the twr - daq was operational . this surface waveform digitization stored more information such as the induced charge and the arrival times of individual pulses . as the twr - daq was also faster than the previous daq , the trigger threshold could be reduced . trigger thresholds from 8 to 13 hit oms were used during different years . this upgrade significantly improved the performance of amanda and ultimately allowed to use it to enhance icecube s performance at low energies @xcite . since amanda is about eight times more densely instrumented than icecube and fully surrounded by icecube strings , it offered the potential to increase the low - energy performance of icecube and to be used as the first low - energy core inside a large neutrino telescope . this lead to its integration in the data taking of icecube . the operational integration of amanda into icecube required the establishment of connections between the two detectors for the exchange of trigger information to be able to merge events as well as for an accurate synchronization in time . mapo is about 300 m away from the icecube control lab ( icl ) that houses the icecube surface data acquisition . optical fibers have been used in order to connect the two buildings . moreover , a tcp / ip connection was established for the communication between the buildings . a gps module was installed to synchronize the twrs responsible for the digitization of the amanda waveforms and to synchronize the detector with icecube . the icecube clock was used as reference . an optical fiber connection was used to transmit the amanda trigger signal to icecube . in the integrated mode , amanda and icecube were still triggered separately . since amanda had a lower energy threshold than icecube , a readout of icecube was initiated every time amanda triggered , even if there were not sufficient hits in icecube to produce a trigger by itself . amanda was not read - out in correspondence with icecube triggers . events from amanda and icecube were merged on the basis of a time coincidence . as the duration of amanda records was fixed to 10.24 @xmath18s , while the duration of icecube records was extended if new triggers occurred within the readout window there was a possibility that more than one record in amanda was associated with one record in icecube . in this case , they were all included in the same combined event . the icecube neutrino telescope monitors the entire sky without the need of explicit pointing . the energy- and zenith - dependent sensitivity of the icecube 40-strings configuration are described in @xcite . in previous works , a generic , unbroken @xmath13 signal spectrum up to the pev region has been assumed for the optimization of the data analysis and the evaluation of the detector performance @xcite . this approach achieves the best signal to noise in the energy range above a few tev , since the assumed signal spectrum is significantly harder than the characteristic spectrum of ( background ) atmospheric neutrinos @xmath20 . however , a lower energy threshold is of primary importance for the search for neutrino sources characterized by an energy cutoff or by soft spectra . in the optimization of the analysis , we have used a generic , soft power - law spectrum following @xmath21 and we have also considered the crab nebula spectrum measured by h.e.s.s . @xcite which would correspond to a neutrino spectrum @xmath22 with an exponential cutoff at 7 tev ( provided that all the measured gamma - rays are of hadronic nature ) . this last spectrum is representative of a `` low - energy '' source and its study is very instructive in the understanding of the impact of an energy cutoff on the performance of icecube . we will refer to this spectrum throughout the paper as a `` crab - like '' spectrum . we report in the following on the different searches that have been performed . the location of the solar system in the local spiral arm gives us a particular view of the galaxy . given the vertical scale of the thin disk and the distribution of cold gas in the galaxy , we see most of the galactic accelerators projected in a narrow band close to the galactic plane . in the energy range of interest for the detection of galactic neutrino sources , icecube can explore the northern sky , which includes part of the first quadrant of the galaxy , the whole second quadrant , and a small portion of the third galactic quadrant . the search for neutrino sources in the galactic plane is performed by superimposing a fine grid over the region of the sky within the galactic coordinates @xmath23 , @xmath24 . the step size of the grid is chosen to be smaller than the angular resolution achieved in the analysis reported in section 4.4.2 . in the analysis of ic22+a , a grid of @xmath25 has been used , while a @xmath26 grid has been chosen for the ic40+a , given the improved angular resolution ( see figure [ ang_res ] ) . at each point of the grid , an unbinned maximum likelihood ratio test is performed on all selected events . the likelihood of a composite signal and background hypothesis is compared to the background - only hypothesis , similar to the method described in @xcite , without the inclusion of an energy term in the likelihood . seven particularly interesting sources have been studied individually in this analysis as representatives of different types of galactic accelerators . the interest in these sources is motivated by the observation of a gev - tev gamma - ray counterpart at the time of the analysis . for cygnus x-3 , due to the high variability in the radio and x - ray bands , we have tested the hypothesis of variable neutrino emission and performed a time - dependent analysis . the other sources are listed below and are treated as steady point source candidates . * crab nebula , distance:@xmath152.0 kpc * @xcite : the crab nebula is powered by a pulsar with a spin - down luminosity of @xmath27 . this energy is injected into relativistic particles and magnetic fields @xcite , although the exact composition of the pulsar wind , as well as the mechanism by which the total power of the pulsar is transported and dissipated is not known . it is an efficient particle accelerator , where @xmath28 of the total power of the pulsar is injected into relativistic electrons which emit synchrotron radiation from radio to x - rays @xcite . although it appears as the strongest gamma - ray source in the sky , the ratio between the gamma - ray luminosity at e @xmath29 1 tev and the spin - down luminosity is of the order of @xmath30 @xcite . the simplest interpretation of this is that electrons rapidly lose their energy through synchrotron radiation at lower frequencies , and that the majority of cosmic - rays , if present in a significant proportion , escape from the source without interaction . the constraint on the steady neutrino production in the crab obtained by icecube @xcite is at the level of @xmath31 , a factor of @xmath32 larger than the luminosity in gamma - rays assuming the h.e.s.s . gamma - ray spectrum @xcite ( and its corresponding neutrino spectrum expected in case of an hadronic origin ) extrapolated to the energy range between 400 gev and 40 tev . * cas a , distance:@xmath153.4 kpc * @xcite : this source is a classical shell - type supernova remnant ( snr ) . its high - energy gamma - ray flux was detected by hegra in the energy region between 1 tev to 10 tev without any evidence of an energy cutoff @xcite , and detected by magic down to 250 gev following a power - law spectrum @xmath33 and with an integrated photon flux above 1 tev of @xmath34 @xcite . * ic 443 , distance:@xmath151.5 kpc * @xcite : ic 443 is an asymmetric shell - type snr , where part of the shell is impacting on a molecular cloud , accelerating particles to very high energies in the process . tev gamma - rays are observed arriving from the molecular line emission region , giving support to a hadronic origin of the tev gamma - rays . the spectrum measured in the energy range from 100 gev to 1.6 tev is well fitted by a very steep power - law @xmath35 @xcite . the integrated photon flux above 1 tev obtained by extrapolation is @xmath36 . * w51 , distance:@xmath156.0 kpc * @xcite : this source has been detected in gev gamma - rays by the fermi - lat telescope @xcite , at tev energies by h.e.s.s . @xcite and by magic @xcite . the high - energy emission is thought to arise from the interaction between a composite snr ( w51c ) with a molecular cloud present in the region . the high luminosities observed in gev gamma - rays , greater than @xmath9 , and the hints of a hadronic origin for the gamma - ray spectrum makes this an interesting target for icecube despite its large distance . the magic collaboration recently extended the spectrum from the highest fermi / lat energies to 5 tev and finds that the spectral index of the source follows a single power law with an index of @xmath37 @xcite . * ls i+61 303 , distance:@xmath152.0 kpc * @xcite : this source is a high - mass x - ray binary with a compact object in an eccentric orbit around a be star . the nature of the compact object is not known , and both a pulsar wind model and a microquasar model have been suggested for this source . magic detected very high energy emission modulated with the orbital period @xcite . the highest significant detection is obtained around apastron , at orbital phases 0.6 - 0.7 , with a spectrum following dn / de @xmath38 , at e @xmath29 300 gev . no tev emission is observed at periastron , although significant gamma - ray absorption in the strong radiation field of the be star is expected in this case @xcite . this scenario is supported by the detection of mev - gev gamma - rays by the fermi satellite @xcite , which may result from the cascade process in @xmath39 . the hypothesis of particle injection along the whole orbit is then a plausible option . this , together with the considerable amount of both matter and radiation from the companion star available for cosmic ray interactions , makes this source an interesting candidate for steady neutrino emission . for the search of periodic neutrino emission from binary systems performed by icecube , we refer to @xcite . * ss 433 , distance:@xmath155.0 kpc * @xcite : ss 433 is a confirmed microquasar and a black hole candidate in orbit around a massive star . the source exhibits two oppositely directed relativistic jets which are thought to eject material at a rate larger than @xmath40 @xcite . it is the only x - ray binary system in which hadrons have been found in the jet @xcite . the entire source is embedded within a nebulous structure ( w50 ) which is thought to be the expanding supernova shells of the progenitor star of the black hole in ss 433 . the source has been searched for by the hegra , magic , and cangaroo - ii cherenkov telescopes @xcite , resulting in upper limits for the gamma - ray emission from both the inner system and the different interaction regions with the w50 nebula . strong gamma - ray absorption is expected from this system , due to the periodic eclipses through the companion star as well as attenuation due to the precession of the accretion disk envelope @xcite . as in the case of ls i+61 303 , the presence of a significant amount of target material for cosmic ray interactions as well as the possibility of a higher energy emission than what is inferred from gamma - ray observations due to absorption , makes this an interesting candidate for a neutrino source . ss 433 has also been tested for possible periodic neutrino emission in @xcite . the cygnus region is roughly located within galactic longitudes @xmath41 and latitudes @xmath42 , where our line of sight is directed nearly along the local spiral arm of the galaxy @xcite . at a distance of approximately 5 kpc our line of sight has left the local arm and crosses the perseus arm , and even the outer arm further away ( @xmath010 kpc ) . here many different sources are located at various distances superimposed in a relatively small area in the sky , resulting in a complex region which harbors some of the closest and most massive regions of star formation in the galaxy . the vast majority of the molecular gas detected in the cygnus region is concentrated on the local arm @xcite , at distances between 1 - 3 kpc . one of the most massive giant molecular cloud complexes in the galaxy resides within this region , at a distance of @xmath151.7 kpc . it is thought to be the birth place of the massive cyg ob2 association , and probably also cyg ob9 and cyg ob1 , as well as of a number of less massive star clusters with young stars or ongoing star formation @xcite . the strong stellar winds and radiation pressure of the massive stars in the cygnus region have strongly influenced the spatial distribution of the molecular gas in the region , displacing and compressing the gas forming filamentary structures and dense clumps which surround the less dense environment of the cluster , in which the gas has been evacuated . if high - energy particles are generated within the stellar associations , they can interact with the ultraviolet radiation fields producing tev gamma - rays through the inverse compton and p@xmath43 processes . however , protons and nuclei can travel longer distances than electrons , and they may also leave the photon dominated regions around the massive star clusters and interact with the nearby molecular clouds . the resulting neutrino flux map would then reflect the complicated distribution of the gas in the region . it is also worth noting that the injection of cosmic - rays may take place at multiple locations due to presence of several particle accelerators inside the cygnus region . the existence of these accelerators is confirmed by the observation of strong tev gamma - ray emission throughout an area of approximately @xmath44 @xcite . the potential for icecube to observe neutrinos from this region has been discussed in @xcite , based on the measured tev gamma - ray flux . due to the complexity of the possible spatial distribution of events within the cygnus region , we have applied an analysis of the spatial correlations between neutrino events to search for an astrophysical neutrino signal in an extended region . if the intensity fluctuations of a possible neutrino signal throughout the region follow a certain correlation structure , this may show up as a significant deviation from the random distribution of atmospheric neutrino events . in the icecube analysis , we use the two - point correlation function formalism introduced by peebles @xcite and co - workers @xcite to study the large - scale matter distribution in the universe @xcite . in particle astrophysics , the correlation function has been used to search for anisotropies in the spatial distribution of cosmic - rays @xcite and neutrinos @xcite . here we use the approach to search for neutrinos inside an area of @xmath45 centered on the most active part of the cygnus complex in tev gamma - rays . we define our test statistic in terms of a clustering function , @xmath46 , which is the excess or deficit in the number of event pairs within a certain distance with respect to the background - only hypothesis ( similar to a cumulative correlation function , see e.g. @xcite and references therein , @xcite for estimators of the two - point correlation function based on pair counting ) : @xmath47 where @xmath48 is the distance between two events , and @xmath49 , @xmath50 , where the sum is over all non - repeated pairs in , respectively , the real data sample and in a sample randomized in azimuth ( representative of a pure background case ) . in our case , @xmath51 only if either the event @xmath52 or the event @xmath53 , or both , are within the region under study , and it is equal to zero otherwise . it is worth noting that with this definition we measure both the intensity of the process that generated the observed neutrino event pattern as well as its correlation structure . [ cygx3_flares ] the high - energy sky presents strong variability in time . if neutrino emission from a particular object is expected to vary with time and if it is in coincidence with electromagnetic emission , it is advantageous to include the time information in the data analysis . one such case is the microquasar ( cyg x-3 ) at a distance of @xmath54kpc . coinciding with the period of data taking considered in this analysis , first observations of production of high - energy photons within the system of cyg x-3 were published by the fermi @xcite and agile @xcite satellite missions independently . the reported gamma - ray fluxes are in the energy band between 100 mev and 100 gev ( fermi ) , following a power - law of spectral index @xmath55 , and between 100 mev and 10 gev ( agile ) with spectral index @xmath56 . however , gamma - rays from cyg x-3 are only observed during certain periods of time , probably correlated with strong radio outbursts and certain x - ray emission states of the system @xcite . no tev gamma - rays from cyg x-3 have been detected by cherenkov telescopes like magic and veritas so far , a possible explanation being strong absorption of high - energy photons or limited observation time . upper limits on the integrated gamma - ray flux above 250 gev are at the level of @xmath57 @xcite . assuming a hadronic origin of at least part of the gamma - rays , the high densities in the system of cyg x-3 could provide an environment for copious neutrino production at tev energies , detectable by neutrino telescopes such as icecube @xcite . a previous search for periodic neutrino emission was reported in @xcite . here we perform a search for neutrino emission during , or close to , the observed flaring activity in cyg x-3 . the phenomenology of emission states of cyg x-3 has been studied carefully using radio and x - ray observations @xcite . with radio and x - ray data , there are two ways to identify active periods of cyg x-3 associated with jet ejection : the observation of a radio flux above 1 jy , following @xcite , and the observation of hyper - soft x - ray states and subsequent hardening of the x - ray spectrum , following @xcite . the identification of potential flaring periods of cyg x-3 is thus split into a radio and an x - ray part , as in table [ cygx3_flares ] . for reconstruction of radio flares from x - ray data , an average radio flare was obtained from gaussians fitted to 28 flares seen to rise above 1 jy in radio data . one average radio flare was put at each time of potential radio flaring , as seen in x - ray data ( i.e. spectral hardness > 0.001 within 10 days after a state with hardness < 0.001 ) to obtain a pseudo radio flux density @xmath58 . the resulting search windows from all selections can be overlapping and are combined with a logical or operation , resulting in the final time windows . the utilized radio data were taken with the arcminute microkelvin imager ( ami ) radio telescope @xcite at a frequency of 15 ghz between may 2008 and may 2009 in irregular intervals . x - ray data were obtained from the rossi x - ray timing explorer / all sky monitor ( rxte / asm ) @xcite , using the b band between 3 and 5 kev , and from the burst alert telescope on board the swift satellite ( swift / bat ) @xcite sensitive between 15 - 50 kev . the ratio of bat ( hard x - ray ) to asm b ( soft x - ray ) counts provides the spectral hardness parameter . gamma - ray data from fermi or agile are not explicitly taken into consideration in this analysis . applying the selection criteria described in table [ cygx3_flares ] to the radio and x - ray data from the time when icecube 40-strings was operating ( between mjd 54560 and mjd 54989 ) results in the four time windows indicated in table [ windows ] and figure [ fig_windows ] . even though there are no ami data from the first 2.5 months of this period , an atel @xcite was issued for a strong radio flare around mjd 54574 that is consistent with the time windows selected from x - ray data . cccc & start & stop & duration + & ( mjd ) & ( mjd ) & ( days ) + 1 & 54571.4 & 54582.5 & 11.1 + 2 & 54584.5 & 54607.4 & 22.9 + 3 & 54637.6 & 54649.5 & 11.9 + 4 & 54811.5 & 54824.6 & 13.1 + [ windows ] the data have been analyzed with a maximum likelihood test using a time - dependent version of the unbinned likelihood ratio method @xcite . the search time windows are incorporated into the signal probability density function ( p.d.f . ) of the likelihood function as normalized gaussians with mean located at the window center and fwhm equal to the window duration . during the maximization , the windows were allowed to be shifted up to 20 days to earlier or later times . this allows us to find neutrino emission that comes before or after a radio flare . the value of 20 days is motivated by the hypothesis of emission during the radio quenched state , which can happen up to @xmath59 days before the onset of a major radio flare @xcite . as in the other analyses presented in this work , no energy estimator is used in the likelihood . in the search for neutrinos from the microquasar cygnus x-3 , five searches are performed in total : one with each of the four windows as a hypothesis of a neutrino signal light curve , shifting each window individually , and one search using all four windows , shifting the windows simultaneously . doing this analysis , only about @xmath60 of the discovery flux of a time - integrated search ( that uses no information about the activity of cyg x-3 ) is needed for a @xmath61 discovery , assuming the neutrino emission happens during the windows or within @xmath62 days . the discovery flux with @xmath60 detection probability is @xmath63 for an e@xmath2 spectrum and @xmath64 for an e@xmath1 spectrum . we discuss here the selection of the sample of neutrino candidates for each of the two considered detector configurations . while the event selection is optimized for each of these two configurations separately , the two analyses are both based on the selection of events on the basis of track reconstructions . we therefore first discuss the concepts that are common to the two event selections : the track reconstruction and its application in on- and off - line event selection as well as a special filtering of the amanda data . at energies above @xmath15200 gev , the identification of neutrino candidate events in the icecube and amanda data is based on the selection of well - reconstructed up - going events since the dominant background are muons from cosmic ray air - showers , which reach the detector in the downward direction but are filtered out by the earth in the upward direction , leaving only neutrinos . the icecube and amanda data are divided into two streams which are treated in a similar way . the first stream are events that trigger amanda . as explained above , these events are then complemented with the data collected in icecube and for this reason they are considered as combined events ( c - events ) . the second stream are those events which trigger only icecube ( ico - events ) , either because the number of hits in amanda is below the trigger threshold or because amanda was not active at the time the event was recorded . initial fast track reconstructions and event selections are applied on - line at the south pole since the available bandwidth for satellite transfer of data is limited . for ico - events , a straight track is fit to the data by minimizing the distance between the track and the hits ( linefit ) @xcite . for c - events , a pattern recognition algorithm ( called jams ) has been used @xcite . the on - line filters are based on these reconstructions and the relevant events for the presented analysis are selected by requiring that the first - guess reconstruction is not down - going . at this level , the data are still dominated by atmospheric muons that are mis - reconstructed as up - going . in particular , coincidences between multiple muons from different cosmic - ray air showers can mimic up - going event topologies . after transfer to the north , more sophisticated reconstructions are applied to improve the angular resolution and to provide quality parameters for the rejection of the down - going atmospheric muon background . these reconstructions are maximum likelihood fits which are based on the probability density function for the arrival time of a photon given the track hypothesis @xcite . two likelihood reconstructions have been studied : one based only on the first photon in each optical module ( spe ) and a more complete one which includes the possible presence of multiple photo electrons ( mpe ) . for more details about the spe and mpe reconstructions , we refer to @xcite . a list of variables that are indicative of the quality of the reconstructed event is given in @xcite . in addition to the track reconstruction itself , an estimate of the angular uncertainty of the track reconstruction is obtained for each individual event by the evaluation of the likelihood function near the maximum . this method is described in @xcite . the estimate of the angular uncertainty is used for the event selection as well as in the maximum likelihood ratio test . in the ic40+a analysis , we also make use of an energy reconstruction which is based on the characterization of the energy loss along the particle track . at energies above a few hundred gev , the energy loss of a muon in the ice is proportional to its energy . the energy reconstruction used in this work is presented in @xcite . in contrast to icecube , the waveforms collected with the optical modules of amanda are not digitized directly in the optical modules but are transferred to the surface as analog signals . this introduces two undesired effects in the data . the first one is crosstalk between different cables . large pulses in one optical module can cause a detectable signal within the electronics of cables connected to other optical modules . the second issue concerns the pickup of electromagnetic noise . the pmt signals have to be transferred over a distance between 1.5 km and 2.0 km to the surface . the cables needed for this task are vulnerable to pick up electromagnetic noise as they act as electromagnetic antennae . methods dedicated to the identification of non - particle induced signals based on the waveforms have been developed and result in an efficient separation from particle - induced signals . the integral over the entire collected waveform pulse is used in order to remove crosstalk pulses . since crosstalk pulses do not originate from a charge deposit in the pmt , they consist of ( positive and negative ) fluctuations around the baseline with the total integral close to zero . waveforms from a particle - induced signal in the pmt have a characteristic width , which is wider at the surface due to dispersion in the cables . for amanda strings 5 - 10 , this is typically 250 - 300 ns for single photoelectrons ( spe ) . waveforms from multiple photoelectrons ( mpe ) result from the ( linear ) overlay of many spe waveforms and typically are wider than those . in contrast , noise - induced waveforms are often very spiky , i.e. they have many peaks within the width corresponding to a typical spe pulse . this feature of non - particle - induced waveforms has been used to remove noisy events from the data : events which simultaneously have a median peak rate in the waveforms recorded in amanda strings 5 - 10 which is incompatible with a pmt signal ( above 5 mhz ) and a high number ( more than 20 ) of waveforms in these strings , are considered non - particle induced and are removed from the data set . figure [ flarecheck ] illustrates this cut . both analyses presented in this paper first apply cross - talk cleaning and the above technique for rejection of non - particle induced events to the amanda data , before further event selections are made . data have been collected from may 31 , 2007 until april 4 , 2008 when icecube was operating in a 22-string configuration . the lifetime of the ic22+a run is 276 days , including 143 days of amanda operating in stable mode . the unusually long downtime of amanda during this period was caused by various hardware failures during may 2007 ( trigger system ) and during august 2007 ( high voltage supply system ) . in this section , the event selection from the trigger level up to the final analysis level is described and the characteristics of the combined neutrino sample are highlighted . as explained above , icecube and amanda are triggered separately in the combined detector mode . the trigger rate of ic22 is 550 hz , while amanda triggers at 200 hz . seasonal variations affect the trigger rate by about @xmath65 . the overall trigger rate of the combined ic22+a detector after correction for overlaps between the two triggers is 640 hz . at trigger level , the data are strongly dominated by down going atmospheric muons induced by cosmic - ray air showers outnumbering atmospheric neutrinos by a factor of about @xmath66 . the on - line filter reduces the data volume for satellite transfer and has a passing rate for reconstructed up - going or horizontal events of 22 hz for ico - events and 8 hz for c - events , producing a total event rate of 30 hz . 1.6 larger than without amanda . the last bins of the ratio plot are affected by limited statistics.,title="fig:",width=336 ] 1.6 larger than without amanda . the last bins of the ratio plot are affected by limited statistics.,title="fig:",width=336 ] after the rejection of down - going reconstructed events based on first guess reconstructions , the data are still dominated by mis - reconstructed atmospheric muons and further event selections are needed in order to arrive at a sample of events dominated by atmospheric neutrinos . a typical quality parameter used in this analysis is the number of un - scattered photons , so - called `` direct hits , '' which are characterized by a small time residual with respect to the expectation of the geometry of the emitted cherenkov cone . in order to optimize the retention of lower energy events we have employed a multivariate approach . in this approach , a signal likelihood is defined as the product of the signal likelihood from the considered variables and is compared to the likelihood of the background hypothesis . according to the neyman - pearson lemma @xcite , this criterion leads to the best possible discrimination power for the given set of variables if there are no correlations between them . for correlated variables , as in our case , this criterion turns out to still be powerful . we use experimental data dominated by atmospheric muons to describe the background , and simulated neutrinos weighted to an atmospheric neutrino spectrum in order to model the signal . c - events and ico - events are treated independently here . for ico - events , further cuts are used to reject coincident air shower muons . these cuts are based on the smoothness of the distribution of the hits along the track and on reconstructions performed on subsets of hits . for c - events , the rate of coincident muons is significantly lower due to the smaller size of amanda . while a tighter time - window cleaning helped to further reject the coincident muons in the combined events stream , no dedicated cuts / reconstructions have been used to remove these . the final cuts were optimized for the best discovery potential , and finally tracks are selected if the angular resolution estimator returned a value lower than @xmath67 . a harder cut on this parameter did not lead to an improved discovery potential . the resulting neutrino sample contains about 1.8 times more events than the search presented in @xcite . in total , 8727 events are selected , of which 3430 are c - events . these event numbers correspond to a data rate of @xmath68 hz for ic22+a and to @xmath69 hz in ic22 only mode . the effective area of the ic22+a analysis is shown in figure [ a_eff_ic22ama ] . below energies of a few tev , there is a strong improvement when including amanda , with respect to the performance of a low - energy optimized analysis using icecube only . the energy distribution of atmospheric neutrinos for this selection according to simulations is shown in figure [ atm_e_ic22_ama ] . the estimated angular resolution is given by a median of @xmath70 for atmospheric neutrinos and @xmath71 for a crab - like spectrum . about @xmath65 of the final events are expected to be mis - reconstructed muon background . the minimum detectable flux is more than an order of magnitude above the neutrino emission expected from the crab assuming that the h.e.s.s . observations are consistent with a model of @xmath72 interactions . while with this expectation , a positive detection is unlikely but not excluded as the photon flux may be absorbed , the analysis is valuable as a starting point for future improvements with the full icecube detector and the deepcore sub - detector . using an e@xmath2 spectrum for the comparison , the discovery flux in this analysis is between 15% and 33% lower than the one in @xcite , depending on the declination . this clearly demonstrates the improvement obtained by the combined use of a low - energy core and the optimization of the event selection for softer spectra . an additional improvement is obtained by the lower number of trials in the galactic plane scan versus the scan of a hemisphere or the whole sky but is not quantified here . following the explorative analysis strategy developed for the ic22+a , a similar analysis has been conducted on the larger data sample collected with the combined ic40+a detector from april 5 , 2008 to may 20 , 2009 . both parts of the combined icecube - amanda detector operated very stably during this time period . for ic40 about 375 days of data were collected and used in this analysis and for amanda about 306 days . the main causes for downtime were scheduled operations in the course of the integration of new strings into the detector . moreover , the decommissioning of amanda began a few weeks before the completion of the ic40 run . the event selection is in many aspects similar to the one applied to the ic22+a data as the targeted energy range is the same as well as the physics driving the analysis . again , different cut criteria are developed for c- and ico - events . for what concerns this analysis , the trigger logic was kept identical to the one in the previous season . the on - line filter selected about 20 hz of track - like ico - events and 3 hz of up - going c - events . similar to the previous combined analysis , ico - events are selected by a series of one - dimensional cuts on event quality parameters and combined with a multivariate classification based on the neyman - pearson lemma @xcite . the probability density functions for five quality parameters are generated from down going atmospheric muon - dominated data as background and from up going atmospheric neutrino simulation as signal , and combined in the cut . ( the fraction of atmospheric neutrino events in the data is still only about 4% and the data can therefore be regarded as dominated by background atmospheric muons . ) the five quality parameters used in this analysis are : the quality parameter of the likelihood reconstruction , an estimate of the angular uncertainty of the likelihood reconstruction obtained by the evaluation of the likelihood space around the maximum , and three variables which describe the amount and distribution of unscattered light in the event . a pmt pulse from un - scattered light is characterized by a small time residual with respect to the expectation from the geometry of the cherenkov cone . the number of pmt pulses with time residuals between -15 ns and 75 ns , the maximum distance between their projections on the reconstructed track and the smoothness of their distribution along the track of the particle are used in the event selection . more information about these variables is reported in @xcite . the distribution of the resulting cut variable is shown in figure [ ic40ama_np ] for data and for atmospheric neutrino simulation as well as for two example signal neutrino spectra . an optimization of the discovery potential for a soft e@xmath2 spectrum results in an optimal cut value of 1.0 . c - events are first cleaned as described in section [ sec4.1 ] . subsequently , a series of one - dimensional cuts is applied . this series of cuts has been challenged versus other more sophisticated cut strategies and demonstrated to perform well enough in the separation of the atmospheric muon background and the neutrino signal . while the analysis is optimized for soft spectra such as e@xmath2 or the crab - spectrum , it is nevertheless desirable to retain a good efficiency for very high - energy neutrino events as well . with a cut optimized on a very soft spectrum however , the retention of high - energy events is not necessarily optimal as these deposit much more light in the detector and may thus have event topologies that are not caught in the low - energy event selection . to remedy this , additional criteria are included in the event selection if the signature is likely to be induced by a very high - energy neutrino . these additional events are selected in both streams with a series of one - dimensional cuts based among others on their reconstructed energy . the ic40+a analysis also uses maximum likelihood track reconstructions . for low energetic combined icecube - amanda events , a single photo electron ( spe ) pdf is used and for all other events , a multiple photoelectron ( mpe ) pdf is used . the total number of selected neutrino candidates is 19,797 in the entire lifetime of 375 days . the purity of the atmospheric neutrino sample is estimated to be 97 - 98% . of the selected neutrino candidates , 81.3% are ico - events selected with the multivariate neyman - pearson likelihood ratio cut . 2.4% are additional ico - events with high estimated energies . the remaining 16.3% events are c - events . despite the larger lifetime of amanda in 2008/2009 with respect to the previous year , the fraction of c - events in this analysis is smaller than in the previous one . this is partially due to the larger size of the icecube detector but also to the higher purity of the ic40+a sample . the resulting energy distribution as derived from atmospheric neutrino simulation is shown in figure [ ic40ama_mcen ] . the selected c - events peak at lower energies than the ico - events . also the effective area , reported in figure [ ic40_eff_area_comparison ] , shows the effect of amanda at the lower energies . with the larger size of the detector , the effective area is improved significantly with respect to the ic22 analysis . figure [ ic40_azimuth_declination ] shows the distribution of the declinations of the events in the final neutrino sample . the angular resolution obtained with this selection of events is shown in figure [ ang_res ] , together with the angular resolution of the ic22+a analysis . the sensitivity and discovery potential for e@xmath2 neutrino spectra are reported in figure [ sensitivity ] . neutrino spectrum includes on average events with higher neutrino energies and better angular reconstruction . with the ic40+amanda configuration , 90% of the neutrino events with an e@xmath1 neutrino spectrum are in the energy interval from 2.4 tev to 750.0 tev while the energy region from 0.2 tev to 20.5 tev contains 90% of the neutrino events if an e@xmath2 spectrum is considered.,scaledwidth=40.0% ] neutrino spectra . the sensitivity is calculated with the method of feldman and cousins @xcite with a systematic uncertainty of @xmath73 on the neutrino flux included using the method defined in @xcite and modified in @xcite.,width=336 ] as a result of the integrated use of the amanda detector within icecube , we have obtained a significant improvement in the retention of neutrino induced events below a few tev ( see figure [ ic40_eff_area_comparison ] ) . this region is of importance for sources with soft or cutoff spectra . the tests described above resulted in no evidence for significant deviation from the background - only hypothesis . in the absence of detection of an astrophysical neutrino signal , upper limits on the muon neutrino flux from the considered regions of the galaxy have been determined . all the upper limits have been derived for soft neutrino spectra , with and without energy cutoff . the 90% confidence level limits ( @xmath74 ) are calculated using the method of @xcite i.e. , @xmath75 . systematic uncertainties have been included in the limit determination using the method defined in @xcite with the modification in @xcite . the upper limits are calculated for the total of the muon neutrino and antineutrino flux reaching the earth , assuming that no other neutrino flavors contribute to the possible signal . for a source producing muon and electron neutrinos in the ratio of about @xmath76 typical of pion production from pp or p - gamma interaction , neutrino oscillations with a large mixing angle @xmath77 and long baseline result in approximate equipartition of flavors . this analysis is to some extent sensitive to @xmath78 as well , mainly due to the decay of a @xmath79 into a @xmath18 with a branching ratio of @xmath80 . taking into account these effects and the details of energy losses , the contribution of @xmath78 is estimated to be an additional 10 - 16% of the @xmath81 contribution for icecube @xcite and amanda analyses @xcite . + after completion of this analysis , a slight over - prediction of the muon neutrino flux has been observed by comparing the results with an improved montecarlo simulation not available before . the intensity of the effect varies in declination and energy and it is estimated to be less than 30% . as a consequence , all upper limits reported in this work as well as in @xcite , @xcite are slightly over - constraining . in figures [ ic22ama_scanresult ] and [ ic40ama_scanresult ] , we present the results of the scan of the galactic plane with ic22+a and with ic40+a . we note here that the considered region of the galactic plane covers only around 1/23 of the whole sky . this restriction in the tested area results in a lower trial factor compared to a @xmath82-map . under the conservative assumption of a uniform angular resolution , the effective number of trials is expected to be a factor 23 lower than in an all - sky scan like the one realized in @xcite . assuming that the relation between the post - trial p - value @xmath83 and the pre - trial p - value @xmath84 is @xmath85 , the effective number of trials in the ic40+a search is about @xmath86 . the reported post - trial p - values have been obtained by performing the analysis on randomized , and therefore signal - free , data samples . in the analysis of ic22+a , the lowest background probability at point source angular scales is found at galactic coordinates @xmath87 , with a pre - trial p - value of 0.37% . considering the intrinsic trials of the scan by analyzing randomized data samples , an equal or higher significance in at least one of the scanned locations is found in 95% of the cases . the most significant point - like spot in the analysis of ic40+a , with a pre - trial p - value of 0.09% , is found at galactic coordinates @xmath88 . an equal or higher significance is found in @xmath89 of the randomized data samples . grid point , the result of the maximum likelihood analysis is shown in terms of its estimated significance ( @xmath90 pre - trial p - value ) . the reconstructed event locations are also indicated as black dots . the most significant excess of events is located at galactic coordinates @xmath91 and has a pre - trial p - value of @xmath92 ( -log(p - value)=3.03 ) . accounting for the trials induced by the scanning of many points inside the galactic plane , the p - value of this test is @xmath89 . below : upper limits map for a signal spectrum proportional to e@xmath2.,width=576 ] upper limits on the neutrino emission have been calculated for the six pre - selected candidate neutrino sources which have been studied for steady neutrino emission are summarized in table [ sourcelistlimitse3 ] for an e@xmath2 spectrum without cutoff . these limits have been obtained from the ic40+a analysis . l c c c c c c object & r.a . & dec & @xmath93 & pre - trial p - value & @xmath74 + crab nebula & @xmath94 & @xmath95 & 0 & - & 7.3 + lsi + 61 303 & @xmath96 & @xmath97 & 1.6 & 0.25 & 8.3 + w51 & @xmath98 & @xmath99 & 0.6 & - & 8.3 + casa & @xmath100 & @xmath101 & 0 & - & 5.9 + ss433 & @xmath102 & @xmath103 & 0 & - & 9.8 + ic443 & @xmath104 & @xmath105 & 0 & - & 7.3 + [ sourcelistlimitse3 ] in ic22+a , the highest excess in the candidate list was observed at the position of the crab nebula , with a pre - trial p - value of @xmath106 ( @xmath107 post - trial ) . in ic40+a , the most significant clustering of events was observed for lsi + 61 303 . the estimated number ( best - fit value ) of signal events @xmath108 from this location is 1.6 and the observation corresponds to a pre - trial p - value of 25% . accounting for the trials from testing six different positions , the post - trial p - value of this search is @xmath109 , i.e. @xmath109 of randomized data samples show a similar or stronger accumulation of events around one of the six objects . the clustering function @xmath46 described in sec . [ method_cygnus_region ] has been computed for events with galactic coordinates within @xmath110 and @xmath111 for both ic22+a and ic40+a data . figure [ cluster ] shows the clustering function of the events in each sample ; that is , the ratio of the number of event pairs separated by angular distance @xmath48 or less , with respect to the average number of such pairs for randomized events ( i.e. the average case would therefore be located at @xmath112 ) . in the figure , we show the 2@xmath113 - 3@xmath113 ( ic22+a sample ) and @xmath114 ( ic40+a ) levels of the distribution of @xmath46 under the hypothesis of a random distribution of events . the observed values of @xmath46 are represented together with measurement errors obtained by bootstrapping @xcite . no significant concentration of events is seen at any of the angular scales tested . the result obtained on the ic22+a data sample shows a positive fluctuation at the level of 2.3 @xmath113 . this result contains already the correction of the trials obtained via scrambling and associated to the observation at different angular scales . the region considered showed an excess with respect to the background expectation , which translates in excess values of @xmath46 at all angular scales , but no significant structure is observed . the image of the cygnus region obtained in ic22+a is shown in figure [ cygnusic22 ] on the left . however , the ic40+a analysis yields an under - fluctuation within the analyzed area , showing a rather dispersed distribution of events with respect to the average background case and its image is in figure [ cygnusic22 ] on the right . the conclusions from both the galactic plane scans and the correlation analysis are therefore that the variations in the event density in the @xmath115 region analyzed are consistent with background fluctuations . the under - fluctuation observed in ic40+a provides restrictive upper limits to point - like neutrino emission above 500 gev from the cygnus region . upper limits have been computed from the measured value of the clustering function at @xmath116 using a representative e@xmath117 power - law model under the assumption of a point - like neutrino signal located anywhere in the region . that is , assuming that the spatial correlation of signal events in the region is given only by the psf of the analysis . the upper limits from the cygnus region are at the level of @xmath118 for an e@xmath117 spectrum . in the flare search analysis of cygnus x-3 , no evidence for a signal is found in the neutrino sample for any of the sliding search windows . the smallest pre - trial p - value is @xmath119 ( resulting from the search with window 1 ) . after correction for the trials , we arrive at a probability of @xmath120 that this observation occurs in a background only sample ( final p - value ) . upper limits on neutrino emission from cyg x-3 during 20 days before and after the time windows have been determined using the method proposed by feldman & cousins @xcite and are given in table [ tab : upperlimits ] . figure [ fig : cygx-3-neutrino - times ] shows the neutrino events close to cyg x-3 as a function of time . cccccccc search & shift & & exp . & & pre - trial & + window & ( days ) & & bg . evts . & & p - value & e@xmath2 & e@xmath1 + 1 & @xmath121 & @xmath122 & @xmath123 & @xmath124 & @xmath125 & @xmath126 & @xmath127 + 2 & @xmath128 & @xmath122 & @xmath129 & @xmath130 & @xmath131 & @xmath132 & @xmath133 + 3 & @xmath134 & @xmath135 & @xmath136 & @xmath137 & @xmath138 & @xmath139 & @xmath140 + 4 & @xmath141 & @xmath142 & @xmath136 & @xmath137 & @xmath143 & @xmath144 & @xmath145 + all & @xmath146 & @xmath147 & @xmath148 & @xmath149 & - & @xmath150 & @xmath151 + [ tab : upperlimits ] from cyg x-3 and the position of the shifted search windows , each window being shifted individually . the height of the line depicts the @xmath152 of the spatial event weight ( higher line means closer to the source ) . the dashed lines show the unshifted positions of the windows . the arrows indicate the shift.,scaledwidth=70.0% ] in this paper , we have presented dedicated searches for high - energy neutrino emission in the galaxy . these analyses have been performed on data collected with two partial configurations of the icecube neutrino telescope operating in conjunction with its predecessor amanda . in the two data taking periods considered here , icecube operated in a 22-string and in a 40-string configuration , and amanda was an integrated part of icecube in both seasons . results from several searches have been presented . we have performed a scan of the galactic plane with the aim to discover point - like neutrino emission in the part of the milky way which is located in the northern hemisphere . since no significant local clustering of events has been observed , upper limits for soft - spectra neutrino emission from the galaxy have been reported . a search that is sensitive for many possible morphologies of neutrino emission , including for example the presence of several weak or extended sources , has been performed for the cygnus region of the galactic plane , yielding restrictive upper limits . both a strong tev gamma - ray source ( mgro j2019 + 37 ) and a tev diffuse component have been measured in the cygnus region with the milagro detector from the area defined by galactic latitude @xmath153 and galactic longitude @xmath154 @xcite . the diffuse flux has been measured by milagro over a region of @xmath150.02 sr and the total gamma - ray flux ( diffuse and mgro j2019 + 37 ) measured by milagro accounts for @xmath155 tev cm@xmath1 s@xmath157 at 12 tev assuming a differential source spectrum of e@xmath117 . under the hypothesis that the region is transparent to gamma - rays , the milagro measurements can be used to estimate the maximal , associated neutrino flux @xcite . assuming that all the high - energy gamma - rays reported by milagro come from decays of @xmath158 produced in proton - proton interactions , and using the same e@xmath117 spectrum adopted in @xcite , the upper limits derived from the ic40+a analysis are only a factor of @xmath152 @xcite above this estimate of the maximal neutrino flux from inside the cygnus region . this implies that icecube has the potential to detect neutrinos or to constrain the nature of the gamma - ray emission in one of the most active parts of the galaxy in the next few years . finally , a dedicated time - optimized search from the direction of the binary system cygnus x-3 has been performed based on multi - wavelength observations . upper limits for neutrino emission during specific episodes of enhanced radio and x - ray activity have been determined for this binary system . we have presented the first neutrino point source searches which use a more densely instrumented sub - array inside a large neutrino telescope . the capability to improve the performance in the energy range below @xmath010 tev in this way has been demonstrated . using this capability , we have for the first time optimized a search for neutrino point sources particularly for the more steeply falling energy spectra expected for galactic neutrino sources . amanda , the sub - array used in this work , has been decommissioned in 2009 and is now succeded by icecube - deepcore @xcite , an advanced low - energy extension of icecube . the searches presented here have demonstrated that it is possible to improve the sensitivity to galactic sources with early energy cut - offs and steeper spectra than e@xmath1 using a denser core array even at the expense of a larger atmospheric neutrino background . as a detector specifically built to enhance the sensitivity of icecube at low energies , deepcore is positioned in the deep center of the detector where the ice is clearest . this also allows to use the outer strings of icecube as an atmosperic muon veto and thus to go beyond the approach taken in this work to improve the sensitivity in the energy range below @xmath010 tev . we acknowledge the support from the following agencies : u.s . national science foundation - office of polar programs , u.s . national science foundation - physics division , university of wisconsin alumni research foundation , the grid laboratory of wisconsin ( glow ) grid infrastructure at the university of wisconsin - madison , the open science grid ( osg ) grid infrastructure ; u.s . department of energy , and national energy research scientific computing center , the louisiana optical network initiative ( loni ) grid computing resources ; national science and engineering research council of canada ; swedish research council , swedish polar research secretariat , swedish national infrastructure for computing ( snic ) , and knut and alice wallenberg foundation , sweden ; german ministry for education and research ( bmbf ) , helmholtz alliance for astroparticle physics ( hap ) , deutsche forschungsgemeinschaft ( dfg ) , research department of plasmas with complex interactions ( bochum ) , germany ; fund for scientific research ( fnrs - fwo ) , fwo odysseus programme , flanders institute to encourage scientific and technological research in industry ( iwt ) , belgian federal science policy office ( belspo ) ; university of oxford , united kingdom ; marsden fund , new zealand ; australian research council ; japan society for promotion of science ( jsps ) ; the swiss national science foundation ( snsf ) , switzerland .

the role of environmentally induced gas stripping in driving galaxy evolution in groups remains poorly understood . here we present extensive _ chandra _ and very large array mosaic observations of the hot and cold interstellar medium within the members of the nearby , x - ray bright ngc2563 group , a prime target for studies of the role of gas stripping and interactions in relatively small host halos . our observations cover nearly all group members within a projected radius of 1.15 mpc ( ) of the group center , down to a limiting x - ray luminosity and hi mass of erg s and , respectively . the x - ray data are consistent with efficient ram pressure stripping of the hot gas halos of early - type galaxies near the group core , but no x - ray tails are seen and the limited statistics preclude strong conclusions . the hi results suggest moderate hi mass loss from the group members when compared to similar field galaxies . six of the 20 hi detected group members show hi evidence of ongoing interactions with other galaxies or with the intragroup medium . suggestive evidence is further seen for galaxies with close neighbors in position velocity space to show relatively low hi content , consistent with tidal removal of hi . the results thus indicate removal of both hot and cold gas from the group members via a combination of ram pressure stripping and tidal interactions . we also find that 16 of the 20 hi detections occur on one side of the group , reflecting an unusual morphological segregation whose origin remains unclear .

1201.1291

understanding the physical processes involved in the evolution of galaxies is a key goal of extragalactic astronomy . although stellar ( or halo ) mass is emerging as playing a fundamental role for galaxy evolution , environmental influences may also have an impact @xcite . this is particularly relevant in dense environments , in which galaxies may experience a wide range of externally driven processes such as mergers , tidal interactions , and gas stripping due to interactions with ambient gas . all of these processes may act to remove a substantial fraction of the interstellar medium ( ism ) , or rapidly consume or eject it through interaction induced star formation and nuclear activity , eventually leading the galaxy to transition from blue and star - forming to red and quiescent @xcite the relative importance of these various processes should itself be a function of environment , with mergers and tidal interactions dominating in small galaxy groups ( e.g. , @xcite ) , and with galaxy harassment , ram pressure stripping , and starvation ( the cut - off of the supply of cold gas for star formation from warm / hot gas in the halo ) becoming more efficient in massive clusters exhibiting high galaxy velocities @xcite . _ chandra _ observations do show that a lower fraction of early - type galaxies in clusters contain hot halos than their counterparts in groups @xcite , consistent with expectations of the hot halo stripping efficiency being higher in more massive systems . evidence for ram pressure stripping of the _ cold _ ism component in rich clusters is also well - established through many hi studies @xcite , suggesting that interactions between galaxies and the intracluster medium may generally be an important route to removing galactic gas in the most massive systems . the situation is less clear in smaller groups , despite these representing much more typical galaxy environments . naively , galaxy galaxy interactions and mergers should be relatively more important , given the lower galaxy velocities and intergalactic medium densities in groups . indeed , the results of @xcite and @xcite show that hot x - ray halos are retained around the majority of @xmath4 early - types in the central regions of groups , and these halos are not strongly x - ray underluminous compared to those of field galaxies . hence , removal of hot halo gas by ram pressure must be very modest in groups , at least for massive galaxies . nevertheless , numerical simulations suggest that starvation may still occur for moderate - luminosity galaxies on their first passage through even fairly small groups @xcite . viscous stripping of galactic gas through kelvin helmholtz instabilities , the efficiency of which depends only linearly on galaxy velocity @xcite , could also play a role in groups even when brute - force ram pressure is unimportant @xcite . for the cold gas , a number of group galaxies show evidence for extended hi that has been stripped from their host ( e.g. , @xcite ) , and many are deficient in hi compared to similar field galaxies @xcite . some of these objects represent strong candidates for ram pressure stripping of both cold and hot ism @xcite . however , the hi properties of spirals in these environments can often be equally well explained by tidal encounters @xcite , and so it is still not fully clear which process , if any , dominates the gas removal from typical group galaxies . quantifying the importance of the various mechanisms acting on group galaxies requires detailed observations of the ism in these galaxies , both within individual groups and across systems displaying a range of global properties . to identify signatures of ongoing gas removal and how these may depend on local group environment , the ism must be probed on spatial scales of individual galaxies across the full system , with a sensitivity extending down to moderate - luminosity galaxies . exploring the role of hot gas stripping , whether induced by ram pressure or galaxy galaxy interactions , is currently largely limited to early - type galaxies or starbursting spirals , however , since only these galaxy types generally contain x - ray detectable halo gas @xcite . in more quiescent late - type spirals , gas stripping can be far more efficiently explored using hi data . combining x - ray and hi data therefore allows one to study the ism over the full galaxy morphological range . here we combine x - ray mosaicing observations from _ chandra _ with analogous hi observations from the very large array ( vla ) to study the hot and cold ism in the nearby , x - ray bright ngc2563 group . our goal is to search for evidence of ongoing or recent gas stripping ( we here use this term to mean any externally induced removal of gas from galaxies , including via tidal stripping ) , explore possible mechanisms involved , and understand their environmental dependence . while x - ray @xcite and optical @xcite observations of this group already exist , a unique aspect of the present study is that both our x - ray and hi observations cover the entire group ( out to a projected radius of @xmath5 mpc , well beyond the estimated virial radius ) , while still benefiting from the sub - kpc spatial resolution of _ chandra _ at the target redshift . combined with the sensitivity of the vla to low surface brightness diffuse hi emission ( with a synthesized beam width of @xmath6 in the most compact `` d - array '' conguration employed here ) , this enables a detailed investigation of the spatial variation in global ism properties across the full group environment . we describe the overall properties of the group in section [ sec , group ] . the x - ray and hi analyses are detailed in section [ sec , analysis ] , and results are presented in section [ sec , results ] . section [ sec , discuss ] considers the evidence for recent and past ism stripping in the group , and our results and conclusions are summarized in section [ sec , summary ] . we assume @xmath7 km s@xmath2 mpc@xmath2 , @xmath8 , and @xmath9 . the target redshift of @xmath10 then corresponds to a luminosity distance of 68.1 mpc , and @xmath11 on the sky to a projected distance of 19.2 kpc . uncertainties are quoted at the @xmath12 level unless otherwise specified . extensive optical spectroscopy of the ngc2563 field exists from the studies of @xcite . the field is also covered by the sloan digital sky survey ( sdss ) , providing images and magnitudes for all group members , and flux - calibrated spectra for most of them . from the measured redshifts , we determined group membership using the rostat package @xcite , considering all galaxies with known recessional velocities within @xmath13 km s@xmath2 of the group mean . we then calculated the biweight estimators of velocity mean and dispersion . objects with velocities beyond @xmath14@xmath15 from the mean were discarded , and the process repeated iteratively until convergence . this technique identified 64 group members within a projected radius of @xmath16 ( @xmath17 mpc ) from the group center , including two newly discovered members from our hi observations ( see section [ sec , hi ] ) . the resulting mean redshift and radial velocity dispersion of the group based on these 64 galaxies are @xmath18 and @xmath19 km s@xmath2 , respectively . based on objects classified as galaxies in the sdss survey , the group membership is 100% ( 98% ) spectroscopically complete down to @xmath20 ( @xmath21 ) within @xmath22 of the group center . group members were morphologically classified independently by jsm and aiz using sdss @xmath23-band images . the two authors classifications agreed within one hubble type in all cases , and our adopted morphologies represent the average of the two results . we use sdss @xmath24-band magnitudes as a measure of the blue luminosity , and , for comparison to previous studies , @xmath25 magnitudes from the 2-micron all sky survey ( 2mass ) as a rough estimate of galaxy stellar mass . the latter magnitudes are available for most of the group members , except for a few galaxies in close pairs . in these cases , the @xmath25 magnitude was estimated from the sdss @xmath23-band magnitude using @xmath23@xmath26 for early - type galaxies , and @xmath23@xmath27 for late - types , based on the average values for the group members with both magnitudes available . the scatter in each case is @xmath280.20 . as none of the galaxies without @xmath25 magnitudes is detected in x - rays or hi , adopting these relationships has minimal impact on our conclusions . magnitudes were converted to solar luminosities assuming @xmath29 and @xmath30 after correcting for galactic extinction . we further adopt @xmath31 @xcite and @xmath32 @xcite . while a detailed x - ray mass analysis of the group is beyond the scope of this paper , having a rough estimate of the group virial radius @xmath33 is instructive for the discussion to follow . this can be obtained from the hot gas temperature of @xmath34 kev derived by @xcite , combined with the scaling relation of @xcite . this would suggest @xmath35 kpc , and hence @xmath36 kpc in the adopted cosmology , assuming a navarro white potential @xcite with a typical concentration parameter of @xmath3710 . here @xmath38 is the radius enclosing a mean density of @xmath39 times the critical value . the distribution of group members in ngc2563 allows most of them to be observed with 14 acis - i pointings : a @xmath40 central mosaic to cover the inner @xmath41 of the group and five outer pointings . a total of 54 of the 64 confirmed group members were observed by _ chandra_. the majority of the remaining galaxies are optically faint and therefore not likely to be strong x - ray sources . the 14 pointings were done between june 2007 and march 2008 for 30 ks each , except for the central pointing which was observed for 50 ks to achieve a similar contrast for galaxies in the x - ray bright central region . all observations were conducted in vfaint telemetry mode . the level 1 event files were processed with ciao v4.0 and caldb v3.4.3 , filtering out bad pixels and applying the latest gain files . events with _ asca _ grades 1 , 5 , and 7 , and a non - zero status flag were discarded . finally , background flares were screened for using the routine lc_clean.sl in ciao , excluding time periods with background count rates greater than @xmath42 of the quiescent rate . flaring was not a problem for any of our observations . source detection was performed using the `` mexican hat '' wavelet source detection algorithm wavdetect in ciao , with scales of 1 , 2 , 4 , 8 , 16 and 32 pixels , using both soft ( 0.32.0 kev ) , hard ( 2.07.0 kev ) , and `` full '' band ( 0.37 kev ) event lists . the detection threshold was set to limit the number of false detections to @xmath43 per ccd . as summarized in table [ tab : x ] , a total of 17 confirmed group members were detected , including the largest elliptical , ngc2563 itself . this galaxy is coincident with the group center as defined by the peak of the diffuse group x - ray emission . as it is not straightforward to cleanly separate the emission of this galaxy from that of the intragroup medium , the central galaxy is treated as a special case for the remainder of the paper . llrrrrrrcccll ngc2563 & e & 0 & 11.43 & 10.54 & 1547 & @xmath44 & @xmath45 & @xmath46 & @xmath47 & @xmath48 & 1 , tp & e + ngc2562 & s0/a & 91 & 11.21 & 10.35 & 50 & @xmath49 & @xmath50 & 0.7@xmath51 & @xmath52 & @xmath53 & 2 , p & e + ngc2560 & s0/a & 218 & 11.11 & 10.21 & 139 & @xmath54 & @xmath55 & 0.7@xmath51 & @xmath56 & @xmath57 & 2 , p & e + cgcg119 - 069 & e & 238 & 10.13 & 9.49 & 9 & @xmath58 & @xmath59 & 0.7@xmath51 & 1.7@xmath51 & @xmath60 & 3 , u & p ? + ugc04344 & sc & 238 & 10.14 & 9.91 & 17 & @xmath61 & @xmath62 & 0.7@xmath51 & 1.7@xmath51 & @xmath63 & 3 , u & e ? , sf + ugc04332 & sapec & 264 & 10.94 & 10.03 & 117 & @xmath64 & @xmath65 & 0.7@xmath51 & @xmath66 & @xmath67 & 2 , p & e , sy2 + ngc2569 & e & 307 & 10.71 & 9.99 & 18 & @xmath68 & @xmath69 & 0.7@xmath51 & 1.7@xmath51 & @xmath70 & 3 , ut & e ? + ugc04329 & scpec & 439 & 10.72 & 10.06 & 57 & @xmath71 & @xmath72 & 0.7@xmath51 & @xmath73 & @xmath74 & 2 , p & e , sf + ic2293 & sbbc & 470 & 10.37 & 9.89 & 9 & @xmath75 & @xmath76 & 0.7@xmath51 & 1.7@xmath51 & @xmath77 & 3 , ut & p ? , sf + 2mj082236 & s0/a & 542 & 9.99 & 9.32 & 13 & @xmath78 & @xmath79 & 0.7@xmath51 & 1.7@xmath51 & @xmath80 & 3 , u & e ? , sf + ngc2557 & sb0 & 567 & 11.15 & 10.29 & 162 & @xmath81 & @xmath82 & @xmath83 & 1.7@xmath51 & @xmath84 & 1 , tp & e , li + ugc04324 & sab & 664 & 10.71 & 9.92 & 35 & @xmath85 & @xmath86 & 0.7@xmath51 & 1.7@xmath51 & @xmath87 & 3 , u & e , sf + ngc2558 & sb & 739 & 11.03 & 10.26 & 45 & @xmath88 & @xmath89 & 0.7@xmath51 & @xmath90 & @xmath91 & 2 , p & e , li + ic2338 & sbcd & 848 & 10.27 & 9.87 & 59 & @xmath92 & @xmath93 & 0.7@xmath51 & 1.7@xmath51 & @xmath94 & 2 , tp & e , sf + ic2339&sbcpec & 858 & 10.48 & 10.02 & 32 & @xmath95 & @xmath96 & 0.7@xmath51 & 1.7@xmath51 & @xmath97 & 1 , tp & e , sf + cgcg119 - 047 & sab & 923 & 10.62 & 10.04 & 26 & @xmath98 & @xmath99 & 0.7@xmath51 & 1.7@xmath51 & @xmath100 & 3 , u & p ? , sf + ic2341 & e / s0 & 929 & 10.91 & 10.19 & 32 & @xmath101 & @xmath102 & 0.7@xmath51 & 1.7@xmath51 & @xmath103 & 3 , u & e , li for all detected group members , source spectra and associated response files were extracted within circular regions extending to where the galaxy surface brightness becomes consistent with the local background level . background spectra were extracted in surrounding annuli , and results were fitted in the 0.37.0 kev band using xspec v.12.3 @xcite . as many of the sources have few counts , the maximum likelihood based c - statistic @xcite was used to determine best - fit parameters . the goodness - of - fit was verified using the @xmath104 statistic for all sources with sufficient photon statistics to allow meaningful constraints from spectra accummulated in bins of at least 20 net counts . the x - ray emission in luminous early - types is generally dominated by that of hot gas and low - mass x - ray binaries ( e.g. , @xcite ) . we modeled any hot gas emission in the group members using the _ mekal _ optically - thin thermal plasma model @xcite in xspec , assuming solar abundance ratios . any x - ray binary / active galactic nucleus ( agn ) component was modeled with a power law . in all fits , the absorbing hydrogen column density was fixed at the galactic value ( @xmath105 @xmath106 ; @xcite ) , since the limited photon statistics generally precluded useful constraints on @xmath107 from spectral fitting itself . following @xcite and @xcite , the metallicity in the _ mekal _ model was fixed at @xmath108 , the mean value found for galactic hot gas in the large cluster galaxy sample of @xcite . this left four free model parameters : gas temperature @xmath109 , power - law index @xmath110 , and two normalizations . however , due to limited number of counts , the @xmath111 uncertainties on @xmath109 and @xmath110 were unconstrained for most of the galaxies . in such cases we first fixed @xmath110 at 1.7 , as is typical for x - ray binary and agn spectra , and , if necessary , also @xmath109 at 0.7 kev . based on the photon statistics and spectral results , the detected galaxies were then divided into three categories , as summarized in col . ( 12 ) of table [ tab : x ] : \(1 ) galaxies for which the normalization of both components was well - constrained at the @xmath112 level ( three galaxies in total ; well - constrained here means that the @xmath112 uncertainties are finite , and that the normalization is non - zero within those uncertainties ) . for the central galaxy ngc2563 , all of the four parameters are well constrained , but its emission is not easily deblended from that of the intragroup medium . the other two galaxies are ngc2557 , where @xmath110 was fixed , and ic2339 , where also @xmath109 was fixed . best - fit values and @xmath12 errors for all fit parameters were determined using markov chain monte carlo ( mcmc ) simulations in xspec . the values reported were obtained from the median value of the chain . we note for completeness that the standard deviation leading to a parameter being here considered well - constrained ( which is based on the change in the adopted fit statistic when the parameter is varied , using xspec s standard `` error '' command ) is not necessarily identical to that resulting from the mcmc simulations . \(2 ) galaxies with @xmath113 counts but for which the normalization of one or both components remained unconstrained within the @xmath112 uncertainties ( six galaxies in total ) . in the cases where the @xmath112 lower limit on the thermal luminosity @xmath114 was consistent with zero , whereas that on the power - law component was not , the spectrum was assumed to be consistent with a power - law only . ic2338 is consistent with displaying both a thermal and a power - law component , but , unlike galaxies in category ( 1 ) , its @xmath112 upper limit on @xmath114 remained unconstrained . using mcmc , we report the corresponding @xmath12 upper limit . for the remaining galaxies in this category , the best - fit @xmath110 was consistent with @xmath115 as expected from x - ray binaries or an agn , whereas the best - fit temperature was much higher ( @xmath116 kev ) than expected for hot galactic gas . hence , a power - law component likely dominates their x - ray output . for these sources , we fixed the best - fit power - law model and added a @xmath117 kev thermal component . the limit on @xmath114 was determined from the @xmath12 upper limit of the normalization of the _ mekal _ model . \(3 ) galaxies detected with @xmath118 net counts ( eight in total ) . in these cases , statistics are insufficient to allow a robust determination of the nature of their x - ray emission . a @xmath12 upper limit to their @xmath114 was estimated assuming a @xmath117 kev _ mekal _ model , but these galaxies are labeled with u " ( for unknown spectral composition ) in col . ( 12 ) of table [ tab : x ] . to help clarify the nature of these sources , we also considered their hardness ratios @xmath119 ( here the 27 kev/0.32 kev flux ratio ) . this ratio is 0.020.05 for a _ mekal _ model with @xmath1171.0 kev , and 0.350.47 for a power - law spectrum with @xmath1202.0 . for @xmath121 , the dominant component is ambiguous , but some contribution from a thermal component is allowed . although these classifications are only tentative , given the poisson uncertainties on @xmath119 , we do note that all group members spectroscopically confirmed to require only a power - law show @xmath122 , whereas @xmath123 for the galaxies with a spectroscopically identified thermal component . in table [ tab : x ] , a t " is listed after u " in col . ( 12 ) if @xmath124 , to indicate that the galaxy may contain thermal emission . the remaining galaxies are labeled with either tp " , meaning that both thermal and power - law components are likely present ( @xmath121 ) , or p " , indicating consistency with a power - law only ( @xmath125 ) . to further aid in identifying the nature of galactic x - ray emission , we used its spatial extent to help discriminate between nuclear ( agn ) and galaxy - wide emission ( e.g. , hot gas halos , x - ray binaries ) . for each detected group member , we fitted the 0.52 kev brightness profile with a gaussian , and classified it as extended if the @xmath12 lower bound of its full width at half - maximum ( fwhm ) was @xmath126% larger than that of the local point spread function ( psf ) extracted at peak source energy . overall , most of the detected group members appear extended in x - rays , and these are labeled with an `` e '' in col . ( 13 ) of table [ tab : x ] , or with a `` p '' ( point - like ) otherwise . for most sources with @xmath118 counts the extent remains poorly constrained , however , and col . ( 13 ) then includes a question mark . for the x - ray undetected group members , we followed the procedure of @xcite and @xcite to estimate luminosity upper limits . count rates were measured within a circular aperture of @xmath127 kpc from the optical galaxy center , in a few cases expanded to include the @xmath128 encircled energy radius of the local psf at @xmath129 kev to account for psf smearing . the upper limit on @xmath130 was derived from the poisson @xmath111 upper limit on the count rate @xcite assuming a @xmath117 kev _ mekal _ model . the hi analysis was based on two different observing runs at the vla in its most compact , 1 km ( d - array ) , configuration . the first run , done in 1999 , covered the entire group with a @xmath131 point mosaic . individual pointings were separated by @xmath132 , fully sampling the primary beam with its fwhm of @xmath133 . to probe as wide a velocity range as possible with sufficient velocity resolution , we used a total bandwidth of 3.125 mhz and four intermediate frequency ( if ) channels with two frequency settings and slightly overlapping bands . the total velocity range covered was @xmath281100 km s@xmath2 with a resolution of 20 km s@xmath2 . the data were calibrated using standard aips procedures , and then imported into miriad for mosaicing and joint deconvolution , turning the individual pointings into one mosaiced cube for each if . the continuum was then subtracted from the @xmath134@xmath135 data using a fit through the line - free channels identified within the cube . however , inspection of the final cube showed that our velocity coverage was insufficient , with several group members detected close to the edge of the band and their h i velocity range only partly probed . to remedy this , ngc2563 was reobserved in 2007 , using a 14-point mosaic that covered an area identical to that of our _ chandra _ mosaic . to also cover a larger velocity range , we used a total bandwidth of 6.25 mhz with two ifs and no online hanning smoothing . each pointing was observed at two separate frequencies , resulting in a total velocity coverage of 2100 km s@xmath2 . each pointing and velocity setting was calibrated separately , and cubes were made using aips . overlapping frequencies were then combined in the image plane , resulting in a 101channel cube per pointing . after identifying channels containing hi , the continuum was iteratively subtracted from each cube using a linear fit in the image plane through the line free channels . the resulting cubes were then cleaned , combined into one , and the output corrected for the primary beam response . where cubes overlapped , they were averaged with weighting that accounted for the location within the pointing and for the distance from the center of each field . although interference generated by the vla rendered data unusable within a narrow velocity range ( 43314374 km s@xmath2 ) , the only group member affected is an sm dwarf with @xmath136 , so this has little impact on our results . instrumental parameters and specifics of the cubes for both data sets are summarized in table [ tab , vla ] , including their velocity ranges ( heliocentric , optical definition ) . since complete velocity coverage is important here , the 36-point mosaic was used exclusively to search for hi emission , combining it with the central part of the 14-point mosaic . all remaining hi analysis employed the 14-point mosaic only . this was searched for neutral hydrogen using three different methods . first the 101channel cube was divided into groups of 25 , smoothed spatially to half resolution , and hanning smoothed in velocity . this was then used as a mask for the full - resolution cube , blanking all pixels below @xmath112 in the smoothed cube and summing over 25 channels . next we searched the cube by eye , stepping through the velocity channels at different speeds and comparing our detections with the first method . finally , we plotted spectra integrated over a few beams at the location of the optically identified group members . this did not give any additional detections , so our search was essentially an optically blind one . .parameters of the vla d - array observations[tab , vla ] [ cols="<,^,^",options="header " , ] hi is detected only in galaxies of types sa and later , including in all 15 such galaxies with @xmath137 . figure [ fig , hi_mosaic ] shows contours from our noise free h i mosaic overlayed on the _ chandra _ mosaic of the full group , and figure [ fig : sdss_hi ] displays an hi / optical overlay for each hi detected galaxy . the latter figure reveals four significant hi extensions in the group associated with six galaxies ( ic2238/ic2339 , ugc04324/cgcg119 - 040 , cgcg119 - 047 and cgcg119 - 051 , all labeled with an asterisk in the figure ) . this implies that @xmath138% of our hi detected group members display h i evidence for being involved in an ongoing interaction . the galaxies ic2338 and ic2339 represent a particularly close pair , and the hi quantities quoted in table [ tab : hi ] therefore apply for the entire system . the absence of detectable hi among the early - type group members is consistent with previous studies finding hi above our detection limits in only a small fraction of early - types @xcite . since we have performed an optically blind search for hi across the entire group , we can furthermore rule out the existence of any optically dark hi clouds not associated with any galaxy down to our hi mass limit . to investigate to what extent the group members may be deficient in hi , we estimated hi deficiencies def@xmath139 following the usual prescription , @xmath140 the expected hi masses were computed from the best - fit relationship between total hi mass and galaxy optical diameter @xmath141 derived for galaxies in the field @xcite . @xmath141 is here defined as the galaxy major axis measured from the palomar observatory sky survey blue prints . where not already available from the uppsala general catalog of galaxies @xcite , we measured @xmath141 directly from these prints . as the @xcite field sample was restricted to rather large spirals of types sa through sc , only group members of those types and with @xmath142 kpc were considered here . note , as mentioned above , that this includes three galaxies not detected in hi , and these are also listed in table [ tab : hi ] . for the ic2338/2339 pair , a single deficiency was calculated from the total observed and expected hi masses given the optical sizes of both galaxies . the resulting deficiencies listed in table [ tab : hi ] span a wide range , but given the uncertainties in @xmath141 and optical morphology , as well as the dispersion in the hi masses of field galaxies , objects with observed hi masses within a factor of a few of the expected values can not be considered deficient . however , there are six galaxies with def@xmath143 , consistent with an hi deficiency of at least a factor of three . one of these is ugc04386 , but def@xmath139 for this object is not well - established , in part because it resides at the eastern edge of the hi mosaic where the data are noisier , and in part because it is viewed edge - on and is of uncertain morphological type . although our main goal is to study the ism in individual group members , it is also relevant to establish which galaxies are embedded within detectable intragroup gas . to do so , a radial surface brightness profile was extracted from the unsmoothed exposure - corrected _ chandra _ mosaic , centered on ngc2563 itself , and with all other bright sources masked out . the background level was estimated from the four outermost pointings , which are all centered at @xmath144@xmath145 from the x - ray peak and show consistent full - chip count rates . the resulting background - subtracted profile , shown in figure [ fig : surfbright ] , reveals diffuse emission out to at least @xmath146 ( @xmath147 kpc in projection ) . this is a few arcminutes further than the maximum extent determined from _ rosat _ observations @xcite , and roughly corresponds to our estimate of @xmath148 kpc . modeling the profile as a sum of two @xmath149models yields @xmath150 , @xmath151 kpc , and @xmath152 , @xmath153 kpc , for the inner and outer component , respectively . with a reduced @xmath154 , the fit is poor however , in part because the profile steepens beyond that of a @xmath149model at large radii , as also seen in other groups and clusters @xcite . the relatively high central surface brightness could potentially compromise our ability to detect individual galaxies in the group core . we have attempted to circumvent this issue by requiring a longer central _ chandra _ exposure ( 50 ks ) . in addition , the surface brightness has dropped from its central value by three orders of magnitude already at @xmath155 kpc . apart from the central galaxy , there are no optically luminous group members within the central few arcmin , so our x - ray detection limits are impacted minimally by the presence of intragroup gas . this is corroborated by the fact that we find no systematic radial variation in limiting @xmath114 for the 54 group members covered by _ chandra_. a detailed analysis of the intragroup medium in ngc2563 will be the subject of future work . with our hi and x - ray results extending well beyond the group virial radius , we can now explore how the ism properties of individual galaxies vary with position across the entire group , and in particular to what extent there is evidence for ongoing or recent ism stripping from the group members . the detection of x - ray tails with _ chandra _ has provided direct evidence for ram pressure stripping in both groups @xcite and clusters @xcite , and similar inferences have been made from the presence of hi tails within these systems @xcite . in x - rays , however , the detection rate of these features is generally very small ( @xmath156 ) , suggesting that strong stripping of hot ism is either very rare or proceeds very rapidly . no evidence for such activity in the form of x - ray tails is seen within ngc2563 , but this is not surprising given the low detection rate of thermal coronae within the sample . in hi , we identify two tails associated with the relatively isolated cgcg119 - 047 and cgcg119 - 051 , both with hi extending beyond the optical disk in the direction opposite of the group center ( figure [ fig : sdss_hi ] ) . neither of these objects is hi deficient , so any stripping activity must have recently commenced . their respective radial velocities relative to the group mean of 200 and 320 km s@xmath2 imply mach numbers of @xmath157 and @xmath158 for an ambient gas temperature of @xmath159 kev . although both galaxies are currently beyond the radius to which intragroup gas is detected ( @xmath160 and @xmath161 kpc ) , their hi morphologies are suggestive of ongoing ram pressure ( or viscous ) stripping . multi - wavelength observations and higher - resolution hi data would be required to confirm this ( e.g. , @xcite ) . another relatively isolated spiral , ic2293 at @xmath162 kpc , displays no hi tail but is highly deficient ( def@xmath163 ) , and is also a candidate for recent ram pressure stripping . to test for the _ global _ importance of ram pressure stripping within the group , we next consider the radial distribution of ism detections . to put all galaxies on an equal footing , their thermal x - ray luminosities and hi masses were normalized by the galaxy @xmath164 . excluding again the central galaxy ngc2563 , evidence for hot gas is seen in only five of the group members , and figure [ fig , lxlk ] shows that all of these reside relatively far from the group center , with four out of five located beyond the radius @xmath165 kpc to which intragroup gas is detected . in contrast , no hot ism component is found within @xmath166 kpc , where several of the optically luminous early - types reside . this includes ngc2562 at @xmath167 kpc , the closest bright galaxy to the group center , with an upper limit to @xmath114 which is an order of magnitude below the value of log@xmath168 suggested by the @xmath164@xmath114 relation for early - types in groups @xcite . to further illustrate this point , we note that , given our x - ray detection limits ( section [ sec , x ] ) , the @xcite relation would suggest that we should detect all early - type members brighter than log@xmath169 , irrespective of their hot gas content . in practice , only 8/14 ( @xmath170% ) are detected , even if including ngc2563 itself and excluding sa galaxies . the bottom panel in figure [ fig , lxlk ] shows the cumulative fraction of group members covered by _ chandra _ above this @xmath164 that is x - ray detected , and the fraction that also contains evidence for a thermal component . while the former remains constant with @xmath171 , the thermal fraction drops toward the group core . although subject to large poisson errors , this result is consistent with hot ism having been stripped within the dense group core , while the more distant galaxies retaining a thermal component have yet to experience peak ram pressure during their orbit ( but see also @xcite ) . in contrast , galaxies with a detectable _ cold _ ism component are distributed more evenly with @xmath171 . this is illustrated in figure [ fig , mhi_lk](a ) , which compares our measured hi masses to those of similar galaxies in the field . a handful of galaxies show evidence of a significant shortfall in hi content ( i.e. def@xmath172 ) , but the values scatter broadly around def@xmath173 and show no systematic dependence on @xmath171 . a kolmogorov test was performed to test whether the inferred deficiencies are , in fact , consistent with being drawn from a gaussian parent distribution centered at def@xmath174 with some width @xmath175 . we find that for @xmath175 in the range 0.30.5 , this probability is at least @xmath176% , regardless of whether the upper / lower limits and the highly deficient cgcg119 - 040 are included . there is thus no strong evidence from this that the large sa sc group members are globally hi deficient . however , we present evidence in section [ sec , glob ] that the late - type population as a whole _ is _ deficient relative to the field . the estimated hi deficiencies depend on assumed galaxy morphology and the somewhat uncertain comparison values for field galaxies . a more robust measure of the relative hi content within the group members might be provided by @xmath177 . the dependence of this quantity on @xmath171 is shown in figure [ fig , mhi_lk](b ) . note that at fixed @xmath171 , lower - mass galaxies would be more easily ram pressure stripped , but they are also likely to be more hi rich to begin with . to take this into account , the galaxies were divided into three bins of comparable @xmath164 . figure [ fig , mhi_lk](b ) confirms that even at `` fixed '' @xmath164 , there is no systematic radial dependence of hi content within the group , and hence that ram pressure stripping of cold ism is unlikely to be globally important in this system . the presence of hi tails associated with close galaxy pairs is commonly taken as a sign of tidal encounters @xcite . ngc2563 contains two galaxy pairs with clear hi extensions indicating such encounters . the most prominent of these is associated with ic2338 and ic2339 , a spiral pair separated by only @xmath178 kpc in projection and @xmath179 km s@xmath2 in velocity . the sdss images show evidence for a stellar bridge between the galaxies , supporting a tidal origin for the hi tails . an hi extension is also connecting ugc04324 with cgcg119 - 040 ; the latter represents the most hi deficient group member ( deficient by a factor of @xmath180 ) , with all its hi associated with the h i bridge . while no corresponding stellar feature is seen in the relatively shallow sdss images , the hi morphology is strongly suggestive of a tidal encounter . to explore whether recent removal of cold ism through tidal interactions can generally explain the observed hi deficiencies within the group , we show in figure [ fig : hidef](a ) the hi deficiencies as a function of projected distance @xmath181 to the nearest neighbor . if excluding the closely interacting ic2338/2339 pair and the upper / lower limits , a kendall correlation test suggests an anticorrelation at the @xmath182 level . including the upper / lower limits at their nominal values strengthens the correlation significance to @xmath183 . the correlation with @xmath181 remains significant at the @xmath112 level when instead considering the normalized hi masses in figure [ fig : hidef](b ) . while only indicative , this result is consistent with some hi having been removed in galaxy galaxy interactions within the group . since strong tidal interactions require proximity of two galaxies in both position and velocity space , a plot of hi content against @xmath181 may be subject to large scatter due to some galaxies being close in projection only . for each galaxy , we therefore also consider the minimum value of @xmath184 where @xmath185 and @xmath186 are the separations in projection and radial velocity between _ any _ two group members , and @xmath187 mpc and @xmath188 km s@xmath2 are the corresponding maximum values between all galaxies in ngc2563 . strong interactions would require @xmath189 , and , if generally prominent in removing hi , would imply a correlation between hi content and @xmath190 for small values of the latter . for ngc2563 , one might expect interactions to be important up to at least @xmath191 , as this would correspond to two galaxies at the same @xmath192 separated by @xmath193 kpc , or to a physically `` overlapping '' pair with @xmath194 km s@xmath2 . as suggested by figure [ fig : hidef](c ) and ( d ) , there is some evidence of such a correlation for @xmath195 among the individually hi detected galaxies . however , this is significant at less than @xmath112 , so we can only conclude that our results are at least consistent with some hi removal due to galaxy galaxy interactions . repeating this analysis for the x - ray detected group members reveals no indication of a systematic trend in @xmath196 with either @xmath181 or @xmath190 . however , many of these galaxies have x - ray luminosities only slightly above our completeness limit , so deeper observations of a larger sample would be required to confirm this result within groups in general . our comprehensive vla coverage of ngc2563 enables a census of the global amount and distribution of hi in the group , and provides further evidence of recent hi mass loss from the group members . based on measurements from the hi parkes all sky survey ( hipass ) , @xcite have quantified the typical hi mass for late - type galaxies ( sb and later ) in the general `` field '' with @xmath197 : @xmath198 , \label{eq , hi}\ ] ] where @xmath199 and @xmath200 . to compare this to results for ngc2563 , we used estimates of @xmath201 for the group members from sdss , available for all but six of the 64 members ; for those six , we assume a @xmath202band stellar mass - to - light ratio of 0.55 , which is the average for the members with both @xmath164 and @xmath201 available . equation ( [ eq , hi ] ) would then suggest a total hi mass of the relevant group members of @xmath203 , whereas the value inferred from table [ tab : hi ] is @xmath204 , when including the average @xmath111 upper limit of @xmath205 for the undetected members . this suggests that globally , the @xmath206 late - types in the group have suffered mild hi mass loss with respect to similar field galaxies . compared to estimates of def@xmath139 , this result may represent a more robust estimate of global hi deficiency within the group , as it includes also smaller galaxies , depends less sensitively on the inferred morphology and optical size of individual group members , and takes the dispersion seen for field galaxies into account . we also derive a total stellar mass in the group of @xmath207 for all the 64 spectroscopically identified members , and a corresponding total range of @xmath208 . this implies @xmath2090.062 for the full group ( down to our 98% spectroscopic completeness limit of @xmath210 ) , under the plausible assumption that our vla observations are not missing any large - scale diffuse hi . future comparison of this result for an x - ray bright group to other groups and clusters could provide further information on the nature of gas removal in these environments . on the largest scales , the distribution of hi detected galaxies in figure [ fig , hi_mosaic ] reveals a peculiar lopsidedness , with 16 of the 20 detections occurring on the western side of the group . this is not simply due to asymmetric spatial vla coverage , as demonstrated in figure [ fig : hichannel ] , nor to a larger fraction of hi deficient galaxies in the east . the wide velocity coverage of our vla data ( @xmath211 ) further renders it unlikely that a significant fraction of potential group members have been missed on the eastern side of the system . the lopsided hi distribution thus reflects a real asymmetry in the distribution of late - type members , with nearly all the large spirals located in the western half of the group . the origin of this asymmetry remains unclear . one possibility is that the western spirals are part of an infalling subgroup , but echoing an earlier conclusion by @xcite , we find no evidence for kinematic substructure in the group using the technique pioneered by @xcite . another possible scenario is that galaxy morphologies have been modified by strong interactions with a similarly asymmetric intragroup medium . this can be ruled out , however , because the intragroup gas distribution in ngc2563 appears fairly symmetric on large scales and , if anything , seems more strongly extended to the west ( see figure [ fig : mosaic ] and @xcite ) . finally , exploiting the extensive sdss coverage in this field , we have also examined the overall galaxy distribution within a projected distance of @xmath212 mpc of the group center , and find no evidence for a general large - scale enhancement in galaxy density on the western side of the group . further studies of other groups on very large scales may help establish how common such morphological segregations are in these systems . our extensive _ chandra _ and vla coverage of the ngc2563 galaxy group has enabled us to probe both the hot and cold ism within a nearby group out to @xmath213 times the estimated virial radius . the main aim has been to characterize the evidence for galactic gas removal in this system , and understand the mechanisms involved . although the limited number of galaxies preclude strong conclusions based on a single group , our results suggest that both hot and cold gas have been stripped from some of the group members , and that both ram pressure stripping and gas removal via galaxy galaxy interactions are all occurring simultaneously within this one system . this is based on the following evidence : \(i ) the thermal x - ray deficiency of optically luminous early - type galaxies in the group core suggests recent ram pressure stripping of their hot gas . the central group galaxy aside , no hot ism is detected within @xmath166 kpc from the group core , whereas four out of the five non - central galaxies with such a component reside beyond the radius to which intragroup gas is detected . the thermal luminosity of the bright early - type closest to the group core ( ngc2562 ) is at least an order of magnitude below that expected for typical group galaxies of the relevant @xmath164 @xcite . \(ii ) comparison of the hi content of the late - type group members to that of similar galaxies in the field suggests that the former are , on average , mildly deficient in hi . this points to one or more mechanisms removing cold ism within the group . \(iii ) ram pressure ( or viscous ) stripping of cold gas is suspected in a few cases . the hi data reveal one relatively isolated spiral ( ic2293 ) which is hi deficient by more than a factor of three . another two such spirals show hi morphologies suggestive of an ongoing ram pressure interaction , with hi tails pointing away from the group core ( cgcg119 - 047 and 119 - 051 in figure [ fig : sdss_hi ] ) . however , these are not ( yet ) hi deficient by the usual definition , indicating that significant interactions with the intragroup medium may have only recently commenced . \(iv ) ongoing galaxy galaxy interactions removing hi are also strongly suggested in at least two cases . the two most prominent h i tails / extensions in the group occur within close galaxy pairs , and the most hi deficient group member ( cgcg119 - 040 , deficient by a factor of @xmath180 ) is a member of one of these . suggestive evidence is further seen for galaxies with close neighbors in position velocity space to show relatively low hi content , consistent with tidal stripping of hi . the inference that ram pressure stripping of hot galactic gas may have occurred in the central group regions would be in line with other _ chandra _ studies of group and cluster galaxies @xcite , and with simulations which suggest that such stripping can be efficient even in small galaxy groups @xcite . however , it is worth noting that a large fraction of early - type galaxies generally do retain halos even in x - ray bright groups , including some in ngc2563 , and that such halos are not generally underluminous compared to those of galaxies in the field @xcite . despite the detection of a hot intragroup medium within ngc2563 out to @xmath214 , there is also no global evidence for ram pressure stripping of the _ cold _ ism within the group . specifically , no radial trends are seen in the stellar mass normalized hi content among the 20 hi detected group members . the indicative result that ram pressure may affect only the hot ism in typical group galaxies is consistent with existing numerical studies @xcite . galaxy interactions remain the main candidate for removing cold ism within ngc2563 and explaining the global hi deficiency inferred for the late - type group members . other studies have also implied that tidal encounters have an important impact on the hi properties of galaxies in groups @xcite . in addition , such encounters may work to enhance the susceptibility of a galaxy to ram pressure stripping , by perturbing the distribution of the cold gas @xcite . nevertheless , despite the detection of two new group members from our optically blind hi search , we find no evidence within the ngc2563 group for isolated , optically dark h i clouds that might represent previously removed material . given the small number of luminous galaxies in a single group , similar studies of a larger group sample will be required to better understand how the group environment impacts galaxy evolution . for example , we note that the inferred hi - to - stellar mass ratio of @xmath215 inferred for this x - ray bright group may provide a useful benchmark for the hi content of dynamically evolved groups . future comparison of this result to those of x - ray faint systems , richer clusters , and groups at higher redshift ( of which the upcoming square kilometre array should detect many thousands ) could further improve our understanding of the nature and epoch of gas removal in dense environments . such studies would also help to address the commonality of the highly lopsided distribution of hi seen in ngc2563 , which reflects a puzzling galaxy morphological segregation within the group . we are grateful to the referee for a careful and constructive report which significantly improved the presentation of our results . we thank christy tremonti for providing the emission - line ratios for the sdss spectra . this research has made use of the nasa / ipac extragalactic database ( ned ) . jr acknowledges support by the carlsberg foundation . xb thanks prof . zhang , s. nan , and r. fengyun from tsinghua center for astrophysics ( thca ) for discussions on _ chandra _ analysis . support for this work was provided by the national science foundation under grant number 0607643 to columbia university , and by the national aeronautics and space administration through chandra award number g07 - 8134x issued by the chandra x - ray observatory center , which is operated by the smithsonian astrophysical observatory for and on behalf of the national aeronautics space administration under contract nas8 - 03060 . arnaud , k. a. 1996 , in asp conf . ser . 101 , astronomical data analysis software and systems v , ed . g. h. jacoby & j. barnes ( san francisco , ca : asp ) , 17 bailin , j. , & ford , a. 2007 , , 375 , l41 baldry , i. k. , balogh , m. l. , bower , r. g. , glazebrook , k. , nichol , r. c. , bamford , s. p. , & budavari , t. 2006 , , 373 , 469 baldwin , j. a. , phillips , m. m. , & terlevich , r. 1981 , , 93 , 5 balogh , m. l. , navarro , j. f. , & morris , s. l. 2000 , , 540 , 113 barnes , j. e. 1989 , , 338 , 123 beers , t. c. , flynn , k. , & gebhardt , k. 1990 , , 100 , 32 bureau , m. , & carignan , c. 2002 , , 123 , 1316 burstein , d. , krumm , n. , & salpeter , e. e. 1987 , , 94 , 883 cash , w. 1979 , , 228 , 939 cayatte , v. , kotanyi , c. , balkowski , c. , & van gorkom , j. h. 1994 , , 107 , 1003 chamaraux , p. , balkowski , c. , & gerard , e. 1980 , , 83 , 38 chung , a. , van gorkom , j. h. , kenney , j. d. p. , & vollmer , b. 2007 , , 659 , l115 chung , a. , van gorkom , j. h. , kenney , j. d. p. , crowl , h. , & vollmer , b. 2009 , , 138 , 1741 cortese , l. , catinella , b. , boissier , s. , boselli , a. , & heinis , s. 2011 , , 415 , 1797 crowl , h. h. , kenney , j. d. p. , van gorkom , j. h. , & vollmer , b. 2005 , , 130 , 65 cucciati , o. , et al . 2010 , , 524 , a2 davis , d. s. , keel , w. c. , mulchaey , j. s. , & henning , p. a. 1997 , , 114 , 613 de mello , d. f. , smith , l. j. , sabbi , e. , gallagher , j. s. , mountain , m. , & harbeck , d. r. 2008 , , 135 , 548 di serego alighieri , s. , et al . 2007 , , 474 , 851 dressler , a. , & shectman , s. a. 1988 , , 95 , 985 evoli , c. , salucci , p. , lapi , a. , & danese , l. 2011 , , 743 , 45 fabbiano , g. 2006 , , 44 , 323 fabbiano , g. , kim , d .- w . , & trinchieri , g. 1992 , , 80 , 531 finoguenov , a. , ponman , t. j. , osmond , j. p. f. , & zimer , m. 2007 , , 374 , 737 gastaldello , f. , buote , d. a. , humphrey , p. j. , zappacosta , l. , bullock , j. s. , brighenti , f. , & mathews , w. g. 2007 , , 669 , 158 gehrels , n. 1986 , , 303 , 336 jeltema , t. e. , binder , b. , & mulchaey , j. s. 2008 , , 679 , 1162 kalberla , p. m. w. , burton , w. b. , hartmann , d. , arnal , e. m. , bajaja , e. , morras , r. , poppel , w. g. l. 2005 , , 440 , 775 kantharia , n. g. , ananthakrishnan , s. , nityananda , r. , & hota , a. 2005 , , 435 , 483 kauffmann , g. , et al . 2003a , , 341 , 54 kauffmann , g. , et al . 2003b , , 346 , 1055 kauffmann , g. , white , s. d. m. , heckman , t. m. , mnard , b. , brinchmann , j. , charlot , s. , tremonti , c. , & brinkmann , j. 2004 , , 353 , 713 kawata , d. , & mulchaey , j. s. 2008 , , 672 , l103 kenney , j. d. p. , van gorkom , j. h. , & vollmer , b. 2004 , , 127 , 3361 kern , k. m. , kilborn , v. a. , forbes , d. a. , & koribalski , b. 2008 , , 384 , 305 kewley , l. j. , dopita , m. a. , sutherland , r. s. , heisler , c. a. , & trevena , j. 2001 , , 556 , 121 kilborn , v. a. , forbes , d. a. , koribalski , b. s. , brough , s. , & kern , k. 2006 , , 371 , 739 kilborn , v. a. , forbes , d. a. , barnes , d. g. , koribalski , b. s. , brough , s. , & kern , k. 2009 , , 400 , 1962 kim , d .- w . , & fabbiano , g. 2004 , , 611 , 846 koopmann , r. a. , et al . 2008 , , 682 , l85 levy , l. , rose , j. a. , van gorkom , j. h. , & chaboyer , b. 2007 , , 133 , 1104 liedahl , d. a. , osterheld , a. l. , & goldstein , w. h. 1995 , , 438 , l115 mcconnachie , a. w. , venn , k. a. , irwin , m. j. , young , l. m. , & geehan , j. j. 2007 , , 671 , l33 mamon , g. a. 2007 , in eso symp . , groups of galaxies in the nearby universe , ed . i. saviane , v. ivanov , & j. borissova ( berlin : springer ) , 203 mayer , l. , mastropietro , c. , wadsley , j. , stadel , j. , & moore , b. 2006 , , 369 , 1021 mewe , r. , gronenschild , e. h. b. m. , & van den oord , g. h. j. 1985 , , 62 , 197 mihos , j. c. , & hernquist , l. 1996 , , 464 , 641 miller , m. c. , & colbert , e. j. m. 2004 , international journal of modern physics d , 13 , 1 montero - dorta , a. d. , & prada , f. 2009 , , 399 , 1106 moore , b. , lake , g. , quinn , t. , & stadel , j. 1999 , , 304 , 465 morganti , r. , et al . 2006 , , 371 , 157 mulchaey , j. s. , & jeltema , t. e. 2010 , , 715 , l1 mulchaey , j. s. , & zabludoff , a. i. 1998 , , 496 , 73 mulchaey , j. s. , davis , d. s. , mushotzky , r. f. , & burstein , d. 2003 , , 145 , 39 murphy , e. j. , kenney , j. d. p. , helou , g. , chung , a. , & howell , j. h. 2009 , , 694 , 1435 navarro , j. f. , frenk , c. s. , & white , s. d. m. 1997 , , 490 , 493 nilson , p. 1973 , acta universitatis upsaliensis . nova acta regiae societatis scientiarum upsaliensis - uppsala astronomiska observatoriums annaler , uppsala : astronomiska observatorium , 1973 nulsen , p. e. j. 1982 , , 198 , 1007 osmond , j. p. f. , & ponman , t. j. 2004 , , 350 , 1511 osullivan , e. , forbes , d. a. , & ponman , t. j. 2001 , , 328 , 461 osullivan , e. , ponman , t. j. , & collins , r. s. 2003 , , 340 , 1375 peng , y .- j . , et al . 2010 , , 721 , 193 quilis , v. , moore , b. , & bower , r. 2000 , science , 288 , 1617 rasmussen , j. , ponman , t. j. , & mulchaey , j. s. 2006 , , 370 , 453 rasmussen , j. , ponman , t. j. , verdes - montenegro , l. , yun , m. s. , & borthakur , s. 2008 , , 388 , 1245 rasmussen , j. , sommer - larsen , j. , pedersen , k. , toft , s. , benson , a. , bower , r. g. , & grove , l. f. 2009 , , 697 , 79 read , a. m. , & ponman , t. j. 2001 , , 328 , 127 roberts , t. p. 2007 , , 311 , 203 schrder , a. , drinkwater , m. j. , & richter , o .- g . 2001 , , 376 , 98 sengupta , c. , balasubramanyam , r. , & dwarakanath , k. s. 2007 , , 378 , 137 solanes , j. m. , giovanelli , r. , & haynes , m. p. 1996 , , 461 , 609 solanes , j. m. , manrique , a. , garca - gmez , c. , gonzlez - casado , g. , giovanelli , r. , & haynes , m. p. 2001 , , 548 , 97 sun , m. , jones , c. , forman , w. , vikhlinin , a. , donahue , m. , & voit , m. 2007 , , 657 , 197 sun , m. , voit , g. m. , donahue , m. , jones , c. , forman , w. , & vikhlinin , a. 2009 , , 693 , 1142 tremonti , c. a. , et al . 2004 , , 613 , 898 tllmann , r. , breitschwerdt , d. , rossa , j. , pietsch , w. , & dettmar , r .- j . 2006 , , 457 , 779 verdes - montenegro , l. , yun , m. s. , williams , b. a. , huchtmeier , w. k. , del olmo , a. , & perea , j. 2001 , , 377 , 812 vikhlinin , a. , kravtsov , a. , forman , w. , jones , c. , markevitch , m. , murray , s. s. , & van speybroeck , l. 2006 , , 640 , 691 zabludoff , a. i. , & mulchaey , j. s. 1998a , , 496 , 39 zabludoff , a. i. , & mulchaey , j. s. 1998b , , 498 , l5 zabludoff , a. i. , & mulchaey , j. s. 2000 , , 539 , 136

in this paper , we present a low - complexity algorithm for detection in high - rate , non - orthogonal space - time block coded ( stbc ) large - mimo systems that achieve high spectral efficiencies of the order of tens of bps / hz . we also present a training - based iterative detection / channel estimation scheme for such large stbc mimo systems . our simulation results show that excellent bit error rate and nearness - to - capacity performance are achieved by the proposed multistage _ likelihood ascent search _ (-las ) detector in conjunction with the proposed iterative detection / channel estimation scheme at low complexities . the fact that we could show such good results for large stbcs like and stbcs from cyclic division algebras ( cda ) operating at spectral efficiencies in excess of 20 bps / hz ( even after accounting for the overheads meant for pilot based training for channel estimation and turbo coding ) establishes the effectiveness of the proposed detector and channel estimator . we decode perfect codes of large dimensions using the proposed detector . with the feasibility of such a low - complexity detection / channel estimation scheme , large - mimo systems with tens of antennas operating at several tens of bps / hz spectral efficiencies can become practical , enabling interesting high data rate wireless applications . large - mimo systems , low - complexity detection , channel estimation , non - orthogonal space - time block codes , high spectral efficiencies .

0809.2446

current wireless standards ( e.g. , ieee 802.11n and 802.16e ) have adopted mimo techniques @xcite-@xcite to achieve the benefits of transmit diversity ( using space - time coding ) and high data rates ( using spatial multiplexing ) . they , however , harness only a limited potential of mimo benefits since they use only a small number of transmit antennas ( e.g. , 2 to 4 antennas ) . significant benefits can be realized if large number of antennas are used ; e.g. , large - mimo systems with tens of antennas in communication terminals can enable multi - giga bit rate transmissions at high spectral efficiencies of the order of _ several tens of bps / hz_. key challenges in realizing such large - mimo systems include low - complexity detection and channel estimation , rf / if technologies , and placement of large number of antennas in communication terminals . our focus in this paper is on low - complexity detection and channel estimation for large - mimo systems . spatial multiplexing ( v - blast ) with large number of transmit antennas can offer high spectral efficiencies , but it does not give transmit diversity . on the other hand , well known orthogonal space - time block codes ( stbc ) have the advantages of full transmit diversity and low decoding complexity , but they suffer from rate loss for increasing number of transmit antennas @xcite,@xcite,@xcite . however , _ full - rate , non - orthogonal stbcs from cyclic division algebras ( cda ) _ @xcite are attractive to achieve high spectral efficiencies in addition to achieving full transmit diversity , using large number of transmit antennas . for example , a @xmath2 stbc matrix from cda has 1024 symbols ( i.e. , 32 complex symbols per channel use ) , and using this stbc along with 16-qam and rate-3/4 turbo code offers a spectral efficiency of 96 bps / hz . while maximum - likelihood ( ml ) decoding of orthogonal stbcs can be achieved in linear complexity , ml or near - ml decoding of non - orthogonal stbcs with large number of antennas at low complexities has been a challenge . channel estimation is also a key issue in large - mimo systems . in this paper , we address these two challenging problems ; our proposed solutions can potentially enable realization of large - mimo systems in practice . sphere decoding and several of its low - complexity variants are known in the literature @xcite-@xcite . these detectors , however , are prohibitively complex for large number of antennas . recent approaches to low - complexity multiuser / mimo detection involve application of techniques from belief propagation @xcite , markov chain monte - carlo methods @xcite , neural networks @xcite,@xcite,@xcite , etc . in particular , in @xcite,@xcite , we presented a powerful hopfield neural network based low - complexity search algorithm for detecting large - mimo v - blast signals , and showed that it performs quite close to ( within 4.6 db of ) the theoretical capacity , at high spectral efficiencies of the order of tens to hundreds of bps / hz using tens to hundreds of antennas , at an average per - symbol detection complexity of just @xmath3 , where @xmath4 and @xmath5 denote the number of transmit and receive antennas , respectively . in this paper , we present @xmath6 ) a low - complexity near - ml achieving detector , and @xmath7 ) an iterative detection / channel estimation scheme for large non - orthogonal stbc mimo systems having tens of transmit and receive antennas . our key contributions here can be summarized as follows : 1 . we generalize the 1-symbol update based _ likelihood ascent search _ ( las ) algorithm we proposed in @xcite,@xcite , by employing a low - complexity multistage multi - symbol update based strategy ; we refer to this new algorithm as multistage las ( @xmath0-las ) algorithm . we show that the @xmath0-las algorithm outperforms the basic las algorithm with some increase in complexity . we propose a method to generate soft outputs from the @xmath0-las output vector . soft outputs generation was not considered in @xcite,@xcite . the proposed soft outputs generation for the individual bits results in about 1 to 1.5 db improvement in coded bit error rate ( ber ) compared to hard decision @xmath0-las outputs . 3 . assuming i.i.d . fading and perfect channel state information at the receiver ( csir ) , our simulation results show that the proposed @xmath0-las algorithm is able to decode large non - orthogonal stbcs ( e.g. , @xmath1 and @xmath2 stbcs ) and achieve near single - input single - output ( siso ) awgn uncoded ber performance as well as near - capacity ( within 4 db from theoretical capacity ) coded ber performance . 4 . using the proposed detector , we decode and report the simulated ber performance of ` perfect codes ' @xcite-@xcite of large dimensions . 5 . presenting a ber performance and complexity comparison of the proposed cda stbc/@xmath0-las detection approach with other large - mimo / detector approaches ( e.g. , stacked alamouti codes / qostbcs and associated interference canceling receivers reported in @xcite ) , we show that the proposed approach outperforms the other considered approaches , both in terms of performance as well as complexity . we present simulation results that quantify the loss in ber performance due to spatial correlation in large - mimo systems , by considering a more realistic spatially correlated mimo fading channel model proposed by gesbert _ et al _ in @xcite . we show that this loss in performance can be alleviated by providing more receive dimensions ( i.e. , more receive antennas than transmit antennas ) . 7 . finally , we present a training - based iterative detection / channel estimation scheme for large stbc mimo systems . we report ber and nearness - to - capacity results when the channel matrix is estimated using the proposed iterative scheme and compare these results with those obtained using perfect csir assumption . the rest of the paper is organized as follows . in section [ sec2 ] , we present the stbc mimo system model considered . the proposed detection algorithm is presented in section [ sec3 ] . ber performance results with perfect csir are presented in section [ sec4 ] . this section includes the results on the effect of spatial correlation , ber performance of large perfect codes , and comparison of the proposed scheme with other large - mimo architecture / detector combinations . the proposed iterative detection / channel estimation scheme and the corresponding performance results are presented in section [ sec5 ] . conclusions are presented in section [ sec6 ] . consider a stbc mimo system with multiple transmit and multiple receive antennas . an @xmath8 stbc is represented by a matrix @xmath9 , where @xmath10 and @xmath11 denote the number of transmit antennas and number of time slots , respectively , and @xmath12 denotes the number of complex data symbols sent in one stbc matrix . the @xmath13th entry in @xmath14 represents the complex number transmitted from the @xmath6th transmit antenna in the @xmath15th time slot . the rate of an stbc , @xmath16 , is given by @xmath17 . let @xmath5 and @xmath18 denote the number of receive and transmit antennas , respectively . let @xmath19 denote the channel gain matrix , where the @xmath13th entry in @xmath20 is the complex channel gain from the @xmath15th transmit antenna to the @xmath6th receive antenna . we assume that the channel gains remain constant over one stbc matrix duration . assuming rich scattering , we model the entries of @xmath20 as i.i.d @xmath21 denotes a circularly symmetric complex gaussian distribution with mean zero and variance @xmath22 . ] . the received space - time signal matrix , @xmath23 , can be written as @xmath24 where @xmath25 is the noise matrix at the receiver and its entries are modeled as i.i.d @xmath26 , where @xmath27 is the average energy of the transmitted symbols , and @xmath28 is the average received snr per receive antenna @xcite , and the @xmath13th entry in @xmath29 is the received signal at the @xmath6th receive antenna in the @xmath15th time slot . in a linear dispersion ( ld ) stbc , @xmath14 can be decomposed into a linear combination of weight matrices corresponding to each data symbol and its conjugate as @xcite @xmath30 where @xmath31 is the @xmath6th complex data symbol , and @xmath32 are its corresponding weight matrices . the detection algorithm we propose in this paper can decode general ld stbcs of the form in ( [ systemmodelstbclp ] ) . for the purpose of simplicity in exposition , here we consider a subclass of ld stbcs , where @xmath14 can be written in the form @xmath33 from ( [ systemmodel ] ) and ( [ systemmodelstbclpx ] ) , applying the @xmath34 operation matrix @xmath35 $ ] , where @xmath36 is the @xmath6th column of @xmath37 , @xmath38 is a @xmath39 vector defined as @xmath40^t$ ] , where @xmath41^t$ ] denotes the transpose operation . ] we have @xmath42 if @xmath43,@xmath44,@xmath45,@xmath46 are matrices such that @xmath47 , then it is true that @xmath48 , where @xmath49 denotes tensor product of matrices @xcite . using this , we can write ( [ systemmodelvec ] ) as @xmath50 where @xmath51 is the @xmath52 identity matrix . further , define @xmath53 , @xmath54 , @xmath55 , and @xmath56 . from these definitions , it is clear that @xmath57 , @xmath58 , @xmath59 , and @xmath60 . let us also define a matrix @xmath61 , whose @xmath6th column is @xmath62 , @xmath63 . let @xmath64 , whose @xmath6th entry is the data symbol @xmath31 . with these definitions , we can write ( [ systemmodelvec1 ] ) as @xmath65 each element of @xmath66 is an @xmath67-pam or @xmath67-qam symbol . @xmath67-pam symbols take discrete values from @xmath68 , where @xmath69 , and @xmath67-qam is nothing but two pams in quadrature . let @xmath70 , @xmath71 , @xmath66 , and @xmath72 be decomposed into real and imaginary parts as @xmath73 further , we define @xmath74 , @xmath75 , @xmath76 , and @xmath77 as @xmath78^t , & \,\,\,\,\ , { \bf y}_r = [ { \bf y}_i^t \hspace{2 mm } { \bf y}_q^t ] ^t , \nonumber \\ { \bf h}_r = \left(\begin{array}{cc}\widetilde{{\bf h}}_i \hspace{2 mm } -\widetilde{{\bf h}}_q \\ \widetilde{{\bf h}}_q \hspace{5 mm } \widetilde{{\bf h}}_i \end{array}\right ) , & \,\,\,\,\ , { \bf n}_r = [ { \bf n}_i^t \hspace{2 mm } { \bf n}_q^t ] ^t.\end{aligned}\ ] ] now , ( [ systemmodelvec2 ] ) can be written as @xmath79 henceforth , we work with the real - valued system in ( [ systemmodelreal ] ) . for notational simplicity , we drop subscripts @xmath16 in ( [ systemmodelreal ] ) and write @xmath80 where @xmath81 , @xmath82 , @xmath83 , and @xmath84 . the channel coefficients are assumed to be known only at the receiver but not at the transmitter . let @xmath85 denote the @xmath67-pam signal set from which @xmath86 ( @xmath6th entry of @xmath87 ) takes values , @xmath88 . now , define a @xmath89-dimensional signal space @xmath90 to be the cartesian product of @xmath91 to @xmath92 . the ml solution is given by . \hspace{10 mm } ( \mbox{11.a } ) \label{stbc}\ ] ] ] @xmath93 whose complexity is exponential in @xmath12 @xcite . we focus on the detection of square ( i.e. , @xmath94 ) , full - rate ( i.e. , @xmath95 ) , circulant ( where the weight matrices @xmath96 s are permutation type ) , non - orthogonal stbcs from cda @xcite , whose construction for arbitrary number of transmit antennas @xmath10 is given by the matrix in ( 11.a ) given at the bottom of this page @xcite : in ( 11.a ) , @xmath97 , @xmath98 , and @xmath99 , @xmath100 are the data symbols from a qam alphabet . when @xmath101 and @xmath102 , the stbc in ( 11.a ) achieves full transmit diversity ( under ml decoding ) as well as information - losslessness @xcite . when @xmath103 , the code ceases to be of full - diversity ( fd ) , but continues to be information - lossless ( ill ) @xcite,@xcite . high spectral efficiencies with large @xmath10 can be achieved using this code construction . for example , with @xmath104 transmit antennas , the @xmath2 stbc from ( 11.a ) with 16-qam and rate-3/4 turbo code achieves a spectral efficiency of 96 bps / hz . this high spectral efficiency is achieved along with the full - diversity of order @xmath105 . however , since these stbcs are non - orthogonal , ml detection gets increasingly impractical for large @xmath10 . consequently , a key challenge in realizing the benefits of these large stbcs in practice is that of achieving near - ml performance for large @xmath10 at low detection complexities . our proposed detector , termed as the _ multistage likelihood ascent search ( @xmath0-las ) detector _ , presented in the following section essentially addresses this challenging issue . the proposed @xmath0-las algorithm consists of a sequence of likelihood - ascent search stages , where the likelihood increases monotonically with every search stage . each search stage consists of several sub - stages . there can be at most @xmath0 sub - stages , each consisting of one or more iterations ( the first sub - stage can have one or more iterations , whereas all the other sub - stages can have at most one iteration ) . in the first sub - stage , the algorithm updates one symbol per iteration such that the likelihood monotonically increases from one iteration to the next until a local minima is reached . upon reaching this local minima , the algorithm initiates the second sub - stage . in the second sub - stage , a 2-symbol update is tried to further increase the likelihood . if the algorithm succeeds in increasing the likelihood by 2-symbol update , it starts the next search stage . if the algorithm does not succeed in the second sub - stage , it goes to the third sub - stage where a 3-symbol update is tried to further increase the likelihood . essentially , in the @xmath106th sub - stage , a @xmath106-symbol update is tried to further increase the likelihood . this goes on until @xmath107 either the algorithm succeeds in the @xmath106th sub - stage for some @xmath108 ( in which case a new search stage is initiated ) , or @xmath109 the algorithm terminates . the @xmath0-las algorithm starts with an initial solution @xmath110 , given by @xmath111 , where @xmath112 is the initial solution filter , which can be a matched filter ( mf ) or zero - forcing ( zf ) filter or mmse filter . the index @xmath113 in @xmath114 denotes the iteration number in a sub - stage of a given search stage . the ml cost function after the @xmath12th iteration in a given search stage is @xmath115 let us assume that we update the @xmath11th symbol in the @xmath116th iteration ; @xmath11 can take value from @xmath117 for @xmath67-pam and @xmath118 for @xmath67-qam . the update rule can be written as @xmath119 where @xmath120 denotes the unit vector with its @xmath11th entry only as one , and all other entries as zero . also , for any iteration @xmath12 , @xmath121 should belong to the space @xmath90 , and therefore @xmath122 can take only certain integer values . for example , in case of 4-pam or 16-qam @xmath123both have the same signal set @xmath124 , @xmath122 can take values only from @xmath125 . using ( [ ck ] ) and ( [ updatek ] ) , and defining a matrix @xmath126 as @xmath127 we can write the cost difference as @xmath128 where @xmath129 is the @xmath11th column of @xmath130 , @xmath131 , @xmath132 is the @xmath11th entry of the @xmath133 vector , and @xmath134 is the @xmath135th entry of the @xmath126 matrix . also , let us define @xmath136 and @xmath137 as @xmath138 with the above variables defined , we can rewrite ( [ ck+1minusck ] ) as @xmath139 where @xmath140 denotes the signum function . for the ml cost function to reduce from the @xmath12th to the @xmath116th iteration , the cost difference should be negative . using this fact and that @xmath136 and @xmath137 are non - negative quantities , we can conclude from ( [ costdiff ] ) that the sign of @xmath122 must satisfy @xmath141 using ( [ conc1 ] ) in ( [ costdiff ] ) , the ml cost difference can be rewritten as @xmath142 for @xmath143 to be non - positive , the necessary and sufficient condition from ( [ costdiff2 ] ) is that @xmath144 however , we can find the value of @xmath137 which satisfies ( [ nsc1 ] ) and at the same time gives the largest descent in the ml cost function from the @xmath12th to the @xmath116th iteration ( when symbol @xmath11 is updated ) . also , @xmath137 is constrained to take only certain integer values , and therefore the brute - force way to get optimum @xmath137 is to evaluate @xmath143 at all possible values of @xmath137 . this would become computationally expensive as the constellation size @xmath67 increases . however , for the case of 1-symbol update , we could obtain a closed - form expression for the optimum @xmath137 that minimizes @xmath143 , which is given by ( corresponding theorem and proof are given in the appendix ) @xmath145 where @xmath146 denotes the rounding operation , where for a real number @xmath147 , @xmath148 is the integer closest to @xmath147 . if the @xmath11th symbol in @xmath121 , i.e. , @xmath149 , were indeed updated , then the new value of the symbol would be given by @xmath150 however , @xmath151 can take values only in the set @xmath152 , and therefore we need to check for the possibility of @xmath151 being greater than @xmath153 or less than @xmath154 . if @xmath155 , then @xmath137 is adjusted so that the new value of @xmath151 with the adjusted value of @xmath137 using ( [ newdpk ] ) is @xmath153 . similarly , if @xmath156 , then @xmath137 is adjusted so that the new value of @xmath151 is @xmath154 . let @xmath157 be obtained from @xmath158 after these adjustments . it can be shown that if @xmath159 is non - positive , then @xmath160 is also non - positive . we compute @xmath160 , @xmath161 @xmath162 . now , let @xmath163 if @xmath164 , the update for the @xmath116th iteration is @xmath165 where @xmath166 is the @xmath167th column of @xmath126 . the update in ( [ eqx ] ) follows from the definition of @xmath133 in ( [ ck+1minusck ] ) . if @xmath168 , then the 1-symbol update search terminates . the data vector at this point is referred to as ` 1-symbol update local minima . ' after reaching the 1-symbol update local minima , we look for a further decrease in the cost function by updating multiple symbols simultaneously . the motivation for trying out multiple symbol updates can be explained as follows . let @xmath169 denote the set of data vectors such that for any @xmath170 , if a @xmath106-symbol update is performed on @xmath171 resulting in a vector @xmath172 , then @xmath173 . we note that @xmath174 , because any number of symbol updates on @xmath175 will not decrease the cost function . we define another set @xmath176 . note that @xmath177 , and @xmath178 , i.e. , @xmath179 is a singleton set with @xmath175 as the only element . it is noted that if the updates are done optimally , then the output of the @xmath106-las algorithm converges to a vector in @xmath180 . also , @xmath181 . for any @xmath182 , @xmath183 and @xmath184 , it can be seen that @xmath171 and @xmath175 will differ in @xmath185 or more locations . the probability that @xmath186 increases with increasing snr , and so the separation between @xmath187 and @xmath87 will monotonically increase with increasing @xmath106 . since @xmath188 , and @xmath189 decreases monotonically with increasing @xmath106 , there will be lesser non - ml data vectors to which the algorithm can converge to for increasing @xmath106 . therefore , the probability of the noise vector @xmath190 inducing an error would decrease with increasing @xmath106 . this indicates that @xmath106-symbol updates with large @xmath106 could get near to ml performance with increasing complexity for increasing @xmath106 . in this subsection , we present the update algorithm for the general case where @xmath106 symbols , @xmath192 , are updated simultaneously in one iteration . @xmath106-symbol updates can be done in @xmath193 ways , among which we seek to find that update which gives the largest reduction in the ml cost . assume that in the @xmath116th iteration , @xmath106 symbols at the indices @xmath194 of @xmath121 are updated . each @xmath195 , @xmath196 , can take values from @xmath197 for @xmath67-pam and @xmath198 for @xmath67-qam . further , define the set of indices , @xmath199 . the update rule for the @xmath106-symbol update can then be written as @xmath200 for any iteration @xmath12 , @xmath121 belongs to the space @xmath90 , and therefore @xmath201 can take only certain integer values . in particular , @xmath202 , where @xmath203 . for example , for 16-qam , @xmath204 , and if @xmath205 is -1 , then @xmath206 = @xmath207 . using ( [ ck ] ) , we can write the cost difference function @xmath208 as @xmath209 where @xmath210 , which can be compactly written as @xmath211 , where @xmath212 denotes the cartesian product of @xmath213 , @xmath214 through to @xmath215 . for a given @xmath216 , in order to decrease the ml cost , we would like to choose the value of the @xmath106-tuple @xmath217 such that the cost difference given by ( [ eqpqk ] ) is negative . if multiple @xmath106-tuples exist for which the cost difference is negative , we choose the @xmath106-tuple which gives the most negative cost difference . unlike for 1-symbol update , for @xmath106-symbol update we do not have a closed - form expression for @xmath218 which minimizes the cost difference over @xmath212 , since the cost difference is a function of @xmath106 discrete valued variables . consequently , a brute - force method is to evaluate @xmath219 over all possible values of @xmath217 . approximate methods can be adopted to solve this problem using lesser complexity . one method based on zero - forcing is as follows . the cost difference function in ( [ eqpqk ] ) can be rewritten as @xmath220 where @xmath221^t$ ] , @xmath222^t$ ] , and @xmath223 , where @xmath224 and @xmath225 . since @xmath219 is a strictly convex quadratic function of @xmath226 ( the hessian @xmath227 is positive definite with probability 1 ) , a unique global minima exists , and is given by @xmath228 however , the solution given by ( [ selectktwosuboptsolnrk ] ) need not lie in @xmath212 . so , we first round - off the solution as @xmath229 where the operation in ( [ selectktwosuboptsolnrndk ] ) is done element - wise , since @xmath230 is a vector . further , let @xmath231^t$ ] . it is still possible that the solution @xmath232 in ( [ selectktwosuboptsolnrndk ] ) need not lie in @xmath212 . this would result in @xmath233 for some @xmath15 . for example , if @xmath234 is @xmath67-pam , then @xmath233 if @xmath235 or @xmath236 . in such cases , we propose the following adjustment to @xmath237 for @xmath196 : @xmath238 after these adjustments , we are guaranteed that @xmath239 . therefore , the new cost difference function value is given by @xmath240 . it is noted that the complexity of this approximate method does not depend on the size of the set @xmath212 , i.e. , it has constant complexity . through simulations , we have observed that this approximation results in a performance close to that of the brute - force method for @xmath241 and 3 . defining the optimum @xmath216 for the approximate method as @xmath242 , we can write @xmath243 the @xmath106-update is successful and the update is done only if @xmath244 . the update rules for the @xmath133 and @xmath121 vectors are given by @xmath245 the complexity of the proposed @xmath0-las algorithm comprises of three components , namely , @xmath246 computation of the initial vector @xmath110 , @xmath247 computation of @xmath248 , and @xmath249 the search operation . figure [ fig1 ] shows the per - symbol complexity plots as a function of @xmath250 for 4-qam at an snr of 6 db using mmse initial vector . two good properties of the stbcs from cda are useful in achieving low orders of complexity for the computation of @xmath110 and @xmath248 . they are : @xmath246 the weight matrices @xmath96 s are _ permutation type _ , and @xmath247 the @xmath251 matrix formed with @xmath252-sized @xmath253 vectors as columns is a _ scaled unitary matrix_. these properties allow the computation of mmse / zf initial solution in @xmath254 complexity , i.e. , in @xmath3 per - symbol complexity since there are @xmath255 symbols in one stbc matrix . likewise , the computation of @xmath248 can be done in @xmath256 per - symbol complexity . the average per - symbol complexities of the 1-las and 2-las search operations are @xmath257 and @xmath258 , respectively , which can be explained as follows . the average search complexity is the complexity of one search stage times the mean number of search stages till the algorithm terminates . for 1-las , the number of search stages is always one . there are multiple iterations in the search , and in each iteration all possible @xmath259 1-symbol updates are considered . so , the per - iteration complexity in 1-las is @xmath257 , i.e. , @xmath260 complexity per symbol . further , the mean number of iterations before the algorithm terminates in 1-las was found to be @xmath257 through simulations . so , the overall per - symbol complexity of 1-las is @xmath257 . in 2-las , the complexity of the 2-symbol update dominates over the 1-symbol update . since there are @xmath261 possible 2-symbol updates , the complexity of one search stage is @xmath262 , i.e. , @xmath257 complexity per symbol . the mean number of stages till the algorithm terminates in 2-las was found to be @xmath263 through simulations . therefore , the overall per - symbol complexity of 2-las is @xmath264 . these can be observed from fig . [ fig1 ] , where it can be seen that the per - symbol complexity in the initial vector computation plus the 1-las/2-las search operation is @xmath257/@xmath264 ; i.e. , 1-las and 2-las complexity plots run parallel to the @xmath265 and @xmath266 lines , respectively . with the computation of @xmath248 included , the complexity order is more than @xmath255 . from the slopes of the plots in fig . [ fig1 ] , we find that the overall complexities for @xmath267 and 32 are proportional to @xmath268 and @xmath269 , respectively . -las algorithm in decoding non - orthogonal stbcs from cda . mmse initial vector , 4-qam , snr = 6 db . , width=345 ] for the special case of ill - only stbcs ( i.e. , @xmath103 ) , the complexity involved in computing @xmath110 and @xmath248 can be reduced further . this becomes possible due to the following property of ill - only stbcs . let @xmath270 be the complex @xmath251 matrix with @xmath253 as its @xmath6th column . the computation of @xmath110 ( or @xmath248 ) involves multiplication of @xmath271 with another vector ( or matrix ) . the columns of @xmath271 can be permuted in such a way that the permuted matrix is block - diagonal , where each block is a @xmath272 dft matrix for @xmath103 . so , the multiplication of @xmath271 by any vector becomes equivalent to a @xmath4-point dft operation , which can be efficiently computed using fft in @xmath273 complexity . using this simplification , the per - symbol complexity of computing @xmath248 is reduced from @xmath256 to @xmath258 . computing @xmath110 using mmse filter involves the computation of @xmath274 . the complexity of computing the vector @xmath275 is @xmath276 , and the complexity of computing @xmath277 is @xmath278 . in the case of ill - only stbc , because of the above - mentioned property , the complexity of computing @xmath277 gets reduced to @xmath258 from @xmath278 . so the total complexity for computing @xmath110 in ill - only stbc is @xmath279 , which gives a per - symbol complexity of @xmath280 . so , the overall per - symbol complexity for 1-las detection of ill - stbcs is @xmath264 . we propose to generate soft values at the @xmath0-las output for all the individual bits that constitute the @xmath67-pam/@xmath67-qam symbols as follows . these output values are fed as soft inputs to the decoder in a coded system . let @xmath281 $ ] , @xmath282 denote the detected output symbol vector from the @xmath0-las algorithm . let the symbol @xmath283 map to the bit vector @xmath284^t$ ] , where @xmath285 , and @xmath286 , @xmath287 and @xmath288 . let @xmath289 denote the soft value for the @xmath15th bit of the @xmath6th symbol . given @xmath171 , we need to find @xmath290 . note that the quantity @xmath291 is inversely related to the likelihood that @xmath171 is indeed the transmitted symbol vector . let the @xmath171 vector with its @xmath15th bit of the @xmath6th symbol forced to + 1 be denoted as vector @xmath292 . likewise , let @xmath293 be the vector @xmath171 with its @xmath15th bit of the @xmath6th symbol forced to -1 . then the quantities @xmath294 and @xmath295 are inversely related to the likelihoods that the @xmath15th bit of the @xmath6th transmitted symbol is + 1 and -1 , respectively . so , if @xmath296 is + ve ( or -ve ) , it indicates that the @xmath15th bit of the @xmath6th transmitted symbol has a higher likelihood of being + 1 ( or -1 ) . so , the quantity @xmath297 , appropriately normalized to avoid unbounded increase for increasing @xmath4 , can be a good soft value for the @xmath15th bit of the @xmath6th symbol . with this motivation , we generate the soft output value for the @xmath15th bit of the @xmath6th symbol as @xmath298 where the normalization by @xmath299 is to contain unbounded increase of @xmath300 for increasing @xmath4 . the rhs in the above can be efficiently computed in terms of @xmath301 and @xmath126 as follows . since @xmath292 and @xmath293 differ only in the @xmath6th entry , we can write @xmath302 since we know @xmath293 and @xmath292 , we know @xmath303 from ( [ cd1 ] ) . substituting ( [ cd1 ] ) in ( [ soft ] ) , we can write @xmath304 if @xmath305 , then @xmath306 and substituting this in ( [ plus ] ) and dividing by @xmath307 , we get @xmath308 if @xmath309 , then @xmath310 and substituting this in ( [ minus ] ) and dividing by @xmath307 , we get @xmath311 it is noted that @xmath301 and * g * are already available upon the termination of the @xmath0-las algorithm , and hence the complexity of computing @xmath300 in ( [ cd3 ] ) and ( [ cd4 ] ) is constant . hence , the overall complexity in computing the soft values for all the bits is @xmath312 . we also see from ( [ cd3 ] ) and ( [ cd4 ] ) that the magnitude of @xmath300 depends upon @xmath303 . for large - size signal sets , the possible values of @xmath303 will also be large in magnitude . we therefore have to normalize @xmath300 for the turbo decoder to function properly . it has been observed through simulations that normalizing @xmath300 by @xmath313 resulted in good performance . in @xcite , we have shown that this soft decision output generation method , when used in large v - blast systems , offers about 1 to 1.5 db improvement in coded ber performance compared to that achieved using hard decision outputs from the @xmath0-las algorithm . we have observed similar improvements in stbc mimo systems also . in all coded ber simulations in this paper , we use the soft outputs proposed here as inputs to the decoder . in this section , we present the uncoded / turbo coded ber performance of the proposed @xmath0-las detector in decoding non - orthogonal stbcs from cda , assuming perfect knowledge of csi at the receiver . in all the ber simulations in this section , we have assumed that the fade remains constant over one stbc matrix duration and varies i.i.d . from one stbc matrix duration to the other . we consider two stbc designs ; @xmath246 ` fd - ill ' stbcs where @xmath101 , @xmath102 in ( 11.a ) , and @xmath247 ` ill - only ' stbcs where @xmath103 . the snrs in all the ber performance figures are the average received snr per received antenna , @xmath28 , defined in sec . [ sec2 ] @xcite . we have used mmse filter as the initial filter in all the simulations . in fig . [ fig2 ] , we plot the uncoded ber performance of the proposed 1- , 2- , and 3-las algorithms in decoding ill - only stbcs ( @xmath314 , @xmath315 , @xmath1 , @xmath2 stbcs ) for @xmath316 and 4-qam . siso awgn performance ( without fading ) and mmse - only performance ( i.e. , without the search using las ) are also plotted for comparison . it can be seen that mmse - only performance does not improve with increasing stbc size ( i.e. , increasing @xmath250 ) . however , it is interesting to see that , when the proposed search using las is performed following the mmse operation , the performance improves for increasing @xmath250 , illustrating the performance benefit due to the proposed search strategy . for example , though the las detector performs far from siso awgn performance for small number of dimensions ( e.g. , @xmath317 stbcs with 32 and 128 real dimensions , respectively ) , its large system behavior at increased number of dimensions ( e.g. , @xmath1 and @xmath2 stbcs with 512 and 2048 real dimensions , respectively ) effectively renders near siso awgn performance ; e.g. , with @xmath318 , for bers better than @xmath319 , the las detector performs very close to siso awgn performance . we also observe that 3-las performs better than 2-las for @xmath320 , and 2-las performs better than 1-las . since close to siso awgn performance is achieved with 1- , 2- , or 3-symbol update itself , the cases of more than 3-symbol update , which will result in increased complexity with diminishing returns in performance gain , are not considered in the performance evaluation . . 4-qam , @xmath321 bps / hz . _ ber improves as @xmath250 increases and approaches siso awgn performance for large @xmath250_. , width=360 ] [ fig2 ] . 4-qam , @xmath321 bps / hz , 1-las detection . _ ill - only stbcs perform almost same as fd - ill stbcs . _ , width=360 ] [ fig3 ] in fig . [ fig3 ] , we present uncoded ber performance comparison between fd - ill versus ill - only stbcs for 4-qam at different @xmath250 using 1-las detection . the ber plots in fig . [ fig3 ] illustrate that the performance of ill - only stbcs with 1-las detection for @xmath316 and 4-qam are almost as good as those of the corresponding fd - ill stbcs . a similar closeness between the performance of ill - only and fd - ill stbcs is observed in the turbo coded ber performance as well , which is shown in fig . [ fig8 ] for a @xmath1 stbc with 4-qam and turbo code rates of 1/3 , 1/2 and 3/4 . this is an interesting observation , since this suggests that , in such cases , the computational complexity advantage with @xmath103 in ill - only stbcs can be taken advantage of without incurring much performance loss compared to fd - ill stbcs . while the stbc design in ( 11.a ) offers both ill and fd , _ perfect codes _ under ml decoding can provide coding gain in addition to ill and fd @xcite-@xcite . decoding of perfect codes has been reported in the literature for only up to 5 antennas using sphere / lattice decoding @xcite . the complexity of these decoders are prohibitive for decoding large - sized perfect codes , although large - sized codes are of interest from a high spectral efficiency view point . we note that , because of its low - complexity attribute , the proposed @xmath0-las detector is able to decode perfect codes of large dimensions . in figs . [ fig4 ] and [ fig5 ] , we present the simulated ber performance of perfect codes in comparison with those of ill - only and fd - ill stbcs for up to 32 transmit antennas using 1-las detector . , 4-qam , @xmath321 bps / hz , 1-las detection . _ for small dimensions ( e.g. , @xmath314 , @xmath322 , @xmath315 ) , perfect codes with 1-las detection perform worse than ill - only stbcs . for large dimensions ( e.g. , @xmath1 , @xmath2 ) , ill - only stbcs and perfect codes perform almost same . _ , width=379 ] [ fig4 ] in fig . [ fig4 ] , we show uncoded ber comparison between perfect codes and ill - only stbcs for different @xmath250 and 4-qam using 1-las detection . the @xmath314 and @xmath322 perfect codes are from @xcite , and the @xmath315 , @xmath1 and @xmath2 perfect codes are from @xcite . from fig . [ fig4 ] , it can be seen that the 1-las detector achieves better performance for ill - only stbcs than for perfect codes , when codes with small number of transmit antennas are considered ( e.g. , @xmath323 ) . while perfect codes are expected to perform better than ill - only codes under ml detection for any @xmath4 , we observe the opposite behavior under 1-las detection for small @xmath4 ( i.e. , ill - only stbcs performing better than perfect codes for small dimensions ) . this behavior could be attributed to the nature of the las detector , which achieves near - optimal performance only when the number of dimensions is large , keeping @xmath250 . ] , and it appears that , in the detection process , las is more effective in disentangling the symbols in stbcs when @xmath103 ( i.e. , in ill - only stbcs ) than in perfect codes . the performance gap between perfect codes and ill - only stbcs with 1-las detection diminishes for increasing code sizes such that the performance for @xmath2 perfect code and ill - only stbc with 4-qam are almost same and close to the siso awgn performance . in fig . [ fig5 ] , we show a similar comparison between perfect codes , ill - only and fd - ill only stbcs when larger modulation alphabet sizes ( e.g. , 16-qam ) are used in the case of @xmath1 and @xmath2 codes . it can be seen that with higher - order qam like 16-qam , perfect codes with 1-las detection perform poorer than ill - only and fd - ill stbcs , and that ill - only and fd - ill stbcs perform almost same and close to the siso awgn performance . the results in figs . [ fig4 ] and [ fig5 ] suggest that , with 1-las detection , owing to the complexity advantage and good performance in using @xmath103 , ill - only stbcs can be a good choice for practical large stbc mimo systems @xcite,@xcite . , * 16-qam * , @xmath324 bps / hz , 1-las detection . _ for larger modulation alphabet sizes ( e.g. , 16 qam ) , perfect codes with 1-las detection perform poorer than ill - only and fd - ill stbcs_. , width=360 ] [ fig5 ] in @xcite , choi _ et al _ have presented an iterative soft interference cancellation ( isic ) scheme for multiple antenna systems , derived based on maximum a posteriori ( map ) criterion . we compared the performance of the isic scheme in @xcite with that of the proposed 1-las algorithm in detecting @xmath314 , @xmath315 and @xmath1 ill - only stbcs with @xmath250 and 4-qam . figure [ fig6 ] shows this performance comparison . in @xcite , zero - forcing vector was used as the initial vector in the isic scheme . however , performance is better with mmse initial vector . since we used mmse initial vector for @xmath325-las , we have used mmse initial vector for the isic algorithm as well . also , in @xcite , 4 to 5 iterations were shown to be good enough for the isic algorithm to converge . in our simulations of the isic algorithm , we used 10 iterations . two key observations can be made from fig . [ fig6 ] : @xmath246 like the @xmath325-las algorithm , the isic algorithm also shows large system behavior ( i.e. , improved ber for increasing @xmath250 ) , and 2 ) the proposed 1-las algorithm outperforms the isic algorithm by about 3 to 5 db at @xmath319 uncoded ber . in addition , the complexity of the isic scheme is higher than the proposed scheme ( see the complexity comparison in table i ) . . 4-qam , @xmath321 bps / hz . mmse initial vectors for both 1-las and isic . 1-las performs significantly better than isic in @xcite . _ , width=345 ] [ fig6 ] next , we compare the proposed large - mimo architecture using stbcs from cda and @xmath0-las detection with other large - mimo architectures and associated detectors reported in the literature . large - mimo architectures that use stacking of multiple small - sized stbcs and interference cancellation ( ic ) detectors for these schemes have been investigated in @xcite,@xcite,@xcite . here , we compare different architecture / detector combinations , fixing the total number of transmit / receive antennas and spectral efficiency to be same in all the considered combinations . specifically , we fix @xmath326 and a spectral efficiency of 32 bps / hz for all the combinations . we compare the following seven different architecture / detector combinations which use the same @xmath326 and achieve 32 bps / hz spectral efficiency ( see table i ) : @xmath246 proposed scheme using @xmath1 ill - only stbc ( rate-16 ) with 4-qam and 1-las detection , @xmath247 @xmath1 ill - only stbc ( rate-16 ) with 4-qam and isic algorithm in @xcite with 10 iterations , @xmath249 four @xmath314 stacked qostbcs ( rate-1 ) with 256-qam and ic algorithm presented in @xcite , @xmath327 eight @xmath328 stacked alamouti codes ( rate-1 ) with 16-qam and ic algorithm in @xcite , @xmath329 @xmath1 v - blast scheme ( rate-16 ) with 4-qam and sphere decoder ( sd ) , @xmath330 @xmath1 v - blast scheme ( rate-16 ) with 4-qam and zf - sic detector , and @xmath331 @xmath1 v - blast scheme ( rate-16 ) with 4-qam and isic algorithm in @xcite . we present the ber performance comparison of these different combinations in fig . we also obtained the complexity numbers ( in number of real operations per bit ) from simulations for these different combinations at an uncoded ber of @xmath332 ; these numbers are presented in table i , along with the snrs at which @xmath332 uncoded ber is achieved . the following interesting observations can be made from fig . [ fig7 ] and table i : [ cols="^,^,^,^ " , ] [ tab3 ] in @xcite , through theoretical capacity bounds it has been shown that , for a given @xmath5 , @xmath333 and snr , there is an optimum value of @xmath4 that maximizes the capacity bound ( refer figs . 5 and 6 in @xcite , where the optimum @xmath4 is shown to be greater than @xmath5 in fig . 5 and less than @xmath5 in fig . 6 ) . for example , for @xmath334 , @xmath335 , and snr = 10 db , the capacity bound evaluated using ( [ trg_bound ] ) with optimum power allocation for @xmath336 is 19.73 bps / hz , whereas for @xmath267 the capacity bound reduces to 17.53 bps / hz showing that the optimum @xmath4 in this case will be less than @xmath5 . we demonstrate such an observation in practical systems by comparing the simulated coded ber performance of two systems , referred to as system - i and system - ii , using 1-las detection and iterative detection / estimation scheme . the parameters of system - i and system - ii are listed in table ii . @xmath5 and @xmath333 are fixed at 16 and 48 , respectively , in both systems . system - i uses 16 transmit antennas and @xmath1 stbc , whereas system - ii uses 12 transmit antennas and @xmath337 stbc . since the pilot matrix is @xmath338 , the pilot duration @xmath339 is 16 and 12 , respectively , for system - i and system - ii . optimum pilot / data power allocation and 4-qam modulation are employed in both systems . system - i uses rate-1/2 turbo code and system - ii uses rate-3/4 turbo code . with the above system parameters , the spectral efficiency achieved in system - i is @xmath340 bps / hz , whereas system - ii achieves a higher spectral efficiency of @xmath341 bps / hz . in fig . [ fig15 ] , we plot the coded ber of both these systems using 1-las detection and iterative detection / estimation . from the simulation points shown in fig . [ fig15 ] , it can be observed that system - ii with a smaller @xmath4 and higher spectral efficiency in fact achieves a certain coded ber performance at a lesser snr compared to system - i . for example , to achieve @xmath319 coded ber , system - i requires an snr of about 8.9 db , whereas system - ii requires only 8.6 db . this implies that because of the reduction of throughput due to pilot symbols ( by a factor of @xmath342 for a given @xmath333 and @xmath343 ) , a larger @xmath4 does not necessarily mean a higher spectral efficiency . such an observation has also been made in @xcite based on theoretical capacity bounds . the proposed detection / channel estimation scheme allows the prediction of such behavior through simulations , which , in turn , allows system designers to find optimum @xmath4 and stbc size to achieve a certain spectral efficiency in large stbc mimo systems . d training - based systems , one with a larger @xmath4 than the other for a given @xmath5 and @xmath333 . _ with @xmath334 , @xmath335 and optimum power allocation in both systems , system - ii with @xmath336 achieves a higher spectral efficiency ( 13.5 vs 10.33 bps / hz ) while achieving @xmath319 coded ber at a lesser snr ( 8.6 vs 8.9 db ) than system - i with @xmath267 . _ , width=384 ] [ fig15 ] we presented a low - complexity algorithm for the detection of high - rate , non - orthogonal stbc large - mimo systems with tens of antennas that achieve high spectral efficiencies of the order of several tens of bps / hz . we also presented a training - based iterative detection / channel estimation scheme for such large stbc mimo systems . our simulation results showed that the proposed 1-las detector along with the proposed iterative detection / channel estimation scheme achieved very good performance at low complexities . with the feasibility of low - complexity high - performance receivers , like the proposed detection / channel estimation scheme , large - mimo systems with tens of antennas at high spectral efficiencies can become practical , enabling interesting high data rate wireless applications ( e.g. , wireless iptv / hdtv distribution ) . this can motivate the inclusion of large - mimo architectures ( e.g. , @xmath337 , @xmath1 mimo systems , including those using stbcs from cda ) into wireless standards like ieee 802.11n / vht and ieee 802.16/lte - a in their evolution to achieve high data rates at increased spectral efficiencies . the @xmath137 in ( [ optdp ] ) minimizes @xmath143 in ( [ costdiff2 ] ) and this minimum value is non - positive . _ proof : _ let @xmath344 . then @xmath345 , where @xmath346 , and so we can write @xmath347 if @xmath137 were unconstrained to be any real number , then the optimal value of @xmath137 is @xmath348 , which would lie between @xmath349 and @xmath350 ( as per ( [ proofopt2 ] ) ) . since @xmath143 is quadratic in @xmath137 , it is unimodular , and hence the optimal point ( with @xmath137 constrained ) would be either @xmath349 or @xmath350 . using ( [ costdiff2 ] ) and ( [ proofopt2 ] ) , we can evaluate @xmath351 to be @xmath352 since @xmath136 is a positive quantity , the sign of @xmath353 depends upon the sign of @xmath354 . if @xmath355 , then @xmath356 , and therefore @xmath350 is the optimal value of @xmath137 . similarly , when @xmath357 , @xmath349 is the optimal value of @xmath137 . therefore , it follows that indeed the rounding solution given by ( [ optdp ] ) is optimal . @xmath143 is non - positive for all values of @xmath137 between zero and @xmath358 . if @xmath357 , then @xmath349 is optimal , and , from ( [ proofopt2 ] ) , we know that @xmath359 , and therefore @xmath360 . hence @xmath361 @xmath362 is non - positive . similarly , if @xmath355 , then @xmath350 is optimal , and @xmath356 . however , since @xmath349 is always less than @xmath363 , @xmath364 is non - positive and therefore @xmath365 @xmath362 is non - positive . we would like to thank the editor , prof . r. calderbank , for handling the review process . we would like to thank the anonymous reviewers for their critical and useful comments , and for motivating us to compare the performance and complexity of the proposed scheme with those of other large - mimo architectures / detectors . b. a. sethuraman , b. sundar rajan , and v. shashidhar , `` full - diversity high - rate space - time block codes from division algebras , '' _ ieee trans . inf . theory _ , vol . 10 , pp . 2596 - 2616 , october 2003 . k. vishnu vardhan , saif k. mohammed , a. chockalingam , b. sundar rajan , `` a low - complexity detector for large mimo systems and multicarrier cdma systems , '' _ ieee jsac spl . iss . on multiuser detection , for adv . commun . & networks _ , vol . 473 - 485 , april 2008 . belfiore , g. rekaya , and e. viterbo , `` the golden code : a @xmath328 full - rate space - time code with non - vanishing determinants , '' _ ieee trans . inform . theory _ 4 , pp . 1432 - 1436 , april 2005 . f. oggier , j .- c . belfiore , and e. viterbo , _ cyclic division algebras : a tool for space - time coding , _ foundations and trends in commun . and theory , vol . 1 , pp . 1 - 95 , now publishers , 2007 . j. kazemitabar and h. jafarkhani , `` multiuser interference cancellation and detection for users with more than two transmit antennas , '' _ ieee trans . on commun . 574 - 583 , april 2008 . n. prasad , m. k. varanasi , l. venturino , x. wang , `` an analysis of the mimo - sdma channel with space - time orthogonal and quasi - orthogonal user transmissions and efficient successive cancellation decoders , '' _ ieee trans . inf . theory _ , 5427 - 5446 , december 2008 . m. vu , and a. paulraj , `` optimal linear precoders for mimo wireless correlated channels with nonzero mean in space time coded systems , '' _ ieee trans . sig . proc . 54 , no . 6 , pp . 2318 - 2332 , june 2006 . k. t. phan , s. a. vorobyov , and c. tellambura , `` precoder design for space - time coded systems with correlated rayleigh fading channels using convex optimization , '' _ ieee trans . 814 - 819 , february 2009 . a. zaki , saif k. mohammed , a. chockalingam , and s. sundar rajan , `` a training - based iterative detection / channel estimation scheme for large non - orthogonal stbc mimo systems , '' _ ieee icc2009 _ , june 2009 . belfiore and a. m. cipriano , `` space - time coding for noncoherent channels , '' book chapter in _ space - time wireless systems : from array processing to mimo communications , edited by h. blcskei , d. gesbert , c. b. papadias , and a .- j . van der veen , _ cambridge univ . press , 2006 . h. li , s. m. betz , and h. v. poor , `` performance analysis of iterative channel estimation and multiuser detection in multipath ds - cdma channels , '' _ ieee trans . 1981 - 1993 , may 2007 . b. hu , i. land , l. rasmussen , r. piton , and b. h. fleury , `` a divergence minimization approach to joint multiuser decoding for coded cdma , '' _ ieee jl . on sel . areas in commun . 432 - 445 , april 2008 . m. loncar , r. r. muller , j. wehinger , c. f. mecklenbrauker , and t. abe , `` iterative channel estimation and data detection in frequency - selective fading mimo channels , '' _ eur . telecommun_. , vol . 15 , no . 459 - 470 , september / october 2004 . h. zhu , b. farhang - boroujeny , and c. schlegel , `` pilot embedding for joint channel estimation and data detection in mimo communication systems , '' _ ieee commun . letters _ , vol . 7 , no . 1 , pp . 30 - 32 , january 2003 . p. s. rossi and r. r. muller , `` joint twofold - iterative channel estimation and multiuser detection for mimo - ofdm systems , '' _ ieee trans . wireless commun . , _ vol . 7 , no . 11 , pp . 4719 - 4729 , november 2008 . v. shashidhar , b. sundar rajan , and b. a. sethuraman , `` information - lossless space - time block codes from crossed - product algebras , '' _ ieee trans . inform . theory _ 3913 - 3935 , september 2006 . saif k. mohammed received his b.tech degree in computer science and engineering from the indian institute of technology , new delhi , india , in 1998 . from 1998 to 2000 , he was employed with philips inc . , bangalore , as an asic design engineer . from 2000 to 2003 , he worked with ishoni networks inc . , santa clara , ca , as a senior chip architecture engineer . from 2003 to 2007 , he was employed with texas instruments , bangalore as systems and algorithms designer in the wireless systems group . since 2006 , he is pursuing his doctoral degree in electrical and communications engineering at the indian institute of science , bangalore , india . his research interests include low - complexity detection , estimation and coding for wireless communications systems . ahmed zaki received the b.e . degree in electronics and communication engineering from osmania university , hyderabad , india , in 2007 , and the m.e . degree in telecommunication from the indian institute of science , bangalore , india , in 2009 . his research interest lies in the area of wireless communications , including receiver design and channel estimation for large - mimo systems , mimo - ofdm , multiuser communications , and algorithm design . a. chockalingam was born in rajapalayam , tamil nadu , india . he received the b.e . ( honors ) degree in electronics and communication engineering from the p. s. g. college of technology , coimbatore , india , in 1984 , the m.tech degree with specialization in satellite communications from the indian institute of technology , kharagpur , india , in 1985 , and the ph.d . degree in electrical communication engineering ( ece ) from the indian institute of science ( iisc ) , bangalore , india , in 1993 . during 1986 to 1993 , he worked with the transmission r & d division of the indian telephone industries limited , bangalore . from december 1993 to may 1996 , he was a postdoctoral fellow and an assistant project scientist at the department of electrical and computer engineering , university of california , san diego . from may 1996 to december 1998 , he served qualcomm , inc . , san diego , ca , as a staff engineer / manager in the systems engineering group . in december 1998 , he joined the faculty of the department of ece , iisc , bangalore , india , where he is a professor , working in the area of wireless communications and networking . chockalingam is a recipient of the swarnajayanti fellowship from the department of science and technology , government of india . he served as an associate editor of the ieee transactions on vehicular technology from may 2003 to april 2007 . he currently serves as an editor of the ieee transactions on wireless communications . he served as a guest editor for the ieee jsac special issue on multiuser detection for advanced communication systems and networks . he is a fellow of the institution of electronics and telecommunication engineers , and a fellow of the indian national academy of engineering . b. sundar rajan ( s84-m91-sm98 ) was born in tamil nadu , india . he received the b.sc . degree in mathematics from madras university , madras , india , the b.tech degree in electronics from madras institute of technology , madras , and the m.tech and ph.d . degrees in electrical engineering from the indian institute of technology , kanpur , india , in 1979 , 1982 , 1984 , and 1989 respectively . he was a faculty member with the department of electrical engineering at the indian institute of technology in delhi , india , from 1990 to 1997 . since 1998 , he has been a professor in the department of electrical communication engineering at the indian institute of science , bangalore , india . his primary research interests include space - time coding for mimo channels , distributed space - time coding and cooperative communication , coding for multiple - access , relay channels and network coding with emphasis on algebraic techniques . rajan is an associate editor of the ieee transactions on information theory , an editor of the ieee transactions on wireless communications , and an editorial board member of international journal of information and coding theory . he served as technical program co - chair of the ieee information theory workshop ( itw02 ) , held in bangalore , in 2002 . he is a fellow of indian national academy of engineering and recipient of the iete pune center s s.v.c aiya award for telecom education in 2004 . also , dr . rajan is a member of the american mathematical society .

let be a compact , connected symplectic manifold with a hamiltonian action of a compact-dimensional torus . suppose that is an anti - symplectic involution compatible with the-action . the real locus of is , the fixed point set of . duistermaat uses morse theory to give a description of the ordinary cohomology of in terms of the cohomology of . there is a residual action on , and we can use duistermaat s result , as well as some general facts about equivariant cohomology , to prove an equivariant analogue to duistermaat s theorem . in some cases , we can also extend theorems of goresky - kottwitz - macpherson and goldin - holm to the real locus .

math0107151

atiyah observed in @xcite that if @xmath0 is a compact symplectic manifold and @xmath7 a hamiltonian action of an @xmath1-dimensional torus @xmath4 on @xmath0 , then the cohomology groups of @xmath0 can be computed from the cohomology groups of the fixed point set @xmath8 of @xmath7 . explicitly , if @xmath9 is the moment map for @xmath7 , then a generic component @xmath10 of @xmath11 is a perfect bott - morse function . using @xmath10 , we may compute @xmath12 where the @xmath13 are the connected components of @xmath8 and @xmath14 is the bott - morse index of @xmath13 . this result is also true in equivariant cohomology : @xmath15 this is a consequence of atiyah s result and kirwan s equivariant formality theorem for hamiltonian @xmath4-manifolds , as shown in @xcite . in @xcite , duistermaat proved a `` real form '' version of ( [ 1 ] ) . let @xmath16 be an anti - symplectic involution with the property that @xmath17 and let @xmath18 be the fixed point set of @xmath3 . we call @xmath5 the _ real locus _ of @xmath0 . the motivating example of this setup is a complex manifold @xmath0 with a complex conjugation @xmath3 . duistermaat proved that @xmath19 where @xmath20 is the set for which @xmath3 preserves @xmath13 . the @xmath21 coefficients are essential here ; the theorem does not hold with real coefficients . ( see the comments at the end of this section . ) the first of the four main theorems of this paper is an equivariant analogue of ( [ 4 ] ) similar to the equivariant analogue ( [ 2 ] ) of atiyah s result ( [ 1 ] ) . by ( [ 3 ] ) , the group @xmath22 acts on @xmath5 and we will prove the following theorem in section [ se : eqrlocus ] . [ th : additive ] suppose @xmath0 is a symplectic manifold with a hamiltonian action @xmath7 of a torus @xmath23 and an anti - symplectic involution @xmath3 . let @xmath18 denote the real locus of @xmath0 . then the group @xmath24 acts on @xmath5 , and the @xmath24-equivariant cohomology of @xmath5 with @xmath25 coefficients is @xmath26 as above , the subset @xmath20 is the set for which @xmath3 preserves @xmath13 . the idea of the proof will be to derive ( [ 6 ] ) from ( [ 4 ] ) by a simple trick . the second of the main theorems concerns the structure of @xmath27 as a module over the ring @xmath28.\end{aligned}\ ] ] let @xmath29 and @xmath30 denote the classifying space and classifying bundle of @xmath24 . then by the borel definition of equivariant cohomology @xmath31 the cohomology on the right hand side can be computed by the spectral sequence associated with the fibration @xmath32 and we will deduce from this computation the following theorem . [ tfreemod ] the equivariant cohomology @xmath33 is a free module over @xmath34 generated in dimension zero . moreover , as an @xmath34 module , @xmath33 is isomorphic to @xmath35 the idea of the proof is to show that this spectral sequence collapses at its @xmath36 term , that is , that @xmath5 is equivariantly formal . in particular , this implies that @xmath37 is a free module over @xmath38 $ ] . the isomorphisms ( [ 1 ] ) , ( [ 2 ] ) , ( [ 4 ] ) , and ( [ 6 ] ) are all isomorphisms in _ additive _ cohomology . the next two sections of this paper will be concerned with the ring structure of @xmath39 . henceforth , we will assume that the fixed point set @xmath8 is finite and that the one - skeleton of @xmath7 , @xmath40 is of dimension @xmath41 . it is not hard to see that these two assumptions imply that @xmath42 moreover , @xmath43 is point - wise fixed by an @xmath44-dimensional torus @xmath45 , and the diffeomorphism @xmath46 intertwines the action of @xmath47 on @xmath43 with the standard @xmath48 action on @xmath49 . in particular , @xmath43 contains exactly two @xmath4-fixed points . thus , one can describe the intersection properties of the @xmath43 by a graph @xmath50 with edges @xmath51 corresponding to the spheres @xmath43 and vertices @xmath52 . two vertices @xmath53 and @xmath54 are joined by the edge @xmath51 if @xmath55 . furthermore , each edge , @xmath51 is labeled by a weight @xmath56 of @xmath4 , the weight associated with the intertwining homomorphism @xmath57 let @xmath58 be the inclusion of the fixed points into the manifold , and consider the induced map in equivariant cohomology , @xmath59 by a theorem of kirwan @xcite , this map is injective . moreover , @xmath60 and since @xmath61 one can regard an element of @xmath62 as a map @xmath63 goresky , kottwitz , and macpherson @xcite computed the image of @xmath64 , thus determining not just the additive equivariant cohomology of @xmath0 , but in fact the ring structure of this cohomology . @xcite[th : gkm ] a map @xmath65 is in the image of @xmath64 if and only if for each edge @xmath66 of @xmath50 @xmath67 the third of the four main theorems of this paper will be a @xmath25 version of this result for the manifold @xmath5 . we define the one - skeleton of the real locus to be the set @xmath68 assume in addition to the above that @xmath69 and the real locus of the one - skeleton is the same as the one - skeleton of the real locus . we will call a manifold with these properties a _ @xmath70 gkm manifold_. the map @xmath71 is injective , and by factoring ( [ 13 ] ) through the map @xmath72 where @xmath73 , we obtain the following theorem . [ th : realgkm ] suppose @xmath0 is a @xmath74 gkm manifold . an element @xmath75 can be thought of as a map @xmath76 $ ] , and such a map @xmath77 is in the image of @xmath64 if and only if , for each edge @xmath78 of @xmath50 @xmath79,\ ] ] where @xmath80 $ ] is the image of the weight @xmath81 . this completely determines the ring structure of @xmath39 . this theorem is proved independently by schmid @xcite using different techniques . the main examples of @xmath74 gkm manifold include the real loci of non - singular projective toric varieties and real loci of coadjoint orbits , including grassmannian and flag varieties . in the case of toric varieties , this equivariant cohomology ring was computed already by davis and januszkiewicz @xcite , but our description is quite different from theirs . in @xcite , goldin and holm generalize the gkm result to the case where the one - skeleton has dimension at most @xmath82 . assume in addition to the dimension hypothesis that @xmath69 and the real locus of the one - skeleton is the same as the one - skeleton of the real locus . we will call a manifold with these properties a _ @xmath74 gh manifold_. the last of the main theorems is a @xmath25 version of the result of goldin and holm for the real locus @xmath5 . for a subgroup @xmath83 of @xmath24 , we will let @xmath84 will denote the change of coefficient map @xmath85 associated with the inclusion @xmath86 . in section [ se : realgh ] , we will prove the following theorem . [ th : realgh ] suppose that @xmath0 is a @xmath74 gh manifold with @xmath4 fixed points fixed points @xmath87 . let @xmath88 denote the restriction of @xmath89 to the fixed point @xmath90 . the image of the injection @xmath91 is the subalgebra of functions @xmath92 which satisfy @xmath93 for all subgroups @xmath83 of @xmath24 of order @xmath94 and all connected components @xmath95 of @xmath96 , where @xmath97 and @xmath98 are the ( linearly dependent ) weights of the @xmath24 action on @xmath99 . we re - emphasize that duistermaat s techniques only apply to additive cohomology . since we are able to obtain results concerning the ring structure of the equivariant cohomology and its relationship to ordinary cohomology , we also obtain statements about the ring structure of the ordinary cohomology as well . indeed , in many cases , duistermaat s isomorphism ( [ 4 ] ) turns out to give a ring isomorphism . ( see corollaries [ co : ring1 ] and [ co : ring2 ] to theorem [ th : realgkm ] and corollaries [ co : ring3 ] and [ co : ring4 ] to theorem [ th : realgh ] . ) when describing these ring isomorphisms , we will make use of the following notation . the symbol @xmath100 will denote the subring @xmath101 endowed with a new grading wherein a class in @xmath102 is given degree @xmath103 ( and similarly for equivariant cohomology ) . then under suitable hypotheses , the additive isomorphism of duistermaat becomes an isomorphism of graded rings . in section [ se : strings ] , we discuss an application of our main theorems to string theory . the @xmath25-equivariant cohomology ring of @xmath104 with @xmath25 coefficients classifies all possible orientifold configurations of type ii string theories , compactified on @xmath104 . we explain how to compute this cohomolgy ring . the last section of the paper contains some applications of these results to elementary problems in combinatorics . a typical such application is the following . let @xmath50 be the permutahedron , the cayley graph of the symmetric group @xmath105 with edges generated by transpositions . by definition , the vertices of @xmath50 are elements of @xmath105 and two vertices @xmath3 and @xmath7 are joined by an edge if @xmath106 is a transposition . our goal is to attach to each vertex @xmath3 a subset @xmath107 of @xmath108 such that , for all pairs @xmath3 and @xmath7 of adjacent vertices , either @xmath109 or the symmetric difference @xmath110 is @xmath111 , where @xmath112 . let @xmath113 be the real flag variety in @xmath114 dimensions . we will prove that the set of solutions to this problem can be identified with the set @xmath115 where @xmath116 is the @xmath1-fold product of @xmath25 . the results in this section were inspired by a remark of ethan bolker , who pointed out to us that @xmath25 representation theory is simply boolean algebra . we conclude these prefatory remarks with a few comments about the @xmath25 coefficients . we recall witten s recipe for computing the homology of compact manifolds by morse theory . let @xmath5 be a compact manifold , @xmath117 a morse function , and @xmath118 the index @xmath103 critical set of @xmath77 . let @xmath119 be the vector space @xmath120 with basis @xmath118 . equip @xmath5 with a riemannian metric and let @xmath121 be the gradient vector field of @xmath77 . for generic metrics , the stable and unstable manifolds of @xmath121 intersect transversally . in particular , for every critical point @xmath122 , there are a finite number of gradient curves joining @xmath53 to critical points @xmath123 moreover , each of these points can be assigned an intrinsic orientation @xmath124 . now define a boundary operator @xmath125 by setting @xmath126 witten @xcite was the first to explicitly formulate morse theory in this way ; he showed that @xmath127 _ is _ a boundary operator , namely @xmath128 , and that @xmath129 is the homology of the complex @xmath130 . in particular , when all the critical points are of even index , @xmath127 is automatically zero . thus , one gets @xmath131 this fact is key to atiyah s result ( [ 1 ] ) . he observes that if @xmath132 and if @xmath77 is a generic component of the moment map , then @xmath77 is a morse function with critical points all of even index , so ( [ 1 ] ) is a special case of ( [ 16 ] ) . this recipe for computing homology also works in characteristic two ; however , when @xmath25 symmetries are present , the gradient curves joining the points @xmath122 to the points on the list ( [ 15 ] ) often occur in pairs . for instance , for manifolds which satisfy the gkm hypotheses , these pairs of curves correspond to the edges @xmath51 of the graph @xmath50 . that is , each @xmath133 in ( [ 14 ] ) contains of a pair of gradient curves joining the two vertices of @xmath51 . hence , the @xmath74 version of @xmath127 is identically zero . one can obtain duistermaat s result by exploiting this phenomenon . the goal of this article is to push these ideas forward by systematically applying these techniques to the equivariant setting . * acknowledgments*. the authors would like to thank the anonymous referees for useful comments and suggestions . recall @xmath134 is a symplectic manifold with a hamiltonian action @xmath7 of a torus @xmath2 . suppose further that there is an anti - symplectic involution @xmath16 with the property that @xmath135 let @xmath18 be the fixed point set of @xmath3 . we call @xmath5 the _ real locus _ of @xmath0 . recall that duistermaat proved the following equality , computing the ordinary cohomology of the real locus . @xmath136 where @xmath20 is the set for which @xmath3 preserves @xmath13 , and the @xmath14 are the indices of the fixed point sets @xmath13 . we will prove the equivariant analogue ( [ 6 ] ) to this equality , computing the additive structure of the equivariant cohomology : @xmath137 proof of theorem [ th : additive ] consider the product action of @xmath104 on @xmath138 in which each @xmath48 factor acts by multiplication on the corresponding factor of @xmath139 . this is a hamiltonian action . if @xmath140 is the moment map associated with @xmath7 , then the moment map of this product action is @xmath141 , with @xmath142 let @xmath143 . if @xmath144 for every @xmath103 , then @xmath145 and @xmath146 are equivariantly diffeomorphic , so the reduced space @xmath147 is diffeomorphic to @xmath148 . moreover , there is another action of @xmath104 on @xmath149 , namely @xmath7 coupled with the trivial action on @xmath150 . since this commutes with the product action , it induces a hamiltonian action of @xmath104 on @xmath151 . in addition , one gets from @xmath3 an involution @xmath152 of @xmath149 . this induces an anti - symplectic involution @xmath153 on @xmath151 . thus , one can apply duistermaat s theorem to @xmath151 to get a formula for the cohomology of the space @xmath154 in terms of the cohomology of the spaces @xmath155 now @xmath156 is obtained from @xmath157 by attaching cells of dimension @xmath158 and higher . so , for fixed @xmath159 , the sequence @xmath160 stabilizes as @xmath158 grows large , and moreover is equal to the equivariant cohomology of @xmath5 . thus one obtains from ( [ duis ] ) the following real analogue : @xmath161 where @xmath162 . the goal of this section is to determine the structure of @xmath33 as a module over the ring @xmath163 $ ] . we do this by calculating the @xmath36-term of the leray - serre spectral sequence converging to @xmath33 and deducing , by dimensional considerations obtained in the previous section , that the spectral sequence must collapse . this then gives us the desired statement about the @xmath164-term and hence @xmath33 . recall that by definition , we have @xmath165 where @xmath30 is the total space of the universal @xmath24-bundle . denote this fiber product by @xmath166 . we then have a fibration @xmath167 with fiber @xmath5 . let @xmath168 denote the local coefficient system on @xmath29 associated to this fibration . the @xmath36-term we would like to compute is then @xmath169 the computation takes place in two steps . the first step , which is the technical heart of the argument , consists of carrying out the computation in the 1-dimensional case . the remainder of the proof consists of a relatively straightforward exercise in bookkeeping . [ lspect ] let @xmath170 , so that @xmath171 . then @xmath172 is generated over @xmath173 $ ] in degree zero by @xmath174 . moreover , the only relation is given by @xmath175 for @xmath176 and @xmath177 by definition , the cohomology @xmath172 that we would like to compute is the group cohomology @xmath178 with respect to the natural action of @xmath24 on @xmath179 we will henceforth pass back and forth between these two notations without comment . our goal is to understand the cohomology as a module over the cohomology ring @xmath180 corresponding to the trivial action of @xmath24 on @xmath181 denote the non - trivial element of @xmath25 by @xmath182 consider the @xmath24-module @xmath183 with @xmath24-action defined by the equation @xmath184 for all @xmath185 we then get a short exact sequence of @xmath24-modules @xmath186 where @xmath187 and @xmath188 . of course , it is completely essential to this identification of the cokernel of @xmath77 that we work over @xmath25 ; in order for this sequence to be exact , we must have @xmath189 , but this would not be a map of @xmath25-modules if we did not also have @xmath188 . we would like to consider the long exact cohomology sequence associated with ( [ eq : ses ] ) . observe that we have isomorphisms @xmath190\otimes_{{{{\mathbb z}/2{\mathbb z } } } } h^*(x)\\ & \cong & { \mathbb z}[{g_{{\mathbb r}}}]\otimes_{{\mathbb z } } h^*(x ) \\ & \cong & \operatorname{hom}_{{\mathbb z}}({\mathbb z}[{g_{{\mathbb r}}}],h^*(x))\end{aligned}\ ] ] of @xmath24-modules . the first isomorphism is more or less the definition of the left - hand side , the second follows from the fact that @xmath191 is 2-torsion , and the third follows from the finiteness of @xmath24 . thus , the module @xmath183 is co - induced and its higher cohomology vanishes . moreover , the map @xmath192 provides an isomorphism of @xmath191 with @xmath193 so we have @xmath194 and , of course , @xmath195 the long exact sequence in question therefore takes the following form . @xmath196 & { h^*(x)^{{g_{{\mathbb r}}}}}\ar[r ] & { h^*(x ) } \ar[r ] & { h^*(x)^{{g_{{\mathbb r } } } } } \ar ` r[d]`[l ] ` [ llld]`[d][dll ] & \\ & { h^1({g_{{\mathbb r}}};h^*(x ) ) } \ar[r ] & { 0 } \ar[r ] & { h^1({g_{{\mathbb r}}};h^*(x ) ) } \ar ` r[d]`[l ] ` [ llld]`[d][dll ] & \\ & { h^2({g_{{\mathbb r}}};h^*(x))}\ar[r ] & { 0 } \ar[r ] & \dots & } } } \ ] ] the first map in this sequence is the natural inclusion , and the second sends @xmath197 to @xmath198 therefore , @xmath199 is the sought - after quotient of @xmath174 by the subgroup of all elements of the form @xmath200 ; as a result , so is each @xmath201 for @xmath202 recall that @xmath203 $ ] where @xmath204 is a class of degree 1 . the desired result then follows from the fact that the connecting homomorphisms are multiplication by @xmath205 the remainder of the game consists in playing the results of lemma [ lspect ] and theorem [ th : additive ] off of one another . [ cintermediate ] the action of @xmath24 on @xmath191 is trivial . thus , in the case that @xmath206 theorem [ tfreemod ] holds . lemma [ lspect ] computes the @xmath36-term of the leray - serre spectral sequence converging to @xmath33 in the case @xmath207 now , the dimensions of the graded pieces of this @xmath36-term are maximized precisely when @xmath24 acts trivially on @xmath208 moreover , the results of the previous section tell us that the graded pieces of @xmath33 have exactly these maximal dimensions . since the @xmath209-term of a spectral sequence can only be as large as its @xmath36-term , this tells us that the action must be trivial and further that @xmath210 in this case , @xmath211 and the relation @xmath212 is automatically satisfied , so lemma [ lspect ] tells us that @xmath213 is a free module over @xmath214 $ ] generated in degree zero by @xmath208 this completes the proof of theorem [ tfreemod ] in the case @xmath207 the triviality of the action of @xmath24 on @xmath191 for higher - dimensional @xmath24 follows by restricting to arbitrary one - dimensional subtori . the fact that @xmath24 acts trivially on @xmath191 is not new ; it can also be derived from duistermaat s original argument . indeed , duistermaat s isomorphism can be seen to be @xmath24-equivariant , and the result then follows from the connectedness of the torus . using this fact would have somewhat simplified our argument , but we chose to give the above proof so as to avoid appealing to unpublished modifications of the literature . we now have the technical input to handle the general case . [ lgeneral ] let @xmath215 , for any positive integer @xmath1 . then @xmath172 is a free module over @xmath216 $ ] generated by @xmath217 . this is now completely classical . by corollary [ cintermediate ] , the action of @xmath24 on @xmath191 is trivial , so that @xmath168 is actually the constant sheaf @xmath191 . therefore , we have @xmath218\otimes_{{{{\mathbb z}/2{\mathbb z } } } } h^*(x).\ ] ] this completes the proof . our goal is now entirely within reach ; we need only combine the results we have proven so far as in the proof of corollary [ cintermediate ] to establish the @xmath34-module isomorphism @xmath219 proof of theorem [ tfreemod ] lemma [ lgeneral ] tells us that the @xmath36-term of the leray - serre spectral sequence converging to @xmath33 takes precisely the form that @xmath33 itself is asserted to have . however , the results of the previous section tell us that the graded pieces of @xmath33 have the same dimension as those of this @xmath36-term , and hence that the spectral sequence collapses . therefore , @xmath220 so @xmath33 is a free module over @xmath34 generated in dimension zero . thus , its additive structure is as given by ( [ eq : module ] ) . as a result of the collapse of the spectral sequence proved in the previous section , the map @xmath221 is an injection . in the case of the original manifold @xmath0 , the chang - skjelbred theorem @xcite identifies the image of this map . we prove a @xmath25 version of that theorem here . as usual , let @xmath6 be the @xmath1-dimensional `` real torus , '' @xmath0 a @xmath24-manifold , and @xmath222 the fixed point set of the action ; we let @xmath223 denote the inclusion . [ tchang ] suppose that @xmath224 is a free @xmath34-module . for a subgroup @xmath225 , let @xmath226 denote the inclusion . then we have @xmath227 our proof closely models the argument given in @xcite , with appropriate modifications . first of all , recall that @xmath228,\ ] ] with @xmath229 . moreover , we may view each @xmath230 as a linear functional @xmath231 , that is , an element of @xmath232 . this allows us to identify @xmath34 with the symmetric algebra @xmath233 , a fact analogous to the ordinary identification @xmath234 when @xmath4 is a torus with lie algebra @xmath235 . this allows us to view elements of @xmath34 as polynomial functions on @xmath24 . [ lkillk ] let @xmath236 be a subgroup and @xmath237 a @xmath24-equivariant map . if @xmath238 annihilates @xmath239 , then it must also kill @xmath224 . we have a sequence of @xmath24-equivariant maps @xmath240 notice that @xmath241 the sequence ( [ eq : sequence ] ) of @xmath24-spaces therefore gives rise to the following diagram of algebras . @xmath242 & { { h^*_{{g_{{\mathbb r}}}}}}\ar@{->}[l ] \\ & s({k_{{\mathbb r}}}^ * ) \ar@{=}[u ] & s({{g_{{\mathbb r}}}^*})\ar@{=}[u ] } \ ] ] therefore , the map @xmath243 defining the module structure factors as @xmath244 and the proof is complete . this observation furnishes us with the fundamental tool in proving localization theorems for equivariant cohomology . [ pcomplement ] let @xmath5 be a closed @xmath24-invariant submanifold of @xmath0 . for some positive integer @xmath245 , there exist subgroups @xmath246 of @xmath4 , each of which is an isotropy subgroup of some point @xmath247 , such that for any @xmath248 with @xmath249 , the product @xmath250 kills @xmath251 . let @xmath252 be a @xmath24-invariant tubular neighborhood of @xmath5 ; it suffices to prove the desired result for the module @xmath253 . now , given any orbit @xmath254 in @xmath255 with isotropy subgroup @xmath256 , we may find a @xmath24-invariant open neighborhood @xmath257 of @xmath254 admitting a @xmath24-equivariant map @xmath258 . by compactness , we may cover @xmath255 by finitely many such sets @xmath259 . we now show by induction that for all @xmath260 , if @xmath261 are elements of @xmath232 with @xmath249 , then @xmath262 annihilates @xmath263 . the case @xmath264 is simply a restatement of lemma [ lkillk ] . for the inductive step , consider the meyer - vietoris sequence associated to the cover @xmath265 . denoting @xmath266 by @xmath267 , we find an exact sequence @xmath268 now , since @xmath269 , we have a @xmath24-equivariant map @xmath270 , and so the left - hand term of the sequence is killed by @xmath271 . meanwhile , by induction , the right - hand term is annihilated by @xmath272 , and so the product @xmath273 kills the middle term . we will also need a relative version of the same result . [ prelative ] under the hypotheses of proposition [ pcomplement ] , the module @xmath274 is annihilated by the element @xmath275 . of course , the map of pairs @xmath276 is an equivalence , so it suffices to compute @xmath277 . once again , let @xmath252 be a @xmath24-equivariant tubular neighborhood of @xmath5 . we cover @xmath278 by two open sets @xmath279 and @xmath280 ; since , the projection map @xmath281 is an equivalence , so we may identify @xmath274 with the kernel of the map @xmath282 . now , let us write the meyer - vietoris sequence for this cover . @xmath283 but by the above discussion , this gives rise to the following exact sequence . @xmath284 now , proposition [ pcomplement ] applies to both ends of this sequence , so the middle term is killed by @xmath285 . proposition [ prelative ] gives us the basic localization results we will need . [ cloc ] in the setting of proposition [ pcomplement ] , the kernel and cokernel of the map @xmath286 are annihilated by the same element @xmath287 . simply apply proposition [ prelative ] to the exact sequence @xmath288 [ co : inject ] the kernel of the map @xmath289 is torsion , and hence trivial when @xmath224 is a free module . [ killcok ] for every subgroup @xmath236 , there exists a monomial @xmath290 annihilating the cokernel of the map @xmath291 such that no @xmath292 vanishes on @xmath239 . let @xmath293 , and let @xmath294 be its isotropy subgroup . since obviously @xmath295 , there is an @xmath296 with @xmath297 and @xmath298 . but by corollary [ cloc ] , we can find a monomial which is a product of elements of the form @xmath299 and annihilates @xmath300 . finally , we are ready to prove our main result . proof of theorem [ tchang ] first of all , since the map @xmath103 factors as @xmath301 we know that for all @xmath83 , the inclusion @xmath302 holds . for the other direction , recall first that @xmath224 is free ; therefore , by corollary [ co : inject ] the map @xmath64 is injective and we may consequently view @xmath224 as a submodule of @xmath303 . suppose @xmath304 is an @xmath233-basis for @xmath224 . by corollary [ killcok ] , there is a monomial @xmath305 with @xmath306 for every @xmath307 . thus , we may write @xmath308 for unique @xmath309 . now , since @xmath233 is a unique factorization domain , we may divide both sides of this identity by @xmath53 and cancel common factors to obtain the formula @xmath310 where the @xmath311 are uniquely - determined elements of @xmath233 and the @xmath90 are uniquely - determined divisors of @xmath53 such that @xmath311 and @xmath90 are relatively prime . suppose now that @xmath312 were actually in @xmath313 . we may find a subset @xmath314 such that no @xmath315 kills @xmath83 , and @xmath316 annihilates the cokernel of the map @xmath317 therefore , multiplying both sides of ( [ factor ] ) by @xmath54 , we find that @xmath318 with @xmath319 . thus , in ( [ factor ] ) , none of the weights @xmath320 that divide the denominators @xmath90 vanish on @xmath83 . hence , if @xmath321 for all @xmath83 , then @xmath322 for all @xmath103 , and so ( [ factor ] ) tells us that @xmath323 and the proof is complete . now suppose that @xmath95 is a connected component of @xmath324 for some subgroup @xmath83 of @xmath24 of order @xmath94 . let @xmath325 be the inclusion @xmath326 of the fixed points of @xmath95 into @xmath95 . let @xmath327 be the inclusion @xmath328 of the fixed points of @xmath95 into all of the fixed points . then we have the following corollary of theorem [ tchang ] . [ co : gencs ] suppose that @xmath224 is a free @xmath34-module . a class @xmath329 is in the image of @xmath64 if and only if @xmath330 for every subgroup @xmath83 of @xmath24 of order @xmath94 and every connected component @xmath95 of @xmath324 . the proof is analogous to the proof of theorem @xmath331 in @xcite . it follows directly from theorem [ tchang ] . the goal of this section is to prove an analogue of theorem [ th : gkm ] for the real locus @xmath5 of @xmath0 . the proof will require two hypotheses on @xmath5 , namely @xmath332 and @xmath333 where @xmath334 is the one - skeleton of @xmath0 and @xmath335 the one - skeleton of @xmath5 . we will begin by analyzing these conditions and their implications . we first note that the analogues for @xmath0 of the conditions ( [ eq : fixed ] ) and ( [ eq:1skel ] ) , namely @xmath336 and @xmath337 have a very simple and elegant interpretation in terms of the isotropy representations of @xmath4 at fixed points of @xmath0 . [ th : gkmhyp ] the conditions @xmath132 and @xmath338 are satisfied if and only if , for every @xmath339 , the weights @xmath340 , @xmath341 of the isotropy representation of @xmath4 on @xmath342 are pair - wise linearly independent , that is for @xmath343 , @xmath340 is not a multiple of @xmath344 . for the proof of this , see @xcite . when @xmath0 satisfies the two conditions and , we say that @xmath0 is a _ gkm manifold_. let @xmath345 be the weight lattice of @xmath4 . by the @xmath74 reduction of a weight @xmath346 , we mean its image @xmath197 in @xmath347 . we will prove a real analogue of theorem [ th : gkmhyp ] . [ th : realrules ] suppose @xmath0 satisfies the hypotheses of theorem [ th : gkmhyp ] . then the conditions @xmath348 and @xmath349 are satisfied if and only if , for every @xmath339 , the @xmath74 reduced weights , @xmath350 , are all distinct and non - zero . let @xmath351 be a connected component of @xmath222 . then @xmath351 is a @xmath4-invariant symplectic submanifold of @xmath0 , and the action of @xmath4 on it is hamiltonian , so it contains at least one @xmath4-fixed point @xmath53 . however , the hypotheses above imply that the linear isotropy action of @xmath24 on @xmath342 has no fixed points other than the origin . hence , @xmath352 and @xmath353 . this argument applies to all the connected components of @xmath222 , hence the connected components are just the fixed points of @xmath4 , and thus @xmath348 . the proof that @xmath349 is similar . let @xmath83 be a subgroup of @xmath24 of index @xmath41 , and let @xmath351 be a connected component of @xmath324 . then @xmath351 is a @xmath4-invariant submanifold of @xmath0 , and because @xmath354 , it is also @xmath3-invariant . let @xmath355 be a @xmath4-fixed point , and let @xmath356 be the decomposition of @xmath342 into the @xmath41-dimensional weight spaces corresponding to the @xmath340 . by the hypotheses on the reduced weights @xmath350 , either @xmath357 in which case @xmath353 , or @xmath358 for some @xmath103 . let @xmath359 be the character of @xmath4 associated with the representation of @xmath4 on @xmath360 and let @xmath361 . then @xmath362 and @xmath363 thus , by ( [ eq : tang ] ) , @xmath351 is the connected component of @xmath364 containing @xmath53 , and in particular , @xmath351 is contained in @xmath334 . thus , @xmath365 applying this argument to all index @xmath41 subgroups @xmath83 of @xmath24 and all connected components of the fixed point sets of these groups , one obtains the inclusion @xmath366 the reverse inclusion is obvious . this completes the proof . the hypotheses of theorem [ th : realrules ] impose some rather severe restrictions on the manifold @xmath0 . for instance , the cardinality of the set of @xmath74 reduced weights , @xmath347 , is @xmath367 . therefore , since the reduced weights @xmath350 are distinct and non - zero for @xmath368 , we must have that @xmath369 . hence , @xmath370 for example , if @xmath371 , then @xmath372 . this leads us to make the following definition . if @xmath0 is a gkm manifold , and if for every @xmath373 , the @xmath74 reduced weights , @xmath350 , are all distinct and non - zero , we will say that @xmath0 is a _ @xmath74 gkm manifold_. next , we show that relatively few compact _ homogeneous _ symplectic manifolds ( e.g. coadjoint orbits ) are @xmath74 gkm manifolds . consider coadjoint orbits of the classical compact simple lie groups associated with the dynkin diagrams @xmath374 , @xmath375 , @xmath376 and @xmath377 . let @xmath378 , for @xmath379 , be the standard basis vectors of @xmath380 . the positive roots associated to the dynkin diagram @xmath374 consist of @xmath381 so their @xmath74 reductions are distinct and non - zero . however , for @xmath375 , @xmath376 , and @xmath377 , this list of positive roots contains @xmath382 so we conclude each coadjoint orbit of @xmath383 is a @xmath74 gkm space . however , for other compact simple lie groups , _ no _ maximal coadjoint orbit can be a @xmath74 gkm space . on the other hand , on a more positive note , one has if @xmath0 is a non - singular projective toric variety , then it is a @xmath74 gkm space . if @xmath0 is a non - singular toric variety , the weights @xmath340 , @xmath379 , are a @xmath384-basis for @xmath345 , so their images in @xmath347 are a @xmath25 basis of @xmath347 . this theorem , combined with theorem [ th : realgkm ] , gives us a new description of the equivariant cohomology of a real toric variety . the ordinary and @xmath24-equivariant cohomology of these real loci has been computed by davis and januszkiewicz @xcite . their description of these rings is analogous to danilov s description of the ordinary and @xmath4-equivariant cohomology of the original toric varieties . we will now prove a real locus version of the gkm theorem with @xmath25 coefficients . recall from section [ se : intro ] that theorem [ th : gkm ] of gkm characterizes the image of @xmath385 in terms of the weights of the isotropy representations of @xmath4 on the tangent spaces at the fixed points . to prove an analogue of this for the real locus of a symplectic manifold , we must first compute the @xmath25-equivariant cohomology with @xmath25 coefficients of @xmath386 . recall that @xmath48 acts on @xmath49 by @xmath387=[z_o : e^{i\theta}z_1]$ ] . this is a hamiltonian action , with respect to the fubini - study symplectic form on @xmath49 . furthermore , complex conjugation is an anti - symplectic involution on @xmath49 , with fixed point set @xmath386 . there is a residual action of @xmath25 on @xmath388 which reflects @xmath48 about the @xmath389-axis . [ rp1 ] let @xmath390 and @xmath391 denote the fixed points of the @xmath25 action on @xmath386 . then the image of the map @xmath392 is the set of pairs @xmath393 such that @xmath394.\ ] ] it is clear that the constant functions are equivariant classes in @xmath395 furthermore , we know that @xmath396 , and so these are the only equivariant classes . finally , @xmath397 for @xmath398 , and so indeed , the condition stated is the only condition of pairs @xmath399 . theorem [ th : realgkm ] identifies the image of the map @xmath400 in terms of weights of isotropy representations of @xmath24 on the tangent spaces at the fixed points . proof of theorem [ th : realgkm ] the result follows immediately from corollary [ co : gencs ] and lemma [ rp1 ] . the results of this section and the previous section have been proved independently by schmid @xcite . schmid uses an equivariant morse theoretic approach , and consequently the proofs are quite different . as a result of equivariant formality , we get two corollaries of theorem [ th : realgkm ] concerning the relation between the ring structure of the cohomology of @xmath0 and the cohomology of @xmath5 . [ co : ring1 ] suppose that @xmath0 is a gkm manifold and a @xmath74 gkm manifold . then there is a graded ring isomorphism @xmath401 [ co : ring2 ] suppose that @xmath0 is a gkm manifold and a @xmath74 gkm manifold . then there is a graded ring isomorphism @xmath402 note that this last corollary strengthens duistermaat s original result from an isomorphism of vector spaces to an isomorphism of rings . some of the results of this section , most importantly theorem [ th : realgkm ] , are valid not only for the real locus @xmath5 of a hamiltonian @xmath4-manifold , but more generally for any compact @xmath24-manifold @xmath5 which satisfies the following properties : 1 . @xmath5 is equivariantly formal ; 2 . @xmath403 is finite ; and 3 . the weights of @xmath5 satisfy the properties of a @xmath404 gkm manifold . in this situation , we may still characterize the structure of the one - skeleton . theorem [ th : realgkm ] still follows from injectivity and the chang - skjelbred theorem . goldin and holm generalize theorem [ th : gkm ] to the case where the one - skeleton has dimension at most @xmath82 . the goal of this section is to prove a real version of the goldin - holm theorem with @xmath25 coefficients . again , we require the hypotheses that the @xmath405-fixed points of the real locus are the same as the @xmath4-fixed points of @xmath0 as in ( [ eq : fixed ] ) ; and that the real locus of the one - skeleton is the same as the one - skeleton of the real locus , as in ( [ eq:1skel ] ) . finally , we require @xmath406 and @xmath407 if a manifold satisfies these last two hypotheses , we will say that it is a _ gh manifold_. these hypotheses have a nice interpretation in terms of the isotropy representations of @xmath4 at the fixed points of @xmath0 . [ th : ghrealrules ] the conditions @xmath132 and @xmath408 are satisfied if and only if the weights @xmath350 of the isotropy representation of @xmath4 on @xmath342 have the property that every three span a vector subspace of dimension at least two . these hypotheses on @xmath0 have real analogues , namely that @xmath409 and the one - skeleton @xmath335 of @xmath5 is at most @xmath41-dimensional . we will state without proof the following real analogue of theorem [ th : ghrealrules ] . suppose that @xmath0 satisfies the hypotheses of theorem [ th : ghrealrules ] . if the conditions @xmath410 and @xmath411 are satisfied , then for every @xmath339 , the @xmath74 reduced weights @xmath412 are all non - zero , and each element of @xmath413 $ ] appears no more than twice . the proof of this theorem is nearly identical to that of theorem [ th : realrules ] . the hypotheses of this theorem , although weaker than those of theorem [ th : realrules ] , still impose restrictions on the manifold @xmath0 . the cardinality of the set of @xmath414 reduced weights is @xmath367 . since the weights are non - zero , and each weight can appear at most twice , @xmath415 and so @xmath416 for instance , if @xmath371 , @xmath417 . we will now show an example where the condition that the reduced weights be non - zero is _ not _ satisfied . consider @xmath418 with homogeneous coordinates @xmath419 $ ] . let @xmath420 act on @xmath418 by @xmath421 = [ e^{-i\theta}z_0:z_1:e^{i\theta}z_2].\ ] ] this action has three fixed points : @xmath422 $ ] , @xmath423 $ ] and @xmath424 $ ] . the weights at these fixed points are @xmath425 $ } & x , 2x \\ \mbox{$p_2=[0:1:0]$ } & -x , x \\ \mbox{$p_3=[0:0:1]$ } & -2x , -x \end{array}\ ] ] where we have identified @xmath426 with degree one polynomials in @xmath427 $ ] . as cohomology elements , these are assigned degree two . using theorem [ th:4total ] below , we can compute the @xmath48 equivariant cohomology of @xmath418 as follows . the image of the equivariant cohomology @xmath428 in @xmath429\ ] ] is the subalgebra generated by the triples of functions @xmath430 such that @xmath431 \mbox { for every $ i$ and $ j$ , and } \ ] ] @xmath432.\ ] ] however , when we try to compute the @xmath25 equivariant cohomology of @xmath433 , the real locus of @xmath418 , we run into a problem . the @xmath74 reduced weights are given in the table below . @xmath425 $ } & x , 0 , \\ \mbox{$p_2=[0:1:0]$ } & x , x , \\ \mbox{$p_3=[0:0:1]$ } & 0 , x. \end{array}\ ] ] the problem with this @xmath25 action on @xmath433 is that it no longer has isolated fixed points . there is an entire @xmath386 which is fixed by this @xmath25 action . thus , we can not compute the @xmath25 equivariant cohomology of @xmath433 using these methods . we make the following definition , analogous to the definition of @xmath70 gkm manifolds given in section [ se : realgkm ] . suppose that @xmath0 is a gh manifold , and furthermore that @xmath410 and @xmath411 . in this case , we will say that @xmath0 is a _ @xmath74 gh space_. recall the following properties about the @xmath4-equivariant cohomology of manifolds with one - skeleta of dimension at most @xmath82 . these are proved in @xcite , although the reader is cautioned to the different notation used there . first , we compute the @xmath48 equivariant cohomology of a @xmath82-manifold , and then we use this computation to determine the equivariant cohomology of any manifold with one skeleton of dimension at most @xmath82 . [ pr : gkmin4 ] let @xmath5 be a compact , connected symplectic 4-manifold with an effective hamiltonian @xmath48 action with isolated fixed points @xmath434 . the map @xmath435 induced by inclusion is an injection with image @xmath436,\ \sum_{i=1}^d\frac{f_i}{\alpha_1^i\alpha_2^i}\in s({\ensuremath{\mathfrak{s}}}^*)\right.\right\},\ ] ] where @xmath437 and @xmath438 are the ( linearly dependent ) weights of the @xmath439 isotropy action on @xmath440 . [ th:4total ] let @xmath0 be a compact , connected symplectic manifold with an effective hamiltonian @xmath4-action . suppose further that the @xmath4-action has only isolated fixed points @xmath87 and that the one skeleton has dimension at most @xmath82 . let @xmath441 denote the restriction of @xmath442 to the fixed point @xmath90 . the image of the injection @xmath443 is the subalgebra of functions @xmath444 which satisfy @xmath445 for all @xmath446 codimension-1 tori and all connected components @xmath447 of @xmath364 , where @xmath97 ; @xmath98 are the ( linearly dependent ) weights of the @xmath4 action on @xmath448 ; and @xmath449 is inclusion . we can use these computations to compute the @xmath405 equivariant cohomology of a @xmath74 gh manifold . [ le : realgkmin4 ] let @xmath0 be a compact , connected symplectic 4-manifold with an effective hamiltonian @xmath48 action with isolated fixed points @xmath450 . suppose further that @xmath0 is a @xmath74 gh manifold with real locus @xmath5 . the map @xmath451 induced by inclusion is an injection with image @xmath452\ \left| \begin{array}{l}\ f_i - f_j\in x\cdot{{{\mathbb z}/2{\mathbb z}}}[x],\\ \sum_{i=1}^d\frac{f_i}{\alpha_1^i\alpha_2^i}\in { { { \mathbb z}/2{\mathbb z}}}[x]\end{array}\right.\right\},\ ] ] where @xmath437 and @xmath438 are the linearly dependent weights of the @xmath25 isotropy representation on @xmath440 . ( in this case , @xmath453 . ) the map @xmath64 is injective because @xmath5 is equivariantly formal . we know that the @xmath454 must satisfy the first condition because the functions constant on all the vertices are the only equivariant classes in degree @xmath455 , as @xmath456 . the second condition is necessary as a direct result of the @xmath25 version of the localization theorem proved in section [ se : cs ] . notice that this condition gives us one relation in degree @xmath331 cohomology . a dimension count shows us that these conditions are sufficient . as an @xmath457-module , @xmath458 . thus , the equivariant poincar polynomial is @xmath459 as @xmath460 is generated in degree 1 , the @xmath461 degree 1 classes given by the @xmath462 subject to the localization condition generate the entire cohomology ring . thus we have found all the conditions . we now prove theorem [ th : realgh ] , computing the cohomology of any @xmath74 gh manifold . we will show that the image of @xmath463 is the subalgebra of functions @xmath464 which satisfy @xmath465 for all subgroups @xmath83 of @xmath24 of order @xmath94 and all connected components @xmath95 of @xmath96 . proof of theorem [ th : realgh ] this follows immediately from corollary [ co : gencs ] and lemma [ le : realgkmin4 ] . there are two immediate corollaries in this setting , analogous to corollaries [ co : ring1 ] and [ co : ring2 ] . [ co : ring3 ] suppose that @xmath0 is a gh manifold , and that @xmath69 and @xmath466 . then there is a graded ring isomorphism @xmath401 [ co : ring4 ] suppose that @xmath0 is a gh manifold , and that @xmath69 and @xmath466 . then there is a graded ring isomorphism @xmath402 consider the @xmath25 action on @xmath104 , which reflects each copy of @xmath48 . then the equivariant cohomology ring @xmath467 classifies all possible orientifold configurations of type ii string theories , compactified on @xmath104 . see section 3 and appendix c of @xcite for more details . yang - hui he pointed this example out to us . using the results of section [ se : realgkm ] , we can now compute this equivariant cohomology . first , we recognize @xmath104 as the real locus of @xmath468 . this space @xmath0 has a natural @xmath104 action , where the @xmath103th copy of @xmath48 acts in the standard fashion on the @xmath103th copy of @xmath49 . we can compute the @xmath405-equivariant cohomology of this space quite easily . the gkm graph associated to @xmath469 with the @xmath104 action described above is the @xmath1-dimensional hypercube . the vertices correspond to the binary words of length @xmath1 . two binary words are connected by an edge if they differ in exactly one bit . suppose @xmath121 and @xmath470 differ in exactly the @xmath103th bit . then the weight associated to the edge @xmath471 is @xmath230 . thus , when @xmath472 , the gkm graph and weights are shown in the figure below . notice that the reduced weights are all non - zero and are distinct in @xmath473 . thus , we can apply theorem [ th : realgkm ] to compute @xmath474 that is , the equivariant cohomology is the set of functions @xmath475 $ ] such that for every edge @xmath476 , we have @xmath477.\ ] ] we can now consider the copy of @xmath25 sitting diagonally inside @xmath405 . this copy of @xmath25 acts on @xmath104 , and this is the action that originally interested physicists . we can now compute the @xmath25-equivariant cohomology simply by projecting @xmath478\to { { { \mathbb z}/2{\mathbb z}}}[x]=s(({{{\mathbb z}/2{\mathbb z}}})^*)\ ] ] where @xmath230 gets sent to @xmath204 . we will describe below a ( somewhat idealized ) real world application of the results of section [ se : realgkm ] . let @xmath480 be a list of companies whose stocks are being traded in the stock market . let @xmath481 for @xmath368 be investors , and let @xmath482 be the portfolio of @xmath481 . suppose that for certain pairs of investors @xmath481 and @xmath483 , the symmetric difference @xmath484 i.e. the relative status of the portfolios of @xmath481 and @xmath483 , is given . to what extent does this information determine the portfolios @xmath485 ? it is easy to see that it can not _ uniquely _ determine the @xmath485 s . if @xmath486 is one solution to ( [ eq:8.1 ] ) , one gets another solution by taking a fixed subset @xmath487 of @xmath391 and replacing the @xmath485 s by the symmetric difference @xmath488 therefore , we will slightly rephrase this question . let @xmath166 be a collection of two - element subsets of @xmath391 , and suppose that for every @xmath489 , one is given a subset @xmath490 of @xmath391 . list all solutions @xmath486 of the boolean identities ( [ eq:8.1 ] ) . we will say two solutions are identical if they satisfy ( [ eq:8.2 ] ) . note , by the way , that ( [ eq:8.2 ] ) can be rewritten @xmath491 hence , we have defined an equivalence relation . one can inject an element of randomness into this problem by positing that , for @xmath492 , @xmath493 in other words , either the symmetric difference is given by ( [ eq:8.1 ] ) or @xmath494 . again , the problem is to list all possibilities for the @xmath485 s . clearly , the solutions of ( [ eq:8.3 ] ) contain the solutions to ( [ eq:8.1 ] ) ; so by solving ( [ eq:8.3 ] ) , one gets an upper bound on the number of solutions to ( [ eq:8.1 ] ) . moreover , there are a lot of trivial solutions of ( [ eq:8.3 ] ) , namely , @xmath495 where @xmath487 is , as above , a fixed subset of @xmath391 . these can immediately be discarded as potential solutions of ( [ eq:8.1 ] ) . there is an elegant way of reformulating ( [ eq:8.1 ] ) and ( [ eq:8.3 ] ) in the language of @xmath496 arithmetic . let @xmath50 be the graph whose vertices are the @xmath481 s and whose edges are the members of @xmath166 . for every edge @xmath489 , let @xmath497 be the element of @xmath498 whose @xmath159th coordinate is @xmath331 if and only if @xmath499 and @xmath292 the element of @xmath498 whose @xmath159th coordinate is @xmath331 if and only if @xmath500 . then ( [ eq:8.1 ] ) is equivalent to @xmath501 and ( [ eq:8.3 ] ) is equivalent to @xmath502 in particular , ( [ eq:8.3 ] ) becomes an identity of the type described in theorem [ th : realgkm ] , the multiple of @xmath497 on the right being in the degree zero component of @xmath38 $ ] . now let @xmath0 be a gkm manifold acted on by an @xmath1-torus @xmath503 , and let @xmath50 be its associated graph . if the @xmath497 s in ( [ eq:8.6 ] ) are the weights of @xmath24 which were defined in theorem [ th : realgkm ] the results of section [ se : realgkm ] tell us that solutions of ( [ eq:8.6 ] ) can be identified with elements of @xmath504 we will use this observation to determine the solutions of ( [ eq:8.6 ] ) in a couple simple , but interesting , examples . * the complete graph on @xmath1 vertices . * consider @xmath505 as a @xmath104 manifold ( ignoring the fact that the diagonal subgroup of @xmath104 acts trivially ) . the corresponding graph is the complete graph on @xmath1 vertices : its vertices are @xmath506 and every pair of vertices is joined by an edge . the weights @xmath507 from theorem [ th : realgkm ] are just @xmath508 , so the conditions ( [ eq:8.1 ] ) become @xmath509 the solutions of ( [ eq:8.7 ] ) are in one to one correspondence with the elements of @xmath510 or , alternatively , of @xmath511 the elements of the first summand correspond to the trivial solutions of ( [ eq:8.7 ] ) . so if we identify solutions which are equivalent in the sense of ( [ eq:8.2 ] ) , the non - trivial solutions of ( [ eq:8.7 ] ) correspond to non - zero elements of @xmath512 . however , @xmath513 , so there is just one non - trivial solution up to equivalence , and it is given by @xmath514 this is also the unique solution of ( [ eq:8.1 ] ) up to equivalence . * the permutahedron . * let @xmath0 be the complex flag variety @xmath515 , and consider the @xmath104 action on @xmath0 by left multiplication . the graph associated with @xmath0 is the _ permutahedron_. its vertices are the elements of the symmetric group , @xmath105 , and two vertices @xmath3 and @xmath7 are joined by and edge if @xmath106 is a transposition . if @xmath312 is the edge joining @xmath3 to @xmath7 and @xmath106 is the transposition switching @xmath103 and @xmath516 , then as in the previous example , @xmath517 , so the conditions ( [ eq:8.1 ] ) become @xmath518 as in the previous example , the non - trivial solutions to ( [ eq:8.3 ] ) can be identified with the non - zero elements of @xmath519 , and since @xmath520 is the real flag variety , @xmath521 if one thinks of @xmath522 as the quotient of @xmath498 by the diagonal subgroup @xmath523 with @xmath524 , the solutions corresponding to @xmath525 is given by @xmath526 where @xmath527 if and only if the @xmath159th coordinate of @xmath197 is @xmath331 . notice , by they way , if we replace @xmath197 by @xmath528 , ( [ eq:8.11 ] ) becomes @xmath529 where @xmath530 and @xmath531 are the complements of @xmath107 and @xmath487 in @xmath391 , so ( [ eq:8.11 ] ) and ( [ eq:8.12 ] ) are equivalent . it is easy to see by inspection that none of the solutions ( [ eq:8.11 ] ) of ( [ eq:8.3 ] ) are also solutions of ( [ eq:8.1 ] ) . an interesting special case of this example is the complete bipartite graph @xmath532 . in this case , the vertices of @xmath50 are @xmath533 , @xmath534 , @xmath535 and @xmath536 , @xmath537 , and @xmath538 ; the edges are all pairs @xmath539 ; and the sets @xmath490 are @xmath540 in this example , ( [ eq:8.3 ] ) has one non - trivial solution , up to equivalence , namely @xmath541 .

we study a cooperative network with a buffer - aided multi - antenna source , multiple half - duplex ( hd ) buffer - aided relays and a single destination . such a setup could represent a cellular downlink scenario , in which the source can be a more powerful wireless device with a buffer and multiple antennas , while a set of intermediate less powerful devices are used as relays to reach the destination . the main target is to recover the multiplexing loss of the network by having the source and a relay to simultaneously transmit their information to another relay and the destination , respectively . successive transmissions in such a cooperative network , however , cause inter - relay interference ( iri ) . first , by assuming global channel state information ( csi ) , we show that the detrimental effect of iri can be alleviated by precoding at the source , mitigating or even fully canceling the interference . a cooperative relaying policy is proposed that employs a joint precoding design and relay - pair selection . note that both fixed rate and adaptive rate transmissions can be considered . for the case when channel state information is only available at the receiving side ( csir ) , we propose a relay selection policy that employs a phase alignment technique to reduce the iri . the performance of the two proposed relay pair selection policies are evaluated and compared with other state - of - the - art relaying schemes in terms of outage and throughput . the results show that the use of a powerful source can provide considerable performance improvements . cooperative networks , opportunistic relaying , buffer - aided relays , precoding , interference cancellation .

1703.03338

the communication paradigm of cooperative relaying has recently received considerable attention due to its effectiveness in alleviating the effects of multipath fading , pathloss and shadowing , and its ability to deliver improved performance in cognitive radio systems and wireless sensor networks . relay - assisted cellular networks are a promising solution for enhancing coverage and are included in standards , such as ieee 802.16j / m and long term evolution - advanced ( lte - a ) . several protocols for cooperative relaying were presented in @xcite where the gains in transmit and receive diversity were studied . in multi - relay networks , simultaneous transmissions by the relays are in general difficult to handle ; towards this end , opportunistic relay selection has been suggested in @xcite to improve the resource utilization and to reduce the hardware complexity . stemming from the relay selection concept , many studies have proposed improved selection techniques ( see , _ e.g. _ , @xcite ) . traditional hd relaying schemes partition the packet transmission slot into two phases , where the transmission on the source - relay @xmath0 link happens in the first phase , and the transmission on the relay - destination @xmath1 link occurs in the second phase . however , this relaying scheme limits the maximum achievable multiplexing gain to @xmath2 , which also results in bandwidth loss . in order to overcome such multiplexing and bandwidth limitations , several techniques have been proposed in the literature ( see , for example , @xcite and references therein ) . among them , the successive relaying scheme in @xcite incorporates multiple relay nodes and allows concurrent transmissions between source - relay and relay - destination to mimic an ideal full - duplex transmission . however , this scheme targets scenarios with a long distance between the relays and thus inter - relay interference is not considered . an extension of this work is discussed in @xcite , where the authors assume that iri is strong ( in co - located or clustered relays ) and can always be decoded at the affected nodes ; this decoded iri is exploited in a superposition coding scheme that significantly improves the diversity - multiplexing trade - off performance of the system . in earlier studies , relays were assumed to lack data buffers and relay selection was mainly based on the @xmath3 criterion and its variations ( see , for example , @xcite ) . here , the relay that receives the source signal is also the one that forwards the signal to the destination . with the adoption of buffer - aided relays , this coupling is broken , since different relays could be selected for transmission and reception , thus allowing increased degrees of freedom . buffering at the relay nodes has been shown to be a promising solution for cooperative networks and motivates the investigation of new protocols and transmission schemes ( see @xcite and @xcite for an overview ) . @xcite proposed a novel criterion based on @xmath4 relay selection ( mmrs ) , in which the relay with the best source - relay @xmath0 link is selected for reception and the relay with the best relay - destination @xmath1 link is selected for transmission on separate slots . in @xcite , at each slot the best link is selected among _ all _ the available @xmath0 and @xmath1 links , as a part of the proposed @xmath5 policy , thus offering an additional degree of freedom to the network , while buffer - aided link selection was studied in topologies with source - destination connectivity ( see , _ e.g. , _ in @xcite ) , resulting in improved diversity and throughput . more recently , to alleviate the throughput loss of hd relaying , significant focus has been given to minimizing the average packet delay . towards this end , various works have proposed hybrid solutions combining the benefits of mmrs and @xmath5 @xcite , the use of broadcasting in the @xmath0 link @xcite , the prioritization of @xmath1 transmissions @xcite or the selection of the relay with the maximum number of packets residing in its buffer @xcite . however , it was shown that a trade - off exists between delay performance and the diversity of transmission as the number of relays with empty or full buffers increases . to improve throughput and reduce average packet delays , a number of works proposed to employ non - orthogonal successive transmissions by the source and a selected relay . in order to recover the hd multiplexing loss , @xcite suggests to combine mmrs with successive transmissions ( sfd - mmrs ) . as the proposed topology aims to mimic full - duplex relaying , different relays are selected in the same time slot ; however , relays are assumed to be isolated and the effect of iri is ignored . more practical topologies were studied in @xcite where iri exists and is not always possible to be cancelled . for fixed rate transmission , the proposed @xmath6 solution proposed in @xcite , combining power adaptation and interference cancellation , provides a performance close to the upper bound of sfd - mmrs . in addition , kim and bengtsson @xcite proposed buffer - aided relay selection and beamforming schemes taking the iri into consideration ; they consider a model which can be regarded as an example of relay - assisted device - to - device ( d2d ) communications , where the source and destination are low - cost devices with a single antenna and the base stations comprise more powerful relays with buffers and multiple antennas . numerical results show that their approach outperforms sfd - mmrs when interference is taken into consideration , and when the number of relays and antennas increases they approach the performance of the interference - free sfd - mmrs , herein called the _ ideal _ sfd - mmrs . finally , the use of buffer aided relays has been considered in topologies wth full - duplex ( fd ) capabilities @xcite and decoupled non - orthogonal uplink and downlink transmissions @xcite , showing that throughput can be significantly improved due to the increased scheduling flexibility . in many practical considerations ( _ e.g. _ , wireless sensors ) , the relay nodes are hardware - limited to be hd while the source can be a more powerful wireless device with large buffers and multiple antennas . although this observation is not always true ( _ e.g. _ , in d2d communications @xcite ) , it is a reasonable and common practical scenario . towards this end , we study a network which consists of a buffer - aided source with one or multiple antennas , multiple hd buffer - aided relays and a destination . with this setup , we are able to approach ( achieve ) the performance of the ideal sfd - mmrs by adopting a buffer - aided source whose buffer retains replicas of the relay buffers , successfully transmitted data from the source to relays , to facilitate iri mitigation ( cancellation ) ; we relax the assumption of knowing the full csi and instead we allow for csir and limited feedback from the receiver to the transmitter ; under these conditions we propose a relay pair selection scheme that is based on partial phase alignment of signals . the rest of the paper is organized as follows . in section [ sec : model ] , the system model is outlined . the proposed relaying schemes for variable and fixed rate are presented in sections [ sec : policy1 ] and [ sec : policy2 ] , respectively . the performance of the proposed relaying policies in terms of outage and average throughput , along with comparisons with other state - of - the - art relaying schemes are presented in section [ sec : numerical ] . finally , conclusions and a discussion on future possible directions are drawn in section [ sec : conclusions ] . vectors are written in bold lower case letters and matrices in bold capital letters . @xmath7 , @xmath8 and @xmath9 denote the sets of real numbers , complex and natural numbers , respectively . for a matrix @xmath10 ( @xmath11 ) , @xmath12 denotes the entry in row @xmath13 and column @xmath14 . matrix @xmath15 denotes the identity matrix of appropriate dimensions . the trace of a square matrix @xmath16 is denoted by @xmath17 . @xmath18 and @xmath19 denote the transpose and hermitian transpose operations , respectively ; @xmath20 denotes the @xmath21-norm operation . the complex conjugate of a complex number @xmath22 is denoted by @xmath23 ; the real and imaginary parts of a complex number @xmath22 are denoted by @xmath24 and @xmath25 , respectively . we denote by @xmath26 the set of all possible relay - pairs in the relay network , and by @xmath27 its cardinality . a relay pair , denoted by @xmath28 belongs to set @xmath26 if and only if @xmath29 , the receiving relay @xmath30 is not full and the transmitting relay @xmath31 is not empty . we consider a cooperative network consisting of a buffer - aided source @xmath32 with multiple antennas , a set @xmath33 of @xmath34 hd decode - and - forward ( df ) relays with buffers , and a single destination @xmath35 . [ fig : model ] illustrates a simple example with a buffer - aided two - antenna source @xmath32 , two buffer - aided single - antenna hd df relays , and a single - antenna destination @xmath35 . to simplify the analysis , we examine the case where connectivity between the source and the destination is established only via the relays and ignore the direct @xmath36 link ( as in , _ e.g. _ , @xcite ) . the number of data elements in the buffer of relay @xmath37 is denoted by @xmath38 and its capacity by @xmath39 . for the fixed rate transmission policy , we assume that the packets are transmitted at a fixed rate of @xmath40 bits per channel use ( bpcu ) , and the data of each transmission occupies @xmath41 slot in the buffer . first , we provide the signals received at relay @xmath42 and destination @xmath35 . at the destination , at any arbitrary time - slot @xmath43 the following signal is received : @xmath44=h_{td}x[p]+w_d[n ] \ ; , \end{aligned}\ ] ] where @xmath45 $ ] is the signal received and stored in a previous time - slot @xmath46 in the buffer of the now transmitting relay @xmath47 , @xmath48 denotes the channel coefficient from the transmitting relay @xmath31 to the destination @xmath35 , and @xmath49 $ ] denotes the additive white gaussian noise ( awgn ) at the destination in the @xmath43-th time slot , i.e. , @xmath49\sim\mathcal{cn}(0,{\sigma_d^2})$ ] . it must be noted that @xmath45 $ ] was not necessarily received in the previous time - slot ( i.e. , @xmath50 ) . at the same time , the reception of the source s signal by relay @xmath30 is interfered from the transmission of @xmath47 which forwards a previous signal @xmath45 $ ] to the destination . thus , @xmath30 receives @xmath51=\sum_{i\in \mathcal{a } } h_{s_i r } m_i x_{s_i}[n ] + h_{tr } x[p ] + w_r[n ] , \end{aligned}\ ] ] where @xmath52 denotes the index set of transmit antennas at the source , i.e. , @xmath53 , @xmath54 denotes the channel coefficient from the @xmath13-th transmit antenna at the source to the receiving relay @xmath30 , @xmath55 denotes the channel coefficient from the transmitting relay @xmath31 to the receiving relay @xmath30 , @xmath56 $ ] denotes the transmitted signal from the @xmath13-th transmit antenna at the source in the @xmath43-th time slot , @xmath57 $ ] denotes the transmitted signal from the transmitting relay in the @xmath43-th time slot , @xmath58 denotes the @xmath13-th `` beamforming '' weight at the source , and @xmath59 $ ] denotes the awgn at the receiving relay in the @xmath43-th time slot , i.e. , @xmath59\sim\mathcal{cn}(0,{\sigma^2})$ ] . the source @xmath32 is assumed to be saturated ( infinite backlog ) and hence , it has always data to transmit . the buffering memory at the source is organized into @xmath34 queues , that basically contain replicas of the data queues of the relays , in order to exploit it for iri mitigation or cancellation . the operation is assumed to be divided into time slots . in each time slot , the source and a relay simultaneously transmit their own data to mimic fd relaying ( cf . the transmission powers of the source and the @xmath60-th relay are denoted by @xmath61 and @xmath62 , respectively . for notational simplicity , we assume throughout this paper that all devices use a common fixed transmit power level ( i.e. , @xmath63 ) , unless otherwise specified . moreover , we assume that the receivers send short - length error - free acknowledgment / negative - acknowledgment ( ack / nack ) messages over a separate control channel . we assume narrowband rayleigh block fading channels . each channel coefficient is constant during one time slot and varies independently between time slots . for each time slot , the channel coefficient @xmath64 for link @xmath65 follows a circular symmetric complex gaussian distribution with zero mean and variance @xmath66 , i.e. , @xmath67 . thus , the channel power gain @xmath68 follows an exponential distribution , i.e. , @xmath69 . in addition , we assume the awgn at each receiver with variance @xmath70 . with two antennas ( @xmath71 and @xmath72 ) , two hd relays ( the receiving relay is denoted by @xmath30 and the transmitting relay by @xmath31 ) and a destination @xmath35 ; in this example , @xmath73 and @xmath74 . the buffers at @xmath32 consist basically of replicas of the data queues of the relays ; the source has new packets in the source buffer @xmath75 and replicas of the successfully transmitted packets to the relays in a set of copied buffers . ] the goal is to design a precoding matrix at the source such that , at each time slot @xmath43 the @xmath0 link for the relay selected to receive a packet from the source maximizes its sinr by transmitting a linear combination of ( a ) the source s desired signal for the receiving relay @xmath30 and ( b ) the signal of the transmitting relay @xmath31 taken out from the copied buffer at the source . note that maximizing the sinr of the link is equivalent to minimizing the outage probability of the link . thus , this precoding design criterion is relevant for both fixed rate and adaptive rate transmissions . * special case : source with a single antenna . * the received signal at the receiving relay for the @xmath43-th time slot , @xmath76 $ ] , can be expressed as @xmath77 & = h_{sr}\underbrace{\big ( m_{1}x_s[n]+m_{2}x_t[n]\big)}_{\text{source signal}}+h_{tr } x_t[n ] + w_r[n ] \nonumber \\ & = h_{sr } \underbrace{\begin{bmatrix}m_1 & m_2\end{bmatrix}}_{\text{precoding matrix } } \underbrace{\begin{bmatrix}x_s[n ] & x_t[n]\end{bmatrix}^t}_{\text{data from the source}}\nonumber \\ & \quad \quad \quad \quad \quad \quad \quad \quad \quad \quad + h_{tr } x_t[n ] + w_r[n],\end{aligned}\ ] ] where @xmath56 $ ] in eq . is written as @xmath78+m_{2}x_t[n]$ ] , and @xmath79 . each transmitting node uses a fixed power @xmath80 and thus power control issues are not taken into account . the sinr at the receiving relay , denoted by @xmath81 , is given by @xmath82 hence , the following optimization problem can be formulated : [ opt0_singleantenna ] @xmath83 \textrm{s.t . } & ~ |{m}_1|^2 + |{m}_2|^2 \leq 1 . \label{0cond2_sa } \end{aligned}\ ] ] the solution to optimization problem will be given for the general case and the solution to this special case can be found by simplifications _ mutatis mutandis_. * general case : source with multiple antennas . * now , the precoding matrix at the source comprising @xmath84 antennas is @xmath85 . matrix @xmath86 has dimensions @xmath87 , since the transmitted signals are formed as linear combinations of the source signal and the signal transmitted by the active relay . the received signal at the receiving relay for the @xmath43-th time slot , @xmath76 $ ] , can be expressed as @xmath88 & = \sum_{i\in \mathcal{a } } h_{s_i r}\big ( m_{i1}x_s[n]+m_{i2}x_t[n]\big ) \nonumber \\ & \quad \quad \quad \quad \quad \quad \quad \quad \quad \quad + h_{tr } x_t[n ] + w_r[n],\end{aligned}\ ] ] where @xmath56 $ ] in eq . is written as @xmath89+m_{i2}x_t[n]$ ] , and @xmath90 . defining @xmath91 eq . can be written as @xmath92 & = \mathbf{h}_s^h \begin{bmatrix}\mathbf{m}_1 & \mathbf{m}_2\end{bmatrix } \underbrace{\begin{bmatrix}x_s[n ] & x_t[n]\end{bmatrix}^t}_{\text{data from the source}}\nonumber \\ & \quad \quad \quad \quad \quad \quad \quad \quad \quad \quad + h_{tr } x_t[n ] + w_r[n],\end{aligned}\ ] ] from eq . , it can be easily seen that the source node with @xmath84 antennas applies a linear precoding matrix @xmath93 on the transmitted signals such that the iri is reduced , while at the same time the signal from the source to the intended relay is increased.each transmitting node uses a fixed power @xmath80 and thus power control issues are not taken into account . the sinr at the receiving relay , denoted by @xmath81 , is given by @xmath94 hence , the following optimization problem can be formulated : [ opt0 ] @xmath95 \textrm{s.t . } & ~ \mathbf{m}_1^h \mathbf{m}_1 + \mathbf{m}_2^h \mathbf{m}_2 \leq 1 . \label{0cond2 } \ ] ] each transmitting node uses a fixed power @xmath80 and thus power control issues are not taken into account . note that each antenna @xmath13 , @xmath96 , of the source transmits with power @xmath97 . after the precoding , our effective transmitted signal vector is @xmath98 where @xmath99^t$ ] . the power of the effective transmitted signal vector is @xmath100 where @xmath101 since @xmath102 s are independent of each other . hence , @xmath103 . [ propa ] the precoding matrix @xmath86 that solves optimization problem is given by @xmath104 where and @xmath105 . see appendix [ a : proof_propa ] . the last expression in shows that the optimal @xmath106 is a combination of standard maximum ratio transmission across the source antennas , and the single antenna interference suppression solution . proposition [ propa ] shows that the source does not need to cancel the interference completely in order to achieve the maximum sinr at the receiving relay . this result appears counter - intuitive , since interference is the main source of performance degradation and the sinr should be higher when the iri is cancelled . however , this occurs due to the limited power at the source . when the power @xmath80 is high enough , @xmath107 gets very small . hence , from , @xmath108 \displaystyle \frac{\| \mathbf{h}_s \|^2}{| h_{tr } |^2 } , & \text{if } \| \mathbf{h}_s \|^2 < | h_{tr } |^2 . \end{cases}\end{aligned}\ ] ] which means that interference is essentially cancelled if @xmath109 . if the source can change its power , then the power levels of the source and the transmitting relay are not necessarily the same . in what follows , we allow the source to adjust its power . the sinr now becomes @xmath110 the following proposition gives the optimal joint selection of @xmath111 and @xmath61 , that maximizes the sinr , as given in . [ propb ] when @xmath80 and @xmath111 are considered jointly , then the optimal power @xmath112 and @xmath113 are given by @xmath114 where @xmath115 see appendix [ b : proof_propb ] . proposition [ propb ] suggests that in the case which @xmath61 and @xmath111 are jointly optimized , in order to maximize the sinr , the source transmits with @xmath116 and chooses the @xmath111 corresponding to @xmath116 . [ remark2 ] another approach would be to fix the transmission rate as @xmath117 and try to minimize the outage probability of the @xmath0 link . then , the optimization problem becomes [ opt_1 ] @xmath118 \textrm{s.t . } & ~ \mathbf{h}_s^h \begin{pmatrix } \mathbf{m}_1 & \mathbf{m}_2 \\ \end{pmatrix } = \begin{pmatrix } \beta & -\omega h_{tr } \\ \end{pmatrix } , \label{0cond1 } \\ & ~ \mathbf{m}_1^h \mathbf{m}_1 + \mathbf{m}_2^h \mathbf{m}_2 \leq 1 , \label{0cond2 } \\ & ~ \omega \in ( 0,1 ] . \label{0cond3}\end{aligned}\ ] ] it can be easily shown that this is equivalent to maximizing the sinr which leads to maximizing the link rate . since csis of @xmath1 links are assumed to be available at the source and relays , adaptive rate transmission , which implies transmission rate is determined by instantaneous csi at each link , is possible . in this case , the source encodes its own desired codewords based on csi of the selected @xmath0 link and codewords corresponding to the transmitting relay based on csi of the selected @xmath1 link . then , the source combines them via precoding . in the adaptive rate transmission , the main objective is to maximize the average end - to - end rate from the source to the destination , which is equivalently expressed as the average received data rate at the destination @xcite , i.e. , @xmath119 such that @xmath120 where @xmath121 denotes a time window length and @xmath122 and @xmath123 denote the resulting instantaneous link rates depending on time slot @xmath124 once the relay pair @xmath125 is selected . the instantaneous rates are obtained by @xmath126 where @xmath127 and @xmath128 denote the effective sinr / snr ( signal - to - noise - ratio ) of @xmath0 and @xmath129 links at time slot @xmath124 after applying the precoding matrix obtained by proposition [ propa ] , i.e. , @xmath130 \gamma_{t d}(t ) & = \frac{|h_{td}|^2 p}{\sigma^2 } , \end{aligned}\ ] ] where @xmath111 is given in , @xmath39 denotes the maximum length of queue , and @xmath131 and @xmath132 are the queue lengths of receiving and transmitting relays in bpcu at time slot @xmath133 . for the selected relay pair @xmath134 , the queue lengths are updated at the end of time slot as @xmath135 hereafter , based on the designed precoding matrix @xmath86 in section [ sec : policy1 ] , we propose an adaptive rate relay selection policy for maximizing the average end - to - end rate . applying a lagrangian relaxation , the objective function to maximize the average end - to - end rate in for the adaptive rate transmission becomes equivalent to a weighted sum of instantaneous link rates @xcite . the resulting relay selection scheme has the form [ eq : rs_ar ] @xmath136 for predetermined weight factors @xmath137,~k\in\mathcal{k}$ ] , where @xmath138 and @xmath139 denote the instantaneous rates of @xmath0 and @xmath129 links at time slot @xmath124 , respectively . it is worth noting that for finite buffer size , the optimal weight factors can be found using either a subgradient method or a back - pressure algorithm @xcite . according to the back - pressure algorithm , the weight factors can be found as @xmath140 during a training period where @xmath141 and @xmath142 denote the buffer occupancies at time @xmath124 at the @xmath60-th relay and the source , respectively . the details can be found in @xcite . the optimal relay pair selection in requires an exhaustive search with @xmath143 combinations . therefore , the computational complexity of optimal relay pair selection is @xmath144 . the implementation of the proposed scheme requires global csi of the instantaneous channels and the buffer states . the central unit ( _ e.g. _ , the source in this case ) uses this information in order to select the appropriate relay pair . this global csi requires a continuous feedback for each wireless link as we use a continuous monitoring of the ack / nack signaling in order to identify the status of the buffers . although these implementation issues are beyond the scope of this paper , the proposed scheme can be implementable with the aid of various centralized and distributed csi acquisition techniques and relay selection approaches as in @xcite where each relay sets a timer in accordance to the channel quality and through a countdown process the selection of the best relay is performed . additionally , csi overhead can be reduced significantly , through distributed - switch - and - stay - combining as in @xcite . in this section , we relax the assumption of having knowledge of the full csi . instead , we allow for csir knowledge with limited feedback , i.e. , each receiving node has csi of the channel it receives data from and it can provide some information to the transmitting node . more specifically , each ( receiving ) relay feeds back to the source a phase value via a reliable communication link . this approach is closely related to phase feedback schemes proposed and standardized for miso transmission ( see , for example , @xcite ) . the source signal can use this phase value in one of two possible ways : * to mitigate the interfering signal , so that the overall interference is reduced or even eliminated ; * to amplify the interfering signal , so that it can be decoded and removed from the rest of the received signals . the phase value can be quantized into the desired number of bits , using uniform quantization . we aim at recovering the multiplexing loss of the network by having the source and a relay to align phases , such that the iri is reduced . as a result , the proposed phase alignment approach reduces the communication overhead , while introducing a smart quantization that further reduces the complexity of the transmission . since there is no full csi knowledge , as assumed in the scheme proposed in the previous section , no precoding can be applied and a relay - selection policy is proposed based on csir knowledge with limited feedback that aims at taking advantage of the multi - antenna source . selection of the relay - pair is performed by choosing the pair that achieves the maximum end - to - end sinr . since the source has no csi for any of the links between the antennas at the source and the relays , equal power allocation across all antennas is a natural approach at the transmitter side . however , for overhead reduction on csi estimation at the receiver , we consider to use just two of them : one for transmitting the packet to a relay and the other to mitigate the iri . the existence of more than two antennas , however , can increase the diversity gain by choosing a subset of antennas based on csi , while the fact that not all available antennas are included , the overhead for channel information is limited . the same principle can equally well be applied using a single antenna source node , where one and the same antenna is used both for the desired packet and the iri mitigation . assuming two antennas used , the receiving signal at @xmath30 in is given by @xmath145&= h_{s_1 r } m_1 x_{s_1}[n ] + h_{s_2 r } m_2 x_{s_2}[n]\nonumber\\ & \quad+ h_{tr } x[p ] + w_r[n ] , \end{aligned}\ ] ] where we set @xmath146 and @xmath147 . in each time slot , the signal from the second antenna of the source @xmath148 $ ] is used in one of the following two ways : * to minimize the interference caused by the transmitting relay ; this is done by transmitting @xmath45 $ ] with a shifted phase such that the interfering signal and the signal from the second antenna is in anti - phase . the optimal phase for this is given by @xmath149 * to maximize the interference caused by the transmitting relay in order to make the signal strong enough to be decoded first , and hence , eliminate it . the optimal phase for this is given by @xmath150 [ propc ] the phase @xmath151 such that the interfering signal at the receiving relay @xmath42 from the transmitted signal @xmath45 $ ] of the transmitting relay @xmath47 is minimized is given by @xmath152 similarly , the phase @xmath153 such that the interfering signal at the receiving relay @xmath42 from the transmitted signal @xmath45 $ ] of the transmitting relay @xmath47 is maximized is given by @xmath154 see appendix [ c : proof_propc ] . proposition [ propc ] gives the expressions for phase alignment for each of the two approaches considered . by appropriately choosing the phase shift @xmath155 of the signal from one of the source s antennas , the source can minimize or maximize the interfering signal in order to mitigate it or eliminate it completely . note that the optimal value of @xmath155 ( either @xmath151 or @xmath153 ) can be quantized into the desired number of bits , using uniform quantization , and be fed back to the source . since only csir is available , the transmitters use fixed rate transmission . in such a case , the main objective is to minimize the outage probability . independently of nodal distribution and traffic pattern , a transmission from a transmitter to its corresponding receiver is successful ( error - free ) if the sinr of the receiver is above or equal to a certain threshold , called the _ _ capture ratio _ _ @xmath156 . therefore , at the receiving relay @xmath30 for a successful reception when at the same time relay @xmath31 is transmitting we require that @xmath157 and at the destination we require that @xmath158 an outage event occurs at the relay @xmath30 and destination @xmath35 when @xmath159 and @xmath160 , respectively . the outage probability is denoted by @xmath161 , where @xmath13 represents the receiving node and @xmath60 the transmitting node . each relay @xmath13 is able to estimate the sinr for each transmitting relay @xmath60 , denoted by @xmath162 , @xmath163 ( the full pilot protocol needed to the channel estimation is out of the scope of this work ) . we assume that this information can be communicated to the destination . in addition , the destination node can compute its own snr due to each of the transmitting relays , denoted by @xmath164 , @xmath165 . finally , we assume that the destination node has buffer state information and selects the relays for transmission and reception , based on some performance criterion , _ e.g. _ , with the maximum end - to - end sinr ( as it is defined in @xcite ) , through an error - free feedback channel . note that by having the destination to take the decision , no global csi is required at any node . as we have seen in proposition [ propc ] , the source can minimize the interfering signal or maximize it in order to eliminate it by appropriately choosing the phase shift @xmath155 of the signal from one of the source s antennas . it can be easily deduced that at low iri it is beneficial to try to remove the interfering signal , whereas at high iri it is beneficial to amplify the interfering signal and thus eliminate it completely by decoding it first . the receiving relay is able to compute which option gives the highest sinr in each case , since it has knowledge of the channel states and hence , it can decide which phase to feed back to the source at each time slot . the procedure of the proposed algorithm is as follows : by examining _ one - by - one _ the possible relay pairs , first we calculate the power of the signal received at _ d _ which is @xmath166 for an arbitrary relay @xmath31 with non - empty buffer . the receiving relay @xmath30 must be different than the transmitting relay and its buffer should not be full . for each candidate relay @xmath13 for reception , a _ feasibility check _ for interference cancellation ( ic ) is performed , i.e. , @xmath167 if ic is feasible , the candidate relay is examined whether snr at the receiving relay after ic is above the _ capture ratio _ @xmath156 or not , i.e. , once interference is removed becomes @xmath168 if ic is infeasible , interference mitigation ( i m ) is considered . hence , it is examined whether sinr at the receiving relay after i m is above the _ capture ratio _ or not . if the relay denoted by @xmath30 can provide an snr / sinr above the capture ratio after ic / im , it is considered as a candidate receiving relay . for the selected relay pair @xmath134 , the queue lengths are updated at the end of time slot as @xmath169 note that for fixed rate transmission , the queue length is equivalently modeled as the number of packets in the queue . the relay pair selection policy , as before , requires an exhaustive search with @xmath143 combinations imposing a complexity of the order @xmath144 . note , however , that links with sinr / snr above the _ capture ratio _ @xmath156 are suitable for transmission without compromising the performance of the proposed scheme , i.e. , it is not necessary to choose the relay pair that provides the maximum end - to - end sinr , as long as the outage event is avoided . hence , its simplicity allows the selection of a relay pair much faster , provided that the channel conditions are good . this characteristic will allow for more advanced algorithms in the future where the queue length will be a decisive factor on the decision of the relay pair such that certain delay constraints are met . for the time - being a relay pair is selected such that there is no reduction in performance ( either due to outage or full / empty buffers ) . we have developed a simulation setup based on the system model description in section [ sec : model ] , in order to evaluate the performance of the proposed relay pair selection schemes with current state - of - the - art for both adaptive and fixed rate transmission cases according to csi availability at the source . in the simulations , we assume that the clustered relay configuration ensures @xmath170 rayleigh block fading with average channel qualities @xmath171 , @xmath172 , and @xmath173 for all the @xmath0 , @xmath174 , and @xmath1 links , respectively . this assumption simplifies the interpretation of the results and it is used in several studies in the literature @xcite . the @xmath170 case and the related analysis can be considered as a useful guideline for more complex asymmetric configurations , in which power control at the source and relay nodes may be used to achieve symmetric channel configuration for better outage performance @xcite . for adaptive rate transmission , we evaluate the performance of the proposed ba - sp relay selection ( ba - sprs ) scheme in terms of the average end - to - end achievable rate in bpcu . the following upper bound and state - of - the - art relay selection schemes are considered for the performance comparison . * upper bound : the optimal solution of under no iri * hd best relay selection ( hd - brs ) @xcite * hd hybrid relay selection ( hd - hrs ) @xcite * hd @xmath5 relay selection ( hd - mlrs ) @xcite * ideal ( iri interference is assumed negligible ) and non - ideal ( iri interference is taken into consideration ) sfd - mmrs @xcite since in this work we consider multiple antennas at only the source while all conventional schemes above consider a single antennas at the source , we employ maximum ratio transmission ( mrt ) at the source , which maximizes the effective channel gain at each receiving relay , for all conventional schemes as well for making the comparison fair . fig . [ fig : art_rate_snr ] shows the average end - to - end rate for varying snr with two relays and two antennas at the source when @xmath175 , @xmath176 db with various iri intensities : the same intensity ( @xmath177 db ) , stronger iri ( @xmath178 db ) and weaker iri ( @xmath179 db ) . while the ideal sfd - mmrs scheme almost achieves the upper bound , non - ideal sfd - mmrs significantly degrades the performance since the effect of iri is severe so that it can not be neglected . the discrepancy between the proposed ba - sprs scheme and the upper bound increases with snr , since it results in increasing iri . especially , the gap becomes larger for stronger iri condition ( @xmath178 db ) . this is because perfect ic is impossible as iri increases . however , when iri becomes at least @xmath180 db below the @xmath0 and @xmath1 channel qualities , the proposed scheme almost approaches the upper bound . the conventional hd schemes achieve almost half the average end - to - end rate of the ideal sfd - mmrs and the proposed ba - sprs schemes . although the hd - mlrs scheme is a more advanced scheme than hd - brs and hd - hrs schemes in the sense of diversity order , it is the worst for this case with infinite buffer length , since our system setup yields a channel imbalance between the @xmath0 and @xmath1 links due to multiple antennas at only the source , which gives a higher chance to select the @xmath0 link than the @xmath1 link for the hd - mlrs scheme . in other words , for the hd - mlrs scheme , the larger the buffer size , the more source data is buffered at relay buffers , while the throughput is determined by the amount of received data at the destination . from the result , it is shown that the proposed ba - sprs scheme always outperforms the non - ideal sfd - mmrs and hd relay selection schemes regardless of snr and iri intensity . , @xmath175 , @xmath176 db , and @xmath181 , or @xmath180 db . ] fig . [ fig : art_rate_k ] shows the effect of the number of relays . as the number of relays increases , the proposed ba - sprs scheme quickly approaches the upper bound . especially , when @xmath182 and @xmath183 db , it already almost achieves the upper bound while other schemes can never achieve the upper bound except for the ideal sfd - mmrs scheme neglecting iri . therefore , in the case when @xmath184 , an additional relay ( i.e. , making @xmath182 ) contributes significantly to the performance enhancement with the same iri intensity . for stronger iri ( @xmath185 db ) , although there exist some gaps from the upper bound for small number of relays , it is fast recovered with increasing number of relays even under strong iri condition and finally converges to the upper bound . therefore , increasing the number of relays , offering spatial diversity and ignoring hardware and deployment costs , can be a simple solution against strong iri situations . for the same reason stated in fig . [ fig : art_rate_snr ] , the conventional hd - mlrs scheme rather degrades the average end - to - end rate with respect to increasing number of relays , since increasing the number of relays causes a similar effect to that of increasing the buffer size . db , @xmath176 db , and @xmath183 or @xmath180 db . ] fig . [ fig : art_rate_qmax ] shows the average end - to - end rate for different maximum buffer sizes for @xmath182 and @xmath186 db , which limits the instantaneous rate for each link . the ideal sfd - mmrs and the proposed ba - sprs schemes converge to the upper bound as the maximum buffer size increases but they have a cross point at @xmath187 . the proposed ba - sprs scheme achieves rather slightly higher average rate when @xmath188 than the ideal sfd - mmrs scheme neglecting iri . as a result , the proposed ba - sprs scheme is still effective for the finite buffer size which is a more practical setup . all the schemes converge to their own maximum rate when approximately @xmath189 except for the hd - mlrs scheme . as shown in figs . [ fig : art_rate_snr ] and [ fig : art_rate_k ] , the hd - mlrs scheme rather degrades the average end - to - end rate as the buffer size increases due to the effect of channel imbalance between @xmath0 and @xmath1 links . the hd - brs scheme without using a buffer is not affected from the finite buffer size . when @xmath182 , @xmath186 db , and @xmath190 db . ] for fixed rate transmission , we evaluate the proposed ba - pa relay selection ( ba - pars ) scheme in terms of outage probability and average end - to - end rate . we additionally consider the following state - of - the - art scheme proposed for fixed rate transmission : * buffer - aided successive opportunistic relaying ( ba - sor ) with iri cancellation @xcite . for all simulation results with fixed rate transmission , we consider @xmath170 channel condition with the same iri intensity , i.e. , @xmath191 . [ fig : pout_all ] shows the outage probability and @xmath1 links is less than the capture ratio @xmath156 , we define the outage probability as a portion of successfully transmitted packets among the total number of transmitted packets since the previous definition is not rigorous for the case of concurrent transmissions with iri . ] with various snr values for the transmission rate @xmath192 bpcu , three relays ( @xmath182 ) , and infinite length of buffer ( @xmath193 ) . basically , the hd - brs scheme has the worst outage performance due to lack of buffering . the hd - hrs scheme , a hybrid mode of hd - brs and hd - mmrs , is always better than the hd - brs scheme . since the hd - mlrs scheme can achieve a full diversity ( i.e. , @xmath194 diversity order ) for hd transmission , it shows the best performance except for low snr region . at low snr , the ideal sfd - mmrs scheme without taking iri consideration achieves slightly better performance than the hd - mlrs scheme . however , its outage performance is significantly degraded if iri is imposed . the ba - sor scheme achieves a good performance at low and medium snr but it becomes bad at high snr since its relay selection criterion is to maximize the minimum snr of both @xmath0 and @xmath1 links link when the best snr for @xmath0 link is worse than snrs of all @xmath1 links , instead of a random selection , since the @xmath1 link separately contributes on the outage event . ] . the proposed ba - pars scheme achieves similar performance to the ba - sor scheme at low snr but it is not degraded at high snr thanks to a hybrid mode of ic and i m . in addition , assuming a powerful source node such as base station as we stated in section ii , we depict the case of double power at the source for the proposed ba - pars scheme , which shows that the proposed ba - pars scheme can achieve the outage performance of the ideal sfd - mmrs scheme . hence , if extra power at the source is available , the proposed ba - pars scheme can provide the best outage performance . note that the ideal sfd - mmrs and the double powered ba - pars schemes achieve a half diversity gain ( i.e. , @xmath34 diversity order ) of the hd - mlrs scheme but a better power gain at low snr , since for both schemes , a half rate is required at each link to meet the same transmission rate as the hd - mlrs scheme . bpcu , @xmath182 , and @xmath193 . for the proposed ba - pars scheme , we additionally consider double power at the source ( denoted by ` [ 2p ] ' ) to show the case of a powerful source node . ] [ fig : pout_var_buffer ] shows the outage probability of the proposed ba - pars scheme for varying the maximum buffer size when @xmath192 bpcu and @xmath182 . as in @xcite , we assume that half of buffer elements are full at initial phase ( in order to reach the steady - state queue lengths quicker ) . as the maximum buffer size @xmath39 increases , the outage performance is improved and converges to the case of having buffers of infinite length . the convergence occurs at lower buffer sizes at high snr than at low snr , since buffer full / empty events contribute more in outage events at high snr due to sufficiently good received signal strengths ( i.e. , outage events due to bad channel conditions occur rarely and outage events are due to buffer full / empty events ) . when @xmath192 bpcu and @xmath182 . ] fig . [ fig : rate_all_frt ] shows the average end - to - end achievable rate with three different transmission data rates ( @xmath195 and @xmath196 bpcu ) when @xmath182 and @xmath197 . the conventional hd schemes achieve a half of the data rate due to an hd limitation although the hd - mlrs scheme achieves a full diversity in outage performance . while the ideal sfd - mmrs scheme obtains the best performance which achieves full data rates ( 1.5 and 2.5 bpcu ) as snr increases , the non - ideal sfd - mmrs scheme is significantly degraded due to iri , which shows rather worse performance with @xmath198 bpcu than the hd schemes . the ba - sor scheme can achieve the full data rate with @xmath199 bpcu at high snr but significantly degrades with higher data rates . in contrast , the proposed ba - pars scheme can achieve the full data rates for all the cases such that it guarantees the required data rate if a proper data rate is chosen according to snr , even if it has some gaps compared to the ideal sfd - mmrs scheme since the source power is split for ic / im . similarly to the outage performance , if double power at the source is available , the proposed ba - pars scheme can approach the achievable rate of the ideal sfd - mmrs scheme without suffering from iri . and @xmath196 bpcu ) when @xmath182 and @xmath197 . ] in this work , we present two relay - pair selection policies , depending on the available csi in the system , that employ a buffer - aided multi - antenna source , a cluster of hd buffer - aided relays and a destination . in the case of global csi , a linear precoding strategy is applied by the source in order to mitigate iri . a relay pair is selected , such that the average end - to - end rate is maximized . in the case of csir , phase alignment is applied by the source in order to mitigate / cancel iri . a relay pair is selected , such that the maximum end - to - end sinr is achieved . the benefits of this network deployment are demonstrated via a numerical evaluation , where the improved performance is observed with respect to the average end - to - end rate and outage probability , while the conventional non - ideal sfd - mmrs scheme with iri is significantly degraded . part of ongoing research is to investigate scenarios where only statistical information is known about the csi . after algebraic manipulations , optimization problem can be written as [ opt1 ] @xmath200 \textrm{s.t . } & ~ \mathbf{h}_s^h \begin{bmatrix } \mathbf{m}_1 & \mathbf{m}_2 \\ \end{bmatrix } = \begin{bmatrix } \beta & -\omega h_{tr } \\ \end{bmatrix } , \label{1cond1 } \\ & ~ \mathbf{m}_1^h \mathbf{m}_1 + \mathbf{m}_2^h \mathbf{m}_2 \leq 1 , \label{1cond2}\\ & ~ \omega \in ( 0,1 ] . \label{1cond3}\end{aligned}\ ] ] where @xmath105 , and @xmath201 , @xmath111 depend on the choice of @xmath202 . by a simple phase shift we observe that we can choose @xmath203 and @xmath204 , such that @xmath201 is a real number ; i.e. , let @xmath205 $ ] , such that @xmath206 ; then , @xmath207 . hence , optimization problem can be written as [ opt2 ] @xmath208 \textrm{s.t . } & ~ \mathbf{h}_s^h \begin{bmatrix } \mathbf{m}_1 & \mathbf{m}_2 \\ \end{bmatrix } = \begin{bmatrix } \beta & -\omega h_{tr } \\ \end{bmatrix } , \label{2cond1 } \\ & ~ \mathbf{m}_1^h \mathbf{m}_1 + \mathbf{m}_2^h \mathbf{m}_2 \leq 1 , \label{2cond2 } \\ & ~ \omega \in ( 0,1 ] . \label{2cond3}\end{aligned}\ ] ] let @xmath209 $ ] be a fixed value . then , the problem becomes equivalent to maximizing @xmath210 , i.e. , [ opt3 ] @xmath211 \textrm{s.t . } & ~ \mathbf{h}_s^h \begin{bmatrix } \mathbf{m}_1 & \mathbf{m}_2 \\ \end{bmatrix } = \begin{bmatrix } \beta & -\omega h_{tr } \\ \end{bmatrix } , \label{3cond1 } \\ & ~ \mathbf{m}_1^h \mathbf{m}_1 + \mathbf{m}_2^h \mathbf{m}_2 \leq 1 . \label{3cond2 } \end{aligned}\ ] ] suppose @xmath212 , @xmath213 , where @xmath214 will be identified later in the proof . then , by conditions and we have @xmath215 since @xmath201 is maximized , condition should be satisfied with equality . this will emerge in the sequel by contradiction . assume @xmath216 for some @xmath217 . then , since both @xmath218 and @xmath201 are real numbers we can easily deduce that @xmath219 substituting into , @xmath220 ; @xmath201 is maximized when @xmath221 , so condition is satisfied with equality . thus , @xmath222 and @xmath223 now , we need to find @xmath214 and @xmath224 . it is observed in that @xmath214 should be maximized in order to maximize @xmath201 . given that @xmath225 , this is equivalent to minimizing @xmath226 . towards this end , we formulate the following optimization problem : [ opt5 ] @xmath227 by cauchy - schwartz inequality @xmath228 minimizing @xmath229 can be achieved when holds with equality , i.e. , @xmath230 combining and we get @xmath231 since @xmath232 , then @xmath233 combining and , the maximum @xmath201 is given by @xmath234 substituting @xmath214 and @xmath201 into , we have @xmath235 finally , the precoding matrix @xmath86 is given by @xmath236 now , we want to find the value of @xmath111 for which the sinr at the receiving relay is maximized . for both and , it is required that @xmath237 . we substitute into optimization and for simplicity of notation we denote @xmath238 and @xmath239 . hence , optimization problem is written as @xmath240 \textrm{s.t . } & ~ 0 < \omega \leq \min\left\{1,{\sqrt{\frac{a}{b}}}\right\ } , \label{omega : ineq:1}\end{aligned}\ ] ] where @xmath241 denotes the minimum of arguments and the right hand side of comes from the condition that @xmath210 ( and subsequently @xmath242 ) is non - negative . by differentiating @xmath242 with respect to @xmath111 , @xmath243 at a turning point , @xmath244 ; hence , @xmath245 the two roots of are obtained by @xmath246 first , we verify that the roots have real - values by checking if the second order equation in has a positive discriminant , i.e. , @xmath247 let @xmath248 be the smallest root . it can be easily shown that at @xmath248 we attain a maximum , whereas at @xmath249 we attain a minimum . in addition , we need to show that any @xmath250 does not fulfill inequality constraint , so that the maximum of @xmath242 is not obtained on the boundary . we show this by contradiction . for @xmath251 , suppose @xmath252 . then , @xmath253 where @xmath254 stems from the fact that all the eliminated elements are positive . for @xmath255 , suppose @xmath256 . then , @xmath257 where @xmath258 stems from the fact that all the eliminated elements are all positive . hence , @xmath259 , given it fulfills inequality constraint . it is easily shown that @xmath248 is positive , so we will check if @xmath260 . for @xmath261 we need to check if @xmath262 . suppose @xmath263 ; then , @xmath264 which contradicts the fact that @xmath265 . step @xmath266 follows because @xmath267 ( @xmath268 ) . for @xmath269 we need to check if @xmath270 . suppose @xmath271 ; then , @xmath272 which contradicts the fact that @xmath265 . step @xmath273 follows because @xmath274 . it can be easily shown that the choice of @xmath61 changes neither the precoding matrix @xmath86 nor the value of @xmath201 . let @xmath276 and @xmath277 . then , the optimization problem in which power level selection is also possible can be written as @xmath278 \textrm{s.t . } & ~ 0 < \omega \leq \min\left\{1,\sqrt{\frac{a}{b}}\right\ } , \label{omega : ineq:11 } \\ & ~ 0 < x \leq m , \label{omega : ineq:12}\end{aligned}\ ] ] where @xmath279 . differentiating @xmath280 w.r.t . @xmath281 , the value of @xmath281 that maximizes @xmath280 is given by @xmath282 differentiating @xmath280 w.r.t . @xmath111 and by following similar steps to that of proposition [ propa ] , the value of @xmath111 that maximizes @xmath280 is given by @xmath283 the solution of - gives @xmath284 , suggesting that @xmath285 , i.e. , the source uses infinite power . however , power is constrained and this solution is not feasible . by substituting @xmath286 of in , we obtain @xmath287 which is monotonically decreasing with @xmath111 . hence , in order to maximize the sinr with respect to @xmath281 , @xmath111 should be kept as small as possible , while satisfying . this means that the value of @xmath281 that maximizes the sinr is at @xmath288 . let @xmath289 be the value computed via for @xmath288 . for @xmath288 , the value of @xmath111 that maximizes the sinr with respect to @xmath111 , say @xmath290 , can be computed via . then , * if @xmath291 , one can find @xmath292 that increases sinr ( since the maximum of the sinr with respect to @xmath281 is shifted to a lower value ) ; but then , @xmath293 which means that if @xmath291 the maximum value is always obtained at @xmath294 ; * if @xmath295 , then @xmath296 and the hence @xmath288 ; it can not be improved further since @xmath297 is an increasing function with respect to @xmath281 until @xmath298 . hence , @xmath290 finds the optimal value of the sinr with respect to @xmath111 , while it also increases the sinr with respect to @xmath281 . hence , the source should _ always _ transmit with @xmath116 . in case , @xmath116 is small such that is not satisfied and does not yield a feasible @xmath111 , then the maximum sinr is achieved for @xmath111 on the boundary , i.e. , @xmath299 , since @xmath297 is an increasing function with respect to @xmath111 , for @xmath300 ( as given by ) . @xmath275 by triangle inequality @xmath301 the optimization problem @xmath302 is minimized when inequality holds with equality ; this occurs when @xmath303 is in phase with @xmath304 . let @xmath151 the optimal angle @xmath155 for optimization . since @xmath305 , then the minimization yields @xmath306 similarly , by triangle equality @xmath307 the optimization problem @xmath308 is maximized when inequality holds with equality ; this occurs when @xmath303 is in phase with @xmath55 . let @xmath153 the optimal angle @xmath155 for optimization . since @xmath305 , the maximization yields @xmath309 j. n. laneman , d. n. c. tse , and g. w. wornell , `` cooperative diversity in wireless networks : efficient protocols and outage behavior '' , _ ieee trans . inform . theory _ 30623080 , dec . 2004 . i. krikidis , j. s. thompson , s. mclaughlin , and n. grtz , `` max - min relay selection for legacy amplify - and - forward systems with interference , '' _ ieee trans . wireless commun . _ , vol . 8 , pp . 30163027 , june 2009 . z. ding , i. krikidis , b. rong , j. s. thompson , c. wang , and s. yang , `` on combating the half - duplex constraint in modern cooperative networks : protocols and techniques , '' _ ieee trans . wireless commun . 2027 , dec . 2012 . c. wang , y. fan , i. krikidis , j. s. thompson , and h. v. poor , `` superposition - coded concurrent decode - and - forward relaying , '' _ proc . inf . theory ( isit ) _ , toronto , canada , july 2008 , pp . 23902394 . n. nomikos , t. charalambous , i. krikidis , d. n. skoutas , d. vouyioukas , m. johansson , and c. skianis , `` a survey on buffer - aided relay selection , '' _ commun . surveys tuts . 2 , pp . 10731097 , 2016 . i. krikidis , t. charalambous , and j. s. thompson , buffer - aided relay selection for cooperative diversity systems without delay constraints , " _ ieee trans . wireless commun . _ , vol.11 , no.5 , pp.19571967 , may 2012 . t. charalambous , n. nomikos , i. krikidis , d. vouyioukas and m. johansson , `` modeling buffer - aided relay selection in networks with direct transmission capability , '' _ ieee commun . 649 - 652 , april 2015 . m. shaqfeh , a. zafar , h. alnuweiri and m. s. alouini , `` maximizing expected achievable rates for block - fading buffer - aided relay channels , '' _ ieee trans . wireless commun . 9 , pp . 59195931 , sept . 2016 . m. oiwa , c. tosa and s. sugiura , `` theoretical analysis of hybrid buffer - aided cooperative protocol based on max max and max link relay selections , '' _ ieee trans . on vehicular tech . 92369246 , nov . 2016 . n. nomikos , d. vouyioukas , t. charalambous , i. krikidis , p. makris , d. n. skoutas , m. johansson and c. skianis , `` joint relay - pair selection for buffer - aided successive opportunistic relaying , '' _ wiley - blackwell trans . emerg . telecom . _ , vol . 25 , no . 8 , pp . 823834 , aug . n. nomikos , t. charalambous , i. krikidis , d. n. skoutas , d. vouyioukas and m. johansson , `` a buffer - aided successive opportunistic relay selection scheme with power adaptation and inter - relay interference cancellation for cooperative diversity systems , '' _ ieee trans . on commun . 5 , pp . 16231634 , may 2015 . r. simoni , v. jamali , n. zlatanov , r. schober , l. pierucci and r. fantacci , `` buffer - aided diamond relay network with block fading and inter - relay interference , '' _ ieee trans . wireless commun . 73577372 , nov . s. m. kim and m. bengtsson , `` virtual full - duplex buffer - aided relaying relay selection and beamforming , '' _ ieee pers . ( pimrc ) _ , london , united kingdom , sept . 2013 , pp . 17481752 . n. nomikos , t. charalambous , i. krikidis , d. vouyioukas and m. johansson , `` hybrid cooperation through full - duplex opportunistic relaying and max - link relay selection with transmit power adaptation , '' _ ieee int . conf . on comm . ( icc ) _ , june 2014 , pp . 57065711 . r. w. heath jr and a. j. paulraj , `` a simple scheme for transmit diversity using partial channel feedback , '' _ proc . 32th asilomar conf . signals , systems and computers _ , pacific grove , ca , nov . 1998 , pp . 10731078 . k. k. mukkavilli and a. sabharwal and b. aazhang , `` design of multiple antenna coding schemes with channel feedback , '' _ proc . 35th asilomar conf . signals , systems and computers _ , vol . 2 , pacific grove , ca , nov . 2001 , pp . 10091013 . d. benevides da costa and s. aissa , `` end - to - end performance of dual - hop semi - blind relaying systems with partial relay selection , '' _ ieee trans . wireless commun . _ , vol . 8 , no . 8 , pp . 43064315 , aug . 2009 . z. tian , g. chen , y. gong , z. chen , j. chambers , `` buffer - aided max - link relay selection in amplify - and - forward cooperative networks , '' _ ieee trans . vehicular tech . 2 , pp . 553565 , feb . 2015 .

we investigate the structure of the macroscopic-loop amplitude obtained from the two - matrix model at the unitary minimal critical point . we derive a general formula for the-resolvent correlator at the continuum planar limit whose inverse laplace transform provides the amplitude in terms of the boundary lengths and the renormalized cosmological constant . the amplitude is found to contain a term consisting of multiplied by the product of modified bessel functions summed over their degrees which conform to the fusion rules and the crossing symmetry . this is found to be supplemented by an increasing number of other terms with which represent residual interactions of loops . we reveal the nature of these interactions by explicitly determining them as the convolution of modified bessel functions and their derivatives for the case and the case . we derive a set of recursion relations which relate the terms in the-resolvents to those in the-resolvents . = -.5 in = 1.5ex plus 1pt = 11 addtoresetequationsection normalsizesetsizexiipt plus3pt minus3pt@ plus3pt plus3.5pt minus0pt # 1@underline#1 twosidetrue @=12 0.0 in 0.0 in 6.0 in 8.5 in .75 in = 11

hep-th9511220

# 1 ps . @headings oddfootevenfoot oddhead oddhead evenheadoddhead # # 1 ps . @headings = 12 # 1@xmath9 = 11 # 1 ps . @headings oddfootevenfoot oddhead evenheadoddhead # # 1 ps . @headings = 1.5ex plus 1pt = 11 addtoresetequationsection normalsizesetsizexiipt plus3pt minus3pt@ plus3pt plus3.5pt minus0pt # 1@underline#1 @xmath8 twosidetrue @=12 0.0 in 0.0 in 6.0 in 8.5 in .75 in = 11 # 1 ps . @headings oddfootevenfoot oddhead oddhead evenheadoddhead # # 1 ps . @headings = 12 # 1@xmath9 = 11 # 1 ps . @headings oddfootevenfoot oddhead evenheadoddhead # # 1 ps . @headings plus 1000pt minus 1000pt # 1 # 1= to # 1#2#1#2 # 1#2#3 ^ 2 # 1#2 # 3 # 1#1| # 1| # 1 # 1#1 # 1/ # 1| # 1| # 1#1^ = 2 gev / c^2 # 1#2#3 @xmath10 # 1#2#3#4 @xmath11 # 1#2 @xmath12 # 1#2#3#4#5 @xmath13 # 1 # 1#2#3 _ phys . _ * # 1 * , # 2 ( 19#3 ) # 1#2#3 _ phys . lett . _ * # 1 * , # 2 ( 19#3 ) # 1#2#3 _ rev . * # 1 * , # 2 ( 19#3 ) # 1#2#3 _ phys . rep . _ * # 1 * , # 2 ( 19#3 ) # 1#2#3 _ phys . rev . d_*#1 * , # 2 ( 19#3 ) # 1#2#3 _ nucl . phys . _ * b#1 * , # 2 ( 19#3 ) # 1@xmath9 hep - th 9511220 + ou - het 222 + november , 1995 + * macroscopic @xmath0-loop amplitude for minimal models coupled to two - dimensional gravity + fusion rules and interactions * @xmath14 + and + * h. itoyama*@xmath14 + @xmath14 department of physics , + graduate school of science , osaka university , + toyonaka , osaka , 560 japan + matrix models provide an arena in which the notion of integrability is realized as noncritical string theory . at the same time , they produce efficient computation of some quantities which would be very formidable in the continuum framework . computation of macroscopic loop amplitudes@xcite @xcite demonstrates this fact most explicitly : the boundary condition which is hard to solve in the continuum framework @xcite turns out to be related to the most natural quantity in matrix models . let us begin with recalling this . a crude correspondence of matrix models with path integrals of noncritical strings tells us that the connected part of the correlator given by averaging over matrix integrals of the product of singlet correlators < < tr ^q_1 tr ^q_2 tr ^q_n > > _ n , conn is an n - punctured surface swept by a noncritical string . to turn these punctures into holes of a macroscpic size , one first introduces a fixed loop length at the i - th boundary by @xmath15 . we are naturally led to consider the limiting procedure _ n ( _ 1 , _ 2 , _ q_i , a 0 , _ n < < tr ^q_1 tr ^q_2 tr ^q_n > > _ n , conn which defines the macroscopic @xmath0-loop amplitude . here @xmath16 is the renormalized string coupling and @xmath17 is an auxiliary parameter which plays the role of a cutoff . an equivalent and more efficient procedure is to consider the correlator consisting of the product of @xmath0-resolvents @xmath18 , to pick its most singular piece and finally to carry out the inverse laplace transforms over @xmath19 s . this in turn means [ eq : procedure ] a_n ( _ 1 , _ 2 , _ n ) = ( _ j=1^n l_j^-1 ) _ a0 _ n < < _ i=1^n tr > > _ n , conn . here , @xmath20 denotes the inverse laplace transform with respect to @xmath21 such that @xmath22 and @xmath23 denotes the critical value of @xmath19 . in this paper , we will carry out this procedure in depth at the @xmath24 critical point realized by the symmetric potential of the two - matrix model . here @xmath25 . in the next section , we evaluate the connected part of the correlator consisting of the product of @xmath0-resolvents for large @xmath26 just mentioned above and derive a general formula for this object in the continuum planar limit . we exploit the planar solution to the heisenberg algebra and its parametrization provided by @xcite . our formula contains a term distinguishable from others , namely the one which is expressible as the total @xmath27 derivatives . here @xmath3 denotes the renormalized cosmological constant . this structure is familiar from the case of pure two - dimensional gravity . this term is , however , found to be supplemented , for @xmath28 , by an increasing number of other terms with @xmath0 . this latter structure testifies to the existence of interactions which can not be captured by the naive notion of operator product expansion for microscopic loop operators : the macroscopic loop operator will be expanded by these . for that reason , these interactions may be referred to as contact interactions . in section iii , we consider the @xmath27 @xmath29 term . we are successful in representing this terms as the summations over @xmath30 indices with its summand in a form of @xmath0 factorized products . these summations are found to conform to the fusion rules and the crossing symmetry for the dressed primaries of the unitary minimal conformal field theory @xcite . using the formula for the inverse laplace transform found in @xcite , [ eq : inverse ] l^-1 [ ] = - k_k / m ( m ) - _ k / m ( m ) . we determine the complete form for this part of the amplitude in terms of the boundary lengths @xmath2 @xmath31 . the answer reads as & & a_n^ fusion ( _ 1 , _ 2 , _ n ) + & & = - ( ) ^n-2 ( ) ^n-3 . the case @xmath32 has been briefly reported in @xcite . in section iv , we consider the remaining pieces in the formula which represent the residual interactions of loops . for the case @xmath5 and the case @xmath6 , we have succeeded in expressing these in terms of the convolution of modified bessel functions and their derivatives . we , therefore , obtain the complete answer for @xmath33 and the one for @xmath34 , which are eq . ( [ eq : n=4answer ] ) and eq . ( [ eq : n=5answer ] ) respectively . although it is not unlikely that one can determine the full amplitude this way for arbitrary @xmath0 , the proof remains elusive . we will finish with a few remarks concerning with the properties of these residual interactions . in appendix a , we derive a set of recursion relations which are used to evaluate the formula in section iii . these recursion relations relate the expression of the terms appearing in the @xmath0-resolvent to those in the @xmath7-resolvent . these define , therefore , the @xmath0-loop amplitude in terms of @xmath7-loop amplitude through the inverse laplace transforms albeit being implicit . consider in the two - matrix model the connected part of the correlator consisting of the product of @xmath0-resolvents at finite @xmath26 : [ eq : nresolv ] < < tr tr tr > > _ n , conn . here , @xmath35 are the matrix variables and @xmath19 s are eigenvalue coordinates which , in the continuum limit , become laplace - conjugate to loop lengths . we denote by @xmath36 the averaging with respect to the matrix integrations . it should be noted that this expression is at most @xmath37 due to the large @xmath26 factorization of the correlator consisting of the product of singlet operators . in the second quantized notation , ( [ eq : nresolv ] ) is expressible as & & _ n < 0 _ i=1^n : d _ i b^ ( _ i ) b(_i ) : 0 > _ n + & & = _ n < 0 _ i=1^n : b^_k_ib_j_i : 0 > _ n _ i=1^n < k_i j_i > , where b ( ) & = & _ j=0^ < j > b_j , b^ ( ) = _ j=0^<k > b_k^ + b_j > & = & 0 , j= 0,1,2 , + and 0 > _ n & & _ j=0^n-1 b_j^ > . the normal ordering @xmath38 is with respect to the filled sea @xmath39 . we introduce a notation ( z_i ; _ i , , n ) _ z_i^ < j_i - j_i > + _ i= j_i /n = + _ i /n . the evaluation of @xmath40 by the wick theorem provides @xmath41 terms of the following structure : each term is given by the product of @xmath0-kronecker delta s multiplied both by a sign factor and by the product of @xmath0-step functions to ensure that the summations over the @xmath0-indices @xmath42 and @xmath43 are bounded either from below @xmath44 or from above@xmath45 . we denote this product by @xmath46 . these @xmath41 terms are in one - to - one correspondence with the circular permutations of @xmath0 integers @xmath47 , which we denote by @xmath48 . the @xmath49 is an element of @xmath48 . for large @xmath26 , we find [ eq : step1 ] & & ( ) ^n-2 < < _ i=1^n tr > > _ n , conn + & = & _ , , _ _ n ( _ 1 , _ 2 , _ n ; ) sgn ( ) ( _ j=1^n ) _ k=1^n ( ) ^_k + & & ( _ i=1^n _ i ) ^n-2 _ i^=1^n _ _ i^ = + ( 1/n ) . note that in the large @xmath26 limit , we can use @xmath50(z_i ; \lambda_i ) } $ ] in place of @xmath51(z_i;\lambda_i , \lambda , n)$ ] according to the same reason as stated in @xcite . here the ` classical ' function is defined by ( z_i ; _ i ) _ n _ z_i^ , _ i = j_i /n . the @xmath52 denotes the signature associated with the permutation @xmath49 . let us define m ! d_m(z , z^ ) _ 0 ^m ( z^/z)^ = - _ -1 ^m ( z^/z)^ , m= 0 , . in the continuum limit we will be focusing from now on , it is sufficient to use d_m(z , z^ ) d_m(z - z ) . let @xmath53 be @xmath54 or @xmath55 , depending upon whether the restriction on the summation over @xmath56 is bounded from below or from above respectively . it is not difficult to show sgn ( ) _ i=1^nsgn_i ( ) = -1 , for any @xmath49 and @xmath0 . the summations over @xmath42 and @xmath43 can then be performed for all @xmath49 at once , leaving with this minus sign . now we turn to the integrations over @xmath57 @xmath58 . the convergence on the geometric series leads to the successively ordered integrations of @xmath59 for each @xmath49 . by simply picking up a pole of @xmath57 at @xmath60 for @xmath61 and using & & f ( z_i , ) ( ) ^ ( ) + & = & - ( ) ^ ^z_i^ * dz_i f ( z_i , ) , = 0,1 , , we find that eq . ( [ eq : step1 ] ) is written as [ eq : step2 - 1 ] ( ) ^n-2 < < _ i=1^n tr > > _ n , conn = _ i=1^n ( - ) _ n |__i = , where [ eq : step2 - 2 ] _ n & & _ i_1^n ( ) ^n-2 _ _ n d_n-2 ( [ i_1 - ( i_1 ) ] ) _ j ( i_1 ) d_0 ( [ j- ( j ) ] ) + & + & _ i_1 , i_2^n ( ) ^n-3 ( ) d_n-3 ( [ i_1 - ( i_1 ) ] ) d_1 ( [ i_2 - ( i_2 ) ] ) + & & _ j ( i_1 , i_2 ) d_0 ( [ j- ( j ) ] ) + & + & + & + & _ i_1 , i_2 , , i_n-2^n ( ) ( ) ( ) + & & _ j=1^n-2 d_1 ( [ i_j - ( i_j ) ] ) _ j ( i_1 , i_2 , , i_n-2 ) d_0 ( [ j- ( j ) ] ) . the integrals in the equation above are with respect to @xmath62 s and we adopt a notation [ i ] z_i^ * , [ i - j ] z_i^ * - z_j^ * . this expansion is in one to one correspondence with the following expansion of @xmath63 & & ( _ i=1^n x_i ) ^n-2 = _ i_1=1^n x_i_1^n-2 + _ n-2c_n-3((n 5 ) + _ n,4 ) _ i_1 , i_2x_i_1^n-3 x_i_2 + & + & _ n-2c_n-4((n 7 ) + _ n,6 ) _ i_1 , i_2x_i_1^n-4 x_i_2 ^ 2 + & + & _ n-2c_n-4((n 6 ) + _ n,5 ) _ i_1 , i_2 , i_3 x_i_1^n-4 x_i_2 x_i_3 + & + & _ n-2c_n-5((n 9 ) + _ n,8 ) _ i_1 , i_2x_i_1^n-5 x_i_2 ^ 3 + & + & _ n-2c_n-3 _ 3c_2 ( ( n 8 ) + _ n,7 ) _ i_1 , i_2 , i_3x_i_1^n-5 x_i_2 ^ 2 x_i_3 + & + & _ n-2c_n-5((n 9 ) + _ n,6 ) _ i_1 , i_2 , i_3 , i_4 x_i_1^n-5 x_i_2 x_i_3 x_i_4 + & + & + & + & _ i_1 , i_2 , , i_n-2 x_i_1 x_i_2 x_i_n-2 . here the summations without a parenthesis are over @xmath64 different integers @xmath65 @xmath66 . the number of terms appearing is equal to the number of partitions of @xmath67 objects into parts . in order to put eqs . ( [ eq : step2 - 1 ] ) , ( [ eq : step2 - 2 ] ) in a simpler form , let us introduce [ eq : braket ] & & ( cccc m_1 , & m_2 , & , & m_n + i_1 , & i_2 , & , & i_n ) _ n + & & - _ _ n d_m_1 ( [ i_1 - ( i_1 ) ] ) d_m_2 ( [ i_2 - ( i_2 ) ] ) d_m_n ( [ i_k - ( i_n ) ] ) + & & = -__n 1[i_1-(i_1)]^m_1 + 1 1[i_2-(i_2)]^m_2 + 1 1[i_n-(i_n)]^m_n+1 .in particular , ( cccc n-2 , & 0 , & , & 0 + i_1 , & i_2 , & , & 0 ) _ n & & - _ _ n d_n-2 ( [ i_1 - ( i_1 ) ] ) _ j ( i_1 ) d_0 ( [ j- ( j ) ] ) + ( ccccc n-3 , & 1 , & 0 , & , & 0 + i_1 , & i_2 , & i_3 , & , & 0 ) _ n & & - _ _ n d_n-3 ( [ i_1 - ( i_1 ) ] ) d_1 ( [ i_2 - ( i_2 ) ] ) + & & _ j ( i_1 , i_2 ) d_0 ( [ j- ( j ) ] ) + & e.t.c & in the appendix a , we prove that [ eq : sub - recursion ] ( cccc m_1 , & m_2 , & , & m_n + i_1 , & i_2 , & , & i_n ) _ n = 0 _ m _ n-3 , as well as [ eq : recursion ] & & ( ccccccc m_1 , & m_2 , & , & m_k , & 0 , & , & 0 + i_1 , & i_2 , & , & i_k , & i_k+1 , & , & 0 ) _ n + & = & _ = 1^k ( clccccr m_1 , & , & m_-1 , & , & m_k , & 0 , & + i_1 , & , & i _ , & , & i_k , & , & ) _ n-1 + & if & _ m _ in particular , [ eq : n-2,0 ] ( lcr n-2 , & 0 , & + i_1 , & , & ) _ n = ( lcr n-3 , & 0 , & + i_1 , & , & ) _ n-1 = and [ eq : n-3,1 ] ( lcr n-3 , & 1 , & + i_1 , & i_2 & ) _ n & = & ( lcr n-4 , & 1 , & + i_1 , & i_2 , & ) _ n-1 + & + & ( clcr n-3 , & 0 , & + i_1 , & i_2 , & ) _ n-1 . let us introduce graphs in which the factor @xmath68 ^ 2 $ ] is represented by a double line linking circle @xmath69 and circle @xmath70 to handle the quantities defined by eq . ( [ eq : braket ] ) more easily . for example , for n=3 , ( ccc 1 & 0 & 0 + 1 & 2 & 3 ) _ 3 = 1[z^*_1-z^*_2]^2 1[z^*_1-z^*_3]^2 = 2.2 cm = = 0.5 by -0.3ex . using the recursin relation eq . ( [ eq : recursion ] ) and eq . ( [ eq : sub - recursion ] ) we have , for n=4 , & & ( cccc 2 & 0 & 0 & 0 + 1 & 2 & 3 & 4 ) _ 4 = ( ccc 2 - 1 & 0 & 0 + 1 & 2 & 3 ) _ 3 + & & = = 2.5 cm = = 0.5 by -0.3ex = 2.2 cm = = 0.5 by -0.3ex = = 2.5 cm = = 0.5 by -0.3ex , & & ( cccc 1 & 1 & 0 & 0 + 1 & 2 & 3 & 4 ) _ 4 = ( ccc 1 - 1 & 1 & 0 + 1 & 2 & 3 ) _ 3 + ( ccc 1 & 1 - 1 & 0 + 1 & 2 & 3 ) _ 3 + & & = = 2.5 cm = = 0.5 by -0.3ex = 2.2 cm = = 0.5 by -0.3ex + = 2.5 cm = = 0.5 by -0.3ex = 2.2 cm = = 0.5 by -0.3ex + & & = = 2.cm = = 0.5 by -0.3ex + = 2.cm = = 0.5 by -0.3ex , and , for n=5 , & & ( ccccc 3 & 0 & 0 & 0 & 0 + 1 & 2 & 3 & 4 & 5 ) _ 5 = ( cccc 3 - 1 & 0 & 0 & 0 + 1 & 2 & 3 & 4 ) _ 4 + & & = = 2.5 cm = = 0.5 by -0.3ex = 2.5 cm = = 0.5 by -0.3ex = = 2.5 cm = = 0.5 by -0.3ex , & & ( ccccc 2 & 1 & 0 & 0 & 0 + 1 & 2 & 3 & 4 & 5 ) _ 5 = ( cccc 2 - 1 & 1 & 0 & 0 + 1 & 2 & 3 & 4 ) _ 4 + ( cccc 2 & 1 - 1 & 0 & 0 + 1 & 2 & 3 & 4 ) _ 4 + & & = = 2.5 cm = = 0.5 by -0.3ex \ { = 2.cm = = 0.5 by -0.3ex + = 2.cm = = 0.5 by -0.3ex } + & & + = 2.5 cm = = 0.5 by -0.3ex = 2.5 cm = = 0.5 by -0.3ex + & & = = 2.5 cm = = 0.5 by -0.3ex + = 2.5 cm = = 0.5 by -0.3ex + = 2.5 cm = = 0.5 by -0.3ex . from the examples above , it is clear that the graphs for the general case can be written down quite easily . in terms of the quantities defined by eq . ( [ eq : braket ] ) , we obtain a formula for the n - point resolvent : [ eq : result ] ( ) ^n-2 < < _ i=1^n tr > > _ n , conn = _ i=1^n ( - ) _ n |__i = , where [ eq : delta ] & & -1(n-2 ) ! _ n(z^*_1,.z^*_n ) + & & = _ i_1^n ( ) ^n-2 ( cccc n-2 , & 0 , & , & 0 + i_1 , & i_2 , & , & i_n ) _ n + & & + _ m_1+m_2=n-2 _ ( i_1,i_2 ) ( ) ^m_1 ( ) ^m_2 ( ccccc m_1 , & m_2 , & 0 , & , & 0 + i_1 , & i_2 , & i_3 , & , & i_n ) _ n + & & + _ m_1+m_2+m_3=n-2 _ ( i_1,i_2,i_3 ) ( ) ^m_1 ( ) ^m_2 ( ) ^m_3 ( cccccc m_1 , & m_2 , & m_3 , & 0 , & , & 0 + i_1 , & i_2 , & i_3 , & i_4 , & , & i_n ) _ n + & & + + & & + _ ( i_1,,i_n-2 ) ( ) ( ) ( ccccc 1 , & , & 1 , & 0 , & 0 + i_1 , & , & i_n-2 , & i_n-1 , & i_n ) _ n . here @xmath71 and the summation @xmath72 denotes a set of @xmath64 unequal integers from @xmath73 and @xmath74 in the array represents the remaining integers . ( [ eq : result ] ) , ( [ eq : delta ] ) are a part of our main results . it is straightforward to write down the correlator for lower @xmath0 explicitly ; _ 2 & = & _ i=1 ^ 2 , + _ 3 & = & _ i=1 ^ 3 . for @xmath75 , the correlator can be written more compactly using graphs as introduced below : [ eq : n=4 ] _ 4 & = & \ { = 2.5 cm = = 0.5 by -0.3ex } + & + & \ { = 1.8 cm = = 0.5 by -0.3ex } , [ eq : n=5 ] _ 5 & = & ( ) ^2 \ { = 2.7 cm = = 0.5 by -0.3ex } + & + & ( ) \ { = 2.5 cm = = 0.5 by -0.3ex } + & + & \ { = 1.8 cm = = 0.5 by -0.3ex } . in these figures a double line linking circle @xmath69 and circle @xmath70 , a single line having an arrow from circle @xmath69 to circle @xmath70 and a solid circle @xmath69 represent @xmath68 ^ 2 $ ] , @xmath68 $ ] and @xmath76 respectively . the summations are over all possible graphs that have the same topology specified . each graph appears just for once in the summation . note that the links to the external circles are not double lines but the single ones with arrows and that the internal circles are solid circles . for general @xmath0 , @xmath77 is expressed in the same way . the rule is as follows . first , we consider all possible graphs which have @xmath0 circles and @xmath78 links in the same way as in the case @xmath6 . second , if the internal solid circle @xmath69 has @xmath2 links in each graph , the graph is operated by @xmath79 . then the summation over all graphs gives the expression for @xmath77 . isolating the term which comes with the highest number of total derivatives of the bare cosmological constant @xmath80 ( as opposed to @xmath81 ) by _ i_1 ( ) ^n-2 & = & ( ) ^n-3 _ i_1 = 1^n + & - & _ i_1=1^n _ j ( i_1 ) , we obtain another expression for @xmath82 : [ eq : deformed - delta ] & & _ n = ( ) ^n-3 _ i_1^n ( ) ( clcr n-2 , & 0 , & & + i_1 , & , & & ) + & + & _ i_1 , i_2^n ( ) ^n-3 ( ) \ { ( ( n 5 ) + _ n,4 ) ( clcr n-3 , & 1 , & 0 , & + i_1 , & i_2 , & , & ) . + & & . -(n-3 ) ( clcr n-2 , & 0 , & & + i_1 , & i_2 , & & ) } + & + & _ i_1 , i_2^n ( ) ^n-4 ( ) ^2 \ { ( ( n 7 ) + _ n,6 ) ( clcr n-4 , & 2 , & 0 & + i_1 , & i_2 , & & ) . + & & . - ( _ k=0^n-4 _ kc_1 ) ( clcr n-2 , & 0 , & & + i_1 , & i_2 , & & ) } + & + & _ i_1 , ( i_2 , i_3)^n ( ) ^n-4 ( ) ( ) \ { ( ( n 6 ) + _ n,5 ) . + & & ( clcr n-4 , & 1 , & 1 & 0 + i_1 , & i_2 & i_3 & ) . - ( _ k=0^n-4 _ kc_1 ) ( clcr n-2 , & 0 , & 0 , & + i_1 , & i_2 & i_3 , & ) } + & + & _ i_1 , i_2^n ( ) ^n-5 ( ) ^3 \ { ( ( n 9 ) + _ n,8 ) ( clcr n-5 , & 3 , & & + i_1 , & i_2 & & ) . + & & . - ( _ k=0^n-4 _ kc_2 ) ( clcr n-2 , & 0 , & & + i_1 , & i_2 , & & ) } + & + & _ i_1 , i_2 , i_3^n ( ) ^n-5 ( ) ^2 ( ) \ { ( ( n 8 ) + _ n,7 ) . + & & . ( clcr n-5 , & 2 , & 1 , & + i_1 , & i_2 , & i_3 , & ) - ( _ k=0^n-4 _ kc_2 ) ( clcr n-2 , & 0 , & 0 & + i_1 , & i_2 , & i_3 , & ) } + & + & _ i_1 , ( i_2 , i_3 , i_4 ) ^n ( ) ^n-5 ( ) ( ) ( ) \ { ( ( n 7 ) + _ n,6 ) . + & & . ( clccr n-3 , & 1 , & 1 , & 1 , & + i_1 , & i_2 , & i_3 , & i_4 , & ) - ( _ k=0^n-4 _ kc_2 ) ( clccr n-2 , & 0 , & 0 , & 0 , & + i_1 , & i_2 , & i_3 , & ) } + & + & + & + & _ ( i_1 , i_2 , , i_n-2 ) ^n ( ) ( ) \ { ( clclcr 1 , & 1 , & & 1 , & 0 , & + i_1 , & i_2 , & & i_n-2 , & & ) . + & - & . ( clclcr n-2 , & 0 , & & 0 , & 0 , & 0 + i_1 , & i_2 , & & i_n-2 , & & ) - .... - ( clclcr 0 , & 0 , & & n-2 , & 0 & 0 + i_1 , & i_2 , & & i_n-2 , & & ) } . here the summations with a parenthesis @xmath83 are over @xmath64 unequal indistinguished indices . using this formula and the partial fraction , one can put the expression into a form which does not contain a direct link among solid circles . this is sometimes useful in the manipulation we carry out later . the following parametrization of @xmath84 @xcite is important in the next section : [ eq : z= ] z_i^ * = ( 2_i ) . this parametrization is understood together with p_i- p_i^*= [ ] ( z_i ; _ i ) -[]^ * = a m m_i , = ( a m/2)^1/m [ 3loop - o ] and -_*=-(m+1 ) ^2m=-(m+1)a^2 t , m = ( 2t)^1/2 . [ 3loop - p ] we obtain _ _ i = - , = 2 ( ) . the origin of the parametrization eq . ( [ eq : z= ] ) comes from the planar solution to the heisenberg algebra @xcite in @xcite . in fact , eq . ( [ 3loop - o ] ) represents the solution at the @xmath1 critical point . ( @xmath85 and @xmath86^ { * } $ ] represent the critical value to @xmath19 and to @xmath86 $ ] respectively . ) we now discuss the term we have isolated in eq . ( [ eq : deformed - delta ] ) , the expression for the @xmath0-point resolvent , namely the one which comes with the highest number of total derivatives with respect to @xmath80 and is , therefore , familiar from the case of pure two - dimensional gravity . we will exhibit striking properties with this term , using the parametrization noted in eq . ( [ eq : z= ] ) . let us define [ eq : polygon ] p_n(_1 , _ 2 , _ n ) & & _ i_1^n ( ) ( clcr n-2 , & 0 , & , & 0 + i_1 , & i_2 , & , & i_n ) + & = & _ i=1^n . a key manipulation we will use is the partial fraction [ eq : pf ] = + . one can associate a line from @xmath69 to @xmath70 with @xmath87}$ ] . the following identity is responsible for expressing @xmath88 as a sum of the product of @xmath0 factors each of which depends only on @xmath89 : i_k ( , ; m ) & & + & & = -2_j=1^m - k_i=1^k . [ eq : ident ] let us first work out the cases @xmath90 to get a feeling . for @xmath91 , & & p_2(_1 , _ 2 ) = = 2 ( ) i_k_1= 1 ( _ 1 , _ 2 ; m ) + & = & 2 ( ) ( ) _ j=1^m - k_1_i_1=1^k_1 + & = & 2 ( ) ( ) _ j_1=1^m-1 , where @xmath92 . for @xmath32 , we use eq . ( [ eq : pf ] ) for the term containing @xmath93 to create a link @xmath94 $ ] , which is originally absent . this relates @xmath95 to @xmath96 . we find & & p_3 ( _ 1 , _ 2,_3 ) + & = & 2 ( ) + & = & 2 ( ) ( ) ^2 ( _ j_1=1^m - k_1_i_1=1^k_1=1 ) ( _ j_2=1^m - k_2_i_2=1^k_2 ) + & & . here @xmath97 . this can be repeated for arbitrary @xmath0 . in the case @xmath5 , we use the partial fraction for the two terms containing @xmath93 and @xmath98 to create a link @xmath99 $ ] , which is originally absent . this enables us to relate the case @xmath5 to the case @xmath32 . in general , @xmath88 is related to @xmath100 by using the partial fraction for the terms containing @xmath101 to create a link @xmath102 $ ] . we obtain [ eq : answer ] & & p_n(_1 , _ 2 , _ n ) = - ( ) ( ) ^n ( _ = 1^n-1 _ j_=1^m - k__i_=1^k _ ) + & & ( _ ^=1^n-1 ) , where @xmath103 for @xmath104 . ( [ eq : answer ] ) expresses the @xmath105 as a sum of the @xmath0-products of the factor @xmath106 . owing to this property , one can perform the inverse laplace transform immediately , which we will carry out at eq . ( [ eq : fusionanswer ] ) . let us now discuss the restrictions on the summations of @xmath30 integers @xmath107 @xmath108 @xmath29 @xmath109 in eq . ( [ eq : answer ] ) . we write these as a set : & & f_n ( j_1 , i_2 , j_2 i_n-1 , j_n-1 ) + & & \ { ( j_1 , i_2 , j_2 , i_n-1 , j_n-1 ) 1 i _ k _ , 1 j _ m- k _ , = 1,2 , n-1 } + & = & f_2 ( i_1=1 , j_1;k_1=1 ) _ = 2^n-1 _ 2 ( i _ , j_;k _ ) , where _ 2 ( i _ , j_;k _ ) \ { ( i _ , j _ ) 1 i _ k _ , 1 j _ m- k _ , k _ fixed } . we will show that these restrictions on the sums are in fact in one - to - one correspondence with the fusion rules of the unitary minimal models for the diagonal primaries . let us begin with the case @xmath32 . define lcr p_1j_1 + k_1 - i_1 , & p_2 j_2 + k_2 - i_2 , & q_3 j_2 + i_2 - 1 , + a_12 p_1 -1 , & a_23 p_2 -1 , & a_31 q_3 -1 , the inequalities on @xmath110 are found to be equivalent to the following four inequalities : [ eq : trieq ] a_12 + a_23 - a_31 & = & 2(k_2-i_2 ) 0 . + a_12 - a_23 + a_31 & = & 2(i_2 - 1 ) 0 . + - a_12 + a_23 + a_31 & = & 2(j_2 - 1 ) 0 + a_12 + a_23 + a_31 & = & 2(j_2 + k_2 - 2 ) 2(m-2 ) . from the third and the fourth equation of eq . ( [ eq : trieq ] ) , the inequality @xmath111 follows , which is a condition for @xmath112 . defining a set [ eq : d3 ] d_3(a_1 , a_2 , a_3 ) & & \ { ( a_1 , a_2 , a_3 ) _ i(j)^3 a_i - a_j 0 i=1 ~3 , + & & _ i=1 ^ 3 a_i = even 2(m-2 ) } , we state eq . ( [ eq : trieq ] ) as _ 3(j_1 , i_2 , j_2 ) = d_3 ( a_12,a_23,a_31 ) . we also write _ 2(j_1 ) _ 2 ( i_1=1 , j_1;k_1=1 ) _ 2 ( a_12 ) . for the case @xmath91 . ( [ eq : d3 ] ) is nothing but the condition that a triangle be formed which is made out of @xmath113 and @xmath114 and whose circumference is less than or equal to @xmath115 . it is also the selection rule for the three point function of the diagonal primaries in @xmath116-th minimal unitary conformal field theory @xcite . in fact , the fusion rules for diagonal primary fields read as _ i i _ j j _ k k= 0 , if and only if @xmath117 and two other permutations and @xmath118 odd ) @xmath119 hold . this set of rules is nothing but @xmath120 . for the case @xmath5 , introduce @xmath121 . we find _ 2 ( i_3 , j_3;k_3 ) = d_3 ( a_31,a_34,a_41 ) the restrictions on the sum in the case @xmath5 can be understood as gluing the two triangels : [ eq : f4 ] f_4 ( j_1 , i_2 , j_2 , i_3 , j_3 ) & = & d_3 ( a_12,a_23,a_31 ) _ 3 ( a_34,a_41,a_31 ) + & & d_4 ( a_12,a_23,a_34,a_41 ; a_31 ) . the allowed integers on @xmath122 are naturally interpreted as permissible quantum numbers flowing through an intermediate channel . as one can imagine , eq . ( [ eq : f4 ] ) is not the only way to represent the restriction : one can also represent it as & & f_4 ( j_1 , i_2 , j_2 , i_3 , j_3)= d_3 ( a_12,a_24,a_41 ) _ 3 ( a_23,a_34,a_24 ) + & & d_4 ( a_12,a_23,a_34,a_41 ; a_24 ) , which embodies the crossing symmetric property of the amplitude . the restrictions in the general case @xmath0 are understood as attaching a triangle to the case @xmath7 . to see this , define lr p _ = j _ + k _ - i _ , & q _ = j_-1 + i_-1 -1 , + a _ , 1 = q_-1 , & a _ , + 1 = p_-1 , . using @xmath123 @xmath124 , we derive a_n-1,n + a_n,1 - a_n-1,1 & = & 2 ( j_n-1 - 1 ) 0 , + a_n-1,n - a_n,1 + a_n-1,1 & = & 2(k_n-1 - i_n-1 ) 0 , + - a_n-1,n + a_n,1 + a_n-1,1 & = & 2(i_n-1 - 1 ) 0 , + a_n-1,n + a_n,1 + a_n-1,1 & = & 2(j_n-1 + k_n-1 -2 ) 2(m-2 ) . the restriction on @xmath125 and @xmath126 are , therefore , @xmath127 , which is what we wanted to see . all in all , we find [ eq : dn ] & & f_n ( j_1 , i_2 , j_2 , i_n-1 , j_n-1 ) + & = & d_3(a_n-1,n , a_n,1 , a_n-1,1 ) _ n-1 ( j_1 , i_2 , j_2 , i_n-2 , j_n-2 ) + & = & d_3(a_n-1,n , a_n,1 , a_n-1,1 ) _ n-1(a_1,2,a_2,3 , a_n-2,n-1 , a_n-1,1 ; a_3,1 , a_4,1 , a_n-2,1 ) + & & d_n(a_1,2,a_2,3 , a_n-1,n , a_n,1 ; a_3,1 , a_4,1 , a_n-1,1 ) from now on , a shortened notation @xmath128 is understood to represent @xmath129 . putting eq ( [ eq : answer ] ) and eq . ( [ eq : dn ] ) together , we obtain a formula [ eq : fanswer ] & & p_n(_1 , _ 2 , _ n ) + & = & - ( ) ( ) ^n _ d_n ( _ j = 2^n ) , + where @xmath130 means @xmath131 . once again , the fact that the different divisions of @xmath132 into @xmath133 triangles are embodied by this single expression is precisely the statement of the old duality . the object @xmath105 is equipped with @xmath134 and @xmath135 for @xmath136 and any @xmath137 obeys the rule of the triangle specified above . it is , therefore , natural to visualize this as a vertex which connects @xmath0 external legs corresponding to @xmath0 loops . the vertex can be regarded as a dual graph of an n - gon that corresponds @xmath131 . using the formula ( [ eq : inverse ] ) , we perform the inverse laplace transform with respect to @xmath21 @xmath138 . we obtain [ eq : fusionanswer ] & & ( _ j=1^n l_j^-1 ) ( ) ^n-2 ( ) ^n-3 _ i=1^n ( - ) p_n(_1 , _ 2 , , _ n ) + & & = ( -)^n+1 ( ) ^n-2 ( ) ^n-3 . where @xmath139 denotes the inverse laplace transform with respect to @xmath140 . expressing this by @xmath17 and @xmath3 we obtain [ eq : fusionanswerf ] & & a_n^ fusion ( _ 1 , _ 2 , _ n ) + & & = - ( ) ^n-2 ( ) ^n-3 . this is the answer quoted in the introduction . it is straightforward to look at the small length behavior of eq.([eq : fusionanswerf ] ) . this was done in @xcite in the case @xmath32 , using the formula [ eq : kform ] k_(x ) = . the agreement with the approach from the generalized kdv flows @xcite ( see also @xcite . ) has been given . we will not dwell on this point further . our formula in the last section tells how the higher order operators ( gravitational descendants ) in addition to the dressed primaries included in the form of the loop length are constrained to obey the selection rules of cft . the two - matrix model realizing the unitary minimal series coupled to gravity as the continuum limit of the @xmath1 symmetric critical point knows the fusion rules and the duality symmetry in the form of the loop operators . the term we have dealt with in the last section for general @xmath0 is , however , supplemented with an increasing number of other terms with @xmath0 @xmath141 . the existence of such terms itself implies that the knowledge we obtain from the two and the three point functions is not sufficient to determine the full amplitude for @xmath142 . this coincides with the notion of contact interactions familiar from the field theory of derivative couplings as well as in ( super ) string theory @xcite . the counterpart of our approach to this phenomenon in the continuum framework is presumably related to the discusson on the boundary of moduli space . in what follows , we show how to perform the inverse laplace transformation of the resolvents to get loop amplitudes in terms of loop lengths in the case of @xmath5 , @xmath143 . it is necessary to put _ n(_1,,_n ) _ n(z^*_1,,z^*_n ) |__i= in a manageable form to the inverse laplace transform . let us recall that @xmath144 can be inverse laplace transformed immediately . if @xmath145 is expressed as a polynomial of @xmath146 and their derivatives with respect to @xmath80 , the inverse laplace transform can be done immediately . let us pursue this possibility . we also make use of the fact that when one of the loops shrinks and the loop length goes to zero , the @xmath0-loop amplitude must become proportional to the derivative of the @xmath7-loop amplitude with respect to the cosmological constant . we represent this fact by _ n(_1,,_n ) _ n-1(_1,,_n-1 ) . the inverse laplace transformation of @xmath144 is ^-1 = - ( ) ( ) ^n ( ) ^n _ d_n and in the limit @xmath147 we have _ 1-k / m(m ) ( m)^k / m-1 . if the n - th loop shrinks we have , therefore , ^-1 ^-1 . [ eq : shrink - p ] then in the limit @xmath148 , @xmath145 must satisfy the following relation : ^-1 ^-1[eq : shrink - delta ] this relation restricts the possible form of @xmath145 . as we want @xmath145 to be expressed as a polynomial of @xmath149 and their derivatives with respect to @xmath80 , we need to introduce a notation [ eq : notations ] & & ( _ 1 , _ 2 , _ n ) + & & _ _ n p_i_1 ( _ ( 1 ) , , _ ( i_1 ) ) p_i_2(_(j_2 ) , , _ ( j_2 + i_2 - 1 ) ) p_i_(_(n - i _ + 1 ) , , _ ( n ) ) , where @xmath150 represents the permutations of @xmath151 . to be more specific , for example [ eq : sp ] ( _ 1 , _ 2 , _ 3,_4 ) = _ _ 4 p_3(_(1 ) , _ ( 2),_(3 ) ) p_3(_(2 ) , _ ( 3),_(4 ) ) + ( _ 1 , _ 2 , _ 3,_4 ) = _ _ 4 p_4(_(1 ) , _ ( 2),_(3 ) _ ( 4 ) ) p_2(_(3),_(4 ) ) . it is convenient to represent @xmath144 by an n - vertex which connects n external legs . for example for @xmath91 , 3 and 4 & & p_2(_1,_2 ) = 2.5 cm = = 0.5 by -0.3ex , p_3(_1,_2,_3 ) = 2.5 cm = = 0.5 by -0.3ex + & & p_4(_1,_2,_3,_4 ) = 2.5 cm = = 0.5 by -0.3ex the n - vertex can be regarded as a dual graph of the n - gon ( polygon ) which corresponds to @xmath130 . in terms of these vertices , let us express eq . ( [ eq : sp ] ) as follows . ( _ 1 , _ 2,_3,_4 ) & & = 2.5 cm = = 0.5 by -0.3ex + ( _ 1 , _ 2,_3 , _ 4 ) & & = 2.5 cm = = 0.5 by -0.3ex the relation eq . ( [ eq : shrink - p ] ) can be represented , for example , as = 2.5 cm = = 0.5 by -0.3ex = 2.5 cm = = 0.5 by -0.3ex in the case of n=5 . now we are concerned with the case of n=4 first . let us recall that for n=3 _ 3(_1,_2,_3 ) p_3(_1,_2,_3 ) .@xmath152 must include a term which becomes proportional to @xmath153 in the limit @xmath154 , which is @xmath155 . @xmath156 may also include terms which vanish in this limit . such terms must consist of the product of two multi - vertices which have 6 external legs in total . by explicit computation , we find & & _ 4(_1 , _ 2 , _ 3 , _ 4 ) = p_4 ( _ 1 , _ 2,_3 , _ 4 ) -(_1 , _ 2,_3 , _ 4 ) + & & + ( _ 1 , _ 2 , _ 3 , _ 4 ) + & & = = 2.5 cm = = 0.5 by -0.3ex + & & + = 2.5 cm = = 0.5 by -0.3ex , [ eq : n=4delta ] where the prime represents the differentiation with respect to @xmath80 . carrying out the procedure indicated in ( [ eq : procedure ] ) for the case @xmath5 together with eqs . ( [ eq : result ] ) , ( [ eq : n=4delta ] ) , we obtain the complete answer for the macroscopic four loop ampitude : [ eq : n=4answer ] & & a_4 ( _ 1 , _ 4 ) = a_4^fusion ( _ 1 _ 4 ) + & + & t^-2 - ( _ j=1 ^ 4 ) ( _ _ 4 ) \ { - _ d_3 _ d_3^ _ j=1 ^ 4 ( m _ ( j ) ) . + & + & . _ d_4 _ d_2^ _ j=1 ^ 4 ( m _ ( j ) ) } . here @xmath157 , @xmath158 , @xmath159 @xmath160 and @xmath161 . we have introduced @xmath162 , @xmath163 , @xmath164 and @xmath165 and have defined the convolution @xmath166 by ( m ) _ 0^ a(m^ ) b(m ( -^ ) ) . let us now turn to the n=5 case . @xmath167 include a term which is proportional to [ eq : n=5type1 ] p_5(_1,,_5 ) = = 2.5 cm = = 0.5 by -0.3ex corresponding to the first term in eq . ( [ eq : n=4delta ] ) . corresponding to the second term in eq . ( [ eq : n=4delta ] ) , @xmath167 must include a term which consists of the product of two multi - vertices with 7 extenal legs in total . the possible form is [ eq : n=5type2 ] & & a + b + & & + c + d + & & = a = 2.5 cm = = 0.5 by -0.3ex + b = 2.5 cm = = 0.5 by -0.3ex + & & + c = 2.5 cm = = 0.5 by -0.3ex + d = 2.5 cm = = 0.5 by -0.3ex . in the limit @xmath168 , it becomes ( 3a+c ) + ( 3b+d ) + + 2c + 2d . we require this expression to be proportional to the @xmath80-derivative of the second term in eq .([eq : n=4delta ] ) . we find a=(1-c)/3,b=(2+c)/3,d=-1-c . @xmath167 may include terms which vanish in the limit under consideration as well . they must consist of the products of three multi - vertices with 9 external legs in total . as one of the such terms we have [ eq : n=5type3 ] & & [ sp_1234p_34p_45 ] -[sp_1234p_23p_345 ] + [ sp_123p_124p_235 ] + & & = = 2.5 cm = = 0.5 by -0.3ex -2 = 2.5 cm = = 0.5 by -0.3ex + & & + = 2.5 cm = = 0.5 by -0.3ex . ( there are some other combinations which satisfy the conditions . ) @xmath167 must be expressed as a linear combination of the above three types of terms eq . ( [ eq : n=5type1 ] ) , eq . ( [ eq : n=5type2 ] ) and eq . ( [ eq : n=5type3 ] ) if the assumption under consideration is true . by explicit calculation , we have found in fact that @xmath167 can be expressed as a linear combination of eq . ( [ eq : n=5type1 ] ) , eq . ( [ eq : n=5type2 ] ) and eq . ( [ eq : n=5type3 ] ) : & & _ 5(_1 , _ 2 , _ 3 , _ 4 , _ 5 ) = ( ) ^2 p_5(_1 , _ 2 , _ 3 , _ 4 , _ 5 ) + & + & ( _ 1 , _ 2 , _ 3 , _ 4 , _ 5 ) + & - & ( _ 1 , _ 2 , _ 3 , _ 4 , _ 5 ) + & + & ( _ 1 , _ 2 , _ 3 , _ 4 , _ 5 ) . following the same procedure as obtaining eq . ( [ eq : n=4answer ] ) , we find the complete answer for the five loop amplitude : [ eq : n=5answer ] & & a_5 ( _ 1 , _ 5 ) = a_5^fusion ( _ 1 _ 5 ) + ( _ j=1 ^ 5 ) ( _ _ 5 ) + & & . here @xmath169 means the derivative acting only on the left(right ) part of the convolutions . the rest of the notations here are similar to those of the @xmath5 case and will be self - explanatory.),([eq : n=5answer ] ) to obtain the corresponding expression for the microscopic operators . ( use eq . ( [ eq : kform ] ) ) . ] we conjecture that @xmath170 can be represented as a polynomial of @xmath171 and their @xmath3 derivatives for any @xmath0 : the final answer would then be obtained by convolutions of various @xmath172 s and their derivatives . if the conjecture is true in fact , the power counting argument tells us that the j - th term of @xmath173 turns out to be represented by a figure which consists of the products of j multi - vertices with @xmath174 external legs in total . we hope that , for higher loops , @xmath175 can be put in principle in a form such as eqs . ( [ eq : n=4answer ] ) , ( [ eq : n=5answer ] ) in the same manner as we have determined the four and five loops from the lower ones . explicit determination of the full amplitude in this way beyond five loops , however , appears to us still very formidable . we thank atsushi ishikawa for enjoyable collaboration in @xcite . we also thank kenji hamada , keiji kikkawa , and alyosha morozov for helpful discussions on this subject . in this appendix , we prove the recursion relations for ( assume that eq . ( [ eq : simpler1 ] ) and eq . ( [ eq : simpler2 ] ) are true at @xmath0 . without loss of generality , @xmath176 can be taken to be @xmath177 . let us consider the left hand side of eq . ( [ eq : simpler1 ] ) or eq . ( [ eq : simpler2 ] ) in which @xmath0 is replaced by @xmath178 . to compute them we observe that the elements of @xmath179 are generated by associating @xmath0 different ways of inserting @xmath180 $ ] with each element @xmath181 . in the case where @xmath180 $ ] is inserted in between @xmath182 $ ] and @xmath183 $ ] , this contribution is equal to [ eq : cont1 ] - _ _ n 1[1-(1)]^m _ j=2^n 1[j - ( j ) ] . the contributions from the sum of the remaining @xmath78 insertions are found to be equal to [ eq : cont2 ] - _ _ n 1[1-(1)]^m _ j=2^n 1[j - ( j ) ] . here we have used [ eq : frac ] = ( - ) . ) _ n . then from the assumption , eq . ( [ eq : simpler1 ] ) and eq . ( [ eq : simpler2 ] ) are also satisfied when @xmath0 is replaced by @xmath178 . on the other hand for @xmath32 eq . ( [ eq : simpler1 ] ) and eq . ( [ eq : simpler2 ] ) are clearly true , so we have proven the relations . we take @xmath186 @xmath187 without loss of generality . the way in which the elements of @xmath179 are generated is the same as the one given above . in the case where @xmath180 $ ] is inserted in between @xmath188 $ ] and @xmath189 $ ] @xmath190 , the contribution is [ eq : cont1 g ] - _ _ n 1[1-(1)]^m_1 + 1 1[-()]^m _ + 1 [ ( + 1 ) -(+1)]^m_+1 + 1 1 [ k-(k)]^m_k+1 _ j ( 1 , 2 , k)^n-1 1[j - ( j ) ] . the contributions from the sum of the remaining insertions are & - & _ _ n _ ( p_2,p_k ) ^n-1 1 [ 1-(1)]^m_1 + 1 1 [ k-(k)]^m_k+1 + & & _ j ( p_2 , p_k ) ^n-1 1[^j(1 ) - ^j+1(1 ) ] ( - ) . here @xmath192 are such that @xmath193 . using eq . ( [ eq : frac ] ) again , we find that this equals [ eq : cont2 g ] - _ _ n _ = 1^k 1 [ 1-(1)]^m_1 + 1 1 [ -()]^m _ + 1 [ ( + 1 ) -(+ 1)]^m_+1 + 1 1 [ k -(k)]^m_k+1 _ j ( 1 , 2 , k)^n-1 1[j - ] . putting eqs . ( [ eq : cont1 g ] ) and ( [ eq : cont2 g ] ) together , we find [ eq : * * * ] & & ( ) _ n . + then from the assumption eq . ( [ eq : relation3 ] ) at @xmath0 , eq . ( [ eq : relation3 ] ) in which @xmath0 is replaced by @xmath178 is also satisfied . on the other hand for @xmath32 eq . ( [ eq : relation3 ] ) is clearly satisfied , so we have proven eq . ( [ eq : relation3 ] ) and eq . ( [ eq : relation4 ] ) . e. gava and k. narain , _ phys . _ * b263 * ( 1991 ) 213 ; h. itoyama and y. matsuo , _ phys . _ * b262 * ( 1991 ) 233 ; m. staudacher , _ phys . lett . _ * b305 * ( 1993 ) 332 ; l. bonora and c.s . xion , _ nucl . _ * b405 * ( 1993 ) 191 ; _ phys . _ * b347 * ( 1995 ) 41 , hep - th/9405004 ; l. bonora , c. p. constantinidis and c.s . xion , hep - th/9412006 , sissa 189/94/ep , uthep-695 ;

we show that the proportionately spaced emission bands in the dynamic spectrum of the crab pulsar ( hankins t. h. & eilek j. a. , 2007 , apj , 670 , 693 ) fit the oscillations of the square of a bessel function whose argument exceeds its order . this function has already been encountered in the analysis of the emission from a polarization current with a superluminal distribution pattern : a current whose distribution pattern rotates ( with an angular frequency ) and oscillates ( with a frequency differing from an integral multiple of ) at the same time ( ardavan h. , ardavan a. & singleton j. , 2003 , j opt soc am a , 20 , 2137 ) . using the results of our earlier analysis , we find that the dependence on frequency of the spacing and width of the observed emission bands can be quantitatively accounted for by an appropriate choice of the value of the single free parameter . in addition , the value of this parameter , thus implied by hankins & eilek s data , places the last peak in the amplitude of the oscillating bessel function in question at a frequency ( ) that agrees with the position of the observed ultraviolet peak in the spectrum of the crab pulsar . we also show how the suppression of the emission bands by the interference of the contributions from differring polarizations can account for the differences in the time and frequency signatures of the interpulse and the main pulse in the crab pulsar . finally , we put the emission bands in the context of the observed continuum spectrum of the crab pulsar by fitting this broadband spectrum ( over 16 orders of magnitude of frequency ) with that generated by an electric current with a superluminally rotating distribution pattern . [ firstpage ] pulsars : individual ( crab nebula pulsar)radiation mechanisms : non - thermal .

0804.3123

very soon after the discovery of pulsars , it was realized that the very stable periodicity of the mean profiles of their pulses could only result from a source that _ rotates _ , and which therefore posesses a rigidly rotating radiation distribution @xcite . in this paper , we show that this source rotation is not only responsible for the periodicity of the pulses , but also determines the detailed frequency dependence of the emitted radiation . by inferring the values of two adjustable parameters from observational data ( values that are consistent with those of plasma frequency and electron cyclotron frequency in a conventional pulsar magnetosphere ) , and by mildly restricting certain local properties of the source , we are able to account _ quantitatively _ for the emission spectrum of the crab pulsar over 16 orders of magnitude of frequency . the rigid rotation of the overall distribution pattern of the radiation from a pulsar is described by an electromagnetic field whose distribution depends on the azimuthal angle @xmath4 only in the combination @xmath5 , where @xmath0 is the angular frequency of rotation of the pulsar and @xmath6 is the time . as we show in appendix a , maxwell s equations demand that the charge and current densities that give rise to this radiation field should have the same time dependence . therefore , the observed motion of the radiation pattern of pulsars can only arise from a source whose _ distribution pattern _ rotates rigidly , i.e. a source whose average density depends on @xmath4 only in the combination @xmath5 . furthermore , if a plasma distribution has a rigidly rotating pattern in the emission region , then it must have a rigidly rotating pattern everywhere ; we show in appendix b that a solution of maxwell s equations with the time dependence @xmath7 applies either to the entire volume of the magnetospheric plasma distribution or to a region whose boundary is an expanding wave front that will eventually encompass the entire magnetosphere of the pulsar . unless the plasma atmosphere surrounding the pulsar is restricted to an unrealistically small volume , it is therefore an inevitable consequence of the observational data that the macroscopic distribution of electric current in the magnetosphere has a rigidly rotating pattern whose linear speed exceeds the speed of light _ in vacuo _ @xmath8 for @xmath9 , where @xmath10 is the radial distance from the axis of rotation . although special relativity does not allow a charged particle with a non - zero inertial mass to move faster than @xmath8 , there is no such restriction on the speed of propagation of the variations in a macroscopic charge or current distribution . for instance , the distribution pattern of a polarization current with a propagation speed that exceeds @xmath8 can be created by the coordinated motion of aggregates of particles that move slower than @xmath8 @xcite . such a polarization - current density is on the same footing as the current density of free charges in the ampre - maxwell equation , so that its propagating distribution pattern radiates as would any other moving source of electromagnetic fields @xcite ; indeed , such superluminal polarization currents have been demonstrated to be efficient emitters of radiation in the laboratory @xcite . once it is acknowledged that the electric current emitting the observed pulses from pulsars has a superluminally - rotating distribution pattern , results from the published literature on the electrodynamics of superluminal sources @xcite can be applied to pulsars . in this paper , we explain the recently - observed emission bands in the dynamic spectrum of the crab pulsar @xcite using the calculations of @xcite . both the oscillations of the intensity and the frequency dependence of the spacing and width of these bands are described by a bessel function with a single parameter that is already constrained by other observational data on the spectrum of the crab pulsar . this bessel function is characteristic of the spectrum of the radiation by a superluminal polarization current whose distribution pattern rotates ( with an angular frequency @xmath0 ) and oscillates ( with a frequency @xmath1 differing from an integral multiple of @xmath0 ) at the same time . it differs from the bessel function encountered in the analysis of synchrotron radiation only in the relative magnitudes of its argument and its order : while the bessel function describing synchrotron radiation has an argument smaller than its order and so decays exponentially with increasing frequency , the bessel function encountered in @xcite , whose argument exceeds its order , is an oscillatory function of frequency with an amplitude that decays only algebraically . it is this slower decay of the bessel function in question that endows the emission from a superluminally - rotating source with a broad spectrum . the physical mechanism underlying the broadband nature of this emission is focusing in the time domain @xcite : contributions toward the intensity of the radiation made over an extended period of emission time are received during a significantly shorter period of observation time . this paper is organized as follows . section [ newsec2 ] describes how the most radiatively efficient parts of a pulsar magnetosphere are thin filaments within the superluminally rotating part of its current distribution pattern @xcite . a knowledge of the morphology of these filaments @xcite , and the ability of a source that travels faster than its own waves to make multiple contributions to the signal received by an observer at a given instant @xcite , are necessary to understand many of the traits of pulsar observations described later in the paper . section [ newsec3 ] summarizes our earlier analysis @xcite on the frequency spectrum of the radiation from a rotating superluminal source . using these results , we derive various features of the emission bands in section [ newsec4 ] and compare them to the frequency bands seen in the interpulses of the crab pulsar [ figs . 68 , @xcite ] ; using a single input parameter , related to the pulsar s plasma density , we reproduce the observational bands and predict the final ultra - violet emission peak of the crab pulsar . section [ newsec5 ] discusses the suppression of the bands by interference ; this is relevant to the much less frequency - banded microbursts and nanoshots of the crab s main pulses [ figs . 2 and 3 , @xcite ] . section [ newsec6 ] describes the continuum spectrum of the crab pulsar ; by introducing a further input parameter related to the plasma dynamics we are able to account _ quantitatively _ for the whole emission spectrum over 16 orders of magnitude of frequency . section [ newsec7 ] gives a short discussion and summary . the mathematical details of our arguments are presented in appendices a , b and c. at large distances from the source , the radiation field of a superluminally rotating extended source at an observation point @xmath11 is dominated by the emissions of its volume elements that approach @xmath11 along the radiation direction with the speed of light and zero acceleration at the retarded time @xcite . these elements constitute a filamentary part of the source whose radial and azimuthal widths become narrower ( as @xmath12 and @xmath13 , respectively ) , the larger the distance @xmath14 of the observer from the source , and whose length is of the order of the length scale @xmath15 of the source parallel to the axis of rotation @xcite . ( here , @xmath10 , @xmath4 , and @xmath16 are the cylindrical polar coordinates of the source points . ) for an observation point @xmath11 with spherical polar coordinates ( @xmath17 ) that is located in the far zone , this contributing part of the source lies at @xmath18 , and is essentially a straight line parallel to the rotation axis , as shown in fig . [ fig1 ] @xcite . ( the dimensionless coordinate @xmath19 stands for @xmath20 , where @xmath0 , the angular frequency of rotation of the distribution pattern of the source , is the same frequency as that with which the pulsar rotates . ) , the filamentary part of the distribution pattern of the source that approaches the observation point with the speed of light and zero acceleration at the retarded time , the orbit of this filamentary source , and the subbeam formed by the bundle of cusps that emanate from the constituent volume elements of this filament [ after @xcite ] . the subbeam is diffractionless in the polar direction . the figure represents a snapshot corresponding to a fixed value of the observation time @xmath21.,width=302 ] once a source travels faster than its emitted waves , it can make more than one retarded contribution to the field observed at any given instant @xcite . this multivaluedness of the retarded time means that the wave fronts emitted by each of the contributing elements of the source possess an envelope , which in this case consists of a two - sheeted , tube - like surface whose sheets meet tangentially along a spiraling cusp curve ; see fig . [ fig2 ] @xcite . for moderate superluminal speeds , the field inside the envelope receives contributions from three distinct values of the retarded time , while the field outside the envelope is influenced only by a single instant of emission time @xcite . coherent superposition of the emitted waves on the envelope ( where two of the contributing retarded times coalesce ) and on its cusp ( where all three of the contributing retarded times coalesce ) results in not only a spatial but also a temporal focusing of the waves : the contributions from emission over an extended period of retarded time reach an observer who is located on the cusp during a significantly shorter period of observation time @xcite . the field of each contributing volume element of the source is strongest , therefore , on the cusp of the envelope of wave fronts that it emits [ see @xcite and references therein ] . the bundle of cusps generated by the collection of the contributing source elements ( i.e. by the filamentary part of the source that approaches the observer with the speed of light and zero acceleration ) constitutes a radiation subbeam whose widths in the polar and azimuthal directions are of the order of @xmath22 and @xmath23 , respectively @xcite . the overall radiation beam generated by the source consists of a ( necessarily incoherent ) superposition of such subbeams . this beam s azimuthal width is the same as the azimuthal extent of the source , and its polar width , @xmath24 , is determined by the radial extent @xmath25 of the superluminal part of the source @xcite . this will be important in secs . [ newsec4 ] and [ newsec5 ] ; the @xmath26 dependence of the radiation intensity within the overall beam ( i.e. what is observed as the main pulse , interpulse , and other components of the mean pulse ) thus reflects the distribution of the source density around the cylindrical surface @xmath27 from which the main contributions to the field arise . since the cusps represent the loci of points at which the emitted spherical waves interfere constructively [ i.e. represent wave packets that are constantly dispersed and reconstructed out of other waves @xcite ] , the subbeams generated by a superluminal source need not be subject to diffraction as are conventional radiation beams . nevertheless , they have a decreasing angular width ( i.e. are nondiffracting ) only in the polar direction @xcite . their azimuthal width @xmath28 decreases as @xmath29 with distance because they receive contributions from an azimuthal extent @xmath30 of the source that likewise shrinks as @xmath29 . they would have had a constant azimuthal width had the azimuthal extent of the contributing part of the source been independent of @xmath14 . on the other hand , the solid angle occupied by the cusps has a thickness @xmath31 in the direction parallel to the rotation axis that remains of the order of the height @xmath15 of the source distribution at all distances ( see fig . 1 ) . consequently , the polar width @xmath32 of the particular subbeam that goes through the observation point decreases as @xmath33 , instead of being independent of @xmath14 @xcite . because it has a constant linear width parallel to the rotation axis , an individual subbeam subtends an area of the order of @xmath14 , rather than @xmath34 . in order that the energy flux remain the same across all cross sections of the subbeam , therefore , it is essential that the poynting vector associated with this radiation correspondingly decay more slowly than that of a conventional , spherically decaying beam : as @xmath33 , rather than @xmath35 , within the bundle of cusps that emanate from the constituent volume elements of the source and extend into the far zone @xcite . this result , which also follows from the superposition of the linard - wiechert fields of the constituent volume elements of a rotating superluminal source @xcite , has been demonstrated experimentally @xcite . within the distribution pattern of an extended , rotating superluminal source of angular frequency @xmath0 [ after @xcite ] . the point @xmath36 is on a circle of radius @xmath37 , or @xmath38 , i.e. has an instantaneous linear velocity @xmath39 . the cross section is with the plane of @xmath36 s orbit ; dashed circles designate the light cylinder @xmath40 ( @xmath41 ) and the orbit of @xmath36 . ( b ) three - dimensional view of the light cylinder , the envelope of wave fronts emanating from @xmath36 , and the cusp along which the two sheets of this envelope meet tangentially . the cusp is tangent to the light cylinder in the plane of the orbit and spirals outward into the far zone.,width=302 ] the fact that the observationally - inferred dimensions of the plasma structures responsible for the emission from pulsars are less than @xmath42 metre in size @xcite reflects , in the present context , the narrowing ( as @xmath35 and @xmath29 , respectively ) of the radial and azimuthal dimensions of the filamentary part of the source that approaches the observer with the speed of light and zero acceleration at the retarded time . not only do the nondiffracting subbeams that emanate from such filaments account for the nanostructure , and so the brightness temperature , of the giant pulses , but the nonspherical decay of the intensity of such subbeams ( as @xmath33 instead of @xmath35 ) explains why their energy densities at their source appear to exceed the energy densities of both the plasma and the magnetic field at the suface of a neutron star when estimated on the basis of the inverse square law @xcite . in this section we summarize the results of @xcite relevant to pulsars ; the original intent of that paper was to calculate the spectrum of the radiation from a generic superluminal source that has been implemented in the laboratory @xcite . this source comprises a polarization - current density @xmath43 for which @xmath44 with @xmath45 where @xmath46 are the components of the polarization @xmath47 in a cylindrical coordinate system ( @xmath48 ) based on the axis of rotation , @xmath49 is an arbitrary vector that vanishes outside a finite region of the @xmath50 space , and @xmath51 is a positive integer . for a fixed value of @xmath6 , the azimuthal dependence of the polarization ( [ eq:1 ] ) along each circle of radius @xmath10 within the source is the same as that of a sinusoidal wave train with the wavelength @xmath52 whose @xmath51 cycles fit around the circumference of the circle smoothly . as time elapses , this wave train both propagates around each circle with the velocity @xmath53 and oscillates in its amplitude with the frequency @xmath54 . this is a generic source : one can construct any distribution with a uniformly rotating pattern , @xmath55 or @xmath56 , by the superposition over @xmath51 of terms of the form @xmath57 . in the following discussion , we assume that the modulation frequency @xmath54 is positive and different from an exact integral multiple of the rotation frequency @xmath0 [ for the significance of this incommensurablity requirement , see @xcite ] . although formulated in terms of a polarization current ( for which there is manifestly no restriction on the propagation speed of the variations in the current density ) , the results that follow hold true for any current distribution whose density depends on @xmath4 as @xmath58 in @xmath9 . the electromagnetic fields @xmath59 that arise from such a source are given , in the absence of boundaries , by the following classical expression for the retarded four - potential : @xmath60 here , @xmath61 and @xmath62 are the space - time coordinates of the observation point and the source points , respectively , @xmath63 stands for the magnitude of @xmath64 , and @xmath65 designate the spatial components , @xmath66 and @xmath67 , of @xmath68 and @xmath69 in a cartesian coordinate system @xcite . in @xcite , we first calculated the linard - wiechert field that arises from a circularly moving point source ( representing a volume element of an extended source ) with a superluminal speed @xmath70 , i.e. considered a generalization of the synchrotron radiation to the superluminal regime . we then evaluated the integral representing the retarded field of the extended source ( [ eq:1 ] ) by superposing the fields generated by the constituent volume elements of this source , i.e. by using the generalization of the synchrotron field as the green s function for the problem [ see equation ( 19 ) of @xcite ] . for a source that travels faster than @xmath8 , this green s function has extended singularities arising from the constructive interference of the emitted waves on the envelope of wave fronts and its cusp ( see fig . [ fig2 ] ) . correspondingly , the fields @xmath71 and @xmath72 are given , in the frequency domain , by radiation integrals with rapidly oscillating integrands whose phases are stationary on the loci of the coherently contributing source elements : those source elements that approach the observer , along the radiation direction , with the speed of light and zero acceleration at the retarded time . for a radiation angular frequency @xmath73 that appreciably exceeds the angular velocity @xmath0 , hz . the condition @xmath74 will therefore always hold . ] these integrals can be evaluated by the method of stationary phase to obtain the following expression for the electric field of the emitted radiation outside the plane of the source s orbit : @xmath75 in which @xmath76\}\nonumber\\ & & \times q_{\hat\varphi}{\bar q}_r{\bar{\bf q}}_z+\{m\to -m,\omega\to-\omega\ } , \label{eq:6}\end{aligned}\ ] ] @xmath77 @xmath78 or @xmath79 and @xmath80{\hat{\bf e}}_\parallel\nonumber \\ & & + \big[({\bar s}_\varphi\cos\theta_p -{\bar s}_z\sin\theta_p){\bf j}_{n-\omega/\omega}(n)\nonumber\\ & & \qquad-{\rm i}{\bar s}_r\cos\theta_p{{\bf j}^\prime}_{n-\omega/\omega}(n)\big]{\hat{\bf e}}_\perp , \label{eq:9}\end{aligned}\ ] ] with @xmath81 [ see equations ( 15 ) , ( 23 ) , ( 46 ) and ( 66)(68 ) of @xcite ] . here , @xmath82 stand for @xmath83 , @xmath84 are the spherical polar coordintes of the observation point @xmath11 , @xmath85 , @xmath86 and @xmath87 are the upper and lower limits of the radial interval in which the source densities @xmath88 are non - zero , @xmath89 and @xmath90 are the anger function and the derivative of the anger function with respect to its argument , respectively , and @xmath91 ( which is parallel to the plane of rotation ) and @xmath92 comprise a pair of unit vectors normal to the radiation direction @xmath93 ( @xmath94 is the base vector associated with the coordinate @xmath16 ) . the symbol @xmath95 designates a term exactly like the one preceding it but in which @xmath51 and @xmath54 are everywhere replaced by @xmath96 and @xmath97 , respectively . the radiated power per harmonic per unit solid angle is therefore given by @xmath98 where @xmath99 denotes the element @xmath100 of solid angle in the space of observation points . the contribution of the term @xmath95 in equation ( [ eq:6 ] ) has been ignored here because the anger functions for @xmath101 turn out to be exponentially smaller than those for positive @xmath54 ( see appendix c ) . there is no difference between an anger function , @xmath102 , and a bessel function of the first kind , @xmath103 , when @xmath104 is an integer . even for a non - integral value of @xmath104 , the difference between these two functions vanishes , as @xmath105 , if @xmath106 . in the regime @xmath107 , where the radiation frequency @xmath73 appreciably exceeds the rotation frequency , therefore , the anger functions @xmath108 and @xmath109 in equation ( [ eq:9 ] ) can be replaced by the bessel functions @xmath110 and @xmath111 , respectively . if the radiation frequency @xmath73 appreciably exceeds also the modulation frequency @xmath54 , then these bessel functions can in turn be approximated by the following airy functions : @xmath112 , \label{eq:12}\ ] ] @xmath113 , \label{eq:13}\ ] ] where @xmath114 stands for the derivative of the airy function with respect to its argument ( see appendix c ) . hence , the radiated power is given by @xmath115\nonumber\\ & & \times{\rm ai}\left[-\left(\frac{2}{n}\right)^{1/3}\frac{\omega}{\omega}\right]-{\rm i}({\bar s}_\varphi{\hat{\bf e}}_\parallel+{\bar s}_r\cos\theta_p{\hat{\bf e}}_\perp)\nonumber\\ & & \times\left(\frac{2}{n}\right)^{1/3}{\rm ai}^\prime\left[-\left(\frac{2}{n}\right)^{1/3}\frac{\omega}{\omega}\right]\bigg\vert^2 \label{eq:14}\end{aligned}\ ] ] in the regime where the radiation frequency ( @xmath73 ) apprecialy exceeds both the rotation ( @xmath0 ) and modulation ( @xmath54 ) frequencies of the source distribution ( [ eq:1 ] ) . asymptotic values of the amplitudes of @xmath116 $ ] and @xmath117 $ ] are respectively given by @xmath118 and @xmath119 , when @xmath120 , and by the constants @xmath121 and @xmath122 when @xmath123 @xcite . moreover , the quantity @xmath124 , which appears in equation ( [ eq:14 ] ) , is independent of @xmath125 ( @xmath126 ) if @xmath127 , decays as @xmath128 if @xmath129 , and is of the order of @xmath125 if @xmath130 [ see equation([eq:7 ] ) ] . depending on the relative magnitudes of @xmath125 , @xmath51 and @xmath2 , therefore , the amplitude of the radiated power @xmath131 decays with the harmonic number @xmath125 as @xmath132 , where the exponent @xmath133 can assume one of the values @xmath134 , @xmath135 , @xmath136 or @xmath137 , and @xmath138 denotes the frequency dependence of the fourier components of the source densities @xmath88 . the way in which @xmath138 decay with @xmath125 , for large @xmath125 , is determined by the distribution of the source density @xmath49 with respect to @xmath16 [ see equation ( [ eq:10 ] ) ] , and so is model - dependent . ) does not influence these results , because it depends on the source frequencies @xmath139 , not on the radiation frequency @xmath73 [ see @xcite ] ] however , the spacings between the emission bands predicted by equation ( [ eq:14 ] ) turn out to be essentially independent of the rate of decay of the amplitudes of these bands ( see below ) . having assembled the relevant equations , we now compare their predictions with established observational data on the crab pulsar . radiated per harmonic per unit solid angle ( in arbitrary units ) , into the signal linearly polarized parallel to the plane of rotation , versus the harmonic number @xmath125 ( lower axis ) for @xmath140 , @xmath141 and @xmath142 . the upper axis gives observation frequency @xmath143 for the case of the crab pulsar [ @xmath144 hz ] . the spacing , the width , and the number of oscillations in this figure are the same as those of the bands seen in the interpulse data from the crab pulsar over the frequency interval @xmath145 ghz [ see figs . 68 ; @xcite].,width=302 ] one of the most remarkable features of the observational data from the crab pulsar is the presence of well - defined frequency bands within the midpulses @xcite . we now show that these are a natural consequence of a rotating superluminal source and use their spacing to predict the high - frequency emission of the crab pulsar . when the absolute value of their argument appreciably exceeds unity , the airy functions in equation ( [ eq:14 ] ) are rapidly oscillating functions of the harmonic number @xmath125 with peaks that spread further apart with increasing @xmath125 ( fig . [ fig3 ] ) . the rate of increase of the spacing between the peaks in question is determined solely by the value of @xmath2 . we now derive an explicit expression for the spacing @xmath146 of the maxima of the radiated power @xmath131 as a function of @xmath125 and compare the result with the proportionately spaced emission bands observed by @xcite . within the range @xmath147 of harmonic numbers , where they are oscillatory , the airy functions in equation ( [ eq:14 ] ) can be further approximated by @xmath148\simeq\pi^{-1/2}(n/2)^{1/12}(\omega/\omega)^{-1/4}\cos\psi \label{eq:14a}\ ] ] and @xmath149\simeq\pi^{-1/2}(n/2)^{-1/12}(\omega/\omega)^{1/4}\sin\psi \label{eq:14b}\ ] ] where @xmath150 ( see appendix c ) . in this regime , therefore , equation ( [ eq:14 ] ) reduces to @xmath151 in which @xmath152 and @xmath153 represent the contributions towards the radiated field @xmath154 from differing polarizations . let us first consider the contribution towards the radiated power from the component of the field parallel to the plane of rotation . it follows from equation ( [ eq:15a ] ) that @xmath155\cos^2(\psi-\gamma_\parallel)\big\ } , \label{eq:15c}\end{aligned}\ ] ] where @xmath156^{1/2}\im\{{\bar s}_r{{\bar s}_\varphi}^*\}}{\vert{\bar s}_r\vert^2-[2\omega/(n\omega)]\vert{\bar s}_\varphi\vert^2 } , \label{eq:15d}\ ] ] with @xmath157 and @xmath158 denoting the imaginary part and a complex conjugate , respectively . two successive oscillations of @xmath159 in the neighbourhood of any given @xmath125 arise , according to equation ( [ eq:15c ] ) , from a change in the value of the argument of @xmath160 by @xmath161 . therefore , the spacing between two successive peaks of this function is locally given by @xmath162 an equation that yields @xmath146 versus @xmath125 for all @xmath125 . versus @xmath104 employed in fig . 10 of @xcite by multiplying both @xmath146 and @xmath125 by @xmath163 hz , the rotation frequency relevant for the crab pulsar . ] if the source densities @xmath164 have a weak dependence on @xmath125 , then the high - frequency ( @xmath107 ) limits of equations ( [ eq:15c ] ) and ( [ eq:15d ] ) assume the simple forms @xmath165 and @xmath166^{1/2}[a_\parallel+b_\parallel(n - n_0 ) ] , \label{eq:15g}\end{aligned}\ ] ] where @xmath167 and @xmath168\bigg\vert_{n = n_0}. \label{eq:15i}\ ] ] denote the first two coefficients in the taylor expansion of @xmath169 about a local value @xmath170 of @xmath125 . in this case , equation ( [ eq:15e ] ) , together with equations ( [ eq:14c ] ) and ( [ eq:15f])([eq:15i ] ) , yields explicit expressions both for the radiated power @xmath159 and for the spacing @xmath146 between its successive peaks . these expressions are plotted in figs . [ fig3 ] and [ fig4 ] over the interval @xmath171 . in the case of the crab pulsar , where the rotational frequency @xmath172 is approximately @xmath173 hz , the interval of harmonic numbers shown in figs . [ fig3 ] and [ fig4 ] corresponds to a frequency [ @xmath174 interval covering @xmath175 to @xmath176 ghz , i.e. to the interval of frequencies over which @xcite observe the emission bands . we have chosen a value @xmath177 for the more fundamental free parameter @xmath2 that renders the slope @xmath178 predicted by equation ( [ eq:15e ] ) the same as the slope @xmath179 of the line passing through the observational data @xcite even when the phase @xmath180 is zero . can be approximated by an integer everywhere except in the factor @xmath181 that appears in equation ( [ eq:7 ] ) . ] we have then fitted the @xmath146 intercept of the resulting line to that of the data , without changing its slope , by fixing the values @xmath182 and @xmath183 of the parameters that appear in the expression for @xmath180 . the airy functions in equation ( [ eq:14 ] ) stop oscillating once the value of their argument falls below unity @xcite . we shall see in sec . [ newsec6 ] that the value @xmath184 of the parameter @xmath2 thus implied by the data of @xcite places the last peaks of these functions at a frequency [ @xmath185 that agrees well with the position @xmath186 hz of the ultraviolet peak in the observed spectrum of the crab pulsar @xcite . the widths of the emission bands shown in fig . [ fig3 ] also agree with the widths of those that are observed : the number of bands falling within the interval @xmath187 in fig . [ fig3 ] is the same as the number of bands that occur within the corresponding frequency interval ( @xmath188@xmath176 ghz ) in figs . 68 of @xcite . the contribution towards the radiated power from the component of the field perpendicular to the plane of rotation can be written , in analogy with equations ( [ eq:15c ] ) and ( [ eq:15d ] ) , as @xmath189\cos^2(\psi-\gamma_\perp)\big\ } , \label{eq:16}\end{aligned}\ ] ] where @xmath190^{1/2}\im\{({\bar s}_\varphi\cos\theta_p-{\bar s}_z\sin\theta_p){{\bar s}_r}^*\cos\theta_p\}}{\vert{\bar s}_\varphi\cos\theta_p-{\bar s}_z\sin\theta_p\vert^2-[2\omega/(n\omega)]\vert{\bar s}_r\vert^2\cos^2\theta_p}. \label{eq:17}\ ] ] these expressions imply a band spacing with essentially the same characteristics as those discussed above , differing from those in equations ( [ eq:15c ] ) and ( [ eq:15d ] ) only in that @xmath191 and @xmath192 exchange their roles with @xmath193 and @xmath194 , respectively . ( right axis ) of the emission bands shown in fig . [ fig3 ] versus their mean harmonic number @xmath125 ( lower axis ) . the solid line represents the function given in equation ( [ eq:15e ] ) for @xmath140 , @xmath195 and @xmath196 . the upper axis shows observation frequecy @xmath197 relevant for the crab pulsar ; the left axis shows @xmath198 . data points for the crab pulsar are taken from fig . 10 of @xcite.,width=302 ] it will therefore be seen that the salient parameter responsible for the frequency bands in the crab pulsar is the ratio @xmath199 . using @xmath200 hz relevant for the crab yields a modulation frequency @xmath201 khz . a plausible cause for this modulation is plasma oscillations within the pulsar s atmosphere . a plasma frequency of 570 khz implies a free - electron density @xmath202 , a value that is consistent with the inferred properties of the atmospheres of neutron stars . we now turn to the main pulses of the crab pulsar , in which the frequency - banded structure is much less apparent @xcite . if the components of the electric current within the emitting region ( the region in which the rotating distribution pattern of the current approaches the observer with the speed of light and zero acceleration ) are such that @xmath203 and @xmath204 simultaneously contribute towards the value of the field , destructive interference of the contributions with differing polarizations could result in the suppression of the emission bands . this can be easily seen from equation ( [ eq:15 ] ) if we note that , in this case , the radiated power for @xmath107 is proportional to @xmath205 , \label{eq:20}\ ] ] where @xmath206 is a coefficient that , like @xmath180 and @xmath207 , could be expanded into a taylor series about a local value @xmath170 of @xmath125 . figure [ fig5 ] shows the @xmath125 dependence of the right - hand side of equation ( [ eq:20 ] ) for a case in which @xmath208 decays as @xmath128 , @xmath140 ( as in figs . [ fig3 ] and [ fig4 ] ) , @xmath195 , @xmath196 ( as in fig . [ fig4 ] ) , @xmath209 , and @xmath210 can be approximated by @xmath211 with @xmath212 . the emission bands are thus replaced by small - amplitude modulations of the radiation intensity that gradually die out . the behaviour of the radiated power shown in fig . [ fig5 ] over the interval @xmath187 is consistent with those of the short - lived microbursts that are observed over the corresponding frequency interval ( @xmath188@xmath176 ghz ) within the main pulses of the crab pulsar [ see figs . 24 of @xcite ] . per harmonic per unit solid angle ( in arbitrary units ) versus @xmath125 from a region of the magnetosphere in which the contributions from differing polarizations are comparable in magnitude . the dependence of the right - hand side of equation ( [ eq:20 ] ) on @xmath125 is shown for @xmath213 , @xmath209 , @xmath214 , and the same values of @xmath2 , @xmath215 , @xmath216 , and @xmath170 as in fig . the top axis of this figure gives frequency @xmath104 relevant for observations of the crab pulsar.,width=302 ] the radiation described by equation ( [ eq:14 ] ) would be elliptically polarized if either @xmath192 or @xmath191 is dominant , and linearly polarized if @xmath217 is dominant . the high degree of linear polarization of the interpulse in the frequency range @xmath218 ghz @xcite suggests , therefore , that the interpulse arises from a region of the magnetosphere in which the component of the electric current parallel to the rotation axis dominates , and so the radiation is polarized in the direction of @xmath219 , while the main pulse arises from a region in which more than one polarization contributes toward the observed field . this explains not only the difference between the frequency structures of the main pulse and the interpulse , but also the difference between their temporal structures , i.e. the fact that the intensity of the interpulse has a broad and smooth distribution in time , while that of the main pulse is composed of randomly distributed nanoshots @xcite . it would be possible to identify the individual nondiffracting subbeams described in sec . [ newsec2 ] only in the case of a source whose length scale of spatial variations is comparable with @xmath220 ( in the case , e.g. of a turbulent plasma with a superluminally rotating macroscopic distribution ) . the overall beam within which the nonspherically decaying radiation is detectable would then consist of an incoherent superposition of coherent , nondiffracting subbeams with widely differing amplitudes and phases , that would be observed as randomly distributed nanoshots . otherwise , the smooth transition between the adjacent filamentary parts of the source that generate neighbouring subbeams would result in an overall beam that is likewise smoothly distributed . the filamentary sources sampled from an emitting region in which a single component of the electric current dominates would not be sufficiently different from one another in either amplitude or phase for their associated subbeams to be distinguishable . the observed difference between the dispersion measures of the main pulse and the interpulse @xcite further supports the fact that these pulses arise from distinct regions of the magnetosphere . as we have seen in sec . [ newsec2 ] , an observer who is located at the polar angle @xmath221 samples , in the course of a rotation , the distribution of the density of the electric current around the cylinder @xmath27 . thus the profile of the mean pulse reflects the azimuthal distribution pattern of the source . since they only involve a limited range of frequencies , the features of the band emission we have discussed above do not depend on the spectral distributions of the fourier transforms @xmath164 of the source densities @xmath88 sensitively . to put the emission bands in the context of the continuum spectrum of the crab pulsar , however , we would need both a more realistic model of the emitting current distribution and a more explicit description of the dependence of the density of emitting current on frequency . this notwithstanding , the frequencies across which the observed spectrum of the crab pulsar undergoes sharp changes can be identified with the frequencies at which the airy function and the coefficients @xmath124 and @xmath222 in the spectrum of the radiation described by equation ( [ eq:14 ] ) have their critical points . we have already seen that , for the value @xmath223 implied by the observational data of @xcite , the position @xmath224 of the last peak of the airy function in equation ( [ eq:14 ] ) coincides with the ultraviolet peak in the observed spectrum of the crab pulsar @xcite ( fig . [ fig6 ] ) . the coefficient @xmath124 in equation ( [ eq:14 ] ) is independent of @xmath125 ( @xmath126 ) when @xmath127 , decays as @xmath128 when @xmath129 , and is of the order of @xmath125 when @xmath130 [ see equation([eq:7 ] ) ] . hence , if the frequency @xmath225 of the azimuthal variations of the source in the crab pulsar [ see equation ( [ eq:1 ] ) ] falls in the terahertz range , i.e. @xmath226 , then the order of magnitude of @xmath124 would change from @xmath227 to @xmath228 across the terahertz gap in its observed spectrum . the corresponding enhancement of the radiated power by a factor of order @xmath229 is in fact consistent with observational data ( see below and fig . [ fig6 ] ) . a plausible cause for a frequency @xmath230 thz would be cyclotron resonance of free electrons in a magnetic field of around @xmath231 g. a magnetic field of this order at , or close to , the light cylinder , is consistent with the spin - down properties of the crab pulsar @xcite . moreover , it can be seen from equations ( [ eq:8 ] ) and ( [ eq:8a ] ) that the coefficient @xmath222 changes from being independent of @xmath125 to decaying as @xmath232 when @xmath125 increases past @xmath233 , and so the observation point falls within the fresnel zone . given that the crab pulsar is at the distance @xmath234 cm and has a light cylinder with the radius @xmath235 cm @xcite , this would account for the corresponding steepening of the observed spectrum of the crab pulsar at @xmath236 ( fig . [ fig6 ] ) if the radial extent of the emitting plasma observed from earth has the value @xmath237 , i.e. @xmath238 percent of the light - cylinder radius . these generic features of the spectrum of the radiation from a superluminal source , which do not sensitively depend on the detailed distributions of the fourier components @xmath164 of the source densities , are adequately described by the following simplified version of equation ( [ eq:14 ] ) : @xmath239 , \label{eq:22}\ ] ] where @xmath240 stands for the dependence of the source density on @xmath125 . for @xmath140 and a @xmath241 that steepens by the factor @xmath232 across @xmath236 , equation ( [ eq:22 ] ) yields the continuum spectrum shown in fig . [ fig6 ] if we assume that @xmath240 dominates over @xmath124 everywhere , except across @xmath242 where @xmath124 increases by the factor @xmath243 ( due to the above - mentioned resonance with @xmath230 thz ) , and decays as @xmath244 in @xmath245 and as @xmath246 in @xmath247 . note the close correspondence , in fig . [ fig6 ] , of equation ( [ eq:22 ] ) with the observational data on the spectrum of the crab pulsar over 16 orders of magnitude of frequency @xcite . , predicted by equation ( [ eq:22 ] ) , versus @xmath248 for @xmath249 hz appropriate for the crab pulsar . the points show observational data ( where available ) of the spectrum of the crab pulsar @xcite . in the model , the recovery of intensity at the ultraviolet peak at @xmath250 hz is caused by resonant enhancement due to the azimuthal modulation frequency @xmath230 thz . the change in gradient at @xmath251 hz reflects a transition across the boundary of the fresnel zone ( see text ) . , width=302 ] we must stress the model - independent nature of the results we have reported here . the only global property of the magnetospheric structure of the crab pulsar we have invoked is its quasi - steady time dependence : that the cylindrical components @xmath252 of the density of the magnetospheric electric current depend on @xmath4 only in the combination @xmath58 . not only does this property follow from the observational data ( as rigorously shown in appendices a and b ) , but it is one that has been widely used in the published literature on pulsars [ see , e.g. , @xcite ] . calculating the retarded field that is generated by the fourier component associated with the frequency @xmath54 of such a quasi - steady current distribution , we have derived a broadband radiation spectrum with oscillations whose peaks are proportionately spaced in , and decay algebraically with , frequency . by inferring the values of the parameters @xmath201 khz and @xmath230 thz from the observational data and mildly restricting certain local properties of the source densities @xmath164 that appear in the derived expression we have _ quantitatively _ accounted for the following features in the observed spectrum of the crab pulsar : 1 . the spacing of the emission bands , 2 . the frequency at which the extrapolated spacing between these bands would reduce to zero , 3 . the widths of the emission bands ( their number within a given frequency interval ) , 4 . the salient features of the continuum spectrum ( its sharp rise across the terahertz gap , its ultraviolet peak , the change by @xmath42 of the value of its spectral index at @xmath253 hz ) , and 5 . the differences in the polarizations , dispersion measures , and the time and frequency signatures of the main pulse and the interpulse . given that the derived band spacing @xmath146 increases as @xmath254 with the harmonic number @xmath125 [ equations ( [ eq:14c ] ) and ( [ eq:15e ] ) ] , we predict that the repetition of the observations of @xcite at higher frequencies would result in a dependence of the band spacing on frequency that has a local slope steeper than @xmath255 . there should be @xmath256 bands within the frequency interval @xmath257@xmath258 ghz , for example , whose spacings would increase with frequency as @xmath259 . the two significant input parameters required to simulate the crab pulsar s spectrum are a modulation frequency , @xmath201 khz , tentatively attributed to plasma oscillations of electrons in the pulsar s atmosphere and @xmath230 thz , possibly corresponding to cyclotron resonance of electrons in a magnetic field of around @xmath231 g. the latter gives rise to the resonant increase in spectral weight in the ultraviolet ( fig . [ fig6 ] ) . at this point , it is worth considering briefly the influence of the ratio @xmath2 on pulsar spectra in general . given that typical features of the pulsar s spectrum scale as @xmath260 , we might expect that pulsars with slower rotational frequencies ( and therefore small values of @xmath0 ) will have observed spectral intensities weighted towards higher frequencies . by contrast , so - called millisecond pulsars ( with large @xmath0 ) might be expected to have emission concentrated at lower frequencies than the crab . these predictions seem to be borne out both by geminga ( @xmath261 hz ) , which has emission peaked in the ultra - violet to gamma - ray end of the spectrum , and by millisecond pulsars such as 1937 + 21 ( @xmath262 hz ) and 1957 + 20 ( @xmath263 hz ) , both of which show no emission in the ghz range but strong pulses at radiofrequencies @xcite . furthermore , once it is acknowledged that ( as shown in appendices a and b ) the current emitting the observed pulses from pulsars has a superluminally - rotating distribution pattern , the results reported in the published literature on the electrodynamics of superluminal sources @xcite can be used to explain the extreme values of the giant pulses brightness temperature ( @xmath264 k ) @xcite , temporal width ( @xmath265 ns ) and source dimension ( @xmath265 m ) @xcite , as well as the unique characteristics of the average pulses polarization ( their occurrence as concurrent ` orthogonal ' modes with swinging position angles and with nearly 100 per cent linear or circular polarization ) @xcite and spectra ( the range @xmath266 to @xmath267 of their spectral indices and the breadth of their bandwidth , from radio waves to gamma - rays ) @xcite . we conclude by re - emphasizing that all these features of the observed radiation are consequences solely of the fact that the emitting current depends on the coordinate @xmath4 in the combination @xmath5 . this constraint may be inferred directly from the simplest observation , that pulsars have a rigidly - rotating radiation distribution , and maxwell s equations ( see appendices a and b ) ; it leads to a current distribution with a superluminally rotating pattern at a radius @xmath9 responsible for the unique features of pulsar emission . the explicit form of the current distribution plays a role only in accounting for the pulse - to - pulse variations of the features discussed above ( i.e. for the variable ` weather ' in the pulsar magnetosphere , not its stable ` climate ' ) . the plasma processes by means of which the magnetic field of the pulsar couples the rotational motion of the central neutron star to the observationally implied rigid rotation of the distribution pattern of the emitting current have no direct bearing on the results reported in this paper . the salient features of the observational data can be understood in terms of the superluminal emission mechanism without a knowledge of the magnetospheric structure of the pulsar . we are grateful to jean eilek , tim hankins , jim sheckard , john middleditch , joe fasel , and andrea schmidt for helpful discussions . this work is supported by u.s . department of energy grant ldrd 20080085dr , `` construction and use of superluminal emission technology demonstrators with applications in radar , astrophysics and secure communications '' . a.a . also thanks the royal society for support . 99 abramowitz m. , stegun i.a . , 1970 , _ handbook of mathematical functions _ ( dover ) ardavan h. , 1998 , phys rev e , 58 , 6659 ardavan h. , ardavan a. , singleton j. , 2003 , j opt soc am a , 20 , 2137 ardavan h. , ardavan a. , singleton j. , 2004a , j opt soc am a , 21 , 858 ardavan a. , hayes w. , singleton j. , ardavan h. , fopma j. , halliday d. , 2004b , j appl phys , 96 , 7760 ardavan h. , ardavan a. , singleton j. , fasel j. , schmidt , a. , 2007 , j opt soc am a , 24 , 2443 ardavan h. , ardavan a. , singleton j. , fasel j. , schmidt a. , 2008a , j opt soc am a , 25 , 543 ardavan h. , ardavan a. , singleton j. , fasel j. , schmidt a. , 2008b , j opt soc am a , 25 , 780 bessarab a.v . , gorbunov a.a . , martynenko s.p . , prudkoy n.a . , 2004 , ieee trans plasma sci , 32 , 1400 bolotovskii b.m . , ginzburg v.l . , 1972 , sov phys usp , 15 , 184 bolotovskii b.m . , bykov v.p . , 1990 , sov phys usp , 33 , 477 bolotovski b.m . , serov a.v . , 2005 , phys usp , 43 , 903 courant r. , hilbert d. , 1962 , _ methods of mathematical physics _ , vol . ii ( interscience ) ginzburg v.l . , 1972 , sov phys jetp , 35 , 92 gold t. , 1968 , nature , 218 , 731 goldreich p. , julian w.h . , 1969 , ap j , 157 , 869 hankins t.h . , kern j.s . , weatherall j.c . , eilek j.a . , 2003 , nature , 422 , 141 hankins t.h . , eilek j.a . , 2007 , ap j , 670 , 693 jackson j.d . , 1999 , _ classical electrodynamics _ , 3rd ed . ( wiley ) lyne a.g . , graham - smith f. , 2005 , _ pulsar astronomy _ ( cambridge u press ) mestel l. , wright g.a.e . , westfold k.c . , 1976 , mnras , 175 , 257 moffett d.a . , hankins t.h . , 1999 , apj , 522 , 1046 schmidt , a. , ardavan h. , fasel j. , singleton j. , ardavan a. , 2007 , _ proceedings of the 363rd we - heraeus seminar on neutron stars and pulsars _ , becker w. , huang h.h . , 124 singleton j. , ardavan a. , ardavan h. , fopma j. , halliday d. , hayes w. , 2004 , in _ digest of the 2004 joint 29th international conference on infrared and milimeter waves and 12th international conference on terahertz electronics _ ( ieee ) , 591 soglanov v.a . , popov m.v . , bartel n. , cannon w. , novikov a.y . , kondratiev v.i . , altunin v.i . , 2004 , ap j , 616 , 439 as reflected in the highly stable periodicity of the mean profiles of the pulses detected on earth , the pulsar radiation field has a rigidly rotating distribution pattern on average . that is to say , the cylindrical components of the received radiation fields @xmath71 and @xmath72 depend on the azimuthal angle @xmath4 only in the combination @xmath58 : @xmath268 @xmath269 where @xmath270 are the cylindrical polar coordinates based on the axis of rotation , and @xmath0 is the angular frequency of rotation of the observed radiation pattern . an equivalent statement is that the radiation fields @xmath71 and @xmath72 have a quasi - steady time dependence : @xmath271 and @xmath272 for the general solutions of these partial differential equations are given by the expressions on the right - hand sides of equations ( [ eq : a1 ] ) and ( [ eq : a2 ] ) . in the lorenz gauge , the electromagnetic fields appearing in equation ( [ eq:3 ] ) are given by a four - potential @xmath68 that satisfies the wave equation @xmath273 where @xmath65 designate the cartesian spatial components @xmath66 and @xmath67 of the potential @xmath68 and the current density @xmath69 @xcite . the retarded solution to this equation in unbounded space is given by equation ( [ eq:4 ] ) , i.e. by @xmath274 where @xmath275 is the corresponding green s function @xcite . employing equation ( [ eq:3 ] ) to write out @xmath276 in terms of the cylindrical components @xmath277 of the vector potential , we see that the potential also has the time dependence expressed in equations ( [ eq : a1])([eq : a4 ] ) . it follows from @xmath278 , for instance , that @xmath279 where @xmath280 stands for the differential operator appearing in equations ( [ eq : a3 ] ) and ( [ eq : a4 ] ) : @xmath281 this and the corresponding equations for @xmath282 and @xmath283 show that equation ( [ eq : a4 ] ) is satisfied if @xmath284 , i.e. if @xmath277 depend on @xmath4 and @xmath6 only in the combination @xmath58 . on the other hand , the cylindrical components of the vector potential are related to the cylindrical components @xmath285 of the current density via the following spatial part of equation ( [ eq : a6 ] ) : @xmath286 = c^{-1}\int r\,{\rm d}r\,{\rm d}\varphi\,{\rm d}z\,{\rm d}t\,g}\nonumber \\ & & \times\left[\matrix{\cos(\varphi_p-\varphi)j_r+\sin(\varphi_p-\varphi)j_\varphi\cr \cos(\varphi_p-\varphi)j_\varphi-\sin(\varphi_p-\varphi)j_r\cr j_z\cr}\right ] . \label{eq : a9}\end{aligned}\ ] ] applying the operator @xmath287 to both sides of equation ( [ eq : a9 ] ) and making use of the fact that @xmath288 depends on @xmath26 and @xmath21 in the combinations @xmath289 and @xmath290 [ equation ( [ eq : a7 ] ) ] , and so @xmath291 , we find that the resulting equation can be cast in the form @xmath292= c^{-1}\int r\,{\rm d}r\,{\rm d}\varphi\,{\rm d}z\,{\rm d}t\,g}\nonumber \\ & & \times\left[\matrix{\cos(\varphi_p-\varphi){\cal l}j_r+\sin(\varphi_p-\varphi){\cal l}j_\varphi\cr \cos(\varphi_p-\varphi){\cal l}j_\varphi-\sin(\varphi_p-\varphi){\cal l}j_r\cr { \cal l}j_z\cr}\right ] \label{eq : a11}\end{aligned}\ ] ] by means of integrations by parts with respect to @xmath4 and @xmath6 . hence , @xmath293 vanishes if and only if @xmath294 is zero . it follows from equations ( [ eq : a8 ] ) and ( [ eq : a11 ] ) , therefore , that a necessary and sufficient condition for the rigid rotation of the distribution pattern of the radiation field , i.e. @xmath295 , is the corresponding rigid rotation @xmath296 of the distribution pattern of the source density . our purpose here is to show that a solution of maxwell s equations that has the quasi - steady time dependence @xmath297 i.e. has a rotating distribution pattern , applies either to the entire magnetosphere or to an expanding region whose boundary propagates at the speed of light . by considering the initial - boundary - value problem for the wave equation ( [ eq : a5 ] ) with cauchy data satisfying equation ( [ eq : b0 ] ) , we establish that such a solution can not be smoothly matched to other types of solutions of maxwell s equations across a boundary that is confined to a localized region of the magnetosphere . the change of variables @xmath298 where @xmath299 is , for the moment , to be considered an arbitrary function , transforms the @xmath16-component of the wave equation ( [ eq : a5 ] ) into @xmath300 where @xmath301 , @xmath302 , etc . the jacobian of the above transformation is @xmath303 which remains non - zero as long as @xmath299 is not a function that depends on @xmath4 and @xmath6 as in @xmath58 . now suppose that at a given time @xmath6 the surface @xmath304 represents the boundary of a region within which the solution to equation ( [ eq : b2 ] ) satisfies the symmetry ( [ eq : b0 ] ) , i.e. within which @xmath305 is a function of @xmath306 , @xmath307 and @xmath308 only . at all points on this boundary , we then have @xmath309 and @xmath310 where ( [ eq : b5 ] ) is the wave equation under symmetry ( [ eq : b0 ] ) ; ( [ eq : b4 ] ) and ( [ eq : b5 ] ) hold at the boundary of the region in question by virtue of holding inside that region . hence , any other type of solution of the wave equation that would match the symmetric solution across @xmath311 smoothly that represents a field component is not permitted here , for it would entail the introduction of surface charges and currents with infinite densities within the magnetosphere . ] has to be sought by solving the following initial - value problem : given that the cauchy data on the hypersurface @xmath312 are those expressed in ( [ eq : b4 ] ) and ( [ eq : b5 ] ) , what is the solution to the hyperbolic partial differential equation ( [ eq : b2 ] ) beyond @xmath312 ? note that since @xmath313 vanishes at _ all _ points of the hypersurface @xmath312 , its derivatives @xmath314 , @xmath315 and @xmath316 , which are in directions interior to this hypersurface , also vanish at @xmath312 . thus , equation ( [ eq : b2 ] ) in conjunction with the data ( [ eq : b4 ] ) and ( [ eq : b5 ] ) demands that @xmath317 there are two ways in which this requirement could be met : either @xmath311 is a characteristic surface of the wave equation , and hence the first factor in ( [ eq : b6 ] ) vanishes , or at @xmath311 the derivative of @xmath313 in the direction _ normal _ to the boundary is also zero . in the first case , the boundary of the domain in which symmetry ( [ eq : b0 ] ) is satisfied , @xmath311 , will consist of an expanding wave front that propagates at speed @xmath8 . in the second case , the solution outside @xmath311 will also be symmetric , according to the cauchy - kowalewski theorem [ cf . this is because the extension ( by means of a taylor series ) of the data into an integral strip next to the boundary @xmath311 will yield a solution that is again independent of @xmath318 . since the above argument applies also to the new boundary of the region thus extended , it follows that @xmath305 will also be be independent of @xmath318 outside the surface @xmath304 , i.e. will be a function of @xmath319 throughout the magnetosphere . the corresponding results for @xmath320 and @xmath321 may be derived in the same way . in this appendix , we evaluate the leading term in the asymptotic expansion of the anger functions that appear in equation ( [ eq:6 ] ) for a radiation frequency ( @xmath73 ) that appreciably exceeds both the rotation ( @xmath0 ) and modulation ( @xmath54 ) frequencies of the source , though not necessarily the frequency @xmath322 of its spatial oscillations . in this regime , the anger functions @xmath108 and @xmath109 can be approximated by the bessel functions @xmath110 and @xmath111 , respectively [ see equation ( 41 ) of @xcite ] . once we cast these bessel functions into the canonical forms @xmath323 and @xmath324 by introducing the new variables @xmath325 we can write the leading terms in their asymptotic expansions for large @xmath326 as @xmath327 and @xmath328 [ see equations ( 9.3.23 ) , ( 9.3.27 ) of @xcite ] . taking the limits @xmath329 of the coefficients and arguments of the above airy functions , we obtain the expressions in equations ( [ eq:12 ] ) and ( [ eq:13 ] ) . for @xmath330 , the argument of the resulting airy function and its derivative that appear in equations ( [ eq:12 ] ) and ( [ eq:13 ] ) is large , so that we can further use the asymptotic approximations @xmath331 and @xmath332 [ see equations ( 10.4.60 ) and ( 10.4.62 ) of @xcite ] to replace them by the trigonometric functions given in equations ( [ eq:14a ] ) and ( [ eq:14b ] ) . note that if @xmath54 is replaced by @xmath97 , the argument of the above airy functions assume a positive value , and hence the resulting anger functions @xmath333 and @xmath334 decay exponentially , rather than algebraically , with increasing @xmath125 [ see @xcite ] .

we investigate hadron production as well as transverse hadron spectra from proton - proton , proton - nucleus and nucleus - nucleus collisions from 2gev to 21.3tev within two independent transport approaches ( hsd and urqmd ) that are based on quark , diquark , string and hadronic degrees of freedom . the comparison to experimental data on transverse mass spectra from , and c+c ( or si+si ) reactions shows the reliability of the transport models for light systems . for central au+au ( pb+pb ) collisions at bombarding energies above 5 agev , furthermore , the measured transverse mass spectra have a larger inverse slope parameter than expected from the default calculations . we investigate various scenarios to explore their potential effects on the spectra . in particular the initial state cronin effect is found to play a substantial role at top sps and rhic energies . however , the maximum in the ratio at 20 to 30 agev is missed by 40% and the approximately constant slope of the spectra at sps energies is not reproduced either . our systematic analysis suggests that the additional pressure - as expected from lattice qcd calculations at finite quark chemical potential and temperature- should be generated by strong interactions in the early pre - hadronic / partonic phase of central au+au ( pb+pb ) collisions .

nucl-th0402026

the phase transition from partonic degrees of freedom ( quarks and gluons ) to interacting hadrons is a central topic of modern high - energy physics @xcite . in order to understand the dynamics and relevant scales of this transition laboratory experiments under controlled conditions are presently performed with ultra - relativistic nucleus - nucleus collisions . hadronic spectra and relative hadron abundances from these experiments reflect important aspects of the dynamics in the hot and dense zone formed in the early phase of the reaction . here different quark flavors ( @xmath10 ) - as constituents of the final hadrons - provide valuable information on the early high density phase due to the substantial different mass scales for the light ( @xmath11 ) , strange ( @xmath12 ) and charm ( @xmath13 ) or bottom ( @xmath14 ) quarks . furthermore , as has been proposed early by rafelski and mller @xcite in particular the strangeness degree of freedom might play an important role in distinguishing hadronic and partonic dynamics . lattice qcd calculations at vanishing quark chemical potential and finite temperature indicate critical energy densities for the formation of a quark - gluon plasma ( qgp ) of @xmath3 0.7 - 1 gev/@xmath15 @xcite , which might already be achieved at alternating gradient synchrotron ( ags ) energies of @xmath3 10 @xmath0gev for central au+au collisions @xcite . moreover , the recent lattice qcd calculations at finite quark chemical potential @xmath8 @xcite show a rapid increase of the thermodynamic pressure @xmath16 with temperature above the critical temperature @xmath17 for a phase transition ( or cross - over ) to the qgp . the problem is how to relate these numbers to observables from nucleus - nucleus collisions . estimates based on the bjorken formula @xcite for the energy density achieved in central au+au collisions suggest that the critical energy density for the formation of a qgp of @xmath3 0.7 - 1 gev/@xmath15 is by far exceeded during a few fm / c in the initial phase of au+au collisions at relativistic heavy ion collider ( rhic ) energies @xcite . accordingly , a phase transition ( or cross - over ) might already happen at much lower bombarding energies . experimentally , relativistic nucleus - nucleus collisions have been studied at beam energies from 0.1 to 2 @xmath0gev at the schwerionen - synchrotron ( sis ) , from 2 to 11.6 @xmath0gev at the ags and from 20 to 160 @xmath0gev at the super proton synchrotron ( sps ) @xcite . this program has been extended @xcite at rhic with au+au collisions at invariant energies @xmath18 from @xmath19 to 200 gev ( equivalent energies in a fixed target experiment : 0.2 to 21.3 @xmath0tev ) . whereas many previous experimental studies have focused on longitudinal rapidity distributions for hadrons , the transverse mass ( or momentum ) spectra of hadrons are presently in the center of interest . on the one hand a significant suppression of high transverse momentum hadrons in au+au collisions compared to @xmath1 is observed at rhic energies of @xmath20 200 gev @xcite . this suppression is often attributed to the energy loss of highly energetic particles in a hot colored medium ( qgp ) @xcite . in fact , the recent observation by the phenix @xcite , star @xcite and brahms @xcite collaborations , that a similar suppression is not observed in d+au interactions at midrapidities at the same energy , supports this idea . on the other hand , the measured transverse mass ( @xmath21 ) spectra of hadrons ( heavier than pions ) @xmath22 at ags , sps and rhic energies show a substantial _ hardening _ in central au+au collisions relative to @xmath1 interactions ( cf . this hardening of the spectra ( or increase of the inverse slope parameter @xmath9 in ( [ slope ] ) ) for low transverse mass or momentum is commonly attributed to strong collective flow , which is absent in the respective @xmath1 or @xmath2 collisions . the authors of @xcite have proposed to interpret the approximately constant @xmath6 slopes above @xmath23 a@xmath4gev the step as an indication for a phase transition . this interpretation is also based on a rather sharp maximum in the @xmath7 ratio at @xmath3 20 to 30 a@xmath4gev in central of pb+pb collisions ( the horn @xcite ) . however , the present body of experimental data and our theoretical understanding does not allow for a stringent conclusion based only on the step and horn phenomena . in this paper we will address some further theoretical issues , but would like to point out that more experimental data , for example on collisions in the energy range of @xmath24 gev and on weakly interacting hadrons , will be needed to clarify these issues from the experimental side . we will demonstrate in this work , that neither the pressure needed to generate a large collective flow to explain the hard slopes of the @xmath6 spectra with a plateau at sps energies nor the experimental sharp maximum in the @xmath7 ratio at @xmath3 20 to 30 a@xmath4gev is produced in the present transport models by the interactions of hadrons in the expansion phase of the hadronic fireball . some aspects of this work have been published in refs . @xcite before ; however , we here extend the previous studies to rhic energies and consider a wide variety of dynamical scenarios . in particular we will explore two distinct effects : i ) initial state cronin enhancement and ii ) heavy resonance formation on the transverse mass spectra of pions and kaons at all bombarding energies from lower ags to rhic energies . in our studies we use two independent relativistic transport models that employ hadronic and string degrees of freedom , i.e. urqmd @xcite and hsd @xcite . they take into account the formation and multiple rescattering of hadrons and thus dynamically describe the generation of pressure in the hadronic expansion phase . this involves also interactions of leading pre - hadrons that contain a valence quark ( antiquark ) from a primary hard collision ( cf . @xcite ) . in the parton language , the two transport models employed do not include gluon degrees of freedom explicitly nor gluon - gluon and gluon - quark ( antiquark ) interactions . we will use these models to test , whether the description of nucleus - nucleus reactions in terms of pre - hadronic , hadronic and string degrees of freedom and their interactions is sufficient to explain the experimental data . our work is organized as follows : in section ii we briefly recall the ingredients of the hsd and urqmd transport approaches and point out the extensions incorporated in urqmd ( versions 2.0 and 2.1 ) . in section iii the calculated excitation functions for pions , @xmath6 mesons and hyperons for central au+au ( or pb+pb ) collisions from 2 to 21300 a@xmath4gev are shown ( in comparison to the experimental data ) as well as selected particle ratios . section iv is devoted to transverse mass spectra of pions and @xmath6 mesons from @xmath1 to @xmath2 and central @xmath25 reactions in the same energy range . in section v we will discuss various alternative scenarios within the transport calculations to explore their sensitivity to the hadron transverse mass spectra . section vi is devoted to a study of the cronin effect in central au+au collisions , which we attribute to a pre - hadronic initial state scattering effect that also shows up in @xmath2 reactions . section vii includes a comparison of the thermodynamic lagrange parameters @xmath9 and @xmath26 extracted from the urqmd transport model in the central overlap regime of au+au collisions with the experimental systematics on chemical freeze - out configurations in the @xmath27 plane as well as recent lattice qcd results . section viii closes this study with a summary and discussion of open problems . we use two independent relativistic transport approaches for our study that employ hadronic and string degrees of freedom , i.e. urqmd ( ultra - relativistic quantum molecular dynamics ) @xcite and hsd ( hadron - string dynamics ) @xcite in order to obtain an estimate of systematic uncertainties within the transport models . both approaches take into account the formation and decay of strings as well as the multiple rescattering of hadrons in the expansion phase . these models can be considered as appropriate descriptions at least for the final stage of an ultrarelativistic nucleus - nucleus collision , where the strongly interacting hadrons dominate . the open question is , however , if these transport models also incorporate the proper degrees of freedom in the very early phase of heavy - ion collisions . we stress that transport calculations allow for a systematic study of the change in the dynamics from elementary baryon - baryon or meson - baryon collisions to proton - nucleus reactions or from peripheral to central nucleus - nucleus collisions in a unique way without change in parameters . this is of central importance since the assumptions of thermal and chemical equilibrium as used in hydrodynamical or statistical models @xcite do not necessarily hold in all these reactions and the transport studies allow to explore the amount of ( thermal or chemical ) equilibrium reached @xcite . the default urqmd 1.3 transport approach @xcite includes all baryonic resonances up to masses of 2 gev as well as mesonic resonances up to 1.9 gev as tabulated by the particle data group @xcite . for hadronic continuum excitations a string model is used with meson formation times of the order of 1 - 2 fm / c depending on the momentum and energy of the created hadron . for the present study and the particular aim to address also jet production and fragmentation at rhic energies we additionally have incorporated pythia ( v 6.1 ) @xcite in urqmd 1.3 similar to hsd . this novel and extended version we denote as urqmd 2.0 . moreover , to explore the physical consequences of higher mass states as implemented in the relativistic - quantum - molecular - dynamics model ( rqmd ) @xcite , we have introduced in urqmd additional high mass resonances that are explicitly produced and propagated in _ s - channel _ processes with invariant masses up to @xmath28 gev . we do not assume these states to be excited in baryon - baryon collisions . in practical terms : if the mass of the produced mesonic or baryonic resonance is above @xmath29 gev , i.e in the continuum , the formed resonance state is not fragmented as a string object like in urqmd 1.3 or urqmd 2.0 . we recall that a string of invariant mass 2 @xmath30 3 gev essentially decays to a few pions , a vector meson + pions etc . instead , this high mass state is treated as a quasi - particle that decays according to the branching ratios of the resonances with a mass of 1.9 to 2.0 gev ( as implemented in urqmd ) . we recall that baryon resonances @xmath3 1.9 gev dominantly decay to the @xmath31 channel and with some lower probability to the @xmath32 final state . in this way light meson emission is suppressed by about 25% compared to a string of the same invariant mass . we note , that this recipe is still in line with the available @xmath33 data on pion multiplicities . furthermore , since the high mass resonances decay isotropically in their rest frame , this leads to an enhancement of transverse flow ( cf . section v.e ) . furthermore , we use isotropic angular distributions for elastic meson - baryon interactions below @xmath34 gev . in this work we will denote this scenario as urqmd 2.1 . in the hsd approach nucleons , @xmath35 s , n@xmath36(1440 ) , n@xmath36(1535 ) , @xmath37 , @xmath38 and @xmath39 hyperons , @xmath40 s , @xmath41 s and @xmath42 s as well as their antiparticles are included on the baryonic side whereas the @xmath43 and @xmath44 octet states are included in the mesonic sector . inelastic hadron hadron collisions with energies above @xmath45 2.6 gev are described by the fritiof model @xcite ( incorporating pythia v 5.5 with jetset v 7.3 for the production and fragmentation of jets @xcite ) whereas low energy hadron hadron collisions are modeled in line with experimental cross sections @xcite . both transport approaches reproduce the nucleon - nucleon , meson - nucleon and meson - meson cross section data in a wide kinematic range @xcite . therefore , we expect that the hadronic scattering mechanisms are sufficiently under control . however , in nucleus - nucleus reactions further elastic and inelastic processes on the parton level might occur that beyond the interactions of leading quarks / diquarks are not included in the studies below . we mention that ( pqcd - like ) partonic scattering is incorporated e.g. in the multi - phase transport model ( ampt ) @xcite , which is currently employed from upper sps to rhic energies , but is not suitable for lower energies . whereas the underlying concepts of the transport theoretical models are very similar , the actual realizations differ considerably . a systematic analysis of results from both models and experimental data for central nucleus - nucleus collisions from 2 to 160 @xmath0gev has shown that the longitudinal rapidity distributions of protons , pions , kaons , antikaons and hyperons are quite similar in both models and in reasonable agreement with available data @xcite . we here extend the latter studies to rhic energies , but focus on total and midrapidity hadron yields as well as particle ratios . for completeness we include the results from ref . @xcite in figs . [ fig_yield]-[fig_rat ] . figure [ fig_yield ] shows the excitation function of @xmath46 and @xmath47 yields from 5% central ( ags energies , sps at 160 a@xmath4gev and at rhic energies ) , 7% central ( 20 , 30 , 40 and 80 a@xmath4gev ) , 10% central for @xmath47 at 160 a@xmath4gev au+au ( ags and rhic ) or pb+pb ( sps ) collisions in comparison to the experimental data from refs . @xcite ( ags ) , @xcite ( sps ) and @xcite ( rhic ) for midrapidity ( left column ) and rapidity integrated yields ( right column ) . the solid lines with open squares show the results from hsd , the dashed lines with open triangles indicate the calculations with urqmd 1.3 and the long dashed lines with crosses depict the urqmd 2.1 results . we note in passing , that the corrections to the pion and kaon yields due to the weak decay of resonances have not been included in our calculations in line with the experimental analysis of the na49 collaboration ( cf . however , the experimental subtraction of week decays is problematic and might have an impact on the measured kaon to pion ratio as pointed out in ref . this uncertainty has to be kept in mind throughout the following analysis . as already demonstrated in ref . @xcite the agreement between the transport models hsd ( solid lines ) and urqmd 1.3 ( dashed lines ) is rather good as well as with the data available . the exception are the pion yields at the highest ags energy and lower sps energies , which are overestimated by both approaches ( cf . this is demonstrated in more detail in fig . [ fig_yieldb ] where the @xmath48 and @xmath6 yields ( as in fig . [ fig_yield ] ) are displayed in the energy range up to 80 a gev on a linear scale . the overestimation of the pion yields ( in 4@xmath49 ) in this energy range is slightly more pronounced for urqmd 1.3 than for hsd especially at the higher sps energy . on the other hand , both transport models fall slightly low in the @xmath50 yields at 30 and 40 a gev . the results from the novel version urqmd 2.1 ( long dashed lines ) differ from urqmd 1.3 essentially in the following aspect : the pion multiplicity at ags and sps energies is reduced by about 25% ( cf . section ii ) due to the specific decay scheme of higher mass resonances ( and is now in better agreement with the data ) . on the other hand , urqmd 2.1 gives a lower amount of kaons and especially antikaons ( in @xmath51 ) than the default hsd and urqmd 1.3 approaches . at rhic energies the urqmd 1.3 calculations show a @xmath3 30% higher pion yield at midrapidity than hsd while the differences between the 4@xmath49 integrated yields are of the order of 10% , both between the models and compared to the data . the implementation of higher mass states in urqmd 2.1 ( cf . section ii ) together with the jet production described by pythia leads to an overestimation of pions , kaons and hyperons at rhic energies . note , that the maximum in the @xmath52 yield at midrapidity ( lower left part of fig . [ fig_yield ] ) for @xmath3 30 a@xmath4gev is reproduced by both transport approaches . however , this maximum disappears in the 4@xmath49 integrated yields in line with the data ( cf . lower right part of fig . [ fig_yield ] ) . within the transport models the maximum in the @xmath47 yield at midrapidity can be attributed to a large extent to a change from baryon dominated dynamics at ags energies to meson dominated dynamics at the higher sps energies . in fig . [ fig_rat ] we present the excitation function of the particle ratios @xmath53 and @xmath54 from 5% central ( ags energies , sps at 160 a@xmath4gev and at rhic energies ) , 7% central ( 20 , 30 , 40 and 80 a@xmath4gev ) , 10% central for @xmath47 at 160 a@xmath4gev au+au ( ags and rhic ) or pb+pb ( sps ) collisions in comparison to the experimental data from refs . @xcite ( ags ) , @xcite ( sps ) and @xcite ( rhic ) for midrapidity yields ( left column ) and rapidity integrated yields ( right column ) . the solid lines with open squares show the results from hsd , the dashed lines with open triangles indicate the urqmd 1.3 calculations whereas the long dashed lines with crosses correspond to the urqmd 2.1 results . the deviations between the transport models and the data are most pronounced for the midrapidity ratios ( left column ) since the ratios are very sensitive to actual rapidity spectra . the @xmath7 ratio in urqmd 1.3 shows a maximum at @xmath3 8 a@xmath4gev and then drops to a roughly constant ratio of 0.12 at top sps and rhic energies . urqmd 2.1 follows the results from urqmd 1.3 at ags and sps energies but then increases slightly up to rhic energies . in case of hsd a continuously rising ratio with bombarding energy is found for the midrapidity ratios . this rise up to the full rhic energy is essentially due to an underestimation of the pion yield at midrapidity ( cf . fig . 1 in ref . the 4@xmath49 ratio in hsd , which is approximately constant from top sps to rhic energies , is larger than the ratio from urqmd 1.3 due to the lower amount of pion production and a slightly higher @xmath50 yield ( cf . [ fig_yield ] ) . we note that the lower amount of pions in hsd is essentially due to an energy - density cut ( in the local rest frame ) , which does not allow hadron formation when is energy density is above 1 gev/@xmath15 . despite this seemingly good description of the separate particle yields , the experimentally observed maximum in the @xmath7 ratio at 20 - 30 a@xmath4gev is not reproduced in the models . due to strangeness conservation , the same arguments hold for the @xmath55 ratio , where the pronounced experimental maxima are underestimated due to the excess of pions in the transport models at top ags energies ( for hsd ) and above @xmath3 5 a@xmath4gev ( for urqmd 1.3 ) . here urqmd 2.1 performs better in the ratios due to the lower amount of pions produced by the high resonance decays . since the @xmath56 yields are well described by all approaches ( cf . fig . [ fig_yield ] ) the deviations in the @xmath57 ratios at sps and rhic energies in urqmd 1.3 indeed can be traced back to the excess of pions . this suggests that the entropy production , which is reflected in the pion to baryon ratio at these energies , is overestimated in urqmd 1.3 ( and hsd ) above @xmath3 5 a@xmath4gev in central au+au collisions . note , that in urqmd 2.1 the entropy production is reduced by the generation of high mass states in @xmath12-channels , which essentially decay by emitting a lower amount of pions . it has been discussed in the community @xcite , that the maximum in the @xmath7 ratio might be due to isospin effects in au+au ( pb+pb ) collisions arising from a different rapidity dependence of @xmath58 and @xmath59 quarks in @xmath1 , @xmath60 and @xmath61 collisions . we here address this issue and show in fig . [ fig_ratpiav ] the influence of isospin effects on the kaon to pion ratio . the left part of fig . [ fig_ratpiav ] displays the excitation function of the @xmath62 ratio from central au+au ( or pb+pb ) collisions calculated within hsd ( solid line with open squares ) and urqmd 1.3 ( solid line with open triangles ) . the right hand panel shows both the isospin - sensitive @xmath63 ratio ( open symbols ) and the ( approximately ) isospin - insensitive @xmath64 ratio ( solid symbols ) . the hsd results are shown as squares connected by solid lines while the urqmd 1.3 results are represented as dashed lines with triangles . the larger solid symbols indicate the experimental results for @xmath63 from refs . @xcite and the large open triangles correspond to the ratio @xmath65 from ref . @xcite . as seen from fig . [ fig_ratpiav ] , the calculated ( as well as experimentally measured ) isospin - insensitive ratio @xmath66 is only slightly lower than the isospin - sensitive @xmath67 ratio . this result is due to the approximately constant @xmath68 ratio above @xmath69 a@xmath4gev ( l.h.s . of fig . [ fig_ratpiav ] ) in the transport calculations as well as in the data . we thus conclude that isospin effects can not be responsible for the maximum in the @xmath70 ratio ( r.h.s . of fig . [ fig_ratpiav ] ) . despite the relatively good agreement of data and transport calculations in the longitudinal direction @xcite , significant deviations are found in the transverse direction as pointed out in ref . @xcite before . this demonstrates that longitudinal rapidity distributions are less sensitive to the hadronic interaction channels employed in the transport models . we here present a more complete and systematic analysis up to the full rhic energy of @xmath18 = 200 gev and investigate alternative scenarios . we start with benchmark tests for the transport calculations for @xmath1 and @xmath2 reactions . our results for the transverse mass ( or momentum ) spectra of pions and kaons from @xmath1 reactions are presented in fig . [ fig_mtpp ] . the upper part includes the comparison of the differential cross sections for @xmath71 and @xmath72 from @xmath1 reactions at 12 and 24 gev@xmath73 from hsd ( solid lines ) and urqmd 1.3 ( dashed lines ) with the data from ref . the lower left part of fig . [ fig_mtpp ] shows the transverse mass spectra at midrapidity for @xmath74 and @xmath75 from @xmath1 reactions at a bombarding energy of 160 gev from hsd ( solid lines ) and urqmd 1.3 ( dashed lines ) . the thin lines here correspond to fits with the experimental slope parameters @xmath76 mev for @xmath77 , @xmath78 mev for @xmath50 and @xmath79 mev for @xmath56 from ref . @xcite . the lower right part of fig . [ fig_mtpp ] displays the transverse mass spectra at midrapidity for @xmath77 and @xmath56 from @xmath1 reactions at @xmath80 gev from hsd ( solid lines ) , urqmd 1.3 ( dashed lines , multiplied by a factor 2 ) , and urqmd 2.0 ( 2.1 ) ( long dashed lines ) . the full symbols , furthermore , stand for the data from the star collaboration @xcite . the description of the transverse mass spectra for pions and kaons is reasonably good in both approaches up to @xmath18 = 200 gev . we note in passing that the inclusion of pythia ( v 6.1 ) in urqmd 2.0 and urqmd 2.1 is essential in order to achieve a good description of the data since the contribution from jet fragmentation is essential at rhic energies . without jet production and fragmentation the urqmd model ( version 1.3 ) underestimates the high momentum spectra severely ( similar to rqmd @xcite ) . we continue with @xmath2 reactions and show in the left part of fig . [ fig_tpa ] the results for the inverse slope parameters @xmath9 in ( [ slope ] ) for @xmath81 and @xmath56 mesons at midrapidity from @xmath2 reactions at 14.6 gev@xmath73 ( @xmath82=be , al , cu , au ) , whereas the right part includes the results at 450 gev@xmath73 ( @xmath82=be , s , pb ) . the inverse slope parameters from hsd are shown as open symbols connected by solid lines and from urqmd 1.3 ( 2.0 ) by open symbols ( with crosses ) connected by dashed lines . the full symbols ( l.h.s . ) correspond to the midrapidity data ( @xmath83 ) from the e802 collaboration @xcite while the na44 data @xcite at @xmath84 gev@xmath73 for @xmath85 and at @xmath86 gev@xmath73 ( full diamonds ) and @xmath87 gev@xmath73 ( full squares ) for @xmath88 are given on the right - hand side . the overall reproduction of the transverse slope parameters at ags and sps energies for targets from be to pb by both transport models is sufficiently accurate . the inverse slope parameters from urqmd 1.3 and hsd differ for pions and antikaons by about 10% at ags energies . these differences can be traced back to the different production channels in the transport models at lower energies : in hsd the mesons are essentially produced by string decays , which are strongly forward peaked in momentum , whereas in urqmd 1.3 they partly stem from high mass resonance decays with an isotropic decay in momentum space . at sps energies the meson production in both models proceeds via string formation and decay and leads to very similar results . this also holds for rhic energies where additionally jet production and fragmentation contributes with increasing bombarding energy . we note in passing that when including the initial state cronin effect ( cf . section vi ) the inverse slope parameters @xmath9 in fig . [ fig_tpa ] increase by less than 5% even for the pb - target at 450 gev / c . at 14.6 gev/@xmath13 there is no effect ( within statistics ) for all targets from be to au . we continue with nucleus - nucleus collisions and recall that the experimental @xmath89 spectra ( at midrapidity ) from central c+c and si+si collisions at 160 @xmath0gev @xcite are well described by hsd and urqmd 1.3 . this holds for the @xmath77 , @xmath50 as well as @xmath56 spectra ( cf . fig . 1 in ref . @xcite ) . this situation changes for central collisions of heavy nuclei like au+au or pb+pb . in fig . [ fig_mtaa ] we display the calculated transverse mass spectra at midrapidity from hsd ( solid lines ) and urqmd 2.0 ( dashed lines ) for @xmath81 and @xmath90 from central au+au ( pb+pb ) reactions at 4 , 6 , 8 , 11 , 20 a@xmath4gev ( left part ) and at 30 , 40 , 80 , 160 a@xmath4gev and @xmath80 gev ( right part ) with the data ( full and open symbols ) from refs . @xcite.gev have to be considered as preliminary ] at the lowest energy of 4 @xmath0gev the agreement between the transport approaches and the data is still acceptable , however , severe deviations are visible in the @xmath6 spectra at top ags , sps and rhic energies . note that the @xmath91 spectra are reasonably described at all energies while the inverse slope @xmath9 of the @xmath6 transverse mass spectra is underestimated severely by about the same amount in both transport approaches ( within statistics ) . the increase of the inverse @xmath6 slopes in heavy - ion collisions with respect to @xmath1 collisions , which is generated by rescatterings of produced hadrons in the transport models , is only small because the elastic meson - baryon scattering is strongly forward peaked ( in hsd and urqmd 2.0 ) and therefore gives little additional transverse momentum especially at midrapidity . in this section we will investigate different hadron / string scenarios , which might contribute to the increase of the inverse slope parameter of the @xmath6 mesons in central au+au collisions and are not included in the default versions of the transport approaches . the question comes up whether the discrepancies shown in fig . [ fig_mtaa ] might be due to conventional hadronic medium effects . in fact , the @xmath89 slopes of kaons and antikaons at sis energies ( 1.5 to 2 @xmath0gev ) were found to differ significantly @xcite . as argued in @xcite the different slopes could be traced back to repulsive potentials of kaons with nucleons , which lead to a hardening of the @xmath50 spectra , and attractive antikaon - nucleon potentials , which lead to a softening of the @xmath56 spectra . however , the effect of such potentials was calculated within hsd and was found to be of minor importance at ags and sps energies @xcite since the meson densities are comparable to or even larger than the baryon densities at ags energies and above . additional self energy contributions stem from @xmath6 interactions with mesons which become increasingly important for higher bombarding energies ; however , @xmath12-wave kaon - pion interactions are weak due to chiral symmetry arguments and @xmath92-wave interactions such as @xmath93 transitions are suppressed substantially by the approximately thermal pion spectrum . a recent study on the kaon potentials in hot pion matter gives kaon mass shifts of about @xmath94 mev and vector potentials of @xmath95 mev @xcite for a pion gas at temperature @xmath9= 170 mev . we have employed even slightly larger @xmath6 potentials in dynamical hsd calculations and achieved a hardening of the @xmath6 spectra by less than 10% . furthermore , we have pursued the idea of refs . @xcite that the @xmath6 spectra could be hardened by string - string interactions , which increase the effective string tension @xmath96 and thus the probability to produce mesons at high @xmath89 @xcite . a related suggestion has been put forward in @xcite . in order to estimate the largest possible effect of string - string interactions we have assumed that for two overlapping strings the string tension @xmath96 is increased by a factor of two , for three overlapping strings by a factor of three etc . here the overlap of strings is defined geometrically assuming a transverse string radius @xmath97 , which according to the studies in ref . @xcite should be @xmath98 0.25 fm . additionally an overlap of the strings in longitudinal direction is required which , however , does not involve any new parameter since the dynamics of the string ends is directly available from the hsd transport calculations . based on these assumptions ( and @xmath97=0.25 fm as in refs . @xcite ) , we find only a small increase of the inverse slope parameters at ags energies , where the string densities are low ( dashed lines in fig . [ fig_scenar ] in comparison to the default calculations ( solid lines ) ) . at sps and rhic energies the model gives a hardening of the spectra by about 15% which , however , is still too low in comparison to the data in fig . [ fig_mtaa ] . despite of the large number of initial strings in central au+au collisions at sps or rhic energies , their average eigenvolume ( in the rest frame at break up ) is low ( @xmath99 @xmath15 ) compared to the eigenvolume of a nucleon ( @xmath100 @xmath15 ) due to the fast string fragmentation process . consequently the critical density of color neutral strings may be about 7 times higher than the critical density of nucleons with respect to a phase transition induced by percolation @xcite . we have examined another possibility which we denote as parton elastic scattering. to this end we have included elastic scattering of unformed hadrons in the initial phase of the reaction before they actually evolve to their full hadronic size employing the elastic pion - nucleon cross section per quark degree of freedom . the results of these simulations are shown in fig . [ fig_scenar ] by the dotted lines which give a much better reproduction of the experimental spectra at @xmath18 = 200 gev , that are even slightly overestimated . the drawback on the other side is that the hadron rapidity spectra now are overestimated by @xmath3 20% relative to the default calculation ( and the experimental data ) . the increase in the slope parameter especially for @xmath6 spectra at ags and sps energies , however , is only moderate and a significant discrepancy remains with respect to the experimental spectra . following a further idea of ref . @xcite we explore the possibility that the fragmentation of strings from @xmath101 collisions modifies in the dense and hot medium . an extreme modification amounts to assume an isotropic decay of @xmath101 strings as employed in the rqmd model @xcite which leads to a reduction of longitudinal hadron momenta and an enhancement of their transverse momenta . the results of these simulations are displayed in fig . [ fig_scenar ] for the transverse mass spectra of pions , kaons and antikaons ( dot - dashed lines ) . apparently , this modified fragmentation function leads to much better results in comparison to experiment ( cf . dot - dashed lines in fig . [ fig_scenar ] ) . the drawback with this scenario , however , is that the approximate agreement with the longitudinal rapidity distributions for @xmath102 and protons is destroyed as demonstrated in fig . [ fig_y160iso ] for 5% central pb+pb reactions at 160 a@xmath4gev . here the solid lines correspond to the default hsd calculations whereas the dashed lines stand for the calculations with an isotropic decay of @xmath101 strings . obviously , the latter scenario leads to an enhanced stopping of the nuclei and larger meson production , which is not supported very much by the data . we admit , that a slightly more isotropic decay of the @xmath101 strings ( with respect to the default calculations ) would be favored by the experimental data ; however , we do not aim at fitting data here . we mention - without explicit representation - that the comparison shown in fig . [ fig_y160iso ] for central collisions at 160 a@xmath4gev holds qualitatively also at ags and rhic energies . a further possibility to harden the meson and nucleon spectra in a coupled - channel transport approach is to introduce higher mass baryon resonances with masses @xmath103 gev as in rqmd @xcite , that decay isotropically and dominantly couple to the @xmath104 channel and not to multipion + nucleon channels . as described in section ii , the version urqmd 2.1 has been set up to incorporate this possibility and to explore its consequences . since the massive resonances decay isotropically in their rest frame , this leads to enhanced transverse flow as found in the previous subsection . the penalty of enhanced hadron production ( cf . [ fig_y160iso ] ) is cured in urqmd 2.1 since the high mass resonance decays produce 25% less particles as compared to a string of the same invariant mass . in this way one can produce larger hadron transverse momenta in the transport calculations without increasing the meson multiplicities on average . such a prescription fits the data in the ags and sps energy regime very well ( similar to rqmd @xcite ) . however , the open question remains , if such states could be identified in meson - baryon scattering and , furthermore , if they would survive in a dense hadronic medium . in fig . [ fig_mtu21 ] we show the transverse mass spectra at midrapidity for @xmath74 and @xmath90 from 5% central au+au ( pb+pb ) reactions at 11 a@xmath4gev , 160 a@xmath4gev and @xmath80 gev calculated for different scenarios in urqmd : the dashed lines correspond to the default calculations ( urqmd 2.0 ) , the solid lines to simulations with high mass baryon resonances ( urqmd 2.1 ) . indeed , the description of the pion and @xmath6 improves substantially in urqmd 2.1 relative to urqmd 2.0 ( and hsd ) at ags and sps energies . however , the penalty here is that now the spectra at the top rhic energy are overestimated at high transverse mass for pions and antikaons . moreover , the improvement in the spectral slope for pions and kaons at ags and sps energies in urqmd 2.1 does not go along with an improvement in the rapidity spectra as shown in fig . [ fig_y160u21 ] . though the @xmath77 rapidity distribution can be well described for 5% central pb+pb reactions at 160 a@xmath4gev due to the reduction of pion production in the decay scheme of the high mass states , the rapidity spectra of @xmath6 mesons are also lower in urqmd 2.1 ( solid lines ) compared to urqmd 2.0 ( dashed lines ) and the data . the proton stopping at midrapidity is not much affected by the introduction of the high mass states since they do not couple to the baryon - baryon channel . we conclude , that the recipe of high mass states as incorporated in urqmd similar to rqmd does not solve the problem with the @xmath7 ratio . the cronin effect dates back to measurements of transverse hadron spectra from @xmath2 collisions at the sps and isr @xcite and relates to a ( moderate ) enhancement of transverse momentum spectra with increasing target mass @xmath82 . though the microscopic origin of this effect is not fully understood , most current interpretations are based on semi - hard initial state gluon radiation of the propagating quarks in the medium prior to a hard interaction vertex ( cf . e.g. @xcite ) . in this way the initial transverse momentum distribution of partons is extended to higher average @xmath105 . we discard a further discussion of its physical origin and proceed with effective simulations following ref . @xcite . in the latter work it was suggested that such an enhancement of the intrinsic quark transverse momentum spread @xmath105 might be simulated in the transport approach by increasing the average transverse momentum spread of the quarks @xmath105 with the number of previous collisions of primary nucleons @xmath106 as @xmath107 the parameter @xmath108 has been determined in ref . @xcite by comparing to the experimental data for d+au collisions @xcite at @xmath80 gev . this assumption is in line with an independent analysis within the glauber model by papp et al . note , that @xmath106 is not a further parameter of the model but calculated dynamically in the transport approach . whereas the different alternative scenarios discussed in section v relate to various model assumptions , which are not directly controlled by experimental data , the initial state cronin enhancement is an established experimental fact . consequently , it should be incorporated in future transport approaches ( as in ref . @xcite ) . for further details we refer the reader to the latter publication . our results for the transverse mass spectra at midrapidity for @xmath110 and @xmath90 from 5% central au+au ( pb+pb ) reactions at 11 a@xmath4gev , 160 a@xmath4gev and @xmath80 gev as calculated within hsd including the cronin effect ( [ kt ] ) ( solid lines ) are presented in fig . [ fig_cronaa ] in comparison to the default calculations ( dashed lines ) . the experimental data are the same as in fig . [ fig_mtaa ] . the description of the spectra becomes rather good at rhic energies , improves substantially at the sps energy of 160 a@xmath4gev , but does not give any significant change at 11 a@xmath4gev . consequently , the prehadronic cronin effect - as implemented via ( [ kt ] ) - is not responsible for the dramatic increase of the @xmath6 slopes at ags energies . our calculations , however , show that a hardening of the transverse mass spectra should not only be attributed to the presence of collective flow , but also to initial state semi - hard scattering of fast quarks in a dense medium ; the latter effect becomes more important with increasing @xmath18 . a more detailed and quantitative analysis on the competition of initial state cronin enhancement and collective flow from pressure gradients at sps and rhic energies we postpone to an upcoming study . our findings are summarized in fig . [ fig_tcron ] , where the dependence of the inverse slope parameter @xmath9 ( defined by eq . ( [ slope ] ) ) on @xmath18 is shown and compared to experimental data ( partly preliminary ) from refs . @xcite for central au+au ( pb+pb ) collisions ( l.h.s . ) and @xcite for @xmath1 collisions ( r.h.s . ) . the upper and lower solid lines ( with open circles ) on the l.h.s . in fig . [ fig_tcron ] correspond to results from hsd calculations , where the upper and lower limits are due to fitting the slope @xmath9 itself , an uncertainty in the repulsive @xmath6-pion potential or the possible effect of string overlaps . the solid lines with stars correspond to the hsd calculations with the cronin effect ( [ kt ] ) . the dashed lines with open upper triangles represent slope parameters from urqmd 1.3 , the dot - dashed lines with open lower triangles correspond to urqmd 2.0 results , which are well within the limits obtained from the different hsd calculations without the cronin enhancement . the dotted lines with crosses show the urqmd 2.1 results that incorporate the high mass resonance states . the slope parameters from @xmath1 collisions ( r.h.s . in fig . [ fig_tcron ] ) are seen to increase smoothly with energy both in the experiment ( full symbols ) and in the hsd calculations ( full lines with open circles ) . the error bars related to our calculations are due to an uncertainty in extracting the slope parameter from different @xmath89 intervals . the urqmd 1.3 results are shown as open upper triangles connected by the solid line and systematically lower than the slopes from hsd at all energies . when including jet production and fragmentation via pythia in urqmd 2.0 ( dot - dashed lines with open lower triangles ) the results become similar to hsd above @xmath18 = 10 gev demonstrating the importance of jets in @xmath1 reactions at high energy . coming back to the slope parameters of @xmath6 mesons for central au+au / pb+pb collisions ( l.h.s . [ fig_tcron ] ) we find that the cronin initial state enhancement indeed improves the description of the data at rhic energies , however , does not give any sizeable enhancement at ags energies . here urqmd 2.1 ( dashed lines with crosses ) with the high mass resonance states performs better for @xmath50 mesons , but overestimates the @xmath56 slopes at ags as well as upper sps energies . an alternative way of representing the data is provided by the excitation function for the average transverse momentum @xmath111 for pions ( l.h.s . ) and kaons ( r.h.s . ) in fig . [ fig_pt ] for the different transport models in comparison to the data available . hsd : upper and lower solid lines with open circles , urqmd 2.0 : dot - dashed lines with open lower triangles and urqmd 2.1 : dotted lines with crosses ; the data are from refs . @xcite . the dotted line with the open diamonds ( r.h.s . ) shows @xmath112 for @xmath56 from urqmd 2.1 . furthermore , the solid lines with stars correspond to the hsd calculations with the cronin effect , which gives a good description of the average @xmath111 in case of pions and significantly improves the average @xmath111 for kaons as a function of @xmath18 ( with respect to default hsd and urqmd 2.0 calculations ) . the urqmd 2.1 calculations overestimate @xmath111 for pions at all energies by 10 to 15 % as well as for @xmath56 at ags and sps energies in line with the trends from fig . [ fig_tcron ] ; however , the average @xmath113 for @xmath50 mesons is rather well described . this still leaves us with the question of the origin of the rapid increase of the @xmath6 slopes with invariant energy for central au+au collisions at ags energies and the constant slope at sps energies ( the step ) , which is missed in all transport approaches presently employed . we recall that higher transverse particle momenta either arise from repulsive self energies in mean - field dynamics or from collisions , which reduce longitudinal momenta in favor of transverse momenta @xcite . as shown above in fig . [ fig_tcron ] conventional hadron self - energy effects and ( pre-)hadronic and hadronic binary collisions are insufficient to describe the dramatic increase of the @xmath6 slopes as a function of @xmath18 as well as the initial state cronin enhancement . only when including high mass resonances in urqmd 2.1 with a particular decay pattern we may end up with reasonable results for @xmath50 mesons , however , fail more severely for pions as well as antikaons . these failures of 10 to 15% indicate an additional mechanism for the generation of the pressure that is observed experimentally especially at ags energies . we speculate following the previous study @xcite that partonic degrees of freedom should be responsible for this effect already at @xmath3 5 @xmath0gev . our arguments here are based on a comparison of the thermodynamic parameters @xmath9 and @xmath26 extracted from the transport models in the central overlap regime of au+au collisions @xcite with the experimental systematics on chemical freeze - out configurations @xcite in the @xmath27 plane . the solid line in fig . [ fig_qcd ] characterizes the universal chemical freeze - out line from cleymans et al . @xcite whereas the full dots with errorbars denote the experimental chemical freeze - out parameters - determined from the thermal model fits to the experimental ratios - taken from ref . the various symbols ( in vertical sequence ) represent temperatures @xmath9 and chemical potentials @xmath26 extracted from urqmd 1.3 transport calculations in central au+au ( pb+pb ) collisions at 21.3 a@xmath4tev , 160 , 40 and 11 a@xmath4gev @xcite as a function of the reaction time in the center - of - mass ( from top to bottom ) . the open symbols denote nonequilibrium configurations and correspond to @xmath9 parameters extracted from the transverse momentum distributions , whereas the full symbols denote configurations in approximate pressure equilibrium in longitudinal and transverse direction . during the nonequilibrium phase ( open symbols ) the transport calculations show much higher temperatures ( or energy densities ) than the experimental chemical freeze - out configurations at all bombarding energies ( @xmath115 11 a@xmath4gev ) . these numbers are also higher than the tri - critical endpoints extracted from lattice qcd calculations by karsch et al . @xcite ( large open circle ) and fodor and katz @xcite ( star with horizontal error bar ) . though the qcd lattice calculations differ substantially in the value of @xmath26 for the critical endpoint , the critical temperature @xmath17 is closer to 160 mev in both calculations , while the energy density is of the order of 1 gev/@xmath15 or even below . nevertheless , this diagram shows that at rhic energies one encounters more likely a cross - over between the different phases when stepping down in temperature during the expansion phase of the hot fireball . this situation might change at lower sps or ags ( as well as new gsi sis-300 @xcite ) energies , where for sufficiently large chemical potentials @xmath26 the cross over should change to a first order transition @xcite , i.e. beyond the tri - critical point in the ( @xmath27 ) plane . nevertheless , fig . [ fig_qcd ] demonstrates that the transport calculation show temperatures ( energy densities ) well above the phase boundary ( horizontal line with errorbars from ref . @xcite ) in the very early phase of the collisions , where hadronic interactions yield only a pressure @xmath116 ( with @xmath117 denoting the energy density ) . we recall that for massless partonic degrees of freedom we have @xmath118 and consequently a much higher pressure . this argument is well in line with the studies on elliptic flow at rhic energies , that is underestimated by @xmath3 30% at midrapidity in the hsd approach @xcite and by a factor of @xmath29 in the urqmd 1.3 model @xcite . only strong pre - hadronic / partonic interactions might cure this problem . as shown in ref . @xcite , in order to describe the elliptic flow seen experimentally at rhic in a parton cascade model , one has to employ parton cross sections up to 45 mb . however , such strong interactions are incompatible with perturbative qcd , which gives cross sections that are lower by more than an order of magnitude @xcite . we speculate that such strong nonperturbative interactions on the partonic level are responsible for the large pressure generation in the very early phase of intermediate energy nucleus - nucleus collisions . summarizing this work , we point out that baryon stopping and hadron production in central au+au ( or pb+pb ) collisions is quite well described in the independent transport approaches hsd and urqmd 2.0 . also the longitudinal rapidity distributions of protons , pions , kaons , antikaons and hyperons are similar in both models and in reasonable agreement with available data . the exception are the pion rapidity spectra at the highest ags energy and lower sps energies , which are overestimated by both models @xcite . as a consequence the hsd and urqmd 2.0 transport approaches underestimate the experimental maximum of the @xmath7 ratio at @xmath3 20 to 30 a@xmath4gev and urqmd underestimates also the data at rhic energies . this also holds for the new version urqmd 2.1 introduced here additionally , which incorporates high mass resonance states above 2 gev with a particular decay pattern . this recipe cures the problem of the overestimated pion yield at ags and sps energies but also reduces the @xmath6 yield and therefore has only little effect on the @xmath7 ratio . we have found that the inverse slope parameters @xmath9 for @xmath6 mesons from the hsd and urqmd 2.0 transport models are practically independent of system size from @xmath1 up to central pb+pb collisions and show only a slight increase with collision energy . the calculated transverse mass spectra are in reasonable agreement with the experimental results for @xmath1 reactions at all bombarding energies investigated as well as central collisions of light nuclei ( c+c and si+si ) ( cf . ref . the rapid increase of the inverse slope parameters of kaons for central collisions of heavy nuclei ( au+au or pb+pb ) found experimentally in the ags energy range , however , is not reproduced by both models in their default version ( see fig . [ fig_tcron ] ) . we have , furthermore , discussed a couple of scenarios in section v to improve the description of the experimental data . however , none of the alternative scenarios leads to a fully satisfactory reproduction of all the experimental data . it is worth pointing out that a substantial hardening of the @xmath6 slope results from an initial state cronin enhancement at higher sps and rhic energies , that should not be attributed to collective flow ( or initial pressure ) . this cronin enhancement as known experimentally from @xmath119 reactions at sps energies does not signal a phase transition to a quark - gluon - plasma . we thus have argued - in comparison to lattice qcd calculations at finite temperature and baryon chemical potential @xmath26 from refs . @xcite and @xcite as well as the experimental systematics in the chemical freeze - out parameters ( cf . [ fig_qcd ] ) that the missing pressure should be generated in the early phase of the collision by nonperturbative partonic interactions because the strong hadronic interactions do nt produce it in the later stages . the interesting finding of our analysis is , that pre - hadronic / partonic degrees of freedom seem to play a substantial role in central au+au collisions already at ags energies above @xmath3 5 @xmath0gev . this finding is in line with the observation that the excitation function of the proton elliptic flow @xmath120 also indicates a softening of the equation of state with increasing energy at 4 - 5 @xmath0gev in au+au collisions @xcite . however , further detailed experimental and theoretical studies also on collective flow properties from nucleus - nucleus collisions will be necessary to unravel the various aspects of relativistic many - body systems . the authors acknowledge inspiring discussions with k. gallmeister , c. greiner , c. m. ko , z. w. lin , a. peshier , k. redlich , v. d. toneev and h. weber . furthermore , they are indepted to m. gadzicki and j. klay for providing experimental data in numerical form . e.l.b . is supported by deutsche forschungsgemeinschaft ( dfg ) and gsi . s.a.b . acknowledges support by riken , brookhaven national laboratory and doe grants de - fg02 - 96er40945 and de - ac02 - 98ch10886 as well as an outstanding junior investigator award ( doe grant de - fg02 - 03er41239 ) . m.b . and m.r . are supported by gsi . 99 , nucl . phys . a * 715 * , 1 ( 2003 ) . j. rafelski and b. mller , phys . lett . * 48 * , 1066 ( 1982 ) ; p. koch , b. mller and j. rafelski , phys . rep . * 142 * 167 ( 1986 ) . f. karsch _ et al . _ , b * 502 * , 321 ( 2001 ) . h. stcker and w. greiner , phys . rep . * 137 * , 277 ( 1986 ) . w. cassing _ et al . _ , a * 674 * , 249 ( 2000 ) . h. weber , c. ernst , m. bleicher _ et al . _ , phys . b * 442 * , 443 ( 1998 ) . z. fodor and s. d. katz , jhep * 0203 * , 014 ( 2002 ) ; z. fodor , s. d. katz , and k. k. szabo , phys . b * 568 * , 73 ( 2003 ) . bjorken , phys . d * 27 * , 140 ( 1983 ) . s. v.afanasiev _ et al . _ , na49 collaboration , phys . c * 66 * , 054902 ( 2002 ) . v. friese _ et al . _ , na49 collaboration , j. phys . g * 30 * , 119 ( 2004 ) . j. l. nagle and t. s. ullrich , nucl - ex/0203007 . s.s . adler _ _ , phenix collaboration , phys . lett . * 91 * , 072303 ( 2003 ) ; nucl - ex/0308006 . j. adams _ et al . _ , star collab . , lett . * 91 * , 072304 ( 2003 ) ; _ ibid . _ * 91 * , 172302 ( 2003 ) . i. arsene _ lett . * 91 * , 072305 ( 2003 ) ; x. n. wang , phys . c * 58 * , 2321 ( 1998 ) . et al . _ , nucl . part . * 50 * , 37 ( 2000 ) . m. i. gorenstein , m. gadzicki , and k. bugaev , phys . b * 567 * , 175 ( 2003 ) . m. gazdzicki and m. i. gorenstein , acta phys . b * 30 * , 2705 ( 1999 ) . h. weber , e. l. bratkovskaya , w. cassing , and h. stcker , phys . c * 67 * , 014904 ( 2003 ) . e. l. bratkovskaya , s. soff , h. stcker , m. van leeuwen , and w. cassing , phys . * 92 * , 032302 ( 2004 ) . bass _ et al . * 42 * , 255 ( 1998 ) . m. bleicher _ et al . _ , j. phys . g * 25 * , 1859 ( 1999 ) . j. geiss , w. cassing , and c. greiner , nucl . a * 644 * , 107 ( 1998 ) . w. cassing and e. l. bratkovskaya , phys . rep . * 308 * , 65 ( 1999 ) . w. cassing , k. gallmeister , and c. greiner , nucl . a * 735 * , 277 ( 2004 ) . the hsd transport code is available from : http://www.th.physik.uni-frankfurt.de/@xmath3brat/hsd.html p. braun - munzinger , j. stachel , j.p . wessels , and n. xu , phys . b * 344 * , 43 ( 1995 ) ; _ ibid . _ * 365*,1 ( 1996 ) ; p. braun - munzinger , j. cleymans , h. oeschler and k. redlich , nucl . a * 697 * , 902 ( 2002 ) . l. v. bravina _ et al . _ , c * 60 * , 024904 ( 1999 ) . a * 698 * , 383 ( 2002 ) . e. l. bratkovskaya , w. cassing , c. greiner _ et al . a * 675 * , 661 ( 2000 ) . z. w. lin _ et al . _ , a * 698 * , 375 ( 2002 ) . k. hagiwara _ et al . _ , ( review of particle properties ) , phys . d * 66 * , 010001 ( 2002 ) . bengtsson and t. sjstrand , comp . . commun . * 46 * , 43 ( 1987 ) . h. sorge , phys . c * 52 * , 3291 ( 1995 ) . b. andersson _ et al . _ , z. phys . c * 57 * , 485 ( 1993 ) . w. ehehalt and w. cassing , nucl . a * 602 * , 449 ( 1996 ) . et al . _ , e866 and e917 collaboration , phys . b * 476 * , 1 ( 2000 ) ; _ ibid . * 490 * , 53 ( 2000 ) ; phys . c * 58 * , 3523 ( 1998 ) . j. l. klay _ et al . _ , e895 collaboration , phys . c * 68 * , 054905 ( 2003 ) . s. ahmad _ et al . _ , e891 collaboration , phys . b * 382 * , 35 ( 1996 ) ; c. pinkenburg _ et al . _ , e866 collaboration , nucl . a * 698 * , 495c ( 2002 ) . a. mischke _ et al . _ , na49 collaboration , j. phys . g. * 28 * , 1761 ( 2002 ) ; nucl . phys . a * 715 * , 453 ( 2993 ) . f. antinori _ et al . _ , wa97 collaboration , nucl . a * 661 * , 130c ( 1999 ) . m. gazdzicki _ et al . _ , na49 collaboration , nucl - ex/0403023 . d. ouerdane _ et al . _ , brahms collaboration , nucl . a * 715 * , 478 ( 2003 ) ; j. h. lee _ et al . _ , j. phys . g * 30 * , s85 ( 2004 ) . s. s. adler _ _ , phenix collaboration , phys . rev . c * 69 * , 024904 ( 2004 ) ; nucl - ex/0307022 . c. adler _ _ , star collaboration , nucl - ex/0206008 ; o. barannikova _ et al . a * 715 * , 458 ( 2003 ) ; k. filimonov _ et al . _ , hep - ex/0306056 . a. andronic and p. braun - munzinger , hep - ph/0402291 . p. crochet _ _ , fopi collaboration , j. phys . g * 27 * , 267 ( 2001 ) . m. van leeuwen _ et al . _ , na49 collaboration , nucl . a * 715 * , 161 ( 2003 ) . e. l. bratkovskaya , w. cassing and h. stcker , phys . c * 67 * , 054905 ( 2003 ) . h. g. fischer _ et al . _ , na49 collaboration , nucl . a * 715 * , 118c ( 2003 ) . v. blobel _ et al . b * 69 * , 454 ( 1974 ) . v. topor pop , m. gyulassy , j. barrette , c. gale , x.n . wang , n. xu , and k. filimonov , phys . c * 68 * , 054902 ( 2003 ) . t. abbott _ et al . _ , e802 collaboration , phys . d * 45 * , 3906 ( 1992 ) . h. boggild _ et al . _ , na44 collaboration , phys . c * 59 * , 328 ( 1999 ) . a. frster _ et al . _ , kaos collab . , j. phys . g * 28 * , 2011 ( 2002 ) . b. v. martemyanov _ nucl - th/0212064 . s. soff _ et al . _ , lett b * 471 * , 89 ( 1999 ) . j. dias de deus , e. g. ferreiro , c. pajares , and r. ugoccioni , phys . b * 581 * , 156 ( 2004 ) ; hep - ph/0304068 . j. geiss , c. greiner , e.l . _ , phys . lett . * b447 * , 31 ( 1999 ) . e.g. ferreiro , c. pajares , and d. sousa , phys . b * 422 * , 314 ( 1998 ) ; m.a . braun and c. pajares , eur . j. c * 16 * , 349 ( 2000 ) . m. bleicher , c. spieles , c. ernst _ et al . _ , b * 447 * , 227 ( 1999 ) . h. van hecke _ et al . _ , lett . * 81 * , 5764 ( 1998 ) . c.e . copper _ et al . _ , na49 collaboration , nucl . a * 661 * , 362c ( 1999 ) . et al . _ , phys . d * 11 * , 3105 ( 1975 ) . d. antreasyan _ et al . _ , d * 19 * , 764 ( 1979 ) . kopeliovich , j. nemchik , a. schfer and a.v . tarasov , phys . lett . * 88 * , 232302 ( 2002 ) . g. papp , p. levai , and g. fai , nucl - th/0307062 . et al . _ , na44 collaboration , nucl - ex/0202019 . et al . _ , na49 collaboration , j. phys . g * 30 * , 5583 ( 2004 ) . m. kliemant , b. lungwitz , and m. gadzicki , hep - ex/0308002 . et al . _ , e866 collaboration , nucl . a * 610 * , 139c ( 1966 ) . j. bchler _ et al . _ , na49 collaboration , nucl . a * 661 * , 45c ( 1999 ) . w. cassing and u. mosel , prog . 25 * , 235 ( 1990 ) . j. cleymans and k. redlich , phys . c * 60 * , 054908 ( 1999 ) . f. karsch , talk given in _ quark matter 2004 _ , oakland , january 11 - 17 , 2004 ; http://qm2004.lbl.gov an international accelerator facility for beams of ions and antiprotons ; http://www.gsi.de / gsi - future / cdr/. e. v. shuryak , nucl . a * 661 * , 119c ( 1999 ) . m. bleicher and h. stcker , phys . b * 526 * , 309 ( 2002 ) . d. molnar and m. gyulassy , nucl . a * 698 * , 379 ( 2002 ) . p. danielewicz _ lett . * 81 * , 2438 ( 1998 ) . p. k. sahu and w. cassing , nucl . a * 712 * , 357 ( 2002 ) . e.g. nikonov , a.a . shanenko and v.d . toneev , heavy ion phys . * 8 * , 89 ( 1998 ) .

we consider a network of processor - sharing queues with state - dependent service rates . these are allocated according to balanced fairness within a polymatroid capacity set . balanced fairness is known to be both insensitive and pareto - efficient in such networks , which ensures that the performance metrics , when computable , will provide robust insights into the real performance of the system considered . we first show that these performance metrics can be evaluated with a complexity that is polynomial in the system size when we allow for some controlled asymmetry , in the sense that the network contains a fixed number of parts wherein all queues are ` exchangeable ' . this in turn allows us to derive stochastic bounds for a larger class of networks which satisfy less restrictive symmetry assumptions . these results are applied to practical examples of tree data networks and computer clusters . + * keyworkds : * processor - sharing networks , performance , balanced fairness , poly - symmetry

1610.00560

networks of processor - sharing queues with state - dependent service rates have been extensively used to model a large variety of communication and computation systems like content delivery systems @xcite , computer clusters @xcite and data networks @xcite . they are natural models for such systems as they capture the complex interactions between different jobs and also have a promise of analytical tractability of user performance when subject to stochastic loads . indeed , in the past two decades researchers have been able to obtain explicit performance expressions and bounds for several such systems , see @xcite . however , few performance results scale well with the system size . those that do rely on restrictive assumptions related to the topology or the symmetry of the system @xcite . one of the main goals of this paper is to provide scalable performance results for a class of processor - sharing systems which find applications in bandwidth - sharing networks and computer clusters . one of the key features of processor - sharing networks is the allocation of the service rates per queue in each state . a particular class of resource allocations which is more tractable for performance analysis is characterized by the balance property which constraints the relative gain in the service rate at one queue when we remove a job from another queue . processor - sharing networks where the resource allocation satisfies this property are called whittle networks @xcite . in particular , if the service rates are constrained by some capacity set , corresponding to the resources of the system considered , then there exists a unique policy which satisfies the balance property while being efficient , namely balanced fairness @xcite . in this paper we focus on networks which are constrained by a polymatroid capacity set @xcite and operate under balanced fair resource allocation policy . it was proved in @xcite that balanced fairness is pareto - efficient when it is applied in polymatroid capacity sets , which in practice yields explicit recursion formulas . however , if no further assumptions are made on the structure of the system , the complexity of these formulas is exponential with the number of queues . it was proved in @xcite that it can be made linear at the cost of strict assumptions on the overall symmetry of the capacity set and the traffic intensity at each queue . under symmetry in interaction across queues , it was shown in @xcite that the performance is robust to heterogeneity in loads and system configuration under an appropriate scaling regime . however , there is little understanding of performance for scenarios where queues themselves interact in heterogeneous fashion . in this paper , we consider a scenario where the network is divided into a finite number of parts , so that queues are exchangeable within each part and asymmetric across different parts . for such systems , that we call poly - symmetric , we obtain a performance expression with computational complexity which is polynomial in the number of queues . we demonstrate the usefulness of these bounds by applying them to tree data networks , which are representative of backhaul networks , and to randomly configured heterogeneous computer clusters . in addition , we provide a monotonicity bound which allows us to bound performance of systems with capacity regions which are ` closely ' poly - symmetric . the paper is organized as follows . section [ sec : model ] introduces the model and shows that it applies to systems as varied as tree data networks and computer clusters . we also recall known facts about balanced fairness . in section [ sec : polysymmetry ] , we introduce the notion of poly - symmetry and show that it yields explicit recursion formulas for the performance metrics which have a complexity that is polynomial in the number of queues in the network . finally , section [ sec : bounds ] gives stochastic bounds to compare the performance of different systems . we conclude in section [ sec : ccl ] . we consider a network of @xmath0 processor - sharing queues with coupled service rates and we denote by @xmath1 the set of queue indices . for each @xmath2 , jobs enter the network at queue @xmath3 according to some poisson process with intensity @xmath4 and have i.i.d . exponential service requirements with mean @xmath5 , resulting in a traffic intensity @xmath6 at queue @xmath3 . jobs leave the network immediately after service completion . the network state is described by the vector @xmath7 , where @xmath8 is the number of jobs at queue @xmath3 for each @xmath2 . for each @xmath9 , @xmath10 denotes the set of active queues in state @xmath11 . queues have state - dependent service rates . for each @xmath9 , @xmath12 denotes the vector of service rates per queue when the network is in state @xmath11 . the network is characterized by a _ capacity set _ , which is defined as the set of all feasible resource allocations @xmath13 . this capacity set may be specified by practical constraints like the capacities of the links in a data network or the service rates of the servers in a computer cluster . we are interested in queueing networks whose capacity set is a particular type of polytope called a _ polymatroid _ @xcite . [ def : polymatroid ] a polytope @xmath14 in @xmath15 is a _ polymatroid _ if there exists a non - negative function @xmath16 defined on the power set of @xmath17 such that @xmath18 and @xmath16 satisfies the following properties : normalization : @xmath19 , monotonicity : for all @xmath20 , if @xmath21 , then @xmath22 , submodularity : for all @xmath20 , @xmath23 . @xmath16 is called the _ rank function _ of the polymatroid @xmath14 . .45 ( c1 ) ; at ( @xmath24 ) ( c2 ) ; at ( @xmath25 ) ( c ) ; at ( c1 ) ; at ( c2 ) ; at ( c ) ; at ( @xmath26 ) @xmath27 ; ( @xmath28 ) ( @xmath29 ) ( @xmath30 ) ( @xmath31 ) ( @xmath32 ) ( @xmath33 ) ; at ( @xmath34 ) @xmath35 ; ( @xmath36 ) ( @xmath37 ) ( @xmath38 ) ( @xmath39 ) ( @xmath40 ) ( @xmath41 ) ; .45 ( server1 ) 1 ; at ( @xmath42 ) ( server2 ) 2 ; at ( @xmath43 ) ( server3 ) 3 ; at ( @xmath44 ) ( class1 ) 1 ; at ( @xmath45 ) ( class2 ) 2 ; ( class1 ) ( server1 ) ; ( class1 ) ( server2 ) ; ( class2 ) ( server2 ) ; ( class2 ) ( server3 ) ; + .45 ( server1 ) ; at ( server1 ) @xmath46 ; at ( @xmath47 ) ( server2 ) ; at ( server2 ) @xmath48 ; ( queue1 ) at ( @xmath49 ) ; ( [ xshift=--1pt , yshift=-+1pt ] queue1.north west ) rectangle ( [ xshift=+4pt , yshift=-1pt ] queue1.south west ) ; ( [ xshift=--1pt , yshift=- ] queue1.north west ) rectangle ( [ xshift=15pt , yshift= ] queue1.south west ) ; ( [ xshift=15pt ] queue1.north west ) ( [ xshift=15pt ] queue1.south west ) ; ( queue2 ) at ( @xmath50 ) ; ( [ xshift=--1pt , yshift=-+1pt ] queue2.north west ) rectangle ( [ xshift=+4pt , yshift=-1pt ] queue2.south west ) ; ( [ xshift=--1pt , yshift=- ] queue2.north west ) rectangle ( [ xshift=15pt , yshift= ] queue2.south west ) ; ( [ xshift=15pt ] queue2.north west ) ( [ xshift=15pt ] queue2.south west ) ; ( @xmath51 ) node[midway , above ] @xmath52 ( @xmath53 ) ; ( @xmath54 ) node[midway , above ] @xmath55 ( @xmath56 ) ; .45 queues , title="fig : " ] before we specify the resource allocation , we give two examples of systems that fit into this model . the first example is a data network with a tree topology @xcite , representative of backhaul networks . there are @xmath0 users that can generate flows in parallel and we denote by @xmath1 the set of user indices . for any @xmath2 , user @xmath3 generates data flows according to some poisson process with intensity @xmath4 that is independent of the other users . all flows generated by user @xmath3 follow the same route in the network and have i.i.d . exponentially distributed sizes with mean @xmath5 in bits , resulting in a traffic intensity @xmath6 in bit / s . the state of the network is described by the vector @xmath7 , where @xmath8 is the number of ongoing flows of user @xmath3 , for each @xmath2 . we make the following assumptions on the allocation of the resources . the capacity of each link can be divided continually among the flows that cross it . also , the resource allocation per flow only depends on the number of flows of each user in progress . in particular , it is entirely defined in any state @xmath9 by the total capacity @xmath57 allocated to flows of user @xmath3 , for any @xmath2 . under these assumptions , we can represent the data network by a processor - sharing network with @xmath0 queues , one per user . for each @xmath2 , queue @xmath3 contains the ongoing flows of user @xmath3 and its service rate in state @xmath11 is the total capacity @xmath57 allocated to the flows of user @xmath3 . we will now describe the corresponding capacity set . each link can be identified by the set of users that cross it . specifically , we can describe the network by a family @xmath58 of subsets of @xmath17 , where a set @xmath59 is in @xmath58 if and only if there is a link crossed by the flows of all users @xmath60 . we assume that the network is a tree in the following way . the network is called a _ tree _ if for all @xmath61 , @xmath62 implies that @xmath63 or @xmath64 . there is no loss of generality in assuming that @xmath65 , for if not , the network is a forest where each subtree can be considered independently . for each @xmath66 , we denote by @xmath67 the capacity in bit / s of link @xmath68 . we assume that all links are constraining since otherwise we can simply ignore the non - constraining ones . the resource allocation must then satisfy the capacity constraints @xmath69 so that the capacity set is given by @xmath70 figure [ fig : exmodel ] gives the example of a tree data network with @xmath71 users . the routes of the users are given in figure [ fig : extree ] . the flows of each user cross one link that is individual and another that is shared by both users . the representation of this data network as a processor - sharing network is given in figure [ fig : expsnet ] and the corresponding capacity set is given in figure [ fig : expolymatroid ] . it is easy to see that it is a polymatroid for any value of the link capacities . the following theorem generalizes this last remark to any tree data network . [ theo : tree ] the capacity set of a tree data network is a polymatroid with rank function @xmath16 defined by @xmath72 in addition , we have @xmath73 for each @xmath66 . we can certainly assume that @xmath58 contains all the singletons since letting @xmath74 for each @xmath2 does not modify the capacity set @xmath14 . we can easily see that the result remains true if we do not make this assumption . we apply the following lemma which is a direct consequence of theorems 2.5 and 2.6 of @xcite about intersecting - submodular functions on intersecting families . let @xmath58 be a family of subsets of @xmath17 and @xmath75 such that , for all @xmath61 with @xmath62 , we have @xmath76 , @xmath77 and @xmath78 . further assume that @xmath79 , @xmath80 and @xmath58 contains all the singletons of @xmath17 . then the set of solutions in @xmath81 of the equations @xmath82 is given by @xmath83 where @xmath84 is the normalized , submodular function defined by @xmath85 the definition of a tree ensures that @xmath58 satisfies the assumptions of the lemma , with the function @xmath86 defined on @xmath58 by @xmath87 for any @xmath66 . hence , the set of solutions of the capacity constraints in @xmath81 is @xmath88 where @xmath89 is the normalized , submodular function given by @xmath90 note that no claim about the monotonicity of @xmath89 is made above . the capacity set @xmath14 of the data network is then given by @xmath91 . we will deduce from this that @xmath14 is equal to the polymatroid @xmath92 with rank function @xmath16 given by @xmath93 an illustration of this is given in figure [ fig : treeset ] . one can check that this function @xmath16 coincides with the one given in the theorem statement . we first show that @xmath16 is indeed a rank function and in a second time we will prove that @xmath14 is equal to @xmath92 . the normalization of @xmath16 follows from that of @xmath89 . also @xmath16 is non - decreasing by construction . finally , for each @xmath20 , we have @xmath94 for some @xmath95 such that @xmath96 and @xmath97 , and also @xmath98 where the first inequality holds by submodularity of @xmath89 and the second by definition of @xmath16 , since @xmath99 and @xmath100 . hence @xmath16 is submodular . we finally prove that @xmath101 . it is clear that any vector in @xmath92 is also in @xmath91 since @xmath102 for all @xmath103 . conversely , consider @xmath104 . if @xmath105 is not in @xmath92 , then there is @xmath103 so that @xmath106 , which implies that @xmath107 . by definition of @xmath16 , it follows that there is @xmath108 so that @xmath109 is a strict subset of @xmath110 and @xmath111 . but then @xmath112 so that at least one component of @xmath105 is negative . this is a contradiction . [ ex : tree ] figure [ fig : extreegeneral ] gives the example of a tree data network with its capacity set . the routes of the users are given in figure [ fig : extreegeneralgraph ] . each link is labeled with the set of user indices whose flows cross this link . the capacity constraints are @xmath113 the rank function @xmath16 of the capacity set is given by @xmath114 .48 ( c1 ) ; at ( @xmath24 ) ( c2 ) ; at ( @xmath115 ) ( c12 ) ; at ( @xmath116 ) ( c3 ) ; at ( @xmath117 ) ( c ) ; at ( c1 ) ; at ( c2 ) ; at ( c3 ) ; at ( c12 ) ; at ( c ) ; at ( @xmath118 ) @xmath27 ; ( @xmath119 ) ( @xmath29 ) ( @xmath30 ) ( @xmath120 ) ( @xmath121 ) ( @xmath122 ) ( @xmath123 ) ( @xmath124 ) ; at ( @xmath125 ) @xmath35 ; ( @xmath126 ) ( @xmath37 ) ( @xmath38 ) ( @xmath127 ) ( @xmath128 ) ( @xmath129 ) ( @xmath130 ) ( @xmath131 ) ; at ( @xmath132 ) @xmath133 ; ( @xmath134 ) ( @xmath135 ) ( @xmath136 ) ( @xmath137 ) ( @xmath138 ) ( @xmath139 ) ; .4 we consider a cluster of @xmath140 servers which can be pooled to process jobs in parallel . the set of servers is denoted by @xmath141 . there are @xmath0 classes of jobs and we denote by @xmath1 the set of class indices . for any @xmath2 , class-@xmath3 jobs enter the cluster as a poisson process with intensity @xmath4 and have i.i.d . exponential service requirements , resulting in a traffic intensity @xmath6 for class @xmath3 . jobs leave the cluster immediately after service completion . the state of the cluster is described by the vector @xmath7 , where @xmath8 is the number of jobs of class @xmath3 , for each @xmath2 . the class of a job defines the set of servers that can process it . the server assignment is given by a family @xmath142 of subsets of @xmath143 , where @xmath144 denotes the set of servers that can serve class-@xmath3 jobs , for each @xmath2 . equivalently , the server assignment can be described by a bipartite graph @xmath145 called the _ graph of assignment _ of the computer cluster . the service capacity of server @xmath146 is @xmath147 for each @xmath148 . for any set @xmath103 of job classes , we let @xmath149 denote the aggregate capacity available for the classes in @xmath109 . we make the following assumptions on the allocation of the server capacities . servers can be pooled to process jobs in parallel . when a job is in service on several servers , its service rate is the sum of the rates allocated by each server to this job . we also assume that the capacity of each server can be divided continually among the jobs it can serve . finally , the allocation of the service rates per job only depends on the number of jobs of each class in the cluster . in particular , it is entirely defined in any state @xmath150 by the total capacity @xmath57 allocated to class-@xmath3 jobs , for each @xmath2 . under these assumptions , we can describe the evolution of the cluster with a processor - sharing network with @xmath0 queues , one per class . for each @xmath2 , queue @xmath3 contains class-@xmath3 jobs and its service rate in state @xmath11 is the total capacity @xmath57 allocated to class-@xmath3 jobs collectively . it was proved in @xcite that the capacity set of such a cluster is a polymatroid and that the function @xmath16 defined by is its rank function . figure [ fig : exmodel ] gives the example of a computer cluster . the graph of assignment is given in figure [ fig : exmgraph ] which depicts the job classes on the left and the servers on the right . server @xmath35 can serve both classes whereas servers @xmath27 and @xmath133 are specialized . the corresponding processor - sharing network with @xmath71 queues is showed in figure [ fig : expsnet ] . the capacity set of this computer cluster is a polymatroid in @xmath151 drawn in figure [ fig : expolymatroid ] . the vertical and horizontal sides correspond to the individual constraints of classes @xmath27 and @xmath35 , with @xmath152 and @xmath153 . the diagonal side corresponds to the joint constraint on classes @xmath27 and @xmath35 , with @xmath154 . the service rates are allocated by applying balanced fairness @xcite in the polymatroid capacity set @xmath14 introduced in section [ subsec : capacityset ] . for each @xmath2 , let @xmath155 denote the @xmath0-dimensional vector with @xmath27 in position @xmath3 and @xmath156 elsewhere . balanced fairness is defined as the only resource allocation that both satisfies the _ balance property _ @xmath157 and maximizes the resource utilization in the following sense : in any state @xmath9 , @xmath158 and there exists @xmath159 such that @xmath160 the balance property ensures that there exists a balance function @xmath161 on @xmath162 such that @xmath163 and @xmath164 the second condition implies that @xmath161 satisfies the recursion @xmath165 in @xcite it is proved that balanced fairness is pareto - efficient in polymatroid capacity sets , which means that this maximum is always achieved by the set @xmath166 of active queues : @xmath167 since the balance property is satisfied , the processor - sharing network defined in section [ subsec : capacityset ] is a whittle network @xcite . a stationary measure of the network state @xmath168 is @xmath169 where we use the notation @xmath170 for any @xmath9 . substituting ( [ eq : recphi ] ) into this expression yields @xmath171 it is proved in @xcite that the system is stable , in the sense that the underlying markov process is ergodic , if and only if @xmath172 which means that the vector of traffic intensities belongs to the interior of the capacity set . in the rest of the paper , we assume that this condition is satisfied and we denote by @xmath173 the stationary distribution of the network state . by abuse of notation , for each @xmath103 , we denote by @xmath174 the probability that the set of active queues is @xmath109 : @xmath175 for each @xmath2 , let @xmath176 $ ] denote the mean number of jobs at queue @xmath3 and , for each @xmath103 , let @xmath177 $ ] denote the mean number of jobs at queue @xmath3 given that the set of active queues is @xmath109 . by the law of total expectation , we have @xmath178 the following theorem gives a recursive formula for @xmath174 and @xmath179 for any @xmath103 and @xmath2 . it is a restatement of ( * ? ? ? * theorem 4 ) using the same idea as in ( * ? ? ? * proposition 4 and theorem 1 ) . [ theo : recsets ] for each non - empty set @xmath103 , we have @xmath180 let @xmath2 . for each set @xmath103 , we have @xmath181 if @xmath182 , and otherwise @xmath183 observe that allows one to evaluate recursively @xmath184 for each @xmath103 , from which @xmath185 can be computed . similarly , for each @xmath2 , allows one to evaluate recursively @xmath186 for each @xmath103 and each @xmath2 , from which the value of @xmath187 can be deduced . one could then compute performance metrics like the mean delay or the mean service rate per queue from @xmath187 by applying little s law . note that the complexity is exponential in the number of queues . the exponential complexity of the formulas of theorem [ theo : recsets ] is limiting when we want to predict the performance of large - scale systems . to cope with this , we introduce the notion of poly - symmetry , which allows us to obtain formulas with a complexity that is polynomial in the number of queues at the cost of some regularity assumptions on the capacity set and the traffic intensity at each queue . poly - symmetry is a generalization of the notion of symmetry which was considered in @xcite . the following definition will be used subsequently to introduce poly - symmetry . it is easy to check that it defines an equivalence relation on the set @xmath17 of indices . [ def : exch ] let @xmath14 be a polymatroid on @xmath81 and denote its rank function by @xmath16 . let @xmath188 with @xmath189 . we say that indices @xmath3 and @xmath190 are _ exchangeable _ in @xmath14 if @xmath191 like the name suggests , two indices are exchangeable if and only if exchanging these indices does not modify the capacity set . note that the exchangeability of two indices @xmath3 and @xmath190 implies that they have the same individual constraints @xmath192 . the reverse implication is not true when @xmath193 , as we will see in the following example . .23 ( server1 ) ; at ( @xmath194 ) ( server2 ) ; at ( @xmath195 ) ( server3 ) ; at ( @xmath196 ) ( server4 ) ; at ( @xmath44 ) ( class1 ) 1 ; at ( @xmath45 ) ( class2 ) 2 ; at ( @xmath197 ) ( class3 ) 3 ; ( class1 ) ( server1 ) ; ( class1 ) ( server2 ) ; ( class2 ) ( server2 ) ; ( class2 ) ( server3 ) ; ( class3 ) ( server3 ) ; ( class3 ) ( server4 ) ; .5 [ ex : exchindices ] consider the computer cluster with the graph of assignment depicted in figure [ fig : exexchgraph ] , where all servers have the same unit capacity . the corresponding polymatroid capacity set is illustrated in figure [ fig : exexchset ] . we have @xmath198 and @xmath199 , so that indices @xmath27 and @xmath133 are exchangeable . index @xmath35 is not exchangeable with any of the two other indices because @xmath199 while @xmath200 . let us now define poly - symmetry . let @xmath201 and consider a partition @xmath202 of @xmath17 in @xmath203 parts . [ def : poly ] let @xmath14 be a polymatroid in @xmath81 and denote its rank function by @xmath16 . @xmath14 is called _ poly - symmetric _ with respect to partition @xmath204 if for any @xmath205 , all indices in @xmath206 are pairwise exchangeable in @xmath14 . equivalently , a polymatroid @xmath14 is poly - symmetric with respect to a partition @xmath204 if and only if @xmath204 is a refinement of the quotient set of @xmath17 by the exchangeability relation in @xmath14 . it follows directly from the definition that the polymatroid of example [ ex : tree ] is poly - symmetric with respect to partition @xmath207 as soon as @xmath208 , as we can see in figure [ fig : extreegeneralregion ] . also in example [ ex : exchindices ] , the polymatroid is poly - symmetric with respect to partition @xmath209 . for each @xmath205 , let @xmath210 denote the size of part @xmath211 . for any @xmath103 , let @xmath212 denote the vector of sizes of each part of @xmath109 in the partition . the set of these vectors is denoted by @xmath213 we will now give an alternative definition of poly - symmetry which is equivalent to definition [ def : poly ] . it is a generalization of the definition of symmetry given in @xcite . we will use it to express and prove theorem [ theo : recpoly ] . [ def : poly2 ] let @xmath14 be a polymatroid in @xmath81 and denote its rank function by @xmath16 . @xmath14 is called _ poly - symmetric _ with respect to partition @xmath204 if for any @xmath103 , @xmath214 depends on @xmath109 only through the size of @xmath215 for each @xmath205 . equivalently , there exists a componentwise non - decreasing function @xmath216 such that @xmath217 for all @xmath103 . we call @xmath218 the _ cardinality rank function _ of @xmath14 with respect to partition @xmath204 . _ proof of the equivalence._we only prove that definition [ def : poly ] implies definition [ def : poly2 ] ; the reverse implication is clear . for any @xmath20 with @xmath219 , we can write @xmath220 and @xmath221 and we have @xmath222 . we are thus reduced to proving that @xmath223 for all disjoint sets @xmath224 such that @xmath219 . this can be done by descending induction on the cardinality of @xmath225 . .23 ( server1 ) ; at ( @xmath42 ) ( server2 ) ; at ( @xmath43 ) ( server3 ) ; at ( @xmath226 ) ( class1 ) 1 ; at ( @xmath227 ) ( class2 ) 2 ; at ( @xmath228 ) ( class3 ) 3 ; ( class1 ) ( server1 ) ; ( class1 ) ( server2 ) ; ( class2 ) ( server1 ) ; ( class2 ) ( server2 ) ; ( class2 ) ( server3 ) ; ( class3 ) ( server2 ) ; ( class3 ) ( server3 ) ; .5 , title="fig : " ] [ ex : poly ] consider the computer cluster with the graph of assignment depicted in figure [ fig : expolysym ] , where all servers have the same unit capacity . the corresponding capacity set is illustrated in figure [ fig : expolysymregion ] . it is poly - symmetric with respect to partition @xmath229 and the corresponding cardinal rank function @xmath218 is given by @xmath230 , @xmath231 and @xmath232 . let @xmath202 be a partition of @xmath17 . we consider a processor - sharing network with a polymatroid capacity set which is poly - symmetric with respect to @xmath204 . for each @xmath103 , the vector @xmath212 gives the number of active queues in each part of the partition when the set of active queues is @xmath109 . by abuse of notation , for each @xmath205 , we denote by @xmath233 the vector of @xmath234 with @xmath27 in component @xmath211 and @xmath156 elsewhere . like in section [ subsec : bf ] , the resources are allocated by applying balanced fairness in this capacity set under some vector of traffic intensity @xmath235 which satisfies the stability constraints . for simplicity of notation , for each @xmath236 , we denote by @xmath237 the probability that the number of active queues in part @xmath211 is @xmath238 for each @xmath205 : @xmath239 for each @xmath205 , let @xmath240 $ ] denote the mean number of jobs in the queues of part @xmath211 and , for each @xmath236 , let @xmath241 $ ] denote the mean number of jobs in the queues of part @xmath211 given that there are @xmath242 active queues in part @xmath243 for each @xmath244 . the regularity assumptions ensure that , for each @xmath205 , @xmath245 and @xmath246 for each @xmath236 also give the mean numbers of jobs at queue @xmath3 for any @xmath247 . by the law of total expectation , we have @xmath248 the following theorem gives a recursive formula for @xmath237 and @xmath249 that allows one to compute recursively these quantities with a complexity @xmath250 . the proof is given in appendix [ app : recpoly ] . [ theo : recpoly ] consider a network of @xmath0 processor - sharing queues with state - dependent service rates allocated according to balanced fairness in a polymatroid capacity set @xmath14 . assume that @xmath14 is poly - symmetric with respect to partition @xmath204 and denote by @xmath218 the corresponding cardinality rank function . further assume that for each @xmath205 , all queues of @xmath206 receive jobs with the same traffic intensity @xmath251 , i.e. @xmath252 for all @xmath247 . for each @xmath253 , we have @xmath254 let @xmath205 . for each @xmath236 , we have @xmath255 if @xmath256 , and otherwise @xmath257 this result applies to example [ ex : poly ] with the partition @xmath229 as soon as classes @xmath27 and @xmath133 have the same traffic intensity . the set of suitable vectors of traffic intensities appears in figure [ fig : expolysymregion ] . in this theorem , we have assumed that the cardinality rank function @xmath218 was given . given a system like the ones of sections [ subsec : modeltree ] and [ subsec : modelcluster ] which is known to be poly - symmetric with regard to some partition @xmath258 , one could ask if it is also possible to build @xmath218 with a complexity @xmath250 . this is straightforward for a computer cluster . concerning the tree data networks , we can actually apply a method similar to that of the proof of theorem [ theo : tree ] . specifically , we first define recursively a concave function @xmath89 on @xmath259 by @xmath260 , @xmath261 if there is @xmath66 so that @xmath262 , and otherwise @xmath263 from which we can construct @xmath218 by letting @xmath264 we will now see two examples of systems where this result applies . we consider the simple example of a tree data network where each user has an individual access line and all users share a backhaul link which a capacity @xmath225 in bit / s . the user access lines can have @xmath203 different capacities @xmath265 in bit / s . this corresponds to the model introduced in @xcite to predict some performance metrics in internet service provider networks . ( c1 ) ; at ( @xmath266 ) ( c2 ) ; at ( @xmath267 ) ( c3 ) ; at ( @xmath268 ) ( c4 ) ; at ( @xmath269 ) ( c5 ) ; at ( @xmath270 ) ( c ) ; at ( c1 ) ; at ( c2 ) ; at ( c3 ) ; at ( c4 ) ; at ( c5 ) ; at ( c ) ; at ( @xmath26 ) ; ( @xmath118 ) ( @xmath29 ) ( @xmath30 ) ( @xmath271 ) ( @xmath272 ) ( @xmath273 ) ; at ( @xmath34 ) ; ( @xmath125 ) ( @xmath37 ) ( @xmath38 ) ( @xmath274 ) ( @xmath275 ) ( @xmath276 ) ; at ( @xmath277 ) ; ( @xmath132 ) ( @xmath135 ) ( @xmath136 ) ( @xmath278 ) ( @xmath279 ) ( @xmath280 ) ; at ( @xmath281 ) ; ( @xmath282 ) ( @xmath283 ) ( @xmath284 ) ( @xmath285 ) ( @xmath286 ) ( @xmath287 ) ; at ( @xmath288 ) ; ( @xmath289 ) ( @xmath290 ) ( @xmath291 ) ( @xmath292 ) ( @xmath293 ) ( @xmath294 ) ; ( @xmath295 ) ( @xmath296 ) node[align = center ] [ black , midway , xshift=-1.4 cm ] access lines + with capacity @xmath297 ; ( @xmath298 ) ( @xmath299 ) node[align = center ] [ black , midway , xshift=-1.4 cm ] access lines + with capacity @xmath300 ; at ( @xmath301 ) backhaul link + with capacity @xmath225 ; figure [ fig : multisource ] gives a toy example with @xmath302 possible access rates @xmath297 and @xmath300 . there are three users with access rate @xmath297 and two users with access rate @xmath300 . all users are constrained by the backhaul link with capacity @xmath225 . for each @xmath205 we denote by @xmath206 the set of users with access rate @xmath303 . these form a partition of the set @xmath304 of users . theorem [ theo : tree ] ensures that the capacity set of this data network is a polymatroid with rank function @xmath16 given by @xmath305 it is poly - symmetric with respect to partition @xmath202 . the corresponding cardinality rank function @xmath218 is given by @xmath306 we further assume that for each @xmath205 , all users with access line @xmath303 have the same traffic intensity @xmath307 . then the network is stable whenever @xmath308 , and the system meets the conditions of theorem [ theo : recpoly ] . a metric of interest is the mean throughput per user . for each @xmath2 , we denote by @xmath309 and @xmath310 the probability measure and expectation corresponding to the biased stationary distribution @xmath311 . for each @xmath205 and each @xmath247 , the mean throughput perceived by user @xmath3 is then given by @xmath312 = \frac{\mathbb{e}[\phi_i({\bf x})]}{\mathbb{p}\ { { \bf x}_i > 0 \ } } = \frac{\varrho_k}{\mathbb{p}\ { { \bf x}_i > 0 \}},\ ] ] where the second equality holds by the conservation equation @xmath313 $ ] for all @xmath247 . using the notations of section [ subsec : recpoly ] , the mean throughput of the users with access rate @xmath303 is given by @xmath314 where @xmath237 for each @xmath236 can be computed with a complexity @xmath250 by . other performance metrics like the mean congestion rate per user can be computed similarly . let @xmath315 . we consider a computer cluster with @xmath316 servers and @xmath317 classes . all servers have the same unit capacity and all jobs have a unit mean size . the set @xmath17 of classes is partitioned into two subsets @xmath318 and @xmath319 . @xmath318 contains @xmath320 classes that can each be served by @xmath321 servers and @xmath319 contains @xmath321 classes that can each be served by @xmath320 servers . for any @xmath322 , the @xmath3-th class of @xmath318 can be served by the servers @xmath323 for @xmath324 . for any @xmath325 , the @xmath3-th class of @xmath319 can be served by the servers @xmath326 for @xmath327 . figure [ fig : polycluster ] gives a toy example with @xmath328 and @xmath329 . ( server1 ) 1 ; at ( @xmath330 ) ( server2 ) 2 ; at ( @xmath331 ) ( server3 ) 3 ; at ( @xmath332 ) ( server4 ) 4 ; at ( @xmath333 ) ( server5 ) 5 ; at ( @xmath334 ) ( server6 ) 6 ; at ( @xmath335 ) ( class1 ) 1 ; at ( @xmath336 ) ( class2 ) 2 ; at ( @xmath337 ) ( class3 ) 3 ; at ( @xmath338 ) ( class4 ) 4 ; at ( @xmath339 ) ( class5 ) 5 ; ( class1 ) ( server1 ) ; ( class1 ) ( server2 ) ; ( class2 ) ( server3 ) ; ( class2 ) ( server4 ) ; ( class3 ) ( server5 ) ; ( class3 ) ( server6 ) ; ( class4 ) ( server1 ) ; ( class4 ) ( server3 ) ; ( class4 ) ( server5 ) ; ( class5 ) ( server2 ) ; ( class5 ) ( server4 ) ; ( class5 ) ( server6 ) ; any class of @xmath318 shares exactly one server with any class of @xmath319 , and this server is dedicated to these two classes . the rank function of this cluster is thus given by @xmath340 the polymatroid capacity set defined by this rank function is poly - symmetric with respect to partition @xmath341 and the corresponding cardinality rank function is given by @xmath342 for each @xmath343 , assume that all classes in @xmath206 have the same traffic intensity @xmath251 . further assume that the vector of traffic intensities @xmath344 stabilizes the system , that is @xmath345 we can then apply theorem [ theo : recpoly ] with partition @xmath204 to compute the mean number of job of each class with a complexity @xmath346 . we deduce the mean delay @xmath347 of class-@xmath3 jobs for each @xmath2 by little s law : @xmath348 while the property of poly - symmetry is not often satisfied in practice , except in specific cases like the examples of sections 3.3 and 3.4 , it can be used to derive stochastic bounds on most systems , as shown below . the following result will allow us to control the impact of the capacity set on performance . [ theo : mono ] let @xmath349 and @xmath350 . consider two polymatroids @xmath351 and @xmath14 in @xmath81 such that @xmath351 is a subset of @xmath352 and a superset of @xmath353 . let @xmath235 be an element in the interior of @xmath353 and denote respectively by @xmath173 , @xmath354 and @xmath355 the steady state distributions of the processor - sharing networks with capacity sets @xmath351 , @xmath356 and @xmath357 under traffic intensity @xmath235 . then @xmath358 specifically , for each @xmath2 , we have @xmath359 where @xmath187 , @xmath360 and @xmath361 are the mean number of job at queue @xmath3 under distributions @xmath173 , @xmath354 and @xmath355 respectively . denote by @xmath362 and @xmath16 the rank functions of @xmath351 and @xmath14 respectively . let @xmath161 , @xmath363 and @xmath364 denote the balance functions of the resource allocations defined by balanced fairness in the capacity sets @xmath351 , @xmath356 and @xmath357 respectively . we first prove by induction on @xmath365 that @xmath366 the property holds for @xmath367 . let @xmath368 and assume the inequality is valid for any @xmath369 such that @xmath370 . then we have by : @xmath371 where the first inequality holds by the induction assumption and the second holds by the inclusion of @xmath353 into @xmath351 . we prove the other side of the inequality by using the inclusion of @xmath351 into @xmath352 . this completes the proof by induction . it follows that @xmath372 thus for each @xmath9 , we obtain @xmath373 the proof for the other part of the inequality is similar . the second inequality about the mean number of jobs follows by summation . the following sections illustrate how we can apply this result to the models of tree data networks and computer clusters . we first use this result to relax some assumptions of section [ subsec : multisource ] . we still consider @xmath203 groups @xmath374 of users which form a partition of the set @xmath17 . for each @xmath205 , the access rates of the users in @xmath206 may be different but we assume that they are all between @xmath375 and @xmath376 for some @xmath377 . similarly , the capacity of the backhaul link is between @xmath378 and @xmath379 . the corresponding polymatroid capacity set @xmath351 is not poly - symmetric with respect to partition @xmath258 any more but its rank function @xmath362 satisfies : @xmath380 where @xmath16 is the rank function defined by . denoting by @xmath14 the polymatroid defined by @xmath16 , it follows that @xmath351 is a superset of @xmath357 and a subset of @xmath356 . we can thus apply theorem [ theo : mono ] . in the special case where for each @xmath205 , all users of @xmath206 have the same traffic intensity @xmath381 , with @xmath382 , we can use theorem [ theo : recpoly ] to compute the bounds . specifically , let @xmath354 and @xmath355 denote the steady state distributions of the processor - sharing networks with capacity sets @xmath356 and @xmath357 respectively under traffic intensity @xmath235 . for each @xmath205 and each @xmath247 , the mean throughput @xmath383 of user @xmath3 satisfies @xmath384 where @xmath385 and @xmath386 are the mean throughputs under distributions @xmath354 and @xmath355 respectively . we have @xmath387 and by @xmath388 [ [ random - assignment . ] ] random assignment . + + + + + + + + + + + + + + + + + + we consider a cluster with @xmath140 servers and @xmath389 job classes . let @xmath201 and consider a partition @xmath258 of @xmath17 in @xmath203 parts . we use the same notations as in section [ subsec : recpoly ] . let @xmath390 be a vector of positive integers . we randomly assign the servers to the job classes as follows . for any @xmath205 and @xmath247 , the set @xmath391 of servers that can process class-@xmath3 jobs is chosen uniformly and independently at random among the subsets of @xmath141 of cardinality @xmath392 . like in section [ subsec : modelcluster ] , the random assignment is described by the family @xmath393 which defines a random bipartite graph @xmath394 with a deterministic sets of vertices @xmath17 and @xmath143 and a random set of edges . each realization @xmath142 of the random assignment defines a polymatroid capacity set with a rank function given by . this allows us to define a random rank function associated with the random assignment by @xmath395 now let @xmath16 denote the corresponding mean rank function : @xmath396 , \quad \forall a \subset i.\ ] ] the following lemma proves that the polymatroid defined by @xmath16 is poly - symmetric with respect to @xmath204 . [ lem : meanrank ] for each @xmath103 , we have @xmath397 with @xmath398 , @xmath399 and @xmath400 let @xmath236 and consider any set @xmath103 with @xmath401 . we just need to observe that @xmath402 for each @xmath205 such that @xmath403 , the probability that a server can not serve a specific class of @xmath215 is @xmath404 . since the assignments of the classes are independent , it follows that the probability that this server can serve at least one class in @xmath109 is given by @xmath405 . let @xmath406 be a vector of traffic intensities . for any realization @xmath407 of the random assignment which defines a capacity set in the interior of which @xmath235 is , we can study the steady state behavior of the cluster when we apply balanced fairness . the system is stable and the mean number of jobs per queue is given by @xmath408 , \quad \forall i \in i.\ ] ] [ [ asymptotic - poly - symmetry . ] ] asymptotic poly - symmetry . + + + + + + + + + + + + + + + + + + + + + + + + + we consider a sequence of computer clusters with random assignment as defined in the previous section . let @xmath201 and @xmath409 . for each @xmath349 , the @xmath0-th random cluster of the sequence contains @xmath410 servers and @xmath389 job classes . we denote the set of servers by @xmath411 and the set of job classes by @xmath412 . the service rate of server @xmath146 is @xmath413 for each @xmath414 . for simplicity , we consider a partition @xmath415 of @xmath416 in @xmath203 parts of size @xmath0 . we can generalize the result to @xmath203 parts of different sizes as long as the size of each part is linear in @xmath0 . for each @xmath205 and @xmath417 , the set @xmath418 of servers that can process class-@xmath3 jobs is chosen uniformly and independently at random among the subsets of @xmath419 of cardinality @xmath420 . let @xmath421 denote the random graph defined by this random assignment , @xmath422 the corresponding random rank function and @xmath423 its expectation . by lemma [ lem : meanrank ] , for any @xmath349 and @xmath424 , we have @xmath425 = \xi^{(n ) } m^{(n ) } p_a^{(n ) } \quad \text{with } \quad p_a^{(n ) } = 1 - \prod_{k=1}^k \left ( 1 - \frac{d_k^{(n)}}m \right)^{a_k}\ ] ] for all set @xmath426 with @xmath427 , where @xmath428 is the mean server capacity . theorem [ theo : concentration ] below shows that , under the following two assumptions on the server capacities and the degrees of parallelism , the probability that the random rank function is uniformly close to its mean is @xmath429 . the proof is given in appendix [ app : concentration ] . [ ass : servers ] for each @xmath349 , @xmath419 is partitioned into a constant number of groups . each group contains @xmath430 servers which have the same capacity . [ ass : degrees ] for each @xmath205 , the sequence @xmath431 satisfies @xmath432 . [ theo : concentration ] let @xmath350 . under assumptions [ ass : servers ] and [ ass : degrees ] , there exists a sequence @xmath433 such that @xmath434 and for any @xmath349 , @xmath435 and @xmath436 corollary [ coro : concentration ] follows from theorem [ theo : concentration ] . for any @xmath349 , let @xmath437 denote the polymatroid defined by the rank function @xmath423 . @xmath437 is poly - symmetric with respect to the partition @xmath438 . [ coro : concentration ] let @xmath350 . under assumptions [ ass : servers ] and [ ass : degrees ] , the random capacity set resulting from the random assignment is a subset of @xmath439 and a superset of @xmath440 with probability @xmath429 . [ [ performance - metrics . ] ] performance metrics . + + + + + + + + + + + + + + + + + + + + for each @xmath349 , we consider a vector @xmath441 of traffic intensities per part which stabilizes the processor - sharing network with @xmath0 queues and capacity set @xmath440 under balanced fairness , that is , @xmath442 where @xmath443 is the cardinality rank function of the mean capacity set with respect to partition @xmath438 : @xmath444 let @xmath445 denote the state in the @xmath0-th randomized computer cluster when we allocate the resources according to balanced fairness . given a realization @xmath446 which is stabilized by @xmath447 , the mean numbers of jobs per class are given by @xmath448 , \quad \forall i \in i^{(n)}.\ ] ] for each realization @xmath446 which is not stabilized by @xmath447 , we let @xmath449 . this allows us to define the random variables @xmath450 associating theorem [ theo : mono ] with theorem [ theo : concentration ] yields the following result . [ theo : boundcluster ] let @xmath350 . for any @xmath349 , denote by @xmath451 and @xmath452 the stationary distributions of the processor - sharing networks with @xmath0 queues and capacity sets @xmath453 and @xmath454 respectively , when the traffic intensity of the classes in @xmath455 is @xmath456 , for any @xmath205 . let @xmath457 and @xmath458 denote the corresponding mean number of jobs per queue in part @xmath211 , for each @xmath205 . under assumptions [ ass : servers ] and [ ass : degrees ] , we have @xmath459 for each @xmath349 , theorem [ theo : recpoly ] gives formulas to compute @xmath460 and @xmath461 for each @xmath205 with a complexity @xmath462 . using little s law , we can deduce bounds on the mean delay per class . , @xmath463 , @xmath464 , @xmath465 we omit to write the exponents for short . consider a randomized cluster with @xmath466 servers with unit capacity . the set of classes is partitioned into two parts @xmath318 and @xmath319 with @xmath463 classes each . the classes of @xmath318 have a degree @xmath464 and the ones of @xmath319 have a degree @xmath467 . all jobs have a mean size @xmath27 and the arrival rates are proportional to the degrees . the traffic intensity of any class in @xmath319 is thus twice that of any class in @xmath318 . let @xmath468 so that @xmath469 where @xmath218 is the cardinality rank function of the mean capacity set with respect to partition @xmath341 : @xmath470 we can prove that the vector of traffic intensities @xmath471 with @xmath472 for all @xmath473 and @xmath474 for all @xmath475 is on the boundary of the mean capacity set . let @xmath350 . for each @xmath476 , @xmath477 stabilizes the processor - sharing networks with capacity sets @xmath356 and @xmath357 . the bounds on the mean delay that follow from theorem [ theo : boundcluster ] by applying little s law are given by @xmath478 where @xmath479 and @xmath480 are computed with the recursion expressions of theorem [ theo : recpoly ] as follows . for each @xmath481 , @xmath482 for any @xmath343 and @xmath236 , we have @xmath483 if @xmath256 , and otherwise @xmath484 figure [ fig : boundscluster ] gives the bounds obtained as a function of @xmath485 , for different values of @xmath486 . for practicality , we draw the mean service rate per job , which is simply the inverse of the mean delay since all jobs have a unit mean size . in this paper , we consider processor - sharing networks and introduce a poly - symmetry criterion on the structure of the capacity set which ensures that the performance metrics can be computed with a complexity which is polynomial in the number of queues if the traffic intensities per queue are adjusted accordingly . we showed that these formulas can also be used to bound the performance of a system when its capacity set is closely poly - symmetric . we applied these results to tree data networks and computer clusters . we believe there is further scope of enhancing the stochastic bounds by expanding their scope as well as obtaining tighter bounds in some specific scaling regimes . let @xmath253 . by , we have @xmath487 the regularity assumptions ensure that @xmath488 for any @xmath103 with @xmath401 . thus we obtain @xmath489 for any @xmath205 and any @xmath247 , we do the substitution @xmath490 and thus we obtain for any @xmath205 , @xmath491 this proves . let @xmath205 and @xmath253 . by definition of @xmath249 , we have @xmath492 = \sum_{i \in i_k } \mathbb{e}\left [ { \bf x}_i | |i({\bf x})|_\sigma = a \right].\ ] ] it follows that @xmath493 and by , we obtain @xmath494 using the regularity assumptions , this can be rewritten as @xmath495 the first term is given by . the second term is simply @xmath496 finally , for any @xmath247 , we have @xmath497 doing the same substitution as in , we have for any @xmath244 , @xmath498 hence the third term of the sum is equal to @xmath499 where the second equality holds by . when we substitute the three terms by their expressions , we obtain . we give the proof only for the case @xmath302 for ease of notation ; the other cases follow analogously . for now , we assume that for all @xmath349 , all servers have the same capacity @xmath500 for any @xmath414 . let @xmath350 . we will show that there exists a sequence @xmath433 such that @xmath434 and for any @xmath349 , @xmath501 let us first give the main ideas of the proof . like in @xcite , it is divided in three steps . we first provide a bound for @xmath502 for each @xmath426 for @xmath0 large enough . then , for each @xmath503 , we use the union bound to obtain a uniform bound over all sets @xmath103 with @xmath427 . finally , another use of the union bound over all @xmath504 permits to conclude . let @xmath349 , @xmath504 and @xmath426 so that @xmath427 . recall that @xmath505 $ ] with @xmath506 the variables @xmath507 for @xmath414 are negatively associated bernoulli random variables with parameter @xmath508 hence , the following chernoff bounds @xcite apply : @xmath509},\end{aligned}\ ] ] where for any @xmath510 , @xmath511 $ ] is the kl divergence between two bernoulli random variables with parameters @xmath512 and @xmath513 respectively , given by @xmath514 = p \log\left ( \frac{p}{q } \right ) + ( 1-p ) \log\left ( \frac{1-p}{1-q } \right).\ ] ] we also use the following lemmas which will be proved later in appendix [ app : techlemmas ] : [ lem : smallminor ] let @xmath515 . consider a sequence @xmath433 such that @xmath516 and @xmath517 . for large enough @xmath0 , we have @xmath518 [ lem : bigminor ] there exists a positive constant @xmath519 such that @xmath520 \ge -\delta + \epsilon \frac{a_1 d_1^{(n ) } + a_2 d_2^{(n)}}{m^{(n ) } } , \quad \forall n \ge 1 , \quad \forall a \in { \cal n}^{(n)}.\ ] ] consider the sequence @xmath433 given by @xmath521 observe that @xmath434 , @xmath516 and @xmath517 . now let @xmath349 and @xmath504 . we distinguish two cases depending on the value of @xmath522 . by lemma [ lem : smallminor ] , there is a positive constant @xmath525 such that , for large enough @xmath0 , @xmath526 using ( [ eq : smallchernoff ] ) , we deduce that @xmath527 for any @xmath426 such that @xmath427 . the union bound yields @xmath528 since @xmath434 , we obtain for large enough @xmath0 @xmath529 with @xmath530 . combining lemma [ lem : bigminor ] with ( [ eq : bigchernoff ] ) , we deduce that there is a positive constant @xmath533 such that @xmath534 for any @xmath426 such that @xmath427 . since @xmath535 and @xmath516 , we have @xmath536 when @xmath0 is large enough . if @xmath531 , we also have that @xmath537 so that @xmath538 for large enough @xmath0 . the same argument holds by inverting @xmath539 and @xmath540 when @xmath532 , so we conclude that there is a positive constant @xmath541 such that for large enough @xmath0 , we have @xmath542 for any @xmath426 such that @xmath427 . the union bound yields @xmath543 since @xmath544 and @xmath545 , for large enough @xmath0 , we have @xmath546 for some positive constant @xmath547 . combining cases [ subapp : case1 ] and [ subapp : case2 ] , we deduce that there exists a positive constant @xmath548 such that @xmath549 when @xmath0 is large enough . using the union bound again , we obtain @xmath550 since @xmath434 , we conclude that for a constant @xmath551 , we have for large enough @xmath0 @xmath552 finally , when servers are in groups as in assumption [ ass : servers ] , we can break down @xmath422 into a sum of random rank functions , one for each groups . the result follows by showing the concentration in each group as above , and then using the union bound again . let @xmath515 . consider a sequence @xmath433 such that @xmath516 and @xmath517 . for large enough @xmath0 , we have @xmath518 consider the sequence @xmath553 of functions defined on @xmath151 by @xmath554 we have @xmath555 ^ 2 \xrightarrow[n \to \infty ] { } 1 \quad \text{and } \quad f_n\left ( 0 , \frac{2n}{g_n } \right ) = 1 - \left [ \left ( 1 - \frac{d_2^{(n)}}{bn } \right)^\frac{n}{g_n } \right]^2 \xrightarrow[n \to \infty ] { } 1 . \end{aligned}\ ] ] thus , there is @xmath556 so that @xmath557 and @xmath558 for all @xmath559 . then , for any @xmath559 and any @xmath560 , we have @xmath561 where the first two inequalities hold by concavity of @xmath562 . by definition of @xmath563 , @xmath564 { } = { } & ( 1 - \epsilon ) p_a^{(n ) } \log(1 - \epsilon ) + \left ( 1 - ( 1 - \epsilon ) p_a^{(n ) } \right ) \log\left ( 1 - ( 1 - \epsilon ) p_a^{(n ) } \right ) \\ & { } - { } \left ( 1 - ( 1 - \epsilon ) p_a^{(n ) } \right ) \log\left ( 1 - p_a^{(n ) } \right ) . \end{aligned}\ ] ] the first and the second terms are greater than @xmath565 and @xmath566 respectively . with @xmath567 , we obtain @xmath564 \ge - \delta - \left ( 1 - ( 1 - \epsilon ) p_a^{(n ) } \right ) \log\left ( 1 - p_a^{(n ) } \right ) \ge - \delta - \epsilon \log\left ( 1 - p_a^{(n ) } \right ) . \end{aligned}\ ] ] finally , observe that @xmath568 where in the inequality we used the fact that @xmath569 for @xmath343 . hence , we obtain the expected result . t. bonald and a. proutire . a queueing analysis of data networks . in j. r. boucherie and m. n. van dijk , editors , _ queueing networks : a fundamental approach _ , pages 729765 . springer us , boston , ma , 2011 . t. bonald , a. proutire , j. roberts , and j. virtamo . computational aspects of balanced fairness . in r. l. j. charzinski and p. tran - gia , editors , _ providing quality of service in heterogeneous environmentsproceedings of the 18th international teletraffic congress - itc-18 _ , volume 5 of _ teletraffic science and engineering _ , pages 801 810 . elsevier , 2003 . v. joseph and g. de veciana . stochastic networks with multipath flow control : impact of resource pools on flow - level performance and network congestion . in _ proceedings of the acm sigmetrics _ , pages 6172 , 2011

this survey is a guide for the non specialist on how to use rational homotopy theory techniques to get approximations of farber s topological complexity , in particular , and of schwarz s sectional category , in general .

1703.02791

the _ sectional category _ @xcite of a continuous map @xmath0 is the least integer @xmath1 for which there are @xmath2 local homotopy sections for @xmath3 whose domains form an open cover of @xmath4 . if @xmath5 is a path - connected topological space then two important invariants of the homotopy type of @xmath5 can be described through sectional category . the first one is the _ lusternik - schnirelmann ( ls ) category _ of @xmath5 @xcite , @xmath6 , which is the sectional category of the base point inclusion @xmath7 the second one is the _ ( higher ) topological complexity _ @xcite of @xmath5 , @xmath8 , which is the sectional category of the @xmath9 diagonal inclusion @xmath10 in this paper we will only consider , unless stated otherwise , simply connected cw complexes of finite type . as a consequence , every time we write the word _ space _ , we actually mean one of such cw complexes . + if @xmath5 is a space , then there exists a cofibration @xmath11 , called the _ rationalization _ of @xmath5 , which verifies : * @xmath12 is a _ rational space _ , that is , @xmath13 ( or equivalently @xmath14 ) is a rational vector space . * @xmath15 is a _ weak rational homotopy equivalence _ , that is , @xmath16 ( or equivalently @xmath17 ) is an isomorphism . * every continuous map @xmath0 , with @xmath4 a rational space , factors uniquely up to homotopy : + x_0 & + x&y + the main lower bound for sectional category is cohomological . namely , if @xmath3 is a continuous map and @xmath18 is any ring , then @xmath19 where @xmath20 denotes the _ nilpotency _ of an ideal @xmath21 , that is , the longest non trivial product of elements in @xmath21 . often , this bound is not good enough . this is mainly because cohomology does not capture all the homotopic information of a map . however , rational homotopy theory offers a whole new set of algebraic lower bounds which are based on the following inequalities : @xmath22 more precisely , there is a pair of contravariant adjoint functors a_pl & : || where * * cdga * is the category of simply connected commutative differential graded algebras over @xmath23 of finite type ( see section [ sec : sullivansrtc ] ) . * @xmath24 is the category of simply connected cw complexes of finite type . when restricted to rational spaces , these functors actually yield an equivalence in the associated homotopy categories @xcite . this means that all the rational homotopic information of a continuous map @xmath3 can be encoded algebraically through @xmath25 . from this algebraic object , one can deduce approximations for sectional category which are better than the cohomological lower bound , @xmath26 by relaxing the algebraic characterization of @xmath27 . for a deep description of the tools we use , the reader is invited to read the standard reference on rational homotopy theory @xcite . we start describing the algebraic objects we use . + a _ commutative graded algebra _ ( cga ) is a graded @xmath23-vector space @xmath28 together with a degree zero bilinear map @xmath29 , @xmath30 , called _ multiplication _ , verifying for @xmath31 , * @xmath32 and @xmath33 , * @xmath34 , and * @xmath35 . where @xmath36 means that @xmath37 and we say that @xmath38 is _ homogeneous _ of degree @xmath39 . observe that , if @xmath38 has odd degree , then @xmath40 . a _ morphism _ of cga @xmath41 is a linear map of degree zero such that @xmath42 , @xmath43 . derivation _ of degree @xmath44 on @xmath45 is a degree @xmath44 linear map @xmath46 verifying @xmath47 . a _ differential _ on @xmath45 is a derivation @xmath48 of degree @xmath49 such that @xmath50 . + a _ commutative differential graded algebra _ ( cdga ) is a pair @xmath51 where @xmath45 is a cga and @xmath48 is a differential on @xmath45 . a _ morphism _ of cdga is a morphism of cga commuting with differentials . we denote by * cdga * this category . there exists a _ homology _ functor @xmath52 defined as @xmath53 and @xmath54)=[f(z)]$ ] . a morphism of cdga @xmath3 is said to be a _ quasi - isomorphism _ if @xmath55 is an isomorphism . + the _ direct sum _ of two cdgas is a cdga , @xmath56 , with @xmath57 and @xmath58 for @xmath59 and @xmath60 . + the _ tensor product _ of two cdgas is a cdga , @xmath61 , with @xmath62 and @xmath63 . + given a graded @xmath23-vector space @xmath64 , define @xmath65 where @xmath66 and @xmath67 . we have that @xmath68 is a graded algebra with product @xmath69 consider @xmath21 the ideal of @xmath68 generated by elements of the form @xmath70 and define the _ free commutative graded algebra _ generated by @xmath64 : @xmath71 we will write simply @xmath72 to denote the element @xmath73\in\lambda v$ ] . we say that an element @xmath74 has _ word length _ @xmath75 and degree @xmath76 . we will denote by @xmath77 the elements of positive degree of a graded algebra @xmath45 , and by @xmath78 the elements of positive word length of a free graded algebra . we will also employ notations such as @xmath79 to mean @xmath80 , and so on . observe that there is an isomorphism @xmath81 which we often will use implicitly . + the free commutative graded algebra @xmath82 has the following two universal properties:[universalprops ] * given @xmath45 a cga and @xmath83 a linear map of degree zero , then there exists an unique cga morphism @xmath84 verifying @xmath85 for all @xmath86 . * if @xmath87 is a linear map of degree @xmath44 then there exists a unique derivation on @xmath82 of degree @xmath44 , @xmath88 , such that @xmath89 for all @xmath86 . let @xmath51 be a cdga . a _ relative sullivan algebra _ over @xmath51 is a cdga of the form @xmath90 such that @xmath91 , if @xmath59 , and @xmath92 with @xmath93 and @xmath94 if , in addition , the differential @xmath95 satisfies @xmath96 we say that @xmath97 is a _ minimal relative sullivan algebra_. observe that the inclusion @xmath98 is a cdga morphism . in fact , the category * cdga * is a closed model category @xcite ( if we restrict to cdgas @xmath51 for which @xmath99 @xcite ) with the following structure : * weak equivalences : quasi - isomorphisms @xmath100 . * fibrations : surjective morphisms @xmath101 . * cofibrations : inclusions of a cdga @xmath51 into a relative sullivan algebra @xmath102 as a consequence we get the following important fact : + * existence of relative sullivan models : * every cdga morphism @xmath103 can be factored as ( a , d)&&(b , d ) + & ( av , d ) , where either @xmath104 is surjective or @xmath90 is minimal ( but not necessarily both at the same time ) . @xmath51-module homotopy retraction _ for @xmath103 is a chain map @xmath105 such that @xmath106 and @xmath107 for all @xmath59 and @xmath108 . moreover , if @xmath109 is a cdga morphism , we say that it is a _ ( cdga ) homotopy retraction _ for @xmath103 . + * the surjective trick : * every cdga morphism @xmath103 can be factored as ( a , d)&&(b , d ) + & ( av , d ) , * the lifting lemma : * for any solid commutative cdga diagram , there exists a dashed arrow completing the commutative cdga diagram ( a , d)&(b , d ) + ( av , d)&(c , d ) , provided that at least one of the vertical morphisms is a quasi - isomorphism . + since * cdga * is a closed model category , there exists a notion of homotopy of maps @xmath110 which is nicely described in ( * ? ? ? 12 - 14 ) . two cdga morphism @xmath111 are _ weakly equivalent _ when there is a homotopy commutative cdga diagram @xmath112&(\lambda v , d)\ar[d]\ar[l , " \simeq"']\ar[r , " \simeq"]&(a_2,d)\ar[d , " \varphi_2"]\\ ( b_1,d)&(\lambda w , d)\ar[l , " \simeq"']\ar[r , " \simeq"]&(b_2,d ) , \end{tikzcd}\ ] ] where @xmath113 and @xmath114 are sullivan algebras ( see below ) . in the special case that @xmath115 , the initial object of * cdga * , we get _ sullivan _ ( perhaps _ minimal _ ) _ models _ for a cdga @xmath116 , @xmath117 . these objects are very important as they are the fibrant - cofibrant objects in the category * cdga*. for us , unless stated otherwise , cdgas denoted in the form @xmath113 will be assumed to be sullivan algebras , the same applies to relative sullivan algebras @xmath118 . + to be more explicit , a _ sullivan algebra _ is a cdga of the form @xmath119 verifying the _ nilpotence condition _ , that is , @xmath92 with @xmath120 and @xmath121 if moreover , the differential @xmath95 verifies @xmath122 , we say that it is a _ minimal sullivan algebra_. as with relative sullivan models for cdga morphism , the construction of sullivan models for a given cdga , @xmath123 , can be carried out inductively , degree by degree , by adding generators to @xmath64 and defining their differentials to _ kill _ or create the necessary homology classes in order to turn @xmath104 into a quasi - isomorphism . this process is very well explained in ( * ? ? ? 144 ) when @xmath124 . + let @xmath113 be a sullivan algebra and write @xmath125 with @xmath126 . the equation @xmath127 implies that @xmath128 for all @xmath129 . observe then that @xmath130 is a differential on @xmath64 , called the _ linear part _ of @xmath48 , which gives a chain complex @xmath131 . analogously , given a cdga morphism @xmath132 define the _ linear part _ of @xmath103 as the chain complex morphism @xmath133 defined by @xmath134 . + we can now state some important facts about sullivan algebras which will be used throughout this survey ( * ? ? ? 14.13 , thm . 14.12 ) . [ prop : sullivanalgs ] let @xmath132 be a cdga morphism between sullivan algebras . * then @xmath103 is a quasi - isomorphism if and only if @xmath135 is an isomorphism . * if @xmath119 and @xmath114 are minimal , then @xmath103 is a quasi - isomorphism if and only if @xmath103 is an isomorphism . given a space @xmath5 , we say that a cdga @xmath51 is a _ model _ for @xmath5 if @xmath136 is weakly equivalent to @xmath51 , that is , if there is a sullivan algebra @xmath113 and quasi - isomorphisms ( a , d)&(v , d ) & a_pl(x ) . in this case , we say that @xmath113 is a _ sullivan model _ of @xmath5 ( and of @xmath51 ) . these special _ cofibrant _ models are suitable for encoding the rational homotopic information of @xmath5 . in fact , by proposition [ prop : sullivanalgs ] , two minimal models for a space @xmath5 are isomorphic . in fact , we have ( * ? ? ? * cor . 10.10 , thm . 15.11 ) [ th : connectintopol ] if @xmath113 is _ the _ minimal sullivan model for @xmath5 , then * @xmath137 , and * for every @xmath138 there is a natural linear isomorphism @xmath139 . a cdga morphism @xmath103 is said to be a _ model _ for a continuous map @xmath3 if @xmath103 is weakly equivalent to @xmath25 . + we will often use a special notation for describing sullivan algebras which can be understood by the following example : @xmath140 means @xmath113 with @xmath64 spanned by @xmath141 of degrees @xmath142 respectively and with @xmath143 , @xmath144 and @xmath145 . the universal properties of p. show that the previous @xmath113 is well defined . let us compute the minimal sullivan models for spheres . we have to construct a quasi isomorphism @xmath146 . recall that @xmath147 with @xmath148 the fundamental class of @xmath149 . so @xmath64 must have one generator of degree @xmath9 which is a cycle , say @xmath38 . take @xmath150 a cycle in degree @xmath9 representing @xmath148 and define ( ( a),0)&a_pl(s^n ) + a & + here the dimension is crucial . if @xmath9 is odd , then @xmath40 and @xmath151 is an isomorphism . therefore @xmath152 is the minimal model of an odd sphere . however , if @xmath9 is even , then @xmath153 and @xmath104 can not be a quasi - isomorphism , this means that we need to add new generators in @xmath64 that turn @xmath154 into boundaries . so let @xmath155 be a new generator of degree @xmath156 and define @xmath157 . then in @xmath158 we have 1&a&a^2&a^3&a^4 + & x&ax&a^2x&a^3x which shows that @xmath159 . however , we still have to define the quasi - isomorphism @xmath104 to make sure that @xmath158 is the minimal model of an even sphere . observe that , in @xmath160 , @xmath161 , this means that there exists @xmath162 of degree @xmath156 such that @xmath163 . we then define @xmath104 as ( ( a , x),d)&a_pl(s^n ) + a & + x&. observe finally that @xmath164 . this proves that @xmath104 is a quasi - isomorphism showing that @xmath158 is the minimal sullivan model of an even sphere . notice that , as corollary , we get the serre finiteness theorem for the homotopy groups of spheres . in the previous example , there exists also a quasi - isomorphism @xmath165 defined as @xmath166 $ ] and @xmath167 . when this happens we say that the space is _ more precisely , a cdga @xmath51 is said to be _ formal _ when @xmath51 and @xmath168 are weakly equivalent . a space @xmath5 is said to be _ formal _ when @xmath136 is a formal cdga . [ examp : notformalspace ] a cdga which is not formal . consider @xmath169 and observe that , as vector spaces , @xmath170 @xmath171 , [ b ] , [ ax ] , [ bx ] , [ abx]\rangle.\ ] ] for degree reasons , any cdga morphism @xmath172 must satisfy @xmath173 , but then , we would have @xmath174)=[\varphi(ax)]=[\varphi(a)0]=0 $ ] , so @xmath103 can not be a cdga quasi - isomorphism . this proves that @xmath113 is not _ the _ minimal model of the cdga @xmath175 and therefore @xmath113 is not formal . a cdga morphism @xmath103 is said to be _ formal _ when it is weakly equivalent to @xmath176 . a continuous map @xmath3 is said to be _ formal _ if @xmath25 is formal @xcite or equivalently , when @xmath177 is a model for @xmath3 . obviously , if a map @xmath0 is formal then both @xmath5 and @xmath4 are formal spaces , however , the converse is not true as we will see in example [ example : fibrehopffib ] . one of the most frequent ways in rational homotopy theory to construct models is through homotopy pullbacks and pushouts . suppose that either @xmath3 or @xmath178 is a fibration in the following commutative diagram @xmath179\ar[r , " \beta"]&x\ar[d , " f"]\\ y\ar[r , " g"']&z . \end{tikzcd}\ ] ] then the universal property of pullbacks induces a diagram @xmath180\ar[ddr , bend right , " \alpha"']\ar[dr , " { ( \alpha,\beta ) } " description]&&\\ & t\ar[d , " \hat{f}"']\ar[r , " \hat{g}"]&x\ar[d , " f"]\\ & y\ar[r , " g"']&z . \end{tikzcd}\ ] ] in which the square is a ( homotopy ) pullback@xcite . we now explain how to construct cdga models for the previous diagram . suppose we have a commutative cdga square modeling diagram [ diag : commsquare ] ( applying @xmath181 for example ) , @xmath182\ar[r , " f"]&(b , d)\ar[d , " \beta"]\\ ( c , d)\ar[r , " \alpha"']&(d , d ) , \end{tikzcd}\ ] ] where the name of a cdga morhpism is the same as the name of the continuous maps it models ( for instance , @xmath183 is a cdga model for @xmath184 ) . we will often adopt this convention . now choose @xmath3 or @xmath178 and factor it as a cofibration followed by a weak equivalence , say @xmath185 with @xmath186 and @xmath187 . since @xmath188 , diagram [ diag : pullbackwhisker ] is modeled by @xmath189\\ ( a , d)\arrow[ur , " f"]\arrow[r , hookrightarrow , " i"']\arrow[d , " g " ' ] & ( a\otimes\lambda v , d)\arrow[u , " \simeq " , " \theta"']\arrow[d , " \hat{g}"]\\ ( c , d)\ar[drr , bend right , " \alpha"']\arrow[r , hookrightarrow , " \hat{f}"']&(c\otimes\lambda v,\overline{d})\ar[dr , " { ( \alpha,\beta ) } " description]\\ & & ( d , d ) \end{tikzcd}\ ] ] in which the square is a pushout in * cdga * , @xmath190 , @xmath191 , @xmath192 for @xmath59 and @xmath86 . observe also that the induced morphism @xmath193 is given by @xmath194 , @xmath195 . + let @xmath196 be a fibration with fiber @xmath197 . then we have a ( homotopy ) pullback @xmath198\ar[r]\ar[d]&e\ar[d , twoheadrightarrow , " p"]\\ * \ar[r , hook]&b . \end{tikzcd}\ ] ] following previous section , take @xmath199 a relative sullivan model for @xmath200 . then we have a model for the previous diagram @xmath201\ar[d , twoheadrightarrow]&(a\otimes \lambda v , d)\ar[d , twoheadrightarrow , " q"]\\ ( { \mathbb{q}},0)\ar[r , hook ] & ( \lambda v , \overline{d } ) , \end{tikzcd}\ ] ] where the projection @xmath202 is a model for the inclusion @xmath203 ( * ? ? ? * theorem 15.3 ) . [ example : fibrehopffib ] consider the hopf fibration @xmath204 . a cdga model for @xmath200 is the projection @xmath205 , since there is a bijection @xmath206\cong [ ( \lambda(a , x),d),(\lambda(x),0)]$ ] . construct now a relative sullivan model for @xmath207 ( and thus , for @xmath200 ) as follows : since @xmath208)=0 $ ] , introduce a generator @xmath209 of degree @xmath210 and define @xmath211 . now extend @xmath207 to @xmath212 by setting @xmath213 . now observe that @xmath214\rangle$ ] , therefore , @xmath104 is a quasi - isomorphism ( proposition [ prop : sullivanalgs ] ) . this gives a model of the fiber of @xmath3 @xmath215 which is @xmath216 . observe lastly that @xmath217 is trivial but @xmath200 is not , this shows that @xmath200 is a map between formal spaces which is not formal . dually , one can also model homotopy pushouts throught pullbacks provided one of the cdga models is surjective . consider @xmath218\ar[r]&(a , d)\ar[d , twoheadrightarrow , " f"]\\ & ( b , d)\ar[r , " g"']&(x , d ) , \end{tikzcd}\ ] ] where @xmath219 is the sub cdga of the direct sum @xmath220 , @xmath221 if @xmath222 is a cofibration modeled by a surjective cdga morphism @xmath223 , the projection @xmath224 is modeled by the the inclusion @xmath225 the weakness of this approach , in contrast with the models of fibration , is that the model for the cofiber , @xmath226 , need not be a sullivan model . recall @xmath207 the model of the hopf fibration @xmath200 in example [ example : fibrehopffib ] . then a model for the homotopy cofiber of @xmath200 is given by @xmath227 . one can see that the sullivan model for this space is given by @xmath228 . let @xmath51 be a cdga model for a space @xmath5 with @xmath32 , then the _ augmentation _ morphism @xmath229 is a cdga model for the base point inclusion @xmath230 . observe that this inclusion is homotopy equivalent to the based path space fibration @xmath231 , @xmath232 whose fiber is the loop space @xmath233 . now let @xmath234 be the minimal sullivan model for @xmath51 and take a relative sullivan model for the augmentation @xmath235 which is of the form @xmath236 with @xmath237 a degree 1 linear isomorphism . moreover , since the fiber of @xmath238 , @xmath233 , is an h - space , by ( * ? ? ? 143 ) we must have that @xmath239 . now the following homotopy pushout diagrams gives a relative model for @xmath240 @xmath241\ar[d , " \theta " ' , " \simeq"]&(\lambda v\otimes\lambda\hat{v},d)\ar[d , " \theta\otimes{{\rm id}}"',"\simeq"]\ar[ddr , bend left , " \simeq"]\\ ( a , d\ar[drr , bend right , " \epsilon"'])\ar[r , hook]&(a\otimes\lambda\hat{v},\overline{d})\ar[dr , " \simeq"]&\\ & & ( { \mathbb{q}},0 ) , \end{tikzcd}\ ] ] where @xmath242 this construction is called the _ acyclic closure _ of @xmath51@xcite . let @xmath5 be a space and @xmath243 the @xmath9 diagonal map @xmath244 . if @xmath51 is a cdga model for @xmath5 , then the @xmath9-multiplication morphism @xmath245 is a surjective cdga model for @xmath246 ( * ? ? ? + let us now give a model of @xmath247 , the cofibre of @xmath248 . if @xmath59 , denote by @xmath249 the corresponding inclusion of @xmath38 into the @xmath39-th factor . let @xmath250 be a set of generators for @xmath45 , that is , @xmath251 , the ideal of @xmath45 generated by @xmath250 . by ( * ? * section 6 ) , @xmath252 is generated by @xmath253 . we see then that a cdga model for @xmath247 is @xmath254 . + the diagonal inclusion @xmath246 is homotopy equivalent to the path fibration @xmath255 , @xmath256 whose fiber is the product of based loops on @xmath5 , @xmath257 . + let us now construct a cdga model for this fibration . let @xmath113 be a sullivan model for @xmath5 , then by ( * ? ? ? 206 ) a relative sullivan model for the multiplication @xmath258 where @xmath259 is given by @xmath260 , together with @xmath261 , @xmath262 and @xmath263 the map @xmath264 is the degree @xmath265 derivation defined by @xmath266 , @xmath267 . here , subscripts are used to distinguish copies of @xmath64 . observe that @xmath268 therefore a quasi - isomorphism @xmath269 is defined by @xmath270 and @xmath271 . now @xmath272 fits in the homotopy pullback @xmath273 x^i\ar[r]\ar[d , " \pi_3"']&x^i\times x^i\ar[d , " \pi_2\times\pi_2"]\\ x\times x\times x\ar[r , " { { \rm id}}\times\delta_2\times{{\rm id}}"']&x\times x\times x\times x \end{tikzcd}\ ] ] which is modeled through the pushout @xmath274\ar[d , " { { \rm id}}\otimes\mu_2\otimes{{\rm id}}"']\ar[r , hook]&(\lambda v\otimes\lambda v\otimes\lambda\hat{v},d)\otimes(\lambda v\otimes\lambda v\otimes\lambda\hat{v},d)\ar[d]\ar[u , " \theta\otimes\theta " ' , " \simeq " ] \\ ( \lambda v_1\otimes\lambda v_2\otimes\lambda v_3,d)\ar[r , hook]&(\lambda v_1\otimes\lambda v_2\otimes\lambda v_3\otimes\lambda \hat{v}_1\otimes\lambda\hat{v_2},d ) . \end{tikzcd}\ ] ] an inductive argument proves that the relative sullivan model for @xmath275 is given by the inclusion @xmath276 where @xmath277 observe , in particular that the minimal model for the homotopy fibre of @xmath278 is given by @xmath279 this corresponds to the model given in ( * ? ? ? 143 ) for h - spaces and the fact that @xmath280 . + even spheres @xmath149 . let @xmath281 be the minimal model for @xmath149 . then , with previous notation , the multiplication morphism is given by @xmath282 with @xmath283 , @xmath284 , and @xmath285 , @xmath286 . therefore a relative sullivan model for @xmath287 is given by @xmath288 with @xmath289 and @xmath290 . let us now compute the differentials . since @xmath291 we have @xmath292 . we now compute @xmath293 using the fact that @xmath95 and @xmath294 are derivations : * @xmath295 : @xmath296 , * @xmath297 : @xmath298 , * @xmath299 : @xmath300 therefore @xmath301 observe that @xmath302 and that the morphism @xmath303 @xmath304 , @xmath305 , @xmath306 is a cdga quasi - isomorphism since @xmath307 is an isomorphism . we present this section as an overview of the rational homotopy techniques used to study ls category . the rational methods used to study sectional category are generalizations of the ones used for ls category . + we begin stating the flix - halperin theorem ( * ? ? ? * theorem viii ) which gives an algebraic characterization of the ls category of rational spaces . this theorem is key in the development of computation methods for ls category . [ th : felixhalperincharac ] let @xmath5 be a space and @xmath113 a sullivan model for @xmath5 . then the rational ls category of @xmath5 , @xmath308 , is the least @xmath1 for which the cdga projection @xmath309 admits a homotopy retraction . [ example : felixhalpneedsullivan ] let @xmath310 and @xmath311 , with @xmath312 , its minimal model . since @xmath313 and @xmath5 is not contractible , we see that @xmath314 . now take a cdga model for @xmath5 of the form @xmath315 the projection @xmath316 is not homology injective since the fundamental class is represented by @xmath317 . therefore @xmath318 can not have a homotopy retraction . this shows that , in previous theorem , it is necessary to take a sullivan model for the space . one can define new invariants by weakening the requirements on homotopy retractions , namely * the _ module ls category _ of @xmath5 , @xmath319 , as the smallest @xmath1 such that @xmath320 admits a homotopy retraction as @xmath113-module , * the _ rational toomer invariant _ of @xmath5 , @xmath321 , as the least @xmath1 such that @xmath322 is injective . hence [ coro : inequalitiescat ] let @xmath5 be a space and @xmath51 be any cdga model for @xmath5 , then @xmath323 if @xmath5 is formal then one can take @xmath324 and all the above are thus equalities . for the first inequality , suppose @xmath325 and consider @xmath326,[x_1],\dots,[x_m]\in{{\rm h}}^+(x,{\mathbb{q}})\cong { { \rm h}}^+(\lambda v , d).\ ] ] since one can identify @xmath327 , we have @xmath328 and therefore , @xmath329\in \ker { { \rm h}}(\rho_m)$ ] . since , by hypothesis , @xmath322 is injective , we must have @xmath330[x_1]\cdots[x_m]=[x_0x_1\cdots x_m]=0 $ ] . let us now prove the last inequality . suppose @xmath331 , and take a quasi - isomorphism @xmath332 . since @xmath333 , @xmath104 factors through @xmath320 as @xmath334 ( \lambda v , d)\ar[dr , twoheadrightarrow , " \rho_m"']\ar[rr , " \theta " , " \simeq"']&&(a , d)\\ & ( \frac{\lambda v}{\lambda^{>m}v},\overline{d}).\ar[ur , " \overline{\rho}_m " ' ] \end{tikzcd}\ ] ] then taking a relative sullivan model for @xmath320 , the lifting lemma gives a homotopy retraction @xmath109 for @xmath320 : @xmath335\ar[d , hook]\ar[dl , " \rho_m"']&(\lambda v , d)\ar[d , " \theta " , " \simeq"']\\ ( \frac{\lambda v}{\lambda^{>m}v},\overline{d})&(\lambda v\otimes\lambda z , d)\ar[l , " \xi " , " \simeq"']\ar[r , " \overline{\rho}_m\circ\xi"']\ar[ur , dashed , " r"']&(a , d ) \end{tikzcd}\ ] ] [ prop : lscatvodd ] if @xmath5 is a space such that @xmath336 is finite dimensional and concentrated in odd degrees , then @xmath337 by theorem [ th : connectintopol ] , the minimal model of @xmath5 is of the form @xmath338 with @xmath339 in odd degree and @xmath340 . since @xmath341 then @xmath342 now consider the element @xmath343 . for degree reasons , @xmath344 and by the nilpotence condition of a sullivan algebra , @xmath345 can not be a boundary . this shows that @xmath346 has a homotopy retraction and @xmath347 does not as it is not homology injective . the following is a first example of how rational homotopy theory gives better lower bounds for sectional category than the standard cohomological ones . let @xmath5 be the space of example [ examp : notformalspace ] and recall its minimal model @xmath348 with @xmath349 and @xmath350 . by the previous proposition , @xmath351 . however @xmath352 . the following is a surprising result by hess which is a second pillar for the study of rational ls category . [ th : hess ] let @xmath5 be a space and @xmath113 be its minimal model . then the projection @xmath309 admits a cdga homotopy retraction if and only if it admits a homotopy retraction as @xmath113-module . in particular @xmath353 . using this result , flix - halperin - lemaire@xcite proved [ th : felixhalplemai ] let @xmath5 and @xmath4 be spaces . then * @xmath354 , * if @xmath5 is a poincar duality complex ( see section [ ssec : poincareduality ] ) , then @xmath355 in this theorem , the hypothesis of poincar duality is necessary : let @xmath5 be the space modeled by the cdga @xmath356 with @xmath357 , @xmath358 , @xmath359 and @xmath360 . observe that , as vector spaces , @xmath361 and @xmath362,[b],[a]^2,[bx]\rangle$ ] . this shows that @xmath363 . the minimal model of @xmath5 is of the form @xmath113 with @xmath364 , @xmath359 , @xmath365 , @xmath360 and @xmath366 . then @xmath367,[b],[a]^2,[bx+a^2v]\rangle$ ] , and there is no element in @xmath368 representing a non zero class in @xmath369 . this shows that @xmath370 . + we now prove by contradiction that @xmath351 . suppose that @xmath371 admits a cdga homotopy retraction @xmath109 . this implies that @xmath372 but this is absurd since @xmath373 and , for degree reasons , @xmath374)=h(r)([bx+a^2v])=h(r)([b][x])=h(r)([b])0=0.\ ] ] another important consequence of theorem [ th : characterisationsecat ] is the so called mapping theorem ( * ? ? ? * theorem i ) . we include a simple proof based on the proof of ( * ? ? ? 389 ) which we will later on extend to theorem [ th : mappingthtc ] . [ th : mappingthcat ] let @xmath375 be a continuous map . if @xmath376 is injective then @xmath377 . let @xmath113 and @xmath114 be the minimal models of @xmath5 and @xmath4 respectively . then , by theorem [ th : connectintopol ] , the hypothesis tells us that @xmath3 is modeled by a surjective cdga morphism @xmath378 . denote @xmath379 the based path fibration on @xmath5 , so that @xmath380 . we construct a model of the ( homotopy ) pullback @xmath381\ar[d , twoheadrightarrow , " q"']&px\ar[d , twoheadrightarrow , " p"]\\ y\ar[r , " f"']&x \end{tikzcd}\ ] ] ( using the acyclic closure of @xmath113 ) as the homotopy pushout @xmath382\ar[d , hook]\ar[r , " \varphi"]&(\lambda w , d)\ar[d , hook , " j"]\\ { \mathbb{q}}&(\lambda v\otimes \lambda \hat{v},d)\ar[l , " \simeq"]\ar[r]&(\lambda w\otimes \lambda \hat{v } , \overline d ) \end{tikzcd}\ ] ] here , @xmath75 is a cdga model for @xmath207 and @xmath383 verifies @xmath384 and @xmath385 is surjective . now write @xmath386 and consider the projection @xmath387 . observe that @xmath388 is a quasi - isomorphism . by proposition [ prop : sullivanalgs ] , @xmath389 is a cdga quasi - isomorphism . since @xmath390 is trivial , we deduce that @xmath391 is trivial and thus @xmath392 . the result follows since rationalization commutes with limits and @xmath393 . observe that in previous proof , the fact that @xmath394 was crucial in assuring that @xmath389 is a cdga morphism . we start describing the whitehead and ganea characterizations for the sectional category of a continuous map @xmath3 provided that our spaces are cw complexes . we will then translate them to the rational context in the category * cdga*. take @xmath395 a cofibration replacement for @xmath3 . we define the @xmath1-th _ fat wedge _ of @xmath39 as @xmath396 then , @xmath397 is the least @xmath1 for which there exists the dashed map making the following a homotopy commutative diagram & t^m(i ) + x&x^m+1 . here @xmath398 denotes the diagonal map @xmath399 take @xmath196 a fibration replacement for @xmath3 . we define the @xmath1-th _ ganea fibration _ for @xmath3 as @xmath400 where @xmath401 for which summands of the form @xmath402 are dropped , and @xmath403 . see ( * ? ? ? 54 ) for a detailed description of the topology of @xmath404 . then , @xmath405 is the smallest @xmath1 for which @xmath406 admits a section . the way to prove the ganea characterization is to glue the local sections of @xmath200 into a section of @xmath407 by means of a partition of the unity on @xmath408 . then , to prove the whitehead characterization it is sufficient to see that there is a homotopy pullback@xcite @xmath409\arrow[d , " g_p^m"']&t^m(i)\arrow[d , hookrightarrow]\\ x\arrow[r , hookrightarrow , " \delta_{m+1}"']&x^{m+1}. \end{tikzcd}\ ] ] let @xmath410 be a surjective cdga model for a cofibration replacement @xmath411 of a continuous map @xmath3 . then by ( * ? ? ? 1 ) the inclusion of the @xmath1-fat wedge @xmath412 is modeled by the quotient @xmath413 while @xmath414 is modeled by the multiplication @xmath415 . now choose a relative sullivan model for @xmath389 , @xmath416 then diagram [ diag : whitevsgane ] is modeled by the following pushout @xmath417\arrow[d , " \mu " ' ] & ( a^{\otimes m+1}\otimes\lambda w , d)\arrow[d , " \overline{\mu}"]\\ ( a , d)\arrow[r , hookrightarrow , " i_m"']&(a\otimes\lambda w,\overline{d } ) . \end{tikzcd}\ ] ] here , @xmath418 is a relative sullivan model for @xmath406 . we have that @xmath27 is the smallest @xmath1 for which one of the following equivalent conditions hold * * whitehead : * there exists a cdga morphism @xmath419 such that @xmath420 . * * ganea : * there exists a cdga retraction for @xmath418 . one can now impose less restrictive conditions to the existence of morphisms in the characterizations of section [ ssec : firstalgcharac ] to get algebraic lower bounds for sectional category . + let @xmath3 be a continuous map and @xmath421 be a relative sullivan model for @xmath406 ( for example , the one constructed in section [ ssec : firstalgcharac ] ) . if @xmath422 is a retraction of @xmath418 ( that is @xmath423 , for @xmath59 ) then we have the following implications : @xmath109 is a cdga morphism @xmath424 @xmath109 is an @xmath51-module morphism @xmath424 @xmath425 is injective . with previous notation , * the _ module sectional category _ of @xmath3 , @xmath426 , is the least @xmath1 for which @xmath418 admits an @xmath51-module retraction . * the _ homology sectional category _ of @xmath3 , @xmath427 , is the least @xmath1 for which @xmath428 is injective . in previous definition one can replace @xmath418 by any model for @xmath407 and asking the retractions to be homotopy retractions . we can now deduce if @xmath3 is a continuous map , then @xmath429 observe that , if we take @xmath3 as the inclusion of the base point into a based space @xmath5 , we get @xmath430 in contrast with theorem [ th : hess ] , there are maps @xmath3 for which @xmath431 : [ example : stanley ] consider the cdga @xmath432 , then the inclusion @xmath433 admits a retraction as @xmath152-module but not as cdga . in fact , if @xmath109 is a cdga retraction then it must verify @xmath106 , @xmath434 , @xmath435 , but then @xmath109 can not commute with differentials since @xmath436 . let us now define an @xmath51-module retraction as @xmath437 and @xmath438 for @xmath39 odd and @xmath439 for @xmath39 even . since we are particularly interested in topological complexity , by taking @xmath3 as the diagonal map , @xmath246 , we introduce for a space @xmath5 , * the _ module topological complexity _ of @xmath5 is @xmath440 , * the _ homology topological complexity _ is @xmath441 . since @xmath442 , we have @xmath443 most of this section is based on @xcite . let @xmath51 be a cdga . an @xmath51-module is a chain complex @xmath444 together with an action of @xmath45 , @xmath445 verifying @xmath446 . a _ semi - free _ extension of an @xmath51-module @xmath444 is an @xmath51 module of the form @xmath447 where @xmath448 , @xmath449 , @xmath450 and the inclusion @xmath451 is an @xmath51-module morphism . observe that relative sullivan algebras @xmath118 are semi - free extensions of @xmath51 ( * ? ? ? * lemma 14.1 ) . in fact , we have that the category of @xmath51-modules is a closed proper model category ( * ? ? ? * theorem 4.1 ) . + any cdga morphism @xmath452 can be seen as an @xmath51-module morphism by taking in @xmath116 the @xmath51-module structure @xmath453 . a _ semi - free model _ for @xmath103 is an @xmath51-module quasi - isomorphism @xmath454 from a @xmath51-semi - free extension of @xmath51 . a _ semi - free _ model for a continuous map @xmath3 is just a semi - free model for any cdga model of @xmath3 . if @xmath455 is a semi - free model for a map @xmath3 then a semi - free model for the @xmath1-th ganea fibration @xmath406 is @xmath456 here @xmath457 ( in @xmath458 ) is given by @xmath459 for @xmath460 , ... , @xmath461 and @xmath462 with @xmath463 and @xmath464 . if @xmath3 is a map , then @xmath426 is the smallest @xmath1 for which @xmath465 admits an @xmath51-module retraction and @xmath427 is the smallest @xmath1 for which @xmath466 is injective . a finite dimensional commutative graded algebra @xmath467 is said to be a _ poincar duality algebra with formal dimension @xmath9 _ when @xmath468 , @xmath469 and there exists an element @xmath470 such that the map of degree @xmath471 h&(h , ) + a&b^#(ab ) + is an isomorphism , where @xmath472 denotes the dual of @xmath148 . let @xmath473 be a cdga morphism with @xmath474 a poincar duality algebra , then @xmath176 is injective if and only if @xmath103 admits a homotopy retraction as a morphism of @xmath51-modules . if @xmath0 is a map with @xmath475 a poincar duality algebra , then @xmath476 . in particular , if @xmath5 is a poincar duality complex , then @xmath477 . let @xmath3 be a continuous map and @xmath478 a surjective cdga model for @xmath3 . recall the notation from section [ ssec : firstalgcharac ] . then , since @xmath479 , we have a diagram @xmath480\\ ( a^{\otimes m+1},d)\arrow[ur , twoheadrightarrow , " \pi"]\arrow[r , hookrightarrow , " j"]\arrow[d , " \mu " ' ] & ( a^{\otimes m+1}\otimes\lambda w , d)\arrow[u , " \simeq"]\arrow[d , " \overline{\mu}"]\\ ( a , d)\ar[drr , bend right , " \rho_m"']\arrow[r , hookrightarrow , " i"']&(a\otimes\lambda w,\overline{d})\ar[dr , dashed , " \tau"]\\ & & \left(\frac{a}{(\ker\varphi)^{m+1}},\overline{d}\right ) . \end{tikzcd}\ ] ] then taking @xmath481 a relative sullivan model for @xmath320 , the lifting lemma gives a diagram @xmath482\arrow[d , hook , " i"]&(a\otimes\lambda z , d)\arrow[d , twoheadrightarrow , " \simeq"]\\ ( a\otimes\lambda w,\overline{d})\ar[ur , " \overline{\tau}"]\ar[r , " \tau"']&\frac{a}{(\ker\varphi)^{m+1 } } \end{tikzcd}\ ] ] which implies [ prop : upperboudnsecat ] if @xmath320 admits a cdga homotopy retraction then @xmath483 . if @xmath320 admits an @xmath51-module homotopy retraction , then @xmath484 . if @xmath322 is injective , then @xmath485 observe that the opposite implication need not hold ( see example [ example : felixhalpneedsullivan ] ) and this remains true even for topological complexity : let us compute the topological complexity of @xmath486 . the path fibration @xmath487 is modeled by the multiplication morphism @xmath488 with @xmath489 . we remark that @xmath490-[a_2])$ ] , therefore @xmath491 . now , since @xmath492 , we have that @xmath493 and @xmath313 admits a ( homotopy ) retraction . this proves , by proposition [ prop : upperboudnsecat ] , that @xmath494 . another way to show this is through @xmath495 . + now consider another cdga model for @xmath216 , namely the cdga @xmath496 with @xmath497 , with @xmath498 , @xmath499 and @xmath500 . observe that @xmath45 is a free cga but that @xmath51 is _ not _ a sullivan algebra . observe also that @xmath51 is a cdga model for @xmath216 as there is a quasi - isomorphism @xmath501 defined as @xmath502 . + we have then that another cdga model for the path fibration @xmath487 is the multiplication on @xmath51 , @xmath503 . we now see that @xmath504 is not homology injective . in fact , the element @xmath505 can be written as @xmath506 . this means that @xmath507\ne 0 $ ] and that @xmath508)=0 $ ] . if @xmath21 is an ideal of @xmath45 , define the _ homology nilpotency _ of @xmath21 , @xmath509 , to be the smallest @xmath1 such that @xmath510 is contained in an acyclic ideal of @xmath45 , that is , a differential ideal @xmath511 , with @xmath512 . + using previous proposition one can deduce that , if @xmath103 is a surjective cdga model for @xmath3 , then @xmath513 in fact , if @xmath514 with @xmath511 an acyclic ideal of @xmath51 . then , by the five lemma , there is a diagram @xmath515\ar[dr , twoheadrightarrow , " \rho_m"]&\\ \left(\frac{a}{j},\overline{d}\right)&&\left(\frac{a}{(\ker\varphi)^{m+1}},\overline{d}\right)\ar[ll ] \end{tikzcd}\ ] ] which can be used , together with the lifting lemma , to give a homotopy retraction for @xmath320 . + as a consequence , if @xmath3 is a formal map such that @xmath177 is surjective , then @xmath516 . the hypothesis of @xmath177 being surjective is necessary : consider the cdga morphism @xmath517 defined by @xmath518 . let @xmath3 be a continuous map whose model is @xmath103 ( the spatial realization of @xmath103 , for instance ) , then @xmath519 . obviously @xmath520 but @xmath103 does not admit a homotopy retraction . in fact , a relative sullivan model for @xmath103 is given by @xmath521\ar[dr , hook , " j"']&\left(\frac{\lambda ( a , b)}{(a^2+b^2)},0\right)\\ & ( \lambda ( a)\otimes\lambda ( b , v_1,v_2,\ldots ) , d),\ar[u , " \simeq " , " \theta " ' ] \end{tikzcd}\ ] ] with @xmath522 , @xmath523 , @xmath524 , @xmath525 , and @xmath526 for @xmath527 . but , as in example [ example : stanley ] , @xmath75 can not have a cdga retraction . this means that @xmath528 . now suppose that our continuous map @xmath3 admits a homotopy retraction . observe that this is the case of the path fibrations @xmath529 . we now construct a special type of model for such maps which we will call s - models . + we can suppose that @xmath530 is a cofibration and that it admits a strict retraction @xmath531 , so that @xmath532 . now , take a surjective quasi - isomorphism @xmath533 and a relative sullivan model @xmath534&{{\rm a_{pl}}}(y)\\ ( \lambda v , d)\ar[u , twoheadrightarrow,"\simeq"',"\theta"]\ar[r , hook]&(\lambda v\otimes \lambda w , d).\ar[u , " \simeq " , " \xi " ' ] \end{tikzcd}\ ] ] now the lifting lemma @xmath535\ar[d , hookrightarrow]&&(\lambda v , d)\ar[d , twoheadrightarrow , " \simeq " ' , " \theta"]\\ ( \lambda v\otimes\lambda w , d)\ar[urr , dashed , " \varphi'"]\ar[r , " \xi"']&{{\rm a_{pl}}}(y)\ar[r , " { { \rm a_{pl}}}(f)"']&{{\rm a_{pl}}}(x ) , \end{tikzcd}\ ] ] gives a cdga morphism , @xmath536 , which is a model for @xmath3 . now suppose @xmath51 is any cdga model for @xmath5 . then there is a pushout diagram @xmath537)\ar[r , hookrightarrow]&(\lambda v\otimes\lambda w , d)\ar[d , " \simeq"]\ar[ddr , bend left , " \psi\circ\varphi'"]&\\ ( a , d)\ar[drr , bend right , " { { \rm id}}_a"]\ar[r , hookrightarrow]&(a\otimes \lambda w , \overline{d})\ar[dr , dashed , " \varphi"]&\\ & & ( a , d ) , \end{tikzcd}\ ] ] which gives a model @xmath103 for @xmath3 which is a retraction for the inclusion @xmath538 such a model is said to be an _ s - model _ for @xmath3 . + we can now state a generalization of the flix - halperin theorem which lets us compute the topological complexity of rational spaces . [ th : characterisationsecat ] let @xmath3 be a continuous map and @xmath539 an s - model for @xmath3 . then @xmath27 is the smallest @xmath1 for which the cdga projection @xmath540 admits a homotopy retraction . also , * @xmath426 is the smallest @xmath1 for which @xmath320 admits a homotopy retraction as @xmath541-module , and * @xmath427 is the smallest @xmath1 for which @xmath322 is injective . observe that the augmentation is an s - model for the base point inclusion . in this case , previous theorem is just the flix - halperin theorem [ th : felixhalperincharac ] . let @xmath51 be any cdga model for a space @xmath5 . we build an s - model for the path fibration @xmath542 as follows : take @xmath113 a sullivan model for @xmath5 ( not necessarily minimal ) and take a quasi - isomorphism @xmath234 . then the cdga morphism @xmath543 is an s - model for @xmath278 . moreover , as with the multiplication ( @xmath544 ) , one can see that @xmath545 where , as usual , @xmath546 , @xmath547 stands for @xmath548 in the @xmath39-th factor . as a corollary to theorem [ th : characterisationsecat ] , we get [ cor : characterizationtc ] let @xmath5 be a space , @xmath113 a sullivan model for @xmath5 , and @xmath549 a quasi - isomorphism . then @xmath550 is the least @xmath1 such that the projection @xmath551 admits a homotopy retraction . moreover , @xmath552 is the least @xmath1 such that @xmath320 admits a retraction as @xmath553-module and @xmath554 is the least @xmath1 such that @xmath322 is injective . compare ( * ? ? ? * example 6.5 ) and ( * ? ? ? * example 3.5 ) . let @xmath555 where the cells are attached through the iterated whitehead products @xmath556 $ ] and @xmath557 $ ] . by @xcite , the minimal model of @xmath5 is of the form @xmath113 with @xmath558 , @xmath349 , @xmath359 , @xmath350 , @xmath559 and @xmath560,[b],[ax],[bx]\rangle$ ] . therefore , writing @xmath561 the projection @xmath562 is a quasi - isomorphism and an s - model for the diagonal map @xmath563 is @xmath564 now consider the element @xmath565 . a computation yields that @xmath566 . this means that @xmath507\in \ker { { \rm h}}(\rho_2)$ ] is non - zero . therefore @xmath567 . on the other hand , the @xmath568 where @xmath569 is the multiplication . this proves that @xmath570 whilst @xmath571 . finally we have if @xmath5 and @xmath4 are topological spaces , then @xmath572 in this section we will give a slight generalization of the mapping theorem for rational topological complexity of grant - lupton - oprea ( * ? ? ? * theorem 3.2 ) and give a proof using sullivan models . [ th : mappingthtc ] let @xmath573 be continuous maps , @xmath574 , between rational spaces such that @xmath575 are injective . if @xmath576 then @xmath577 since @xmath575 is injective , there is a surjective model for @xmath578 , @xmath579 between minimal sullivan models . the condition @xmath576 implies that for each @xmath580 and @xmath581 , there exist @xmath582 such that @xmath583 , @xmath584 , @xmath585 and @xmath586 . as in the proof of theorem [ th : mappingthcat ] , we model the ( homotopy ) pullback @xmath587 p\ar[r]\ar[d , twoheadrightarrow , " q"']&x^i\ar[d , twoheadrightarrow , " \pi_n"]\\ y_1\times\cdots \times y_n\ar[r , " f_1\times\cdots\times f_n"']&x^n , \end{tikzcd}\ ] ] through the pushout @xmath588\ar[r , twoheadrightarrow , " \varphi_1\otimes\cdots\otimes\varphi_n"]&(\lambda w_1\otimes\cdots\otimes\lambda w _ n , d)\ar[d , hook , " j"]\\ ( \lambda v_1\otimes\cdots\otimes\lambda v_n\otimes\lambda \hat{v},d)\ar[r , twoheadrightarrow]&(\lambda w_1\otimes\cdots\otimes\lambda w_n\otimes\lambda \hat{v},\overline{d } ) , \end{tikzcd}\ ] ] where @xmath39 is the relative model for @xmath278 from section [ sec : modelfordiagonal ] , @xmath75 is a model for @xmath207 , @xmath589 , and @xmath590 . we will prove that @xmath207 is trivial . observe that , by the properties of the @xmath591 s , @xmath592 is surjective . now take @xmath593 . since @xmath594 , the projection @xmath595 is well defined . by construction , @xmath596 is an isomorphism , therefore , by proposition [ prop : sullivanalgs ] , @xmath597 is a quasi isomorphism . since @xmath598 is trivial we have that @xmath207 is trivial , thus @xmath599 but by theorem [ th : felixhalplemai ] , @xmath600 . the author thanks the organizers of _ workshop on tc and related topics _ , held in oberwolfach in march 2016 and acknowledges the belgian interuniversity attraction pole ( iap ) for support within the framework `` dynamics , geometry and statistical physics '' ( dygest ) . j.g . carrasquel - vera , t. kahl , and l. vandembroucq , _ rational approximations of sectional category and poincar duality _ , * 144 * ( 2016 ) , 909915 , http://dx.doi.org/10.1090/proc12722 .

we characterize the maximum controlled invariant ( mci ) set for discrete- as well as continuous - time nonlinear dynamical systems as the solution of an infinite - dimensional linear programming problem . for systems with polynomial dynamics and compact semialgebraic state and control constraints , we describe a hierarchy of finite - dimensional linear matrix inequality ( lmi ) relaxations whose optimal values converge to the volume of the mci set ; dual to these lmi relaxations are sum - of - squares ( sos ) problems providing a converging sequence of outer approximations to the mci set . the approach is simple and readily applicable in the sense that the approximations are the outcome of a single semidefinite program with no additional input apart from the problem description . a number of numerical examples illustrate the approach .

1303.6469

given a controlled dynamical system described by a differential ( continuous - time ) or difference ( discrete - time ) equation , its maximum controlled invariant ( mci ) set is the set of all initial states that can be kept within a given constraint set ad infinitum using admissible control inputs . this set goes by many other names in the literature , e.g. , viability kernel in viability theory @xcite , or @xmath0-invariant set in the linear case @xcite . set invariance is an ubiquitous and essential concept in dynamical systems theory , as far as both analysis and control synthesis is concerned . in particular , by its very definition , the mci set determines fundamental limitations of a given control system with respect to constraint satisfaction . in addition , there is a very tight link between invariant sets and ( control ) lyapunov functions . indeed , sub - level sets of a lyapunov function give rise to invariant sets . conversely , at least in the linear case , any controlled invariant set gives rise to a control lyapunov function , and therefore these sets can be readily used to design stabilizing control laws ; see , e.g. , @xcite for a general treatment and , e.g. , @xcite for applications in model predictive control design . the problem of ( maximum ) controlled invariant set computation for discrete - time systems has been a topic of active research for more than four decades . the central tool in this effort has been the contractive algorithm of @xcite and its expansive counterpart @xcite . for an exhaustive survey and historical remarks see the survey @xcite and the book @xcite . both algorithms , although conceptually applicable to any nonlinear system , have been predominantly applied in a linear setting where they boil down to a sequence of linear programs and polyhedral projections . finite termination of this sequence is a subtle problem and sharp results are available only in the uncontrolled setting where no projections are required @xcite ; for discussion of finite - termination in the controlled case see @xcite . the contractive and expansive algorithms were combined in @xcite to design an algorithm terminating in a finite number of iterations and outputting an @xmath1-accurate inner approximation of the mci set ( with the accuracy measured by the hausdorff distance ) . another line of research , culminating in @xcite , exploits the linearity of the system dynamics in a more systematic way and approximates the maximum ( or minimum ) _ robust _ controlled invariant set by the minkowski sum of a parametrized family of sets . very recently , in continuous time , @xcite developed a parallel algorithm for ellipsoidal approximations of the robust mci set scalable to very high dimensions . computation of low - complexity polyhedral controlled invariant sets was investigated in @xcite and @xcite . in the nonlinear case , a common practice is to exploit the tight connection between invariance and lyapunov functions and seek invariant sets as sub - level sets of a ( control ) lyapunov function ; see , e.g. , @xcite and references therein for recent theoretical developments on the related problem of region of attraction computation and , e.g. , @xcite for practical applications of these techniques . this , however , typically leads to non - convex bilinear optimization problems which are notoriously hard to solve . therefore , one often has to resort to ad - hoc analysis of the specific system at hand , which is typically tractable only in small dimensions ; see @xcite for concrete examples . related in spirit is the localization technique of @xcite for discrete - time _ uncontrolled _ systems , also requiring considerable effort in analysing the system . recently , a general approach using a hierarchy of finite - dimensional linear programs ( lps ) was used in @xcite to design a controller ensuring invariance of a given candidate polyhedral set . in our opinion , although being the current state of the art , this work still suffers from the following drawbacks : 1 ) the sets obtained are convex polytopes ( not general semi - algebraic sets , a fact particularly limiting in the nonlinear case where nonconvex mci sets are common ) ; 2 ) the geometry of the candidate polytopic set must be given a priori ; 3 ) there are no convergence guarantees to the mci set . in this paper , we explicitly address all these points . building upon our previous work @xcite on the computation of the region of attraction ( roa ) for polynomial control systems , in this paper we characterize the maximum controlled invariant ( mci ) set for discrete- as well as continuous - time polynomial systems as the solution to an infinite - dimensional lp problem in the cone of nonnegative measures . the dual of this problem is an infinite - dimensional lp in the space of continuous functions . finite - dimensional _ relaxations _ of the primal lp and finite - dimensional _ approximations _ of the dual lp turn out to be semidefinite programs ( sdps ) also related by duality . the primal relaxations lead to a truncated moment problem while the dual approximations to a sum - of - squares ( sos ) problem . super - level sets of one of the polynomials appearing in the dual sos problem then provide outer approximations to the mci set with guaranteed convergence as the degree of the polynomial tends to infinity . the main mathematical tool we use are the so - called occupation measures which allow us to study the time evolution of the whole ensemble of initial conditions ( described by a measure ) rather than studying trajectories associated to each initial condition separately . the use of measures to study dynamical systems has a very long tradition : see @xcite for probably the first systematic treatment ; for purely discrete - time treatment see ( * ? ? ? * chapter 6 ) . to the best of the authors knowledge our paper is the first one to use occupation measures for mci set ( approximate ) _ computation_. the mci set was previously _ characterized _ using occupation measures in @xcite , but there the characterization is rather indirect and not straightforwardly amenable to computation . apart from the authors work @xcite , the related problem of region of attraction computation was tackled using measures in @xcite . there , however , a very different approach was taken , not using _ occupation _ measures but rather analyzing convergence via discretization of the state - space and propagating the initial distribution by means of a discretized transfer operator . here , instead , we employ the ( discounted ) occupation measure which captures the behaviour of the trajectories emanating from the initial distribution over the infinite time horizon . as a result , our approach requires no discretization and , contrary to @xcite , provides true guarantees ( not in an `` almost - everywhere '' or `` coarse '' sense ) and , more importantly , is applicable in a controlled setting . closely related to the occupation measures used here is the rantzer s density @xcite which was used in @xcite to assess the stability of attractor sets of uncontrolled nonlinear systems . the approach , however , does not immediately yield approximations of the mci set ( or the region of attraction ) and applies to uncontrolled systems only . similar in spirit to our approach , from the dual viewpoint of optimization over functions , are the hamilton - jacobi approaches ( e.g. , @xcite ) . however , contrary to these methods , our approach does not require state - space discretization and comes with convergence guarantees . the contribution of our paper with respect to previous work on the topic can be summarized as follows : * we deal with fully general continuous - time and discrete - time polynomial dynamics under semi - algebraic state and control constraints ; * our approximated mci set is described by ( the intersections of ) polynomial super - level sets , including more restrictive classes ( e.g. polytopes , ellipsoids , etc . ) ; * we provide a convex infinite - dimensional lp characterization of the mci set ; * we describe a hierarchy of convex finite - dimensional sdps to solve the lp with convergence guarantees ; * our approach is simple and readily applicable in the sense that the approximations are the result of a single sdp with _ no additional data _ required apart from the problem description . the contribution with respect to our previous work @xcite can be summarised as follows : * in @xcite we compute the roa , which is a related although different object : it is the set of all of initial conditions that can be steered to a given target set while satisfying state and control constraints . in particular , the mci set differs from the roa in the sense that we do not try to hit any target set at a given time but rather try to keep the state within a given set forever . therefore we had to adapt our technique to deal explicitly with invariance ; * in @xcite we dealt with continuous - time systems only , whereas we can cope , with minor modifications , with discrete - time systems as well ; we choose to describe both the continuous - time and discrete - time setups in parallel precisely to underline these common features ; * in @xcite we considered only a finite time - horizon , whereas here we show how to cope , with the help of discounting , with an infinite horizon . this brought additional technical issues not encountered in finite time . what can be considered a drawback of our approach is the fact that the approximations to the mci set we obtain are from the outside and therefore not invariant . however , accurate outer approximations provide important information as to the performance limitations of the control system and are of practical interest , e.g. , in collision avoidance . therefore we believe that our work bears both theoretical and practical value , and naturally complements existing inner - approximation techniques . the paper is organised as follows . the problem to be solved is described in section [ sec : probstate ] . occupation measures are introduced in section [ sec : om ] . the infinite - dimensional primal and dual lps are described in sections [ sec : primallp ] and [ sec : duallp ] , respectively . the finite - dimensional relaxations with convergence results are presented in section [ sec : lmirelax ] . numerical examples are in section [ sec : numex ] . a reader interested only in the semialgebraic outer approximations of the mci set can consult directly the infinite - dimensional dual lps ( [ vlpd ] ) and ( [ vlpc ] ) and their finite - dimensional approximations ( [ dlmid ] ) and ( [ dlmic ] ) in discrete and continuous time , respectively . measures are understood as signed borel measures on a euclidean space , i.e. , as countably additive maps from the borel sets to the real numbers . from now on all subsets of a euclidean space we refer to are automatically understood as borel . the vector space of all signed borel measures with its support contained in a set @xmath2 is denoted by @xmath3 . the support ( i.e. , the smallest closed set whose complement has a zero measure ) of a measure @xmath4 is denoted by @xmath5 . the space of continuous functions on @xmath2 is denoted by @xmath6 and likewise the space of once continuously differentiable functions is @xmath7 . the indicator function of a set @xmath2 ( i.e. , a function equal to one on @xmath2 and zero otherwise ) is denoted by @xmath8 . the symbol @xmath9 denotes the @xmath10-dimensional lebesgue measure ( i.e. , the standard @xmath10-dimensional volume ) . the integral of a function @xmath11 with respect to a measure @xmath4 over a set @xmath2 is denoted by @xmath12 . sometimes for conciseness we use the shorter notation @xmath13 omitting the integration variable and also the set over which we integrate if they are obvious from the context . the ring of polynomials in ( possibly vector ) variables @xmath14, ,@xmath15 is denoted by @xmath16 $ ] . the approach is developed in parallel for discrete and continuous time . consider the discrete - time control system @xmath17 with a given polynomial vector field @xmath18 with entries @xmath19 $ ] , @xmath20 , and given compact basic semialgebraic state and input constraints @xmath21 with @xmath22 $ ] , @xmath23 $ ] . the maximum controlled invariant ( mci ) set is defined as @xmath24 a control sequence @xmath25 is called _ admissible _ if @xmath26 for all @xmath27 . in words , the mci set is the set of all initial states which can be kept inside the constraint set @xmath2 ad infinitum using admissible control inputs . consider the relaxed continuous - time control system @xmath28 where @xmath29 denotes the convex hull , @xmath18 is a polynomial vector field with entries @xmath19 $ ] , @xmath30 and compact basic semialgebraic state and input constraint sets are defined by @xmath31 with @xmath22 $ ] , @xmath23 $ ] . the meaning of the convex differential inclusion ( [ sysc ] ) is as follows : for all time @xmath32 , the state velocity @xmath33 is constrained to the convex hull of the set @xmath34 . the connection of this convexified ( or relaxed ) control problem ( [ sysc ] ) and the classical control problem @xmath35 is the filippov - wa@xmath36ewski theorem @xcite , which shows that the trajectories of @xmath37 are dense ( in the supremum norm ) in the set of trajectories of the convexified inclusion is closed for every @xmath32 since @xmath18 is continuous and @xmath38 compact ; therefore there is no need to take closure of the convex hull in order to apply the filippov - wa@xmath36ewski theorem . ] ( [ sysc ] ) . therefore , from a practical point of view , there is little difference between the two formulations for the purposes of mci set computation ; see section 3.2 and appendices b and c of @xcite for a detailed discussion on this subtle issue . the simplest assumption under which the mci sets for both systems coincide is @xmath39 being convex for all @xmath40 , which is in particular true for input - affine systems of the form @xmath41 with @xmath38 convex . the maximum controlled invariant ( mci ) set is defined as @xmath42 where @xmath43 is required to be absolutely continuous and stands for `` almost everywhere '' with respect to the lebesgue measure on @xmath44 . in words , the mci set is the set of all initial states for which there exists a trajectory of the convexified inclusion ( [ sysc ] ) which remains in @xmath2 ad infinitum . in this section we introduce the concept of occupation measures which is the centrepiece of our approach . given a discount factor @xmath45 , an initial condition @xmath46 and an admissible control sequence @xmath47 such that the associated state sequence @xmath48 remains in @xmath2 for all time , we define the _ discounted occupation measure _ @xmath49 as @xmath50 for all sets @xmath51 and @xmath52 . in words , the discounted occupation measure measures the ( discounted ) number of visits of the state - control pair trajectory @xmath53 to subsets of @xmath54 . the discounting in the definition of the occupation measure ensures that @xmath55 is always finite ; in fact we have @xmath56 . now suppose that the initial condition is not a single point but an _ _ initial measure _ _ can be thought of as the probability distribution of the initial state , although we do not require the mass of @xmath57 to be normalized to one . ] @xmath58 and an admissible control sequence is associated to each initial condition from the support of @xmath57 in such a way that the corresponding state sequence remains in @xmath2 . then we define the _ average discounted occupation measure _ @xmath59 as @xmath60 the average discounted occupation measure measures the discounted average number of visits in subsets of @xmath54 of trajectories starting from the initial distribution @xmath57 . now we derive an equation linking the measures @xmath57 and @xmath4 . this equation will play a key role in subsequent development and in a sense replaces the dynamics equation ( [ sysd ] ) . to derive this equation fix an initial condition @xmath61 and a control sequence @xmath47 such that the associated state sequence @xmath48 stays in @xmath2 . then for any @xmath62 we have @xmath63 integrating w.r.t . @xmath57 we arrive at the sought equation @xmath64 note that this is an infinite - dimensional linear equation in variables @xmath65 . the following crucial lemma establishes the connection between the support of any initial measure @xmath57 solving ( [ eq : liouvdisc ] ) and the mci set @xmath66 . [ lem : corrdisc ] for any pair of measures @xmath65 satisfying equation ( [ eq : liouvdisc ] ) with @xmath67 and @xmath68 we have @xmath69 . a detailed proof is in appendix a. given an initial condition @xmath46 and a trajectory @xmath70 of the inclusion ( [ sysc ] ) that remains in @xmath2 for all @xmath71 , there exists an admissible time - varying measure - valued relaxed control @xmath72 , @xmath73 , such that @xmath74 almost everywhere with respect to the lebesgue measure on @xmath44 . this follows from the definition of the convex hull ( in fact , for each @xmath32 , @xmath75 can be taken to be a convex combination of finitely many dirac measures ) . then , given a discount factor @xmath76 , we define the _ discounted occupation measure _ @xmath77 as @xmath78 for all sets @xmath51 and @xmath52 . in words , the discounted occupation measure measures the ( discounted ) time spent by the state - control pair trajectory @xmath53 in subsets of @xmath54 . the discounting in the definition of the occupation measure ensures that @xmath55 is always finite ; in fact we have @xmath79 . now suppose that the initial condition is not a single point but an _ _ initial measure _ _ can be thought of as the probability distribution of the initial state , although we do not require the mass of @xmath57 to be normalized to one . ] @xmath58 and a state trajectory that remains in @xmath2 along with an admissible relaxed control is associated to each initial condition from the support of @xmath57 . then we define the _ average discounted occupation measure _ @xmath59 as @xmath80 now we derive an equation linking the measures @xmath57 and @xmath4 . this equation will play a key role in subsequent development and in a sense replaces the dynamics equation ( [ sysc ] ) . to derive the equation , fix an initial condition @xmath61 , a trajectory @xmath81 that remains in @xmath2 with an associated admissible relaxed control @xmath82 . then for any @xmath83 integration by parts yields @xmath84 where the boundary term at infinity vanishes due to discounting and the fact that @xmath2 is bounded . integrating with respect to @xmath57 then gives the sought equation @xmath85 note that this is an infinite - dimensional linear equation in variables @xmath65 . the following crucial lemma establishes the connection between the support of any initial measure satisfying ( [ eq : liouvcont ] ) and the mci set @xmath66 . [ lem : corrcont ] for any pair of measures @xmath65 satisfying equation ( [ eq : liouvcont ] ) with @xmath67 and @xmath68 we have @xmath86 . a detailed proof is in appendix b. in this section we show how the mci set computation problem can be cast as an infinite - dimensional lp problem in the cone of nonnegative measures . as in @xcite , the basic idea is to maximize the mass of the initial measure @xmath57 subject to the constraint that it be dominated by the lebesgue measure , that is , @xmath87 . system dynamics is captured by the equations ( [ eq : liouvdisc ] ) and ( [ eq : liouvcont ] ) for discrete and continuous times , respectively ; state and input constraints are expressed through constraints on the supports of the initial and occupation measure . the constraint that @xmath87 can be equivalently rewritten as @xmath88 for some nonnegative slack measure @xmath89 . this constraint is in turn equivalent to @xmath90 for all @xmath91 . these considerations lead to the following primal lps . the primal lp in discrete time reads @xmath92 where the supremum is over the vector of measures @xmath93 . this is an infinite - dimensional lp in the cone of nonnegative borel measures . the following lemma , which is our main theoretical result , relates an optimal solution of this lp to the mci set @xmath66 . the optimal value of lp problem ( [ rrlpd ] ) is equal to the volume of the mci set @xmath66 , that is , @xmath94 . moreover , the supremum is attained by the restriction of the lebesgue measure to the mci set @xmath66 . the proof follows from lemma [ lem : corrdisc ] by the same arguments as theorem 1 in @xcite . by definition of the mci set @xmath66 , for any initial condition @xmath95 there exists an admissible control sequence such that the associated state sequence remains in @xmath2 . therefore for any initial measure @xmath87 with @xmath69 there exist a discounted occupation measure @xmath4 with @xmath96 and a slack measure @xmath97 with @xmath98 such that the constraints of problem ( [ rrlpd ] ) are satisfied . one such measure @xmath57 is the restriction of the lebesgue measure to @xmath66 , and therefore @xmath99 . the fact @xmath100 follows from lemma [ lem : corrdisc ] . the primal lp in continuous time reads @xmath101 where the infimum is over the vector of measures @xmath93 . this is an infinite - dimensional lp in the cone of nonnegative borel measures . the following lemma , which is our main theoretical result , relates an optimal solution of this lp to the mci set @xmath66 . the optimal value of lp problem ( [ rrlpc ] ) is equal to the volume of the mci set @xmath66 , that is , @xmath94 . moreover , the supremum is attained by the restriction of the lebesgue measure to the mci set @xmath66 . the fact that @xmath57 equal to the restriction of the lebesgue measure to @xmath66 is feasible in ( [ rrlpc ] ) ( and therefore @xmath99 ) follows by the same arguments as in discrete time . the fact that @xmath100 follows from lemma [ lem : corrcont ] . in this section we derive lps dual to the primal lps ( [ rrlpd ] ) and ( [ rrlpc ] ) . since the primal lps are in the space of measures , the dual lps will be on the space of continuous functions . super - level sets of feasible solutions to these lps then provide outer approximations to the mci sets , both in discrete and in continuous time . both duals can be derived by standard infinite - dimensional lp duality theory ; see @xcite for a derivation in a similar setting or @xcite for a general theory of infinite - dimensional linear programming . the dual lp in discrete time reads @xmath102 where the infimum is over the pair of functions @xmath103 . the following key observation shows that the unit super - level set of any function @xmath104 feasible in ( [ vlpd ] ) provides an outer - approximation to @xmath66 . [ lem : v0d ] any feasible solution to problem ( [ vlpd ] ) satisfies @xmath105 and @xmath106 on @xmath66 . given any @xmath95 there exists a sequence @xmath25 , @xmath26 , such that @xmath107 for all @xmath32 . the first constraint of problem ( [ vlpd ] ) is equivalent to @xmath108 , @xmath109 . by iterating this inequality we get @xmath110 since @xmath107 and @xmath2 is bounded . therefore @xmath111 and @xmath112 for all @xmath95 . the following theorem is instrumental in proving the convergence results of section [ sec : lmirelax ] . [ nogapinfd ] there is no duality gap between primal lp problems ( [ rrlpd ] ) on measures and dual lp problem ( [ vlpd ] ) on functions in the sense that @xmath113 . follows by the same arguments as theorem 2 in @xcite using standard infinite - dimensional lp duality theory ( see , e.g. , @xcite ) and the fact that the feasible set of the primal lp is nonempty and bounded in the metric inducing the weak- * topology on @xmath114 . to see non - emptiness , notice that the vector of measures @xmath115 is trivially feasible . to see the boundedness , it suffices to evaluate the equality constraints of ( [ rrlpd ] ) for @xmath116 . this gives @xmath117 and @xmath118 , which , since @xmath119 and all measures are nonnegative , proves the assertion . the dual lp in continuous time reads @xmath120 where the infimum is over the pair of functions @xmath121 . the following key observation shows that the unit super - level set of any function @xmath104 feasible in ( [ vlpc ] ) provides an outer - approximation to @xmath66 . [ lem : v0c ] any feasible solution to problem ( [ vlpc ] ) satisfies @xmath122 and @xmath106 on @xmath66 . given any @xmath95 there exists an admissible relaxed control function @xmath123 , @xmath124 , such that @xmath125 for all @xmath32 . for that @xmath126 we have @xmath127 . then by gronwall s inequality @xmath128 , and consequently @xmath129 since @xmath125 and @xmath2 is bounded . therefore @xmath130 and @xmath112 for all @xmath95 . the following theorem is instrumental in proving the convergence results of section [ sec : lmirelax ] . [ nogapinfc ] there is no duality gap between primal lp problems ( [ rrlpc ] ) on measures and dual lp problem ( [ vlpc ] ) on functions in the sense that @xmath113 . follows by the same arguments as theorem 2 in @xcite using standard infinite - dimensional lp duality theory ( see , e.g. , @xcite ) and the fact that the feasible set of the primal lp is nonempty and bounded in the metric inducing the weak- * topology on @xmath114 . to see non - emptiness , notice that the vector of measures @xmath115 is trivially feasible . to see the boundedness , it suffices to evaluate the equality constraints of ( [ rrlpc ] ) for @xmath116 . this gives @xmath117 and @xmath131 , which , since @xmath76 and all measures are nonnegative , proves the assertion . in this section we present finite dimensional relaxations of the infinite - dimensional lps . both in continuous and discrete time , the relaxations of the primal lps lead to a truncated moment problem which translates to a semidefinite program ( sdp ) that can be solved by freely available software , e.g. , sedumi @xcite or sdpa @xcite . dual to the primal sdp relaxation is a sum - of - squares ( sos ) problem that again translates to an sdp problem . the following discussion closely follows the one in @xcite . we only highlight the main ideas behind the derivation of the finite - dimensional relaxations . the reader is referred to ( * ? ? ? * section 5 ) or to the comprehensive reference @xcite for details . first , since the supports of all measures feasible in ( [ rrlpd ] ) and ( [ rrlpc ] ) are compact , these measures are uniquely determined by their moments , i.e. , by integrals of all monomials ( which is a sequence of real numbers when indexed in , e.g. , the canonical monomial basis ) . therefore , it suffices to restrict the test functions @xmath132 and @xmath133 in ( [ rrlpd ] ) and ( [ rrlpc ] ) to all monomials , reducing the linear equality constraints on measures @xmath57 , @xmath4 and @xmath97 of ( [ rrlpd ] ) and ( [ rrlpc ] ) to linear equality constraints on their moments . next , by the putinar positivstellensatz ( see @xcite ) , the constraint that the support of a measure is included in a given compact basic semialgebraic set is equivalent to the feasibility of an infinite sequence of lmis involving the so - called moment and localizing matrices , which are linear in the coefficients of the moment sequence . by truncating the moment sequence and taking only the moments corresponding to monomials of total degree less than or equal to @xmath134 , where @xmath135 is the relaxation order , we obtain a necessary condition for this truncated moment sequence to be the first part of a moment sequence corresponding to a measure with the desired support . in what follows , @xmath136 $ ] denotes the vector space of real multivariate polynomials of total degree less than or equal to @xmath137 . furthermore , throughout the rest of this section we make the following standard standing assumption : [ compact ] one of the polynomials modeling the sets @xmath2 resp . @xmath38 is equal to @xmath138 resp . @xmath139 with @xmath140 , @xmath141 sufficiently large constants . this assumption is completely without loss of generality since redundant ball constraints can be always added to the description of the _ compact _ sets @xmath2 and @xmath38 . the primal relaxation of order @xmath137 in discrete time reads @xmath142 where the notation @xmath143 stands for positive semidefinite and the minimum is over moment sequences @xmath144 truncated to degree @xmath134 corresponding to measures @xmath4 , @xmath57 and @xmath97 in ( [ rrlpd ] ) . the linear equality constraint captures the two linear equality constraints of ( [ rrlpd ] ) with @xmath145 $ ] and @xmath146 $ ] being monomials of total degree less than or equal to @xmath134 . the matrices @xmath147 are the moment and localizing matrices , following the notations of @xcite or @xcite . in problem ( [ plmid ] ) , a linear objective is minimized subject to linear equality constraints and lmi constraints ; therefore problem ( [ plmid ] ) is a semidefinite program ( sdp ) . the dual relaxation of order @xmath137 in discrete time reads @xmath148 where @xmath149 is the vector of lebesgue moments over @xmath2 indexed in the same basis in which the polynomial @xmath132 with coefficients @xmath104 is expressed . the minimum is over polynomials @xmath150 $ ] and @xmath151 $ ] , and polynomial sum - of - squares @xmath152 , @xmath153 , @xmath154 , @xmath155 and @xmath156 , @xmath157 , of appropriate degrees . in problem ( [ dlmid ] ) , a linear objective function is minimized subject to sum - of - squares ( sos ) constraints ; therefore problem ( [ dlmid ] ) is an sos problem which can be readily cast as an sdp ( see , e.g. , @xcite ) . the primal relaxation of order @xmath137 in continuous time reads @xmath158 where the notation @xmath143 stands for positive semidefinite and the minimum is over moment sequences @xmath144 truncated to degree @xmath134 corresponding to measures @xmath4 , @xmath57 and @xmath97 in ( [ rrlpc ] ) . the linear equality constraint captures the two linear equality constraints of ( [ rrlpc ] ) with @xmath145 $ ] and @xmath146 $ ] being monomials of total degree less than or equal to @xmath134 . the matrices @xmath147 are the moment and localizing matrices , following the notations of @xcite or @xcite . in problem ( [ plmic ] ) , a linear objective is minimized subject to linear equality constraints and lmi constraints ; therefore problem ( [ plmic ] ) is a semidefinite program ( sdp ) . the dual relaxation of order @xmath137 in continuous time reads @xmath159 where @xmath149 is the vector of lebesgue moments over @xmath2 indexed in the same basis in which the polynomial @xmath132 with coefficients @xmath104 is expressed . the minimum is over polynomials @xmath150 $ ] and @xmath151 $ ] , and polynomial sum - of - squares @xmath152 , @xmath153 , @xmath154 , @xmath155 and @xmath156 , @xmath157 , of appropriate degrees . in problem ( [ dlmic ] ) , a linear objective function is minimized subject to sum - of - squares ( sos ) constraints ; therefore problem ( [ dlmic ] ) is an sos problem which can be readily cast as an sdp ( see , e.g. , @xcite ) . in this section we state several convergence results for the finite dimensional relaxations resp . approximations ( [ plmid ] ) , ( [ plmic ] ) resp . ( [ dlmid ] ) , ( [ dlmic ] ) . let @xmath160 and @xmath161 denote an optimal solution to the @xmath162 dual sdp approximation ( [ dlmid ] ) or ( [ dlmic ] ) , and define @xmath163 then , in view of lemmata [ lem : v0d ] and [ lem : v0c ] , we know that @xmath160 over - approximates the indicator function of the mci set @xmath66 on @xmath2 , i.e. , @xmath164 on @xmath2 , and that the sets @xmath165 approximate from the outside the mci set @xmath66 , i.e. , @xmath166 . in the sequel we prove the following : * the optimal values of the finite - dimensional primal and dual problems @xmath167 and @xmath168 coincide and converge to the optimal values of the infinite dimensional primal and dual lps @xmath169 and @xmath170 which also coincide ( in view of theorems [ nogapinfd ] and [ nogapinfc ] ) and are equal to the volume of the mci set . * the sequence of functions @xmath160 converges on @xmath2 from above to the indicator function of the mci set in @xmath171 norm . in addition , the running minimum @xmath172 converges on @xmath2 from above to the indicator function of the mci set set in @xmath171 norm and almost uniformly . * the sequence of sets @xmath165 converges to the mci set @xmath66 in the sense that the volume discrepancy tends to zero , i.e. , @xmath173 . the proofs of the results follow very similar reasoning as analogous results on region of attraction approximations in ( * ? ? ? * section 6 ) . [ lem : nogaprelax ] there is no duality gap between primal lmi problems ( [ plmid ] and [ plmic ] ) and dual lmi problems ( [ dlmid ] and [ dlmic ] ) , i.e. @xmath174 . the argument closely follows the one in ( * ? ? ? * theorem 4 ) and therefore we only outline the key points of the proof . to prove the absence of duality gap , it is sufficient to show that the feasible sets of the primal sdps ( [ plmid ] ) and ( [ plmic ] ) are non - empty and compact . the result then follows by standard sdp duality theory ( see ( * ? ? ? * theorem 4 ) for a detailed argument ) . the non - emptiness follows trivially since the vector of measures @xmath175 is feasible in the primal infinite - dimensional lps ( [ rrlpd ] ) and ( [ rrlpc ] ) and therefore the truncated moment sequences corresponding to these measures are feasible in the primal sdp relaxations ( [ plmid ] ) and ( [ plmic ] ) . to see the compactness observe that the first components ( i.e. , masses ) of the truncated moment vectors @xmath176 , @xmath177 and @xmath178 are bounded . this follows by evaluating the equality constraints of ( [ rrlpd ] ) and ( [ rrlpc ] ) for @xmath179 . indeed , in discrete - time we get @xmath180 and in continuous - time we get @xmath181 ; in addition , in both cases we have @xmath182 and therefore the first components are indeed bounded ( since they are trivially bounded from below , in fact nonnegative , due to the constraints on moment matrices ) . boundedness of the even components of each truncated moment vector then follows from the structure of the localizing matrices corresponding to the functions from assumption [ compact ] . boundedness of the entire truncated moment vectors then follows since the even moments appear on the diagonal of the positive semidefinite moment matrices . the following result shows the convergence of the optimal values of the relaxations to the optimal values of the infinite - dimensional lps . [ thm : pdconv ] the sequence of infima of lmi problems ( [ dlmid ] ) and ( [ dlmic ] ) converges monotonically from above to the supremum of the lp problems ( [ vlpd ] ) and ( [ vlpc ] ) , i.e. , @xmath183 and @xmath184 . similarly , the sequence of maxima of lmi problems ( [ plmid ] ) and ( [ plmic ] ) converges monotonically from above to the maximum of the lp problems ( [ rrlpd ] ) and ( [ rrlpc ] ) , i.e. , @xmath185 and @xmath186 . the monotonicity of the optimal values of the relaxations @xmath167 resp . approximations @xmath168 is evident form the structure of the feasible sets of the corresponding sdps . the convergence of the primal relaxations @xmath187 to @xmath169 follows from the compactness of the feasible sets of the primal sdps ( [ plmid ] ) and ( [ plmic ] ) ( shown in the proof of lemma [ lem : nogaprelax ] ) by standard arguments on the convergence of lasserre s lmi hierarchy ( see , e.g. , @xcite ) . the converge of the optimal value of the dual approximations @xmath168 to @xmath188 then follows from lemma [ lem : nogaprelax ] . the next theorem shows functional convergence from above to the indicator function of the mci set . [ thm : dualconvfun ] let @xmath189 $ ] denote the @xmath104-component of a solution to the dual lmi problems ( [ dlmid ] ) or ( [ dlmic ] ) and let @xmath190 . then @xmath160 converges from above to @xmath191 in @xmath192 norm and @xmath193 converges from above to @xmath191 in @xmath192 norm and almost uniformly . the convergence in @xmath171 norm follows immediately from theorem [ thm : pdconv ] and from the fact that @xmath164 by lemmata [ lem : v0d ] and [ lem : v0c ] . the convergence of the running minima follows from the fact that there exists a subsequence of @xmath194 which converges almost uniformly ( by , e.g. , ( * ? ? ? * theorems 2.5.2 and 2.5.3 ) ) . our last theorem shows a set - wise convergence of the outer - approximations to the mci set . let @xmath195\times { \mathbb r}_{2k}[x]$ ] denote an optimal solution to the dual lmi problem ( [ dlmid ] ) or ( [ dlmic ] ) and let @xmath196 . then @xmath197 , @xmath198 from lemmata [ lem : v0d ] or [ lem : v0c ] we have @xmath199 and @xmath164 ; therefore , since @xmath200 and @xmath201 on @xmath2 , we have @xmath202 and @xmath203 . from theorem [ thm : dualconvfun ] , we have @xmath204 in @xmath192 norm on @xmath2 . consequently , @xmath205 but since @xmath197 for all @xmath137 , we must have @xmath206 and the theorem follows . in this section we present numerical examples that illustrate our results . the primal sdp relaxations were modeled using gloptipoly 3 @xcite and the dual sos problems using yalmip @xcite . the resulting sdp problems were solved using sedumi @xcite ( which , in the case of primal relaxations , also returns the dual solution providing the outer approximations ) . for numerical computation ( especially for higher relaxation orders ) , the problem data should be scaled such that the constraint sets are ( within ) unit boxes or unit balls ; for ease of reproduction , most of the numerical problems shown are already scaled . on our problem class we observed only marginal sensitivity to the values of the discrete- and continuous - time discount factors @xmath207 and @xmath208 and report results with @xmath209 and @xmath210 for all examples presented . for a discussion on the scalability of our approach and the performance of alternative sdp solvers see the conclusion and the acrobot - on - a - cart example below . consider the discrete - time double integrator : @xmath211 with the state constraint set @xmath212 ^ 2 $ ] and input constraint set @xmath213 $ ] . the resulting of mci set outer approximations of degree 8 and 12 are shown in figure [ fig:1d ] ; the approximation is fairly tight for modest degrees . the true mci set was computed using the standard algorithm based on polyhedral projections @xcite . consider the cathala system borrowed from @xcite : @xmath214 the chaotic attractor of this system is contained in the set @xmath215 ^ 2 $ ] . mci set outer approximations are shown in figure [ fig:2d ] ; again , the approximations are relatively tight for small relaxation orders . the true mci set was ( approximately ) computed by gridding . ( 140,180 ) ( 20,20 ) ( 250,20 ) ( 120,10)@xmath40 ( 350,10)@xmath40 ( 165,150)@xmath216 ( 395,150)@xmath217 ( 140,180 ) ( 20,20 ) ( 250,20 ) ( 120,10)@xmath40 ( 350,10)@xmath40 ( 165,150)@xmath218 ( 395,150)@xmath219 consider over @xmath220 , or equivalently over @xmath221 with @xmath222 , the quadratic recurrence @xmath223 with @xmath224 a given complex number and @xmath225 the imaginary unit . the filled julia set is the set of all initial conditions of the above recurrence for which the trajectories remain bounded . the shape of the julia set depends strongly on the parameter @xmath226 . if @xmath226 lies inside the mandelbrot set , then the julia set is connected ; otherwise the set is disconnected . in both cases the boundary of the set has a very complicated ( in fact fractal ) structure . here we shall compute outer approximations of the filled julia set intersected with the unit ball . to this end we set @xmath227 . figure [ fig : j1 ] shows outer approximations of degree 12 for parameter values @xmath228 ( inside the mandelbrot set ) and @xmath229 ( outside the mandelbrot set ) . the `` true '' filled julia set was ( approximately ) obtained by randomly sampling initial conditions within the unit ball and iterating the recurrence for one hundred steps . taking higher degree of the approximating polynomials does not give significant improvements due to our choice of the monomial basis to represent polynomials . an alternative basis ( e.g. chebyshev polynomials see the related discussions in @xcite and @xcite ) would allow us to improve further the outer estimates and better capture the intricate structure of the filled julia set s boundary . ( 140,206 ) ( 5,20 ) ( 235,20 ) ( 85,10)@xmath228 ( 315,10)@xmath229 consider the modified controlled hnon map @xmath230 adapted from @xcite with @xmath212 ^ 3 $ ] and @xmath231 $ ] . we investigate two cases : uncontrolled ( i.e. , @xmath232 ) and controlled with @xmath233 . figure [ fig : henon ] shows outer approximations to the mci set of degree eight for both settings and the `` true '' mci set in the uncontrolled setting ( approximately ) obtained by random sampling of initial conditions inside the constraint set @xmath2 . the outer approximations suggest that , as expected , allowing for control leads to a larger mci set . ( 100,265 ) ( 135,20 ) consider the continuous - time double integrator @xmath234 with state constraint set @xmath212 ^ 2 $ ] and input constraint set @xmath235 $ ] . the resulting mci set outer approximations for degrees 8 and 12 are in figure [ fig:1c ] . the approximations are fairly tight even for relatively low relaxation orders . the true mci set was ( approximately ) computed as in section [ sec : doubleintegdiscrete ] by methods of @xcite after dense time discretization . ( 140,180 ) ( 20,20 ) ( 250,20 ) ( 120,10)@xmath40 ( 351,10)@xmath40 ( 175,160)@xmath216 ( 405,160)@xmath236 as our second example we take the spider - web system from @xcite given by equations @xmath237 with the constraint set @xmath212 ^ 2 $ ] . here we exploit the fact that the system dynamics are captured by constraints on @xmath11 only whereas @xmath104 is merely over approximating @xmath238 , and the fact that outer approximations to the mci set are given not only by @xmath239 but also by @xmath240 . therefore , if low - complexity outer approximations are desired , it is reasonable to choose different degrees of @xmath11 and @xmath104 in ( [ dlmic ] ) high for @xmath11 and lower for @xmath104 and use the set @xmath241 as the outer approximation . that way , we expect to obtain relatively tight low - order approximations . this is confirmed by numerical results shown in figure [ fig:2c ] . the degree of @xmath11 is equal to 16 for both figures , whereas @xmath242 for the left figure and @xmath243 for the right figure . we observe no significant loss in tightness by choosing a smaller degree of @xmath104 . the true mci set was ( approximately ) computed by gridding . ( 140,180 ) ( 20,20 ) ( 250,20 ) ( 120,10)@xmath40 ( 351,10)@xmath40 as our last example we consider the acrobot on a cart system adapted from @xcite , which is essentially a double pendulum on a cart where the inputs are the force acting on the cart and the torque in the middle joint of the double pendulum . the system is sketched in figure [ fig : acroboncart_sketch ] . it is a sixth order system with with two control inputs ; the dynamic equation is given by @xmath244 where @xmath245 and @xmath246 the states @xmath14 , @xmath247 , @xmath248 represent , respectively , the position of the cart ( in meters ) , the angle of the lower rod and the angle of the upper rod of the double pendulum ( both in radians ) ; the states @xmath249 , @xmath250 and @xmath251 are then the corresponding velocities in meters per second for the cart and radians per second for the pendulum rods . the constants are given by @xmath252 , @xmath253 , @xmath254 , @xmath255 , @xmath256 , @xmath257 , @xmath258 , @xmath259 , @xmath260 , @xmath261 , @xmath262 . we are interested in computing the maximum controlled invariant subset of the state constraint set @xmath263 \times [ -\pi/3 , \pi/3 ] \times [ -\pi/3 , \pi/3 ] \times [ -0.5 , 0.5 ] \times [ -5 , 5 ] \times [ -5 , 5].\ ] ] we investigate two cases . first , we consider the situation where only the middle joint is actuated and there is no force on the cart ; therefore we impose the constraint @xmath264 $ ] . second , we consider the situation where we can also exert a force on the cart ; in this case we impose @xmath265 \times [ -1,1]$ ] . naturally , the mci set for the second case is larger ( or at least the same ) as for the first case . this is confirmed ) or ( [ dlmic ] ) . ] by outer approximations of degree four whose section for @xmath266 , @xmath267 , @xmath268 is shown in figure [ fig : acroboncart ] . in order to compute the outer approximations we took a third order taylor expansion of the non - polynomial dynamics even though exact treatment would be possible via a coordinate transformation leading to rational dynamics to which our methods can be readily extended ; this extension is , however , not treated in this paper and therefore we opted for the simpler ( and non - exact ) approach using taylor expansion . before solving the problem we made a linear coordinate transform so that the state constraint set becomes the unit hypercube @xmath269 ^ 6 $ ] . this example , which is the largest of those considered in this paper , took 110 seconds to solve with sedumi for @xmath270 ; the corresponding time with the mosek sdp solver was 10 seconds . using mosek we could also solve this example for @xmath218 ( in 420 seconds ) although there the solver converged to a solution with a rather poor accuracy and therefore we do not report the results . ( 120,120 ) ( 158,0 ) ( 238,15)@xmath14 ( 300,53)@xmath271 ( 260,78)@xmath272 ( 232,78)@xmath247 ( 268,107)@xmath248 ( 140,220 ) ( 128,20 ) ( 165,10)@xmath248 ( 335,20)@xmath247 ( 70,119)@xmath273 ( 110,35)@xmath274 ( 140,28.5)@xmath275 ( 184,22)@xmath276 ( 205,18)@xmath277 ( 242,12)@xmath278 ( 271,16)@xmath274 ( 292.5,25)@xmath275 ( 322.5,31.5)@xmath276 ( 332,41)@xmath277 ( 350,49)@xmath278 ( 114,55)@xmath279 ( 107,87)@xmath280 ( 120,118)@xmath276 ( 115,150)@xmath281 ( 121,182)@xmath282 we derived an infinite - dimensional convex characterization of the maximum controlled invariant ( mci ) set , finite - dimensional approximations of ( the dual of ) which provide a converging sequence of semialgebraic outer - approximations to this set . the outer - approximations are the outcome of a single semidefinite program ( sdp ) with no additional data required besides the problem description . therefore the approach is readily applicable using freely available modeling tools such gloptipoly 3 @xcite or yalmip @xcite with no hand - tuning involved . the cost to pay for this comfort is the relatively unfavourable scalability of the semidefinite programs solved the number of variables grows as @xmath283 , where @xmath10 and @xmath284 are the state and control dimensions and @xmath285 is the degree of the approximating polynomial . therefore , in order for this approach to scale to medium dimensions ( say , more than @xmath286 ) one either has to tradeoff accuracy by taking small @xmath285 or go beyond the standard freely available solvers such as sedumi or sdpa . one possibility is parallelization ; for instance , the free parallel solver sdpara @xcite allows for the approach to scale to larger dimensions . alternatively , one can utilize one of the ( few ) commercial sdp solvers ; in particular , the recently released mosek sdp solver seems to show far superior performance on our problem class , and therefore this may allow the approach to scale to larger dimensions ( see also the discussion following the acrobot - on - a - cart example in section [ sec : acrobotoncart ] ) . finally , one can resort to customized structure - exploiting solutions ; this is a promising direction of future research currently investigated by the authors . at this point it should be emphasized that , to the best of the authors knowledge , all of the existing approaches providing approximations of similar quality experience similar or worse scalability properties . other directions of future research include the extension of the presented approach to inner approximations of mci sets , to stochastic systems and to uncertain systems . partial results on the inner approximations for the related problem of region of attraction computation already exist @xcite , albeit in uncontrolled setting only . we start by embedding our problem in the setting of discrete - time markov control processes ; terminology and notation is borrowed from the classical reference @xcite . let us define a stochastic kernel on @xmath38 given @xmath2 as a map @xmath287 such that @xmath288 is a probability measure on @xmath38 for all @xmath289 and @xmath290 is a measurable function on @xmath2 for all @xmath291 . any such stochastic kernel gives rise to a discrete - time markov process when applied to system ( [ sysd ] ) as a stationary randomized control policy ( a policy which , given @xmath40 , chooses the control action randomly based on the probability distribution @xmath292 , i.e. , @xmath293 for all @xmath291 ) . the transition kernel @xmath294 of this stationary markov process is then given by @xmath295 where @xmath40 is the current state and @xmath296 the successor state . the @xmath32-step transition kernel is then defined by induction as @xmath297 with @xmath298 . given an initial distribution @xmath57 , the distribution of the markov chain at time @xmath32 , @xmath299 , is given by @xmath300 the joint distribution of state and control is then @xmath301 the discounted occupation measure associated to the markov process is defined by @xmath302 note that this relation reduces to ( [ eq : discom ] ) when @xmath303 in order to prove lemma [ lem : corrdisc ] we need the following result that can be found in @xcite . [ lem : corrd ] for any pair of measures @xmath65 satisfying equation ( [ eq : liouvdisc ] ) there exists a stationary randomized control policy @xmath304 such the markov chain obtained by applying this control policy to the difference equation ( [ sysd ] ) starting from initial distribution @xmath57 has the discounted occupation measure equal to @xmath4 . disintegrate @xmath4 as @xmath305 , where @xmath306 denotes the @xmath40-marginal of @xmath4 and @xmath307 is a stochastic kernel on @xmath38 given @xmath2 . according to the discussion preceding lemma [ lem : corrd ] , applying @xmath307 to ( [ sysd ] ) gives rise to a stationary discrete - time markov process with the transition kernel @xmath308 starting form the initial distribution @xmath57 . with this notation , equation ( [ eq : liouvdisc ] ) can be equivalently rewritten as @xmath309 for any measurable @xmath133 ( derivation of equation ( [ eq : liouvdisc ] ) did not depend on the continuity of @xmath11 ) . taking @xmath310 we obtain @xmath311 using relation ( [ eq : liouvdisc2 ] ) with @xmath312 to evaluate the integral w.r.t . @xmath306 on the right hand side of ( [ eq : liouvdisc3 ] ) we get @xmath313 by iterating this procedure we obtain @xmath314 and taking the limit as @xmath315 gives @xmath316 where the third term in ( [ eq : convtozero ] ) converges to zero because @xmath317 , @xmath318 and @xmath306 is a finite measure . hence the @xmath40-marginal of the discounted occupation measure of the markov chain coincides with the @xmath40-marginal of @xmath4 . finally , to establish equality of the whole measures observe that @xmath319 * proof of lemma [ lem : corrdisc ] : * disintegrate @xmath4 to @xmath320 as in the proof of lemma [ lem : corrd ] . then for any @xmath321 we have @xmath322 this relation says that the support of @xmath306 is invariant under @xmath307 and follows from lemma [ lem : corrd ] , from the definition of the occupation measure @xmath4 , from the definition of the support and from the fact that @xmath323 is a probability measure for all @xmath40 . define an admissible stationary deterministic control policy by taking any measurable selection @xmath324 . define further the sequence of probability measures @xmath325 where @xmath326 is a closed ball of radius @xmath327 centered at @xmath328 . then @xmath329 converges weakly- * ( or weakly or narrowly ) to @xmath330 and @xmath331 therefore , @xmath332 where the inequality follows by the portmanteau lemma since the set @xmath333 is closed for all @xmath40 by continuity of @xmath18 . therefore in fact @xmath334 and so @xmath335 for all @xmath336 . therefore @xmath337 is invariant for the closed loop system @xmath338 , where @xmath328 is an admissible deterministic control policy . therefore necessarily @xmath339 . finally , from equation ( [ eq : liouvdisc ] ) clearly @xmath340 and so @xmath341 [ lem : corrcontaux ] for any pair of measures ( @xmath343 ) solving ( [ eq : liouvcont ] ) , there exists a family of trajectories of the convexified inclusion ( [ sysc ] ) starting from @xmath57 such that the @xmath40-marginal of its discounted occupation measure is equal to the @xmath40-marginal of @xmath4 . the proof is based on fundamental results of @xcite and @xcite and on the compactification procedure discussed in @xcite . we begin by embedding the problem in a stochastic setting . to this end , define the extended state space @xmath344 as the one - point compactification of @xmath345 , i.e. , @xmath346 , where @xmath347 is the point compactifying @xmath345 . define also the linear operator @xmath348 by @xmath349 where the domain of @xmath350 , @xmath351 , is defined as @xmath352 in words , @xmath351 is the space all continuously differentiable functions vanishing at infinity such that @xmath353 also vanishes at infinity for all @xmath354 . now consider the relaxed martingale problem @xcite : find a stochastic process @xmath355\times \omega\to e$ ] defined on some filtered probability space @xmath356 and a stochastic kernel @xmath357 ( stationary relaxed markov control ) on @xmath38 given @xmath344 such that * @xmath358 * for all @xmath359 the stochastic process @xmath360 is an @xmath361-martingale ( see , e.g. , @xcite for a definition ) . observe that there exists a countable subset of @xmath351 ( e.g. , all polynomials with rational coefficients attenuated near infinity ) dense in @xmath351 in the supremum norm . next , @xmath351 is clearly an algebra that separates points of @xmath344 and @xmath362 . finally , since @xmath363 is polynomial and hence locally lipschitz , the ode @xmath37 has a solution on @xmath44 for any @xmath364 and any fixed @xmath354 in the sense that if there is a finite escape time @xmath365 , then we define @xmath366 for all @xmath367 . each such solution satisfies the martingale relation ( [ eq : mart ] ) ( with a trivial probability space ) . therefore , @xmath350 satisfies conditions 1 - 3 of @xcite and it follows from theorem 2.2 and corollary 2.2 therein that for any pair of measures satisfying the discounted liouville s equation ( [ eq : liouvcont ] ) , there exists a solution to the above martingale problem whose discounted occupation measure is equal to @xmath4 , that is , @xmath368 where @xmath369 denotes the expectation w.r.t . the probability measure @xmath370 . from the martingale property of ( [ eq : mart ] ) and the definition of @xmath350 we get @xmath371 now let @xmath372 denote the marginal distribution of @xmath373 at time @xmath32 ; that is , @xmath374 then the above relation becomes @xmath375 where we have used fubini s thorem to interchange the expectation operator and integration w.r.t . time . defining the relaxed vector field @xmath376 and rearranging we obtain @xmath377 where the equation holds for all @xmath378 almost everywhere with respect to the lebesgue measure on @xmath44 . the lemma then follows from ambrosio s superposition principle ( * ? ? ? * theorem 3.2 ) using the same arguments as in the proof of lemma 4 in @xcite . * proof of lemma [ lem : corrcont ] : * suppose that a pair of measures @xmath65 satisfies ( [ eq : liouvcont ] ) and that @xmath379 . from lemma [ lem : corrcontaux ] there is a family of trajectories of ( [ sysc ] ) starting from @xmath57 with discounted occupation measure whose @xmath40-marginal coincides with the @xmath40-marginal of @xmath4 . however , this is a contradiction since no trajectory starting from @xmath380 remains in @xmath2 for all times and @xmath381 . thus , @xmath382 and so @xmath86 . the authors are grateful to slvka jadlovsk for providing the acrobot - on - a - cart system and andrea alessandretti for providing the spider - web system . l. ambrosio . transport equation and cauchy problem for non - smooth vector fields . in l. ambrosio ( eds . ) , calculus of variations and nonlinear partial differential equations . lecture notes in mathematics , vol . 1927 , springer - verlag , berlin , 2008 . a. n. daryin , a. b. kurzhanski . parallel algorithm for calculating the invariant sets of high - dimensional linear systems under uncertainty . computational mathematics and mathematical physics 53(1):34 - 43 , 2013 . m. liu , s. zhang , z. fan , m. qiu . @xmath384 state estimation for discrete - time chaotic systems based on a unified model . ieee trans . on systems , man , and cybernetics part b , cybernetics , 42(4):1053 - 1063 , 2012 . j. lfberg . yalmip : a toolbox for modeling and optimization in matlab . in proc . ieee cca / isic / cacsd conference , taipei , taiwan , 2004 . a. majumdar , a. a. ahmadi , r. tedrake . control design along trajectories with sums of squares programming . ieee international conference on robotics and automation ( icra ) , 2013 ( to appear ) . i. plik , t. terlaky , y. zinchenko . sedumi : a package for conic optimization . i m a workshop on optimization and control , univ . minnesota , minneapolis , 2007 . m. putinar . positive polynomials on compact semi - algebraic sets . indiana univ . mathematics journal , 42:969 - 984 , 1993 . r. rajarama , u. vaidya , m. fardadc , b. ganapathysubramanian . stability in the almost everywhere sense : a linear transfer operator approach . journal of mathematical analysis and applications , 368:144 - 156 , 2010 . m. yamashita , k. fujisawa , m. fukuda , k. kobayashi , k. nakta , m. nakata . latest developments in the sdpa family for solving large - scale sdps . in m. anjos , j. b. lasserre ( eds . ) . handbook on semidefinite , cone and polynomial optimization : theory , algorithms , software and applications . springer , ny , usa , chap . 24 , 687 - 714 , 2011 .

0.5 cm we demonstrate how to make rigorous predictions for electroweak matrix elements in nuclear systems directly from qcd . more precisely , we show how to determine the short - distance contributions to low - momentum transfer electroweak matrix elements in the two - nucleon sector from lattice qcd . in potential model descriptions of multi - nucleon systems , this is equivalent to uniquely determining the meson - exchange currents , while in the context of nuclear effective field theory , this translates into determining the coefficients of local , gauge - invariant , multi - nucleon - electroweak current operators . the energies of the lowest - lying states of two nucleons on a finite volume lattice with periodic boundary conditions in the presence of a background magnetic field are sufficient to determine the local four - nucleon operators that contribute to the deuteron magnetic moment and to the threshold cross - section of . similarly , the energy - levels of two nucleons immersed in a background isovector axial weak field can be used to determine the coefficient of the leading local four - nucleon operator contributing to the neutral- and charged - current break - up of the deuteron . this is required for the extraction of solar neutrino fluxes at sno and future neutrino experiments . 1.0 cm -1.5 cm 0.5 cm

hep-lat0403005

a major goal of nuclear physics is to be able to rigorously compute the properties and interactions of nuclei directly from qcd . while qcd is formulated in terms of quarks and gluons , the relevant degrees of freedom of strongly interacting systems probed at low - momentum are the lowest - lying hadrons . while the properties of these hadrons are well known experimentally , and some of their interactions are also well studied ( such as the nucleon - nucleon ( nn ) cross - section ) , in general , direct measurement of the interactions that contribute to nuclear processes is not possible . for instance , there are multi - nucleon interactions induced at the chiral symmetry breaking scale , that will not be directly measured , but which contribute to the interactions of nuclei . the only way of rigorously computing strong interaction observables and electroweak observables involving hadrons , is with lattice qcd . in lattice qcd space - time is discretized and green functions are evaluated in euclidean space . at this point in time the variety of processes that can be addressed with lattice qcd is quite limited . present day computational power restricts the sizes of lattices , the lattice spacings , and the quark masses that can be used in simulations , making the extraction of physical observables non - trivial . even if infinite computing power were to become available tomorrow most of the formal framework with which to calculate most nuclear properties and observables does not yet exist . notable exceptions are the nn and nucleon - hyperon ( ny ) scattering lengths and range parameters @xcite . at the heart of the issue is the maiani - testa theorem @xcite which precludes determination of scattering amplitudes from euclidean - space green functions at infinite volume away from kinematic thresholds . by generalizing a result from non - relativistic quantum mechanics @xcite to quantum field theory , lscher @xcite showed how to extract the @xmath1 scattering amplitude from the energy - levels of two particles in a finite - volume lattice simulation . this technique has been used to determine the low - energy @xmath2 phase shifts directly from qcd , e.g. ref . @xcite . however , only one lattice qcd calculation of the nn scattering lengths @xcite exists , and it is quenched with relatively large quark masses ( for a review see ref . @xcite ) . in this work we establish a framework with which to extract the electroweak matrix elements in the two - nucleon sector . this is only a small step toward a framework with which to determine electroweak matrix elements in arbitrary multi - nucleon systems , however , the two - nucleon sector is pleasantly challenging all by itself . naively this appears to require the computation of hadronic five - point correlators , a daunting task that is beyond the reach of current computers . however , calculation of the energy - levels of two nucleons in a finite volume in the presence of appropriate background electroweak fields allows for these matrix elements to be extracted with little more computing resources than are needed to calculate the scattering parameters . we start by studying the energy - levels of two nucleons immersed in a background magnetic field . clearly a calculation of the deuteron electromagnetic properties will not be heralded as a major accomplishment ( though it would be an impressive test of lattice qcd ) as they are well - known from precise experiments , e.g. ref . . however , they do provide an illustration of how the long - distance and short - distance contributions can be isolated with lattice qcd . this technique is then generalized to weak background fields in order to allow for calculation of the axial matrix element that dominates @xmath3 break - up @xcite , the process used by sno @xcite to determine the flavor composition of the solar neutrino flux . the present - day uncertainty in this matrix element will provide a significant uncertainty in the determination of neutrino mass differences and mixing matrix derived from future experiments . determinations of the electroweak matrix elements will be independent of the hadronic theory used to compute the infrared ( ir ) properties of qcd as long as it is a complete theory consistent with all the symmetries of qcd . in this work we use effective field theory ( eft ) , and in particular , the pionless eft , @xmath4 to study the very low - momentum ( @xmath5 ) interaction of nucleons and electroweak gauge fields @xcite . @xmath4 has been developed over the last few years to describe systems with unnaturally large scattering lengths , as are found in the multi - nucleon sector of qcd . such large scattering lengths result from a fine - tuning that nature has presented to us , more than likely for anthropic reasons , and as a result , the analysis in this work will not apply for arbitrary values of the light - quark masses . in the traditional approach to calculating electroweak observables in nuclear physics , one assumes that all processes are dominated by one - nucleon interactions with the electroweak probe , and that multi - nucleon contributions are small . these multi - nucleon interactions result from currents coupling to the exchange of single mesons , @xmath6 s , @xmath7 s and so forth , the same mesons that are used to construct nucleon - nucleon potentials , do not rely on single meson exchange to describe the short - distance component of the nn potential , instead use some `` best - fit '' functional forms that minimize the @xmath8 of the fit to nn data . ] and are called `` meson - exchange currents '' ( mec s ) limit of qcd , is not true in general . ] . in the context of eft , these matrix elements receive contributions from one - nucleon interactions with the electroweak current , from interactions with the mesons that have masses below the cut - off of the theory ( in the pionless theory there are no mesonic contributions , while in the pionful theory there are contributions from pions ) and from local , two- ( or more ) nucleon electroweak current interactions . these multi - nucleon interactions result from the physics at momentum scales above the cut - off of the eft that has been integrated out . a significant advantage of the eft is that no attempt is made to force these interactions to look like they result from the exchange of mesons they are what they are . however , before the eft can become predictive , the coefficients of the various operators that occur in it must be determined either from experiment or from lattice qcd . as the scattering lengths in both the @xmath9 and @xmath10 channels ( @xmath11 and @xmath12 ) are unnaturally large ( both channels are near their ir fixed - points @xcite ) , @xmath4 can even be used to describe electroweak processes involving the deuteron . @xmath4 can be recast into a somewhat simpler theory with which to perform computations up to a given order in the momentum expansion by introducing dynamical fields @xcite . as s - matrix elements calculated in the usual formulation of @xmath4 and with the formalism are the same at any given order in the expansion and the two frameworks differ only in the ultra - violet ( uv ) , both forms will give identical results for observables in a finite - volume . in this work we will not reproduce the methodology of @xmath4 @xcite , however , we shall discuss relevant aspects of the formalism @xcite . in terms of nucleon and degrees of freedom , the leading order ( lo ) low - energy strong interactions for @xmath13 are described by a lagrange density of the form @xmath14 n \ -\ t_j^\dagger\left[\ i\partial_0 + { \nabla^2\over 4 m}-\delta_3\right ] t^j \ -\ s_a^\dagger\left[\ i\partial_0 + { \nabla^2\over 4 m}-\delta_1\right ] s^a \nonumber\\ & & - y_3 \left[\ t_j^\dagger\ n^t p_3^j n\ + \ { \rm h.c.}\right ] - y_1 \left[\ s_a^\dagger\ n^t p_1^a n\ + \ { \rm h.c.}\right ] \ \ \ , \label{eq : lagst}\end{aligned}\ ] ] where @xmath15 is the nucleon annihilation operator , @xmath16 is the @xmath10 annihilation operator with spin - index @xmath17 , and @xmath18 is the @xmath9 annihilation operator with isospin - index @xmath19 . further , @xmath20 is the spin - isospin projector for two nucleons in the @xmath10 channel with spin - index @xmath17 , while @xmath21 is the spin - isospin projector for two nucleons in the @xmath9 channel with isospin - index @xmath19 , @xmath22 in the lagrange density in eq . ( [ eq : lagst ] ) , a factor of the nucleon mass @xmath23 has been absorbed into the definition of the nucleon fields and similarly a factor of @xmath24 into the fields . the s - wave interactions are enhanced by a factor of the expansion parameter , @xmath25 , and are treated non - perturbatively . however the interactions that introduce mixing with higher partial waves , e.g. @xmath26 mixing , are suppressed by at least @xmath27 and so only s - wave to s - wave interactions are required to the order we are working . in order to recover the scattering amplitudes in both s - wave channels , the constants that appear in eq . ( [ eq : lagst ] ) are @xcite @xmath28 where @xmath29 is the renormalization scale , @xmath30 and @xmath31 are the scattering length and effective range in the @xmath10 channel , and @xmath32 and @xmath33 are the scattering length and effective range in the @xmath9 channel . for the processes under consideration , the local , gauge - invariant , electroweak interactions between two - nucleons and the gauge fields need to be included in addition to the interactions that result from gauging the lagrange density in eq . ( [ eq : lagst ] ) . the additional electromagnetic gauge - invariant operators that contribute to the processes of interest at lo and nlo in the power - counting are @xcite @xmath34 \sigma\cdot { \bf b } n \ -\ { e\over m } \left ( \kappa_0 - { \tilde \ltwo\over r_3 } \right ) i\epsilon_{ijk } \ b_k \nonumber\\ & & \ + \ { e \lone\over m\sqrt{r_1 r_3 } } \ \left [ t_j^\dagger s_3 b_j + { \rm h.c.}\right ] \ \ \ , \label{eq : lagmag}\end{aligned}\ ] ] where @xmath35 and @xmath36 are the proton and neutron magnetic moments , @xmath37 is the electric charge , and @xmath38 is the magnetic field . the magnetic moment interaction of the @xmath10 is written in such a way that if @xmath39 the deuteron magnetic moment is the sum of the neutron and proton magnetic moments @xcite ( at nlo ) @xmath40 where @xmath41 is the binding momentum of the deuteron in an infinite volume with no background fields , and satisfies the truncated effective range expansion , @xmath42 where the deuteron energy is @xmath43 . as spin - dependent interactions are suppressed in the infrared , @xmath44 is expected to be small and this is verified experimentally . to make some connection with traditional nuclear physics , the constants @xmath44 and @xmath45 correspond to multi - nucleon interactions with the electromagnetic field induced by mec s . the breakdown scale of @xmath4 restricts the magnitude of electroweak fields that can be described . in order to have a convergent expansion , the magnetic field must be less than @xmath46 so that the one - nucleon magnetic moment interaction shifts the nucleon mass by less than @xmath47 . further , requiring that the nucleon magnetic polarizability shifts the nucleon mass by less than the magnetic moment ( see eq . ( [ eq : mpupshift ] ) below ) requires that @xmath48 . using the experimental values of @xmath49 and @xmath50 @xcite give a constraint of @xmath51 , which is a somewhat weaker constraint than from the magnetic moment interaction . a more naive , but still reasonable restriction is that @xmath52 mev@xmath53 arrived at by requiring nothing more than that `` @xmath54 is small '' , and it is this more restrictive limit that we adhere to . a direct calculation of multi - nucleon processes such as @xmath55 is impossible on a euclidean space lattice . as briefly mentioned in the introduction , the obstruction is the maiani - testa theorem @xcite which precludes the calculation of s - matrix elements away from kinematic thresholds in an infinite volume . in principle however , by performing lattice calculations of matrix elements of the relevant electroweak operators at unphysical kinematics in appropriate multi - nucleon states , e.g. , @xmath56 , the undetermined coefficients of the low - energy eft ( modified to accommodate the injection of energy by the lattice ) can be determined , leading to an _ ab initio _ calculation of the low - energy dependence of the process in question . unfortunately , by present day standards , calculation of such five - point correlators is prohibitively expensive in computational terms . in order to do such a calculation , the two - nucleon ( six quark ) sources and sinks and the electroweak current operator must be tied together with quark propagators in all possible ways and the corresponding contributions averaged over the ensemble of gauge configurations . naively , this is one or two orders of magnitude more difficult than computing the nn four - point correlator . clearly , a more cost - effective solution to this problem is desired we will show that calculations of electroweak processes , such as @xmath57 and @xmath58 , will be feasible in the near future . our method is to some extent motivated by the earliest lattice calculations of the magnetic moments of the proton and neutron . modern calculations of these quantities are performed by taking nucleon matrix elements of the electromagnetic current operator . however , in the early days of lattice qcd , calculations of nucleon matrix elements were computationally prohibitive ( just as multi - nucleon matrix elements are today ) and the magnetic moments were first extracted by measuring the shift in the nucleon mass when the lattice is immersed in a background magnetic field @xcite . taking the magnetic field to be constant and aligned along the @xmath59-direction , @xmath60 the nucleon masses are shifted , e.g. the mass of a proton with spin aligned in the direction of the field is @xmath61 however , it is the energy of the nucleon that is measured on the lattice , and when immersed in a background magnetic field the energy eigenstates of the proton in an infinite volume are @xmath62 where @xmath63 is an integer denoting the landau - level occupied by the proton , and @xmath64 is the momentum of the proton parallel to the magnetic field . for weak magnetic fields , the spin - dependent energy - splitting is proportional to the magnetic moment of the proton , @xmath65 , while for stronger fields the magnetic polarizability , @xmath66 becomes significant and measurable @xcite . to be complete we note that by using background electric fields , electric polarizabilities @xcite and the electric dipole moment of the nucleon @xcite have been investigated . to apply a spatially constant background magnetic field aligned in the @xmath59-direction in a lattice simulation , one takes a su(3 ) gauge - configuration and modifies the link fields by @xmath67 with @xmath68 where @xmath69 is proportional to @xmath70 where @xmath71 is the lattice spacing , throughout . ] . a vector potential with periodic boundary conditions in the transverse directions requires that @xmath72 , where @xmath73 is the transverse area of the box . this requirement generates large magnetic fields for present day lattices @xcite , large enough to shift the nucleon mass by @xmath74 , and thus too strong to be described by @xmath4 . not requiring periodic boundary conditions on the vector potential introduces an exceptional plaquette that can be placed wherever it does the least `` harm '' in the single nucleon case @xcite . it remains to be seen if this plaquette can be placed somewhere to have minimal impact on multi - nucleon simulations or if the periodic requirement must be enforced . for the weak interactions , and in particular for an axial - vector interaction , the left- and right - handed quarks need to be immersed in different background fields . one way to implement this is to use domain - wall ( kaplan @xcite ) fermions in the simulation and have the background field vary only in the fifth dimension . a background axial field was used in the first lattice calculation of the axial coupling of the nucleon @xcite . by applying such background electroweak fields to nucleons in a finite volume , the energy - levels become sensitive to the various one- and two - nucleon electroweak current operators that contribute to the deuteron properties and interactions , such as @xmath57 , @xmath58 and other related electroweak processes . it is interesting to note that lattice calculations of the energy - levels relevant to the deuteron magnetic ( or weak ) moment ( isoscalar operator ) will be more difficult than those impacting isovector processes as they require the evaluation of disconnected diagrams ( where the operator does not connect to valence quarks ) which are significantly harder to compute , even in the single nucleon sector @xcite . as we will discuss later , it will be necessary to work with asymmetric volumes of dimensions @xmath75 the energy - levels of two - particles in an asymmetric volume have been discussed recently @xcite and we review this analysis and extend a number of relevant aspects in appendix [ app : one ] . it is straightforward to show that the locations of the low - lying energy - levels of two - nucleons interacting in an s - wave on an asymmetric lattice are given by the solutions of @xcite @xmath76 where @xmath77 uniquely describes the infinite - volume s - wave scattering amplitude below inelastic thresholds . the function @xmath78 is given by @xmath79 where @xmath80 , and the limits of the linearly - divergent , three - dimensional sum are given by the ellipsoid @xmath81 . the energy corresponding to a given value of @xmath82 is @xmath83 given the behavior of two - nucleon energy - levels as a function of the volume , the phase shifts ( or at least the first few parameters of the effective range expansion @xmath84 ) can be recovered . to aid in such recovery , one can find analytic , large volume ( @xmath85 ) expansions for the locations of the energy - eigenstates , as done by lscher for the case of @xmath86 @xcite and done by li and liu @xcite for the lowest continuum level in asymmetric volumes . we detail this construction in appendix [ app : one ] , and give explicit expressions for the large volume locations of the low - lying levels in the two - nucleon sector . a small @xmath87 expansion of eq . ( [ eq : exactee ] ) can also be performed @xcite , though we do not do this here . it is straightforward to extend this analysis of two - nucleon energy - levels to include a background electroweak field . the presence of the background fields can give rise to a number of complications such as landau - levels and mixing of two - nucleon states of differing total spin and isospin . for this reason we shall develop the necessary generalizations of the above formulas as we need them . the hamiltonian of a structure - less point particle of charge @xmath37 , moving in a uniform , time - independent magnetic field in the @xmath59-direction is @xmath88 where @xmath89 , and @xmath90 is the angular momentum operator parallel to the magnetic field . in an infinite volume , the energy - eigenvalues are given by the expression in eq . ( [ eq : epup ] ) . however , in a finite volume the oscillator potential is bounded above , and for small enough lattices the potential arising from the magnetic field can be treated as a perturbation . consequently , for a small magnetic field we can work with momentum eigenstates in the transverse direction . requiring that the energy - shifts due to these magnetic interactions are small compared with the inter - level spacing of the two - nucleon momentum eigenstates in the finite volume gives , @xmath91 where @xmath92 is the size of the transverse dimensions . in what follows we work with momentum eigenstates in the transverse directions , and thus are constrained by eq . ( [ eq : llcon ] ) . it is useful to begin with the simplest case to analyze and understand the deuteron magnetic moment . to isolate the two - nucleon contribution from the one - nucleon contribution ( the magnetic moments of the proton and neutron combined in the @xmath93 channel ) , the energy - levels of two nucleons interacting in the @xmath10 channel in a finite volume with a background magnetic field in the @xmath59-direction , @xmath94 , are computed . it is obvious from eq . ( [ eq : mpupshift ] ) that the @xmath95 magnetic sub - states of the nn system will have a linear dependence on the magnetic field for small fields . indeed , the effect of the background magnetic field on the single - nucleon states is to shift their masses from @xmath23 to @xmath96 where @xmath97 are the proton and neutron magnetic moments . these mass - shifts appear as residual mass - terms in the single nucleon propagators that contribute to the infinite - volume nn scattering amplitude , and hence to the position of the energy - levels of two - nucleons in a finite volume is removed by a phase redefinition of the nucleon field in defining the non - relativistic theory . the presence of interactions between different spin and isospin components of the nucleon field prevents a rescaling of each component to completely remove the mass - shifts in eq . ( [ eq : nmassb ] ) . ] . in the weak-@xmath98 limit contributions to the nn energy - eigenvalues that are quadratic and higher - order in @xmath98 , such as the nucleon polarizabilities and modifications arising from the eigenstates in the transverse directions being landau - levels and not momentum plane - waves , can be neglected . since it is a non - relativistic system , a simple recipe can be followed to determine the location of the low - lying energy - eigenstates of two nucleons in a background magnetic field in a finite volume @xcite . construct the inverse scattering amplitude in the background magnetic field , replace the continuum momentum - space loop integral ( arising from the bubble - diagrams ) by a discrete sum over the allowed momenta in the volume and then find the zeros of the real part @xcite . it follows directly from this recipe that the location of the @xmath95 energy - levels are given by the solutions to @xmath99 where @xmath100 and @xmath101 is the effective range expansion in the @xmath10 channel . the one - nucleon contribution to the location of the nn energy - levels is through the term @xmath102 and through @xmath103 in the last term in eq . ( [ eq : mpmone ] ) . the one - nucleon and two - nucleon ( determined by @xmath44 ) terms enter eq . ( [ eq : mpmone ] ) differently . consequently , the short - distance and the long - distance interactions with a magnetic field can be separated , i.e. @xmath44 can be isolated from @xmath103 on the lattice . note that the expression in eq . ( [ eq : mpmone ] ) reduces to the expression in eq . ( [ eq : exactee ] ) in the limit of vanishing magnetic field . for symmetric lattices with @xmath86 , it follows directly from eq . ( [ eq : mpmone ] ) that for @xmath104 there are two states that become the deuteron @xmath95 states in the infinite - volume limit , that have energies @xmath105 \ \mp\ \mu_d b_0 \ \ \ , \label{eq : monebsasym}\end{aligned}\ ] ] in a finite - volume . the deuteron magnetic moment at nlo in the eft is given in eq . ( [ eq : deutmag ] ) , but eq . ( [ eq : monebsasym ] ) is true at all orders in the expansion . the leading finite - volume corrections in eq . ( [ eq : monebsasym ] ) are the same as those at @xmath106 @xcite . although this gives a clean lattice measurement of the deuteron magnetic moment ( itself a very good test of the lattice approach ) , it does not facilitate a separation between the one- and two - nucleon contributions . making an expansion of eq . ( [ eq : mpmone ] ) about the infinite - volume limit , @xmath107 , the energies of the first pair of continuum states in the limit where @xmath108 are @xmath109 \ \ \ , \label{eq : moneasym}\end{aligned}\ ] ] where an effective scattering length is defined to be @xmath110 and the constants @xmath111 for @xmath86 are given in table [ table : c1c2c1pc2pee ] @xcite . the energies of the second pair of continuum states in the limit where @xmath108 are @xmath112 where @xmath113 the constants @xmath114 for @xmath86 are given in table [ table : c1c2c1pc2pee ] @xcite , and @xmath115 is defined in eq . ( [ eq : a3def ] ) . the energies in eqs . ( [ eq : moneasym ] ) and ( [ eq : moneasym2 ] ) closely resemble the well - known lscher formulas @xcite , with some interesting modifications . the levels receive shifts proportional to @xmath103 corresponding to the shifts that two non - interacting nucleons would receive in a magnetic field in an infinite volume . in addition , instead of the physical scattering length entering into the @xmath116-expansion , it is the effective scattering length , @xmath115 that appears . the effective scattering length , defined in eq . ( [ eq : a3def ] ) , depends upon the actual scattering length and the short - distance contribution to the nn interaction due to the background magnetic field . these expressions are not entirely unexpected . in the continuum states the nucleons do not `` see each other '' much , and the short - distance interactions are a perturbation ( as is clear from eqs . ( [ eq : moneasym ] ) and ( [ eq : moneasym2 ] ) ) . the single particle energies dominate and the short - distance interaction , which is the sum of both the purely strong interaction and that induced by the background field , is inserted perturbatively . measuring the energies of the continuum levels over a range of lattice volumes will allow for a clean separation between the one - nucleon and two - nucleon contributions to the deuteron magnetic moment . given that the @xmath10 channel is very close to its non - trivial fixed point ( @xmath117 ) , most choices of magnetic field will move the system away from the fixed point , and hence @xmath118 will typically ( i.e. , for natural values of @xmath44 , @xmath119 fm ) be dramatically smaller than @xmath120 . this suggests that , in principle , lattices even smaller than those discussed in ref . @xcite could be used to study the two - nucleon sector when a background magnetic field is present . however , nature is not as kind as it might have been , suppressing the isoscalar spin - dependent interactions ( proportional to @xmath44 ) at low - momentum by even more than the naive @xmath121 obtained in the large-@xmath122 limit @xcite . the experimental value of the deuteron magnetic moment @xmath123 is very close to the single particle sum of @xmath124 which results in an unnaturally small value of @xmath125 in eq . ( [ eq : deutmag ] ) . therefore , for magnetic fields that allow for an analysis of the two - nucleon sector within @xmath4 , the effective scattering length , @xmath115 , is only moderately different from the actual scattering length , @xmath30 . specifically , for @xmath126 , @xmath115 differs from @xmath30 by @xmath127 ( for the physical value of @xmath44 ) , an amount that should produce observable shifts in the continuum two - particle energy levels calculated in lattice simulations . in fig . [ fig : levelsmpmone ] the energy - levels of the three lowest - lying @xmath95 states are shown , along with the analytic expressions in eqs . ( [ eq : monebsasym ] ) , ( [ eq : moneasym ] ) , ( [ eq : moneasym2 ] ) and their asymmetric generalizations that are valid for @xmath104 . = 2.15 in = 2.15 in 0.15 in .2 in an interesting `` twist '' could be applied to the systems we are considering . it is possible that the light - quark masses , @xmath128 , could be tuned to values for which the physical scattering lengths in the @xmath10 or @xmath9 channels , or both , become infinite @xcite . as both channels have unnaturally large scattering lengths at the physical values of @xmath128 , only small changes in these masses will be required , and hence the values of the operators in the pionless theory describing the unphysical system , will differ from their physical values by only small amount . an advantage of considering such an unphysical system is that @xmath115 will depend only on @xmath44 and its determination on the lattice would provide a measurement of @xmath44 for the physical theory with a reasonably small systematic uncertainty . the radiative capture process @xmath0 , and its inverse reaction ( photo - disintegration of the deuteron ) provide a classic demonstration of the importance of mec s in nuclear physics . the one - nucleon contribution to thermal neutron capture provides only @xmath129 of the experimentally measured cross section . from the eft perspective , it is natural that two - nucleon electromagnetic interactions can contribute to this process , in the same way that they contribute to the deuteron magnetic moment . the cross sections for @xmath0 for cold neutrons and for the photo - disintegration of the deuteron near threshold are both sensitive to @xmath45 . the radiative - capture cross section near threshold is given by @xcite @xmath130 \ \ \ , \label{eq : npdg}\end{aligned}\ ] ] where contributions from multipoles other than @xmath131 or @xmath132 , denoted by the ellipses , are parametrically suppressed at low - energies . near threshold , this process takes the two - nucleons between the @xmath9 channel and the @xmath10 channel and the cross section is dominated by the @xmath131 amplitude @xcite @xmath133 \ . \label{eq : xm1amp}\end{aligned}\ ] ] it is worth stressing that this very simple expression for the cross section ( including the analytic form for the @xmath132 amplitude @xcite ) describes all available low - energy data with high precision @xcite . a background magnetic field aligned in the @xmath59-direction induces a mixing between the @xmath9 , @xmath134 state and @xmath10 , @xmath135 state . therefore , the energy - eigenstates in the background field will be a mixture of both the @xmath9 and the @xmath10 channels , which gives rise to quite interesting behavior . in large volumes , the continuum states , for which the strong interactions play only a small role , are perturbatively close to the single particle eigenstates @xmath136 and @xmath137 , with energies dictated by the single particle interactions with the background field , i.e. linear in @xmath98 for small @xmath98 . hence , the energies of the continuum states will be insensitive to the value of @xmath45 . however , the energies of the near threshold bound - states , the deuteron and the virtual bound di - nucleon state in the @xmath9 channel , for which strong interactions play an important role , will depend quadratically on @xmath98 for small @xmath98 and will have sensitivity to @xmath45 . this results from the fact that these states have equal admixtures of the single particle eigenstates in infinite volumes . thus , it is clear that we will need to observe energy - shifts that are quadratic in @xmath98 in order to extract @xmath45 . this presents a potential problem that we did not encounter for the energy shifts of the @xmath95 states . in the directions transverse to @xmath38 , landau - levels are the single - particle eigenstates rather than momentum plane - waves . as discussed in section [ sec : lqcdbf ] , this is a complication in infinite volumes , but becomes a far more serious issue in finite volumes for systems with h(3 ) symmetry . in order to avoid having to contend with this problem , we eliminate it ; highly asymmetric lattices , for which @xmath138 , so as to satisfy eq . ( [ eq : llcon ] ) , strongly suppress excitations in the transverse direction , leaving the system dominated by fluctuations in the @xmath59-direction . ideally , one would take the transverse dimensions almost to zero to eliminate these contributions almost entirely . however , the pion mass sets a lower - limit of @xmath139 in order for the pionless theory to provide a valid description . on a symmetric lattice , as studied in refs . @xcite , the lowest - lying states are in the @xmath140 representation of @xmath141 and as such their energy is dependent upon nn interactions with @xmath142 @xcite with the notable omission of interactions with @xmath143 . on an asymmetric lattice where @xmath141 is not a symmetry , nn interactions with even angular momentum will contribute . however , in the power - counting of @xmath4 , higher angular momentum contributions are suppressed and can be systematically included when required . as the background field induces mixing between the @xmath9 and @xmath10 channels , the energy - eigenvalues are found by diagonalizing the coupled channels system . it is straightforward to show that these eigenvalues are solutions to @xmath144 \left[\ p\cot\delta_3 -{s_1 + s_2\over 2\pi l } \right ] = \left[\ { e b_0 \lone \over 2 } + { s_1 - s_2\over 2\pi l } \right]^2 \ \ , \label{eq : m0solve}\end{aligned}\ ] ] where @xmath145 @xmath146 and @xmath147 is the @xmath9 effective range expansion . the @xmath148 contributions result from the one - nucleon interactions with the background field , while the two - nucleon background field interactions are described by @xmath45 . the @xmath149 asymptotic region is more complicated in these mixed channels than in a single channel and we do not have practicable analytic expressions for the energies of the lowest - lying levels in this limit . thus , all of our intuition and understanding has come about through numerical solution of eq . ( [ eq : m0solve ] ) . in fig . [ fig : levelsmzero ] , we show the spectra of energy - eigenvalues of the @xmath150 nn system in magnetic fields of @xmath151 and @xmath152 on an asymmetric lattice with @xmath153 as a function of lattice size , @xmath87 . as a function of @xmath98 and @xmath45 , the continuum levels behave precisely as one would expect , and are very insensitive to @xmath45 . however , as anticipated , there are one or two states in the theory ( depending on the choice of parameters ) which are very sensitive to @xmath45 . the location of these states in an infinite volume are finely tuned ( near the ir fixed point ) , and as such , `` squeezing them '' into a volume that is small in the transverse directions destroys this fine - tuning . however , the background field , through the operator with coefficient @xmath45 , can restore the fine - tuning and make the location of the state depend strongly upon @xmath45 . further , the observation of two bound - states in the spectrum that are sensitive to @xmath45 for some values of background field and for some volumes indicates that the virtual bound state in the @xmath9 channel can be moved onto the first sheet by the background magnetic field . = 2.15 in = 2.15 in 0.15 in .2 in in the infinite volume limit one of the @xmath45-dependent states becomes the physical deuteron . however , in a finite volume and in the background field its composition , and the composition of the other states , will be quite complicated , each being a combination of the `` deuteron '' , the @xmath9-di - nucleon, and the continuum states , as deduced from fig . [ fig : levelsmzero ] . as @xmath4 is applicable only at very low - energies , the energy - spectra in fig . [ fig : levelsmzero ] are reliable only for @xmath154 . for @xmath151 , as shown in the left panel of fig . [ fig : levelsmzero ] , there is one `` reliable '' state . ] that moves to higher energy as a function of increasing volume , and is very sensitive to @xmath45 in the physical range of @xmath45 . for lattice sizes between @xmath155 and @xmath156 in the longitudinal direction , this level is one of the low - lying ones in the spectrum , and therefore , somewhat easier to extract from lattice calculations , than if it were high in the spectrum . also present in this spectrum is the curious feature of `` level pair - production '' near the cut - off of the theory . in the absence of a background field there are two solutions to eq . ( [ eq : pole ] ) , one corresponding to the deuteron and one to an unphysical state lying outside the theory , nonetheless both solutions exist , and the one outside the theory is simply discarded as unphysical . the same line of reasoning applies to the deep states in fig . [ fig : levelsmzero ] . we interpret these states as the unbound di - nucleon state in the @xmath9 channel mixing with continuum states and the deuteron through the background magnetic field , bringing an additional state down from the cut - off into the spectrum . its sensitivity to @xmath45 suggests that it is a strongly correlated nn pair , as opposed to predominantly continuum in nature . for @xmath152 , as shown in the right panel of fig . [ fig : levelsmzero ] , the two lowest - lying states in the spectrum are both sensitive to @xmath45 . it might well be the case that a precise measurement of the lowest - lying state in the spectrum at weaker fields is the easiest way to determine @xmath45 from lattice qcd . a general statement can be made about the nn system in background magnetic field . there are values of the magnetic field , lattice asymmetry and lattice size for which the spectrum of energy - eigenvalues allows for the short - distance and long - distance contributions to electromagnetic matrix elements to be separated , as is clear from fig . [ fig : levelsmzero ] . while a lattice extraction of these quantities will be very important to perform and provide a proof in principle of this method , it will not , unfortunately , tell us anything we did nt already know experimentally . of central importance to present day and future neutrino experiments is the ability to separate the flavors of neutrinos entering the detectors . the method to accomplish this implemented by the sno collaboration is to use the weak - disintegration of the deuteron , e.g. @xmath157 ( @xmath158 ) and @xmath159 . these processes `` look '' like the photo - disintegration of the deuteron discussed in the previous section when the electromagnetic field is replaced by a weak gauge field . there are both one - nucleon and two - nucleon contributions to these processes , but unlike the electromagnetic case , the two - nucleon contributions have significant uncertainties associated ( for a nice discussion see ref . @xcite ) with them that will impact the future determinations of the neutrino mass and mixing matrices . therefore a determination of the short - distance contributions to the matrix elements for weak - disintegration of the deuteron from lattice qcd is important . the procedure for determining the two - nucleon weak interactions is similar to that described in the electromagnetic sector with a few minor differences . for the electromagnetic interactions it sufficed to use a background magnetic field . however , for the weak interactions it is convenient to introduce a fictitious @xmath160 weak gauge symmetry under which the up- and down - quarks transform , giving rise to the interactions @xmath161 \ \ \ , \end{aligned}\ ] ] and give vevs to @xmath162 and @xmath163 , such that @xmath164 . therefore , in this background weak field there is an additional contribution to the strong interaction lagrange density of @xmath165 this interaction leads to interactions in the formulation of @xmath4 of the form @xmath166 \ + \ ... \ \ \ , \label{eq : wkeft}\end{aligned}\ ] ] where the ellipses denotes terms higher order in @xmath4 , and @xmath167 is the nucleon axial coupling constant . we note that due to the nature of the interaction , weak `` landau - levels '' are not present in this system . the cross sections for the weak - disintegration of the deuteron and other two - nucleon weak processes depend upon the dimensionless coefficient @xmath168 . these processes have been studied extensively by butler and chen @xcite and also by butler , chen and kong @xcite and our ( renormalization - group invariant ) @xmath168 is related to the constants they introduce . the coefficient of the four - nucleon operator in @xmath4 , @xmath169 , is related to @xmath168 via @xmath170 \ \ \ , \label{eq : l1arel}\end{aligned}\ ] ] where @xmath171 are the coefficients of the zero- and two - derivative strong interaction operators in @xmath4 . numerically , eq . ( [ eq : l1arel ] ) reduces to @xmath172 when the rg - scale @xmath173 is chosen , and where @xmath174 is in units of @xmath175 . calculations have been done in which a chiral expansion of the weak currents are performed in a manner consistent with weinberg s power - counting @xcite . matrix elements of these operators are taken between wave - functions generated with the best modern nn potentials @xcite . this method is somewhat _ ad hoc _ , but it has been demonstrated to be convergent where it has been tested . a calculation of tritium @xmath69-decay in such a framework leads to @xcite @xmath176 . chen , heeger and robertson @xcite have surveyed the experimental and theoretical constraints on @xmath177 and have also provided a model - independent determination of this quantity from the sno and super - kamiokande data , finding @xmath178 . the cross sections for the processes @xmath179 , @xmath180 , and @xmath181 are all given in terms of @xmath182 , in refs . @xcite , and analytic expressions for each can be found there . as these expressions are quite complicated , we do not reproduce them here . numerical values of the cross sections for these processes can be found in refs . @xcite . as an example , at @xmath183 the cross sections are , in units of @xmath184 , @xmath185 where @xmath158 . thus , a determination of @xmath168 at the @xmath186 level translates into an uncertainty in these cross sections at the few percent level . in the presence of a background weak field of the form given in eqs . ( [ eq : wkbkgd ] ) and ( [ eq : wkeft ] ) , the @xmath134 component of the @xmath9 channel mixes with the @xmath135 component of the @xmath10 channel , as is the case in the presence of a background magnetic field . the energy - eigenvalues of @xmath150 nn system in a finite volume with this weak field are solutions to @xmath187 \left[\ p\cot\delta_3 -{s_1 + s_2\over 2\pi l } \right ] \ = \ \left [ { gw\ \lonea \over 4 } - { s_1 - s_2\over 2\pi l } \right]^2 \ \ \ , \label{eq : m0solvewk}\end{aligned}\ ] ] where @xmath188 in analogy with the electromagnetic case , the @xmath189 contributions result from the one - nucleon interactions with the background weak field , while the two - nucleon interactions are described by @xmath168 . as in the electromagnetic case , we have explored the spectrum of this system through numerical solution of eq . ( [ eq : s1s2wk ] ) . the spectra of two - nucleons in background weak fields of strengths @xmath190 and @xmath191 are shown in fig . [ fig : levelsmzerowk ] as a function of @xmath87 . in complete analogy with the electromagnetic case , the continuum levels are insensitive to @xmath168 , while there is one level , or two levels for some parameters , that are sensitive to @xmath168 due to the fine - tuning in the nn sector . for lattices with size @xmath192 and @xmath153 the state that is sensitive to @xmath168 is either the lowest state or one of the lowest states , depending upon the choice of field . over a reasonable range of @xmath168 , the energies of the levels that are sensitive to @xmath168 vary by @xmath193 , which should be measurable . = 2.15 in = 2.15 in 0.15 in .2 in it is also worth pointing out that an asymmetric lattice is not necessary to study these weak interactions . it is the existence of landau - levels in the case of a magnetic field that forced us to consider an asymmetric lattice to suppress excitations in the transverse directions . however , for the weak background field in eqs . ( [ eq : wkbkgd ] ) and ( [ eq : wkeft ] ) the transverse eigenstates are momentum eigenstates , and as such a symmetric lattice is sufficient . it is clear from fig . [ fig : levelsmzerowk ] that such lattice simulations of two - nucleons in a background weak field will allow for a determination of @xmath168 . we find this to be quite remarkable . as a final comment on the weak properties of the two - nucleon sector , one could explore the weak - moment of the deuteron . this too will have contributions from the weak - moments of the nucleons , and also from short - distance interactions with the neutral - current gauge - field . by examining the two - nucleon spectrum in the presence of an isoscalar weak field , the short - distance interaction could be isolated from the long - distance contribution and the deuteron weak moment could be determined . this too would be a spectacular accomplishment for lattice in this work we have shown that lattice simulations of the two - nucleon sector in a finite volume and in the presence of background electroweak fields allow for the isolation of the short - distance , multi - nucleon electroweak currents , traditionally called `` meson - exchange currents '' . in lattice qcd simulations one has many more `` knobs '' to turn to get at the physics one is interested in . in particular , the dimensions of the lattice can be chosen to disturb the fine - tuning in the two - nucleon sector in a controlled way to maximize sensitivity to the short - distance electroweak interactions . remarkably , there can be one or two levels in the spectrum that are highly sensitive to these short - distance interactions , while the continuum states show little sensitivity . consequently , for some background fields and lattice dimensions , the short - distance interactions can be isolated from the long - distance interactions . one would like to see a higher - order analysis of these finite volume observables in @xmath4 including the effects of the landau - levels , one- and two - nucleon polarizabilities and other relevant contributions . further , our analysis has been performed exclusively with the pionless eft , @xmath4 . while the result are rigorous where applicable , this work needs to be extended to a larger range of energies and momenta , which requires solving the pionful theory in a finite volume . progress toward this goal is being made in the zero- and single nucleon sectors , but so far little progress has been made in the multi - nucleon sectors . recently , lee , borasoy and schaefer @xcite have performed the first lattice simulations of the pionful eft using weinberg s power - counting , with an eye to a complete numerical solution of multi - nucleon systems . further , encouraging progress has been made toward setting - up lattice simulations in @xmath4 @xcite . the numerical analysis we have presented in this work has been for the physical values of the strong interaction parameters , and consequently , for the physical values of the quark masses . lattice simulations of the immediate future will not be performed at the physical quark masses , be they quenched , partially - quenched or unquenched . it is possible that the scattering lengths in these unphysical simulations will not be unnaturally large , and the pionless theory will not be applicable . therefore , it is more than likely that the pionful theory will need to be solved to more precisely understand how the scattering parameters depend upon the light quark masses , and to directly extract the interactions that result from physics at the cut - off scale of the pionful theory . this work provides the first steps toward extracting the electromagnetic and weak properties of nuclei from lattice qcd simulations . we encourage lattice practitioners to explore this new area and to determine the resources required to perform such calculations . as mentioned earlier , even if infinite computing power were presently available , one knows very little about what to compute in lattice qcd in order to make predictions about nuclear processes . therefore , we encourage nuclear theorists to extend this work and the work of refs . @xcite so that a rigorous computational framework for nuclear physics is in place when the computing power becomes available . we would like to thank silas beane , paulo f. bedaque , david b. kaplan , david lin , assumpta parreno , steve sharpe and matt wingate for numerous discussions . further , we thank the other members of the nplqcd effort for useful input . the energy - levels and formalism associated with two interacting particles on an asymmetric lattice of finite volume , have recently been studied @xcite . we use the asymmetric lattice to suppress nucleon momentum modes in the directions transverse to the background magnetic field in order to extract the electromagnetic counterterm , @xmath44 , i.e. to minimize the contribution from the @xmath194 and @xmath195 interactions which generate landau - levels in an infinite volume . explicit evaluation of the function @xmath78 defined in eqs . ( [ eq : exactee ] ) and ( [ eq : see ] ) can be accomplished in a variety of ways . extending the chowla - selberg relation @xcite to asymmetric lattices , it is easy to show that for @xmath196 @xmath197 where @xmath198 . in order to facilitate evaluation of @xmath78 , and also to determine the coefficients in the expansion of the energy - levels as a function of lattice size , three sums are evaluated for a given lattice asymmetry . the most convergent sum ( @xmath199 in the uv ) is @xmath200 a second sum is @xmath201 \ -\ { 1\over \eta_1\eta_2 x^4 } \ + \ { 1\over \eta_1\eta_2 } \sum_{{\bf n}\ne { \bf 0 } } { 2 x^2 |\tilde { \bf n}|^2 + x^4\over \ \ , \label{eq : jdefee}\end{aligned}\ ] ] and it is important to note that @xmath202 is independent of the value of @xmath203 , and the expression in eq . ( [ eq : jdefee ] ) holds for @xmath196 , and also converges as @xmath199 . the third sum is @xmath204 \nonumber\\ & = & - 2\pi^2 x \ + \sum_{{\bf m}\ne { \bf 0 } } { 1\over |\overline{\bf m}| } e^{-2\pi |\overline{\bf m}| x } \ -\ { 1\over \eta_1\eta_2 x^2 } \ + \ x^2 j_{\eta_1\eta_2 } \ -\ x^4 k_{\eta_1\eta_2 } \nonumber\\ & & \ + \ { x^6\over \eta_1\eta_2 } \sum_{{\bf n}\ne { \bf 0 } } { 1\over |\tilde { \bf n}|^6 \left ( |\tilde { \bf n}|^2+x^2\right ) } \ \ \ , \label{eq : idefee}\end{aligned}\ ] ] which is also independent of the value of @xmath203 , and the expression in eq . ( [ eq : idefee ] ) holds for @xmath196 . the numerical values of these sums are given in table [ table : ijkee ] for different lattice asymmetries @xmath205 . the next continuum level corresponds to non - zero momentum in the longest direction(s ) , e.g. @xmath209 for @xmath210 . for large @xmath87 , the energy of this level for @xmath210 is @xmath211 \ \ \ , \label{eq : c2ee}\end{aligned}\ ] ] where the phase shift is evaluated at @xmath212 , and @xmath213 is a degeneracy factor , given in table [ table : c1c2c1pc2pee ] . again for other choices of @xmath205 , the corresponding expressions are easily derived . values of the coefficients @xmath214 are given in table [ table : c1c2c1pc2pee ] for some choices of @xmath205 . the large values of some of the @xmath215 s in table [ table : c1c2c1pc2pee ] indicate that it is the `` short '' direction(s ) that are setting the scale of the finite - volume corrections , and not the `` long '' direction(s ) . finally we note that there are curves in the @xmath216@xmath217 plane for which @xmath218 separately vanish , as shown in fig . [ fig : c12zeros ] . the point on this plot that is most relevant to our analysis is @xmath219 for which @xmath220 . such `` magic boxes '' may prove to be useful in lattice calculations as the asymptotic formulas in eqs . ( [ eq : deutee])([eq : c2ee ] ) have better convergence properties at smaller @xmath87 than those associated with arbitrary lattice volumes . analogous contours exist for the @xmath221 in the @xmath216@xmath217 plane . s. r. beane , p. f. bedaque , a. parreno and m. j. savage , arxiv : nucl - th/0311027 . s. r. beane , p. f. bedaque , a. parreno and m. j. savage , arxiv : hep - lat/0312004 . l. maiani and m. testa , _ phys . lett . _ * b245 * , 585 ( 1990 ) . q. r. ahmad _ et al . _ [ sno collaboration ] , phys . rev . * 89 * , 011301 ( 2002 ) [ arxiv : nucl - ex/0204008 ] . d. b. kaplan , m. j. savage and m. b. wise , phys . b * 424 * , 390 ( 1998 ) [ arxiv : nucl - th/9801034 ] . d. b. kaplan , m. j. savage and m. b. wise , nucl . b * 534 * , 329 ( 1998 ) [ arxiv : nucl - th/9802075 ] . u. van kolck , nucl . a * 645 * , 273 ( 1999 ) [ arxiv : nucl - th/9808007 ] . j. w. chen , g. rupak and m. j. savage , nucl . a * 653 * , 386 ( 1999 ) [ arxiv : nucl - th/9902056 ] . m. c. birse , j. a. mcgovern and k. g. richardson , phys . b * 464 * , 169 ( 1999 ) [ arxiv : hep - ph/9807302 ] . m. c. birse , j. a. mcgovern and k. g. richardson , arxiv : hep - ph/9808398 . d. b. kaplan , nucl . b * 494 * , 471 ( 1997 ) [ arxiv : nucl - th/9610052 ] . s. r. beane and m. j. savage , nucl . phys . a * 694 * , 511 ( 2001 ) [ arxiv : nucl - th/0011067 ] . d. b. kaplan , m. j. savage and m. b. wise , phys . c * 59 * , 617 ( 1999 ) [ arxiv : nucl - th/9804032 ] . v. olmos de leon _ et al . _ , eur . j. a * 10 * ( 2001 ) 207 . m. lundin _ et al . _ , phys . lett . * 90 * , 192501 ( 2003 ) [ arxiv : nucl - ex/0204014 ] . g. martinelli , g. parisi , r. petronzio and f. rapuano , phys . b * 116 * , 434 ( 1982 ) . c. w. bernard , t. draper , k. olynyk and m. rushton , phys . rev . lett . * 49 * , 1076 ( 1982 ) ; c. w. bernard , t. draper and k. olynyk , nucl . b * 220 * , 508 ( 1983 ) . h. r. rubinstein , s. solomon and t. wittlich , nucl . b * 457 * , 577 ( 1995 ) ; h. r. rubinstein , s. solomon and t. wittlich , nucl . phys . suppl . * 42 * , 358 ( 1995 ) . l. zhou , f. x. lee , w. wilcox and j. christensen , nucl . phys . proc . suppl . * 119 * , 272 ( 2003 ) . m. burkardt , d. b. leinweber and x. m. jin , phys . b * 385 * , 52 ( 1996 ) . h. r. fiebig , w. wilcox and r. m. woloshyn , nucl . b * 324 * , 47 ( 1989 ) . j. christensen , f. x. lee , w. wilcox and l. m. zhou , nucl . * 119 * , 269 ( 2003 ) [ arxiv : hep - lat/0209043 ] . s. aoki and a. gocksch , phys . lett . * 63 * , 1125 ( 1989 ) [ erratum - ibid . * 65 * , 1172 ( 1990 ) ] ; s. aoki , a. gocksch , a. v. manohar and s. r. sharpe , phys . lett . * 65 * , 1092 ( 1990 ) . d. b. kaplan , phys . b * 288 * , 342 ( 1992 ) [ arxiv : hep - lat/9206013 ] . f. fucito , g. parisi and s. petrarca , phys . b * 115 * , 148 ( 1982 ) . s. gsken , arxiv : hep - lat/9906034 . x. li and c. liu , arxiv : hep - lat/0311035 . f. bedaque , arxiv : nucl - th/0402051 . d. b. kaplan and m. j. savage , phys . b * 365 * , 244 ( 1996 ) [ arxiv : hep - ph/9509371 ] . s. r. beane and m. j. savage , nucl . a * 713 * , 148 ( 2003 ) [ arxiv : hep - ph/0206113 ] . s. r. beane and m. j. savage , nucl . a * 717 * , 91 ( 2003 ) [ arxiv : nucl - th/0208021 ] . e. epelbaum , u. g. meissner and w. glockle , nucl . a * 714 * , 535 ( 2003 ) [ arxiv : nucl - th/0207089 ] . e. braaten and h. w. hammer , phys . lett . * 91 * , 102002 ( 2003 ) [ arxiv : nucl - th/0303038 ] . j. w. chen and m. j. savage , phys . c * 60 * , 065205 ( 1999 ) [ arxiv : nucl - th/9907042 ] . g. rupak , nucl . a * 678 * , 405 ( 2000 ) [ arxiv : nucl - th/9911018 ] . mandula , g. zweig and j. govaerts , _ nucl . * b228 * , 91 ( 1983 ) . m. butler and j. w. chen , nucl . a * 675 * , 575 ( 2000 ) [ arxiv : nucl - th/9905059 ] . m. butler and j. w. chen , phys . b * 520 * , 87 ( 2001 ) [ arxiv : nucl - th/0101017 ] . m. butler , j. w. chen and x. kong , phys . c * 63 * , 035501 ( 2001 ) [ arxiv : nucl - th/0008032 ] . s. weinberg , phys . b * 251 * , 288 ( 1990 ) ; nucl . b * 363 * , 3 ( 1991 ) . d. lee , b. borasoy , t. schaefer , [ arxiv : nucl - th/0402072 ] . d. b. kaplan , m. j. savage and m. b. wise , nucl . b * 478 * , 629 ( 1996 ) [ arxiv : nucl - th/9605002 ] . s. r. beane , p. f. bedaque , m. j. savage and u. van kolck , nucl . a * 700 * , 377 ( 2002 ) [ arxiv : nucl - th/0104030 ] . h. m. muller , s. e. koonin , r. seki and u. van kolck , phys . c * 61 * , 044320 ( 2000 ) [ arxiv : nucl - th/9910038 ] . j. w. chen and d. b. kaplan , arxiv : hep - lat/0308016 . s. r. beane and m. j. savage , phys . d * 67 * , 054502 ( 2003 ) [ arxiv : hep - lat/0210046 ] . e. elizalde , _ commun . phys . _ * 198 * , 83 ( 1998 ) , hep - th/9707257 .

determining the presence of widely separated substellar - mass companion is crucial to understand the dynamics of inner planets in extrasolar planetary systems ( e.g. to explain their high mean eccentricity as inner planets are perturbed by the kozai mechanism ) . we report the results of our _ spitzer_/infrared array camera ( irac ) imaging search for widely separated ( 10 25 ) substellar - mass companions for 14 planet - host stars within 15 pc of the sun . using deep 3.6 and 4.5 m observations in subarray mode , we found one object in the field of 47 uma with [ 3.6][4.5 ] color similar to a t5 dwarf , which is , however , unlikely to share common proper motion with 47 uma . we also found three objects with brown - dwarf - like [ 3.6][4.5 ] color limits in the fields of gj 86 , hd 160691 , and gj 581 , as well as another in the field of hd 69830 for which we have excluded common proper motion . we provide model - based upper mass limits for unseen objects around all stars in our sample , with typical sensitivity to 10 m objects from a projected separation of 50 to 300 au from the parent star . we also discuss our data analysis methods for point - spread - function subtraction , image co - alignment , and artifact subtraction of irac subarray images .

1402.0889

large bodies in the periphery of planetary systems have the potential to wreak havoc on the orbits of inner planets through secular interactions . @xcite posit this effect to be responsible for the migration of hot jupiters ; @xcite describe how planet - planet scattering can result in moderately eccentric planetary orbits ; and @xcite demonstrate how kozai - type interactions @xcite can result in highly eccentric planet orbits , inward of a large planet or brown dwarf . given that conventional planet formation models , including both core accretion @xcite and gravitational instability @xcite , describe planets forming in mostly circular , well - aligned circumstellar disks , it is important to explain why the mean eccentricity of discovered exoplanets is currently 0.19 , with many of the highest eccentricities found in single - planet systems ( 78% with @xmath4 are single planets ) . if secular effects involving a massive companion are responsible , these companions may be observable , if they have not yet been ejected from the system . @xcite conclude that most planet scattering instabilities should occur on timescales comparable to the planet formation timespan , but do acknowledge that triple planet systems in certain configurations could potentially be quasi - stable on timespans from @xmath5 to @xmath6 years , so it is not impossible to have planet ejection occur in an old system . @xcite found brown dwarf companions @xmath7 au from the primary , and @xcite found hd 3651b , a brown dwarf orbiting 476 au from a star with an eccentric planet . however , the same formation models mentioned above do not explain how such a massive companion could form so far from the parent star : that far - separated substellar companions and highly eccentric planets both exist deserves explanation . one such explanation is that all objects form close to their parent stars and end up in their current configurations through scattering events ( @xcite , @xcite ) . whether causing havoc or not , widely separated companion brown dwarfs are interesting in their own right , as relatively few are known : of the 992substellar - mass companion objects discovered , only 36 ( all with mass @xmath8 ) are farther than 10 au from their parent , and 21 of those lie between 10 - 300 au . in part , this is a selection effect : the timescales necessary to find these objects with either radial velocity ( rv ) or transit methods are prohibitive , and all of the aforementioned brown dwarfs have been detected by direct imaging . ground - based direct imaging searches have been successful in finding gas giant companions ( e.g. @xcite , @xcite , @xcite and @xcite ) , but have lower sensitivity to the coolest t- and y - dwarfs ( which have distinctive [ 3.6]@xmath2[4.5 ] colors ) because of telluric absorption in the thermal infrared . space telescopes get around this problem , and wise has been successful in finding many field t- and y - dwarfs @xcite , but is less sensitive to brown dwarfs around stars because of the high contrast ratios involved . studies like @xcite , @xcite , and @xcite have searched for stellar - mass companions to planet - host stars , but are generally not sensitive to brown - dwarf - mass objects . the inner working angle of any direct - imaging companion search is limited by the brightness of the parent star compared to its companion . at optical wavelengths , this is determined by the contrast ratio between the parent star s emission and the companion s reflected light . this can be problematic at close angular distances or if the starlight saturates the detector . however , the contrast ratio in the infrared ( e.g. at 3.6 @xmath1 m ) , where thermal emission dominates over reflected light , is much more manageable . this lower ratio opens up the possibility of searching for planetary mass companions with advanced point spread function ( psf ) subtraction techniques , provided that the psf is both stable and well spatially sampled . for these reasons , we have used the infrared array camera ( irac , @xcite ) onboard the spitzer space telescope @xcite to search for companions around 14 nearby stars . using irac gives the added advantage of being able to characterize t - dwarfs , in particular , using their characteristically red [ 3.6]-[4.5 ] color due to methane absorption bands @xcite . this work builds on our previous irac low - mass companion searches ( e.g. @xcite , @xcite , @xcite ) , which have resulted in the discovery of two brown dwarf companions @xcite , including a t7.5 companion of the exoplanet host star hd 3651 . in @xcite , in particular , we have developed the same technique used in the work presented here , adopting irac shorter frame - time subarray observing mode . this allows us to reduce the primary star saturation and the area with high psf - subtraction residuals , narrowing our inner working angle from 20@xmath0 to 5@xmath0 . building on this previous work , we used irac s subarray mode to probe for @xmath9 companions at separations of 25 to 350 au ( for a star 15 pc distant ) around a sample of 14 nearby planet host stars . herein , we present our target list in section [ targetselection ] , our analysis methods ( section [ dataanalysis ] ) , sensitivity limits ( section [ sensitivitysection ] ) , and companion candidates selected on the basis of their irac colors ( section [ detectedsources ] ) . in section [ discussion ] we discuss our results . because so few brown dwarf companions to exoplanetary systems are known , it would be equally interesting to find a perturbing brown dwarf companion as to find one around an exoplanet system that does not exhibit abnormal eccentricity , or likewise one that is bound to a system known to already have another brown dwarf companion . therefore , we selected 14 target stars to observe ( table [ targetstardata ] ) with only the criteria that they be located within @xmath1015 pc from the sun and have exoplanets detected by radial velocity . this 15 pc distance restriction was chosen to provide the best inner working angle , equivalent to a few tens of au from the target star , and an outer working angle of a few hundreds of au ( due to the subarray field of view ) : a range in which perturbing companions acting through the kozai mechanism are expected . target stars range in age from 0.1 to @xmath11 gyr , estimated from various factors including rotational periods , magnetic activity , galactic velocity , and emission lines ( for more information , see individual notes at the bottom of table [ targetstardata ] ) . lcccl[!htbp ] scr j1845 - 6357 & @xmath12 & @xmath13 & m8.5v & @xmath14 m@xmath3 t6 companion closer than 4.5 au + gj 876 & @xmath15 & @xmath16 & m5.0v & [email protected] , [email protected] , [email protected] , [email protected] + gj 581 & @xmath18 & @xmath19 & m5v & [email protected] , [email protected] , [email protected] , [email protected] + gj 849 & @xmath20 & @xmath21 & m3.5v & [email protected] + gj 436 & @xmath22 & @xmath23 & m3.5v & [email protected] + gj 86 & @xmath24 & @xmath25 & g9v & [email protected] , 1.7@xmath0 wd binary + hd 3651 & @xmath26 & @xmath27 & k0v & [email protected] , t - dwarf companion + 55 cnc & @xmath28 & @xmath29 & g8v & @xmath30 , [email protected] , 0.053 , 0.0002 , [email protected] + hd 69830 & @xmath31 & @xmath32 & g8+v & [email protected] , [email protected] , [email protected] , debris disk , 10@xmath0 candidate m companion + hd 147513 & @xmath33 & @xmath100.4 & g1v & [email protected] , possible 5360 au wd binary + ups and & @xmath34 & @xmath35 & f9v & [email protected] , 0.24 , 0.274 , [email protected] + 47 uma & @xmath36 & @xmath37 & g1v & [email protected] , @xmath38 , @xmath39 + hd 160691 & @xmath40 & @xmath41 & g3iv - v & [email protected] , [email protected] , [email protected] , [email protected] + 51 peg & @xmath42 & @xmath43 & g2.5iva & hot jupiter with @xmath44 + among the selected targets , there are 5 m , 1 k , 7 g , and 1 f stars . all are hosts to planets detected by radial velocity ( eccentricities listed in table [ targetstardata ] ) , with the exception of gj 436 , whose planet was discovered by transit . hd 69830 also has a debris disk @xcite and candidate m - dwarf companion @xcite . gj 86 is binary with a white dwarf ( @xmath102@xmath0 ; @xcite ) , and scr j1845 - 6357 , itself a brown dwarf , has a t - dwarf companion detected by direct imaging at a separation of @xmath101.2@xmath0 @xcite . both companions are below the resolving power of irac . hd 147513 is also binary with a white dwarf @xcite , which , at a projected separation of 5360 au , is outside our field of view . nine of the targets have at least one planet with @xmath45 , which may be an indication of the presence of a perturbing companion . gj 876 has two high - eccentricity planets inward of a 2.3 m@xmath3 planet , which may be an example of the type of secular interactions we are exploring @xcite . included in the sample is hd 3651 , whose planet has the highest eccentricity of our sample ( e=0.63 , @xcite ) , and around which we previously discovered a t - dwarf companion at a projected separation of 480 au ( 43@xmath0 , see @xcite ) . this system may be an example of the type of secular interactions we seek , although @xcite refit the rv data of hd 3651 and claim that instead of one high - eccentricity planet , there are likely two lower - eccentricity ( [email protected] and [email protected] ) planets in a 2:1 period commensurability . similarly , @xmath46 and c and d have high eccentricities ( e=0.24 , e=0.274 , @xcite ) , which @xcite explains as due to a 3:1 resonance between @xmath46 and d and e. @xmath46 and e has a mostly circular orbit , so an external perturber may not be necessary to explain the dynamics of this system . finally , gj 436 has one planet with moderate eccentricity ( e=0.15 , @xcite ) . this eccentricity was revalidated by @xcite and attributed by @xcite to an unseen 22.5m@xmath47 object 0.0285 au from the star . if that object exists , we would not be able to resolve it . ten of our targets were previously observed with irac @xcite to search for brown dwarf companions at large separation ( projected separation of @xmath10100 to @xmath101500 au ) . as these observations had frame times of 30 sec , they heavily saturated the primary star , limiting our inner working angle to 20@xmath0 from the target ( see @xcite ) . for this reason we chose to observe our target stars in subarray mode , allowing for dramatically shorter exposure times and restricting saturation to the innermost one or two pixels of the brightest sources . our observations were executed between sept 2007 and aug 2008 ( pid 40976 ) , and were modeled on our previous observations of @xmath48 eridani and fomalhaut ( pid 30754 , @xcite ) , adapted to the lower brightness of the selected targets . we selected frame times for k , g , and f stars were 0.1 seconds ( 0.08 sec integration time ) , and 0.4 sec ( 0.32 sec integration time ) for m stars . each subarray image is 32@xmath4932 pixels , 1.22@xmath0/pix , and is composed of 64 exposures repeated for each dither position . we used the standard irac nine - point reuleaux dither pattern to obtain non - redundant spatial sampling to build up a total integration time of 921.6 sec per target , with 5 or 20 images ( 320 or 1280 exposures ) per pointing for the 0.4 and 0.1 sec exposures , respectively . the large number of exposures was required to increase the signal - to - noise ratio of potential companions . the total overlap area of the exposures is @xmath50 . only 3.6 @xmath1 m and 4.5 @xmath1 m observations were needed because t- and y - dwarfs have characteristically red [ 3.6]@xmath2[4.5 ] colors , due to the presence of methane absorption bands near 3.3 @xmath1 m @xcite . the total integration time was set to allow the detection of 5 gyr old , 5 m@xmath3 planetary mass bodies at 10 pc from the sun , based on the predicted photon noise from the psf of the primary star @xcite . this was our goal , but we had to compromise for an integration time that resulted in lower sensitivity . once the psf was subtracted , however , the detection limit in the innermost 5@xmath0 was not photon noise , but rather the residual noise from subtracting the psf ( see figure 1 in @xcite ) . outside of 5@xmath0 , these residuals diminish and electronic artifacts dominate the noise profile . this issue was already encountered in ( * ? ? ? * see section 2.1 ) . in this work we have improved our artifact removal technique significantly in order to reach higher sensitivity . as the goal of this project is the detection of faint companions around bright stars , our data reduction procedure is designed to combine the individual exposures into a single high - dynamic - range final image , while suppressing the light from the primary star with an accurate psf subtraction . other direct imaging searches have employed the loci algorithm @xcite or principal component analysis ( @xcite ; @xcite ) for psf subtraction . both of these methods use psf segmentation to compensate for instabilities in the psf by using a library of psfs produced with a large number of roll angles . we have an insufficient number ( only one ) of roll angles to use these methods . furthermore , the thermal stability of irac s psf makes them less necessary , and electronic artifacts are the strongest source of noise , above the psf subtraction residual noise in most of our field of view . instead , our data reduction technique builds upon the work of @xcite . we started from basic calibrated data ( bcd ) images from the spitzer heritage archive , processed with the irac pipeline version s18.18.0 . we then wrote our own custom procedure to combine the nine individual frames from each dither position into a final image , for each star , in each irac band . standard mosaicing procedures are not suitable for our targets because , due to the small field of view of the subarray fields , the lack of background stars in the individual bcds prevents accurate world coordinate system ( wcs ) alignment of individual frames by the irac pipeline coordinate refining routine . as a consequence , mosaics made with the standard spitzer science center mopex software @xcite would be blurred . as mentioned in section [ targetselection ] , for subarray , electronic artifacts dominate the noise profile of psf - subtracted images outside of approximately @xmath51 from the center of the psf . these electronic artifacts ( shown in figure [ electronicartifacts ] ; details provided in section [ elec ] ) are produced by bright stars falling on the detector array . the response of the array to bright sources differs from pixel to pixel , so these artifacts need to be characterized and removed in situ , before the nine dither positions are shifted and co - added for sub - sampling . m. field of view is @xmath52.[electronicartifacts ] ] two different psf subtractions were necessary . the first was performed to remove the light from the central bright star in order to better characterize the electronic artifacts . once characterized , the artifacts were removed from the bcds , allowing a second , cleaner psf subtraction to be performed to characterize the stars and search for companions . the irac psf is stable , but position dependent , differing in shape across the focal plane due to light - path geometry , with the most extreme differences near the edges , where the subarray field is located . because electronic artifact characterization requires a precise psf subtraction at each dither position , and since there is not a reliable , high s / n ratio spitzer irac psf available for subarray mode , we created our own psf for each star , using the other stars in our sample , on a per - dither - position basis . figure [ process ] demonstrates our data reduction procedure . in order to characterize , pixel by pixel , the pattern of the electronic artifacts , we subtracted the psf from each bcd image ( figure [ process ] , panel a ) on the original 32@xmath4932-pixel grid . in @xcite , we used a psf made from archival data ( pid 30666 ) , but this was not ideal due to the different dither pattern used in those observations . for this analysis , we created a psf for each dither position from the other images in our data set . each image was aligned and stacked at the pixel size of the original 32@xmath4932 subarray grid to create a psf ( figure [ process ] , panel b ) for each frame . m. ( a ) 99%-saturation - masked image of gj 581 , single dither position . ( b ) psf created for the same dither position . ( c ) combined electronic artifact correction matrix . ( d ) psf- and artifact - subtracted image . hot pixels in the lower right of each image have been masked off . a detected source is visible to the lower left of gj 581 in frames ( a ) and ( d ) . field of view is @xmath52 . [ process ] ] irac starts to become non - linear at 10,000 and 12,000 dn in bands 1 and 2 , respectively @xcite , so we masked off any pixels higher than 99% of those threshold values . the fwhm of irac s 3.6 @xmath1 m psf is 1.66@xmath0 , or 1.36 native pixels ; assuming the core to be gaussian , a 0.1-pixel alignment error affects up to 11% of the image flux . the wcs coordinate uncertainty for our images is on the order of 1@xmath0 , roughly 0.8 pixels , making the wcs coordinates provided with the frames too imprecise for sub - pixel alignment of the individual frames . as previously mentioned , wcs coordinates could not be refined due to a lack of field objects in the subarray images . a simple 2-d centroid of each image was also inadequate for alignment purposes : the psf core changes shape even for small shifts of the stars on the pixel grid , due to irac under - sampling , pixel - phase effects , and position - dependent focus of the camera @xcite . instead , we calculated the stationary points of the numerical gradient of the core of the psf ( its not - saturated part ) along three evenly spaced ( 120 degrees ) axes and used the intersection of those gradients as the center coordinate for each image , giving a typical alignment precision ( average error ) in 3.6 @xmath1 m of 0.0027@xmath0 and 0.0028@xmath0 in 4.5 @xmath1 m . for comparison , using a centroid for alignment of a 1.22@xmath0/pix image produced an average error of 0.07@xmath0 in 3.6 @xmath1 m and 4.5 @xmath1 m . @xmath46 and was too saturated for this technique to work and required manual adjustment . using these coordinates to align the images together , we created a psf for each dither position of each star out of the other twelve star images in our data set ( omitting the star itself and 55 cnc , which has strong horizontal artifacts and was therefore not included in any psf ) . we normalized each star image to the target image by finding a multiplicative scale factor and a pedestal offset between them . the scale factor was initially estimated as the ratio of the total fluxes of the two images ( this would be incorrect to use as a final value because saturation and field star light differ among images ) , while the initial offset was set to zero . using these initial values , we calculated the slope of the radial profile of the difference of the two images . a subtraction with perfect normalization would have zero slope , and the slope is linearly related to the scale factor . after two more scale factor best guesses " , their respective subtraction slopes were used as the basis for a linear regression to calculate the final scale factor between the two images . using this scale factor , we performed a similar calculation to obtain the linear offset between each image and the target . the resulting aligned , normalized frames were stacked together . any pixels with values greater than @xmath53 above the median were discarded , and the mean of the remaining pixels became the psf for that dither position of that target star ( figure [ process ] , panel b ) . by scaling our model psf to have the same irac flux as vega , we were able to use this same technique to calculate the photometry of each of our target stars ( results given in table [ targetphot ] ) . lccccccccccc[t ] scr j1845 - 6357 & 9.54 & 8.97 & 8.51 & 2007 - 10 - 22 & [email protected] & [email protected] + gj 876 & 5.93 & 5.35 & 5.01 & 2007 - 12 - 23 & [email protected] & [email protected] + gj 581 & 6.71 & 6.10 & 5.84 & 2007 - 09 - 08 & [email protected] & [email protected] + gj 849 & 6.51 & 5.90 & 5.59 & 2007 - 11 - 25 & [email protected] & [email protected] + gj 436 & 6.90 & 6.32 & 6.07 & 2008 - 06 - 16 & [email protected] & [email protected] + gj 86 & 4.79 & 4.25 & 4.13 & 2007 - 11 - 14 & [email protected] & [email protected] + hd 3651 & 4.55 & 4.06 & 4.00 & 2008 - 08 - 21 & [email protected] & [email protected] + hd 69830 & 4.95 & 4.36 & 4.17 & 2007 - 11 - 25 & [email protected] & [email protected] + hd 14751 & 4.41 & 4.03 & 3.93 & 2007 - 09 - 13 & [email protected] & [email protected] + 47 uma & 3.96 & 3.74 & 3.75 & 2007 - 12 - 28 & [email protected] & [email protected] + 55 cnc & 4.77 & 4.27 & 4.02 & 2007 - 11 - 24 & [email protected] & [email protected] + ups and & 3.18 & 2.96 & 2.86 & 2008 - 03 - 09 & [email protected] & [email protected] + 51 peg & 4.66 & 4.23 & 3.91 & 2007 - 12 - 26 & [email protected] & [email protected] + hd 160691 & 4.16 & 3.72 & 3.68 & 2007 - 09 - 13 & [email protected] & [email protected] + after subtracting the dither - position - dependent psf , electronic artifacts become visible in each frame . there are several effects , shown in figure [ electronicartifacts ] . muxstripe " is a jail - bar " pattern caused by excess charge in a pixel ( from a bright source ) unbalancing the multiplexer ( readout ) channel pedestals . this is superimposed with a repeating decay pattern in the lower half of the image . muxbleed " is a horizontal bleeding of charge along the row(s ) and readout channel(s ) containing excess - charge pixels . pulldown " is a central - column bias shift caused by a pixel with excess charge . we characterized these effects in sequence by creating and subtracting a correction matrix of each artifact before moving to the next : first , we masked off a circle with radius out to 10% saturation from the center of the psf in order to reduce the psf subtraction residuals that get included in our calculations , then removed each artifact type in turn . the muxstripe is a difference in pedestal value per multiplexer channel ( every fourth column of the image is read by a different channel ) , affecting the top and bottom of the image ( above and below the bright source ) differently . similarly , the muxbleed affects every fourth pixel of each of the @xmath103 rows surrounding the core of a bright source differently . both of these effects are characterized by the median value of the group of affected pixels , which we evaluated and then assigned to the corresponding area in the correction matrix . for example , to characterize the segment of the muxstripe in the bottom of the image in the first multiplexer channel , the median of all of the pixels comprising columns 0 , 4 , 8 , ... , 28 from rows 0 through the row containing the central star was assigned to the correction matrix in those locations . this was done similarly for the other channels and for the muxbleed . this matrix was then subtracted from the psf - masked image , to prepare for the next artifact to be removed . the decay artifact does not occur in every image , nor every readout channel , but when it occurs , it again affects every fourth column of the lower half ( below the bright source ) of the image similarly . previously @xcite , we had removed this artifact by fitting an exponential or linear function to each affected column , separately . we discovered , however , that the artifacts are better removed by assuming the pattern to be the same exponential function in every affected column and subtracting the median of the pixels in each row of those columns . affected readout channels were flagged manually and the median of each row in those columns was saved as a new correction matrix . this matrix was subtracted from the muxbleed / muxstriped - removed image . the pulldown always occurs in the same column as the core of the target star , and is different above and below the star . when the core of the target star falls within @xmath100.25 pixels of the boundary between pixels , the pulldown occurs on two adjacent columns . we modeled the pulldown as the median of the column(s ) containing the target star , top and bottom separately , then saved it as a third correction matrix . at this point , we added the previous three correction matrices together to obtain a complete matrix of the pixel - dependent electronic artifacts in each bcd ( result shown in figure [ process ] , panel c ) . subtracting this matrix from the psf - subtracted image gives us a visual check to show all artifacts have been removed ( figure [ process ] , panel d ) . to mitigate contamination from residual artifacts in our psfs , we subtracted the artifact correction matrix for each image from its bcd and remade new psfs from the result . because all pixel dependency had been removed , and because all stars were observed with the same dither pattern , the individual dither images could then be co - added to create sub - sampled images and psfs . we co - added the artifact - cleaned images at @xmath54 subsampling ( @xmath55/pix ) , then created a psf for each star using the process described in section [ 1stpsf ] using the other sub - sampled images ( again omitting 55 cnc and the star itself ) . with nine spatially distinct dither positions , each pixel can be sub - sampled into at most nine ( 3@xmath493 ) sub - pixels , so after subtracting each star s model psf from its final mosaic , we rebinned the images back to @xmath56 sub - sampling ( @xmath57 ) for analysis ( figures [ ch1regions ] & [ ch2regions ] ) . combining these images with those from pid 30666 , we created and released a @xmath58/pix subarray psf , available at the spitzer science center website . once the psf and electronic artifacts are removed and the images are co - added , other sources become readily visible ( see figures [ ch1regions ] & [ ch2regions ] ) . given the small number of sources in each image , we visually inspected each and measured the photometry of every area that seemed brighter than the background , using a custom procedure based on daophot . we used an aperture radius of two irac pixels ( 2.44@xmath0 ) where possible , or one ( 1.22@xmath0 ) for sources near the edge of the final images . sky annuli were chosen to fit the background while avoiding excessive psf subtraction noise for the typical source , and had an inner radius of 3.17@xmath0 ( 1.59@xmath0 ) with an outer radius of 5.12@xmath0 ( 2.56@xmath0 ) for the larger ( smaller ) apertures . aperture corrections were calculated as the difference in magnitudes between the flux inside each aperture listed above versus the flux inside a 10-irac - pixel aperture placed on a vega - scaled psf . these corrections in 4.5 @xmath1 m are @xmath59 mag for the larger aperture , @xmath60 mag for the smaller . any source calculated to have a local signal - to - noise ratio less than 3 was rejected . all sources detected in 3.6 @xmath1 m and 4.5 @xmath1 m are shown figures [ ch1regions ] and [ ch2regions ] , respectively . table [ sourcephot ] shows the aperture - corrected photometry of all detected sources in the frame of each target star . for sources detected in only one band , the limiting magnitude at that location in the other band is given as a photometric upper limit . four objects ( 47 uma-1 , hd 160691 - 8 , gj 86 - 1 , and gj 581 - 3 ) have potential colors that can be as red as a t5 or later brown dwarf , due to non - detections in 3.6 @xmath1 m . the subarray field of view is too small for the analysis pipeline to perform automatic pointing refinement . no subarray frames had sufficient 2mass sources to perform manual astrometric calibration , and our observations were taken before the peak up " function was available . therefore , the accuracy by which the wcs coordinates are known is limited by spitzer s star tracker , plus pointing drift , which is especially important for short exposures like subarray mode , and jitter ; the combination of these is on the order of 1@xmath0 . we performed relative astrometry on each source by computing its centroid and calculating the distance and position angle relative to the centroid of the central source ( table [ sourcephot ] ) . because the psf is pixelated and has internal structure , the centroid can converge in a slightly different location depending on the subsampling of the psf due to the source falling on different parts of a pixel . to calculate the error of our relative astrometry , we randomly shifted our 0.24@xmath0/pix model psf 1000 times , each time rebinning to 0.407@xmath0/pix and comparing the known shift to the difference in measured centroids . the average of the resulting distribution is 0.05 pixels , or 0.02@xmath0 . this represents the best - case radial distance error for high signal - to - noise , isolated sources . for more crowded , dim sources , the convergence point of the centroid also depends on the initial guess . to calculate this error , we randomly picked 1000 starting locations within a radius of 2.1 pixels ( half the fwhm of irac s band 2 at 0.407@xmath0/pix ) around each source , and calculated the centroid using a box size of 6 pix . this produced a distribution of between 1 and 6 centroids . " we report the center of the distribution as the source position and the rms spread as the initial - guess error . this error was on the order of 0.1@xmath0 and was added in quadrature with the psf - discretization error to find our total relative astrometry error , reported per source in table [ sourcephot ] . to supplement and confirm our subarray results , we performed follow - up photometry for all stars that had full - frame images available in the spitzer heritage archive ( listed in full " columns of table [ sourcephot ] ) . these observations were performed with 30-second frame times and 5 gaussian dithers . we used iracproc to create mosaics and subtract the target stars psfs , then we performed aperture photometry . to isolate the typical source from artifacts and other sources of noise , we used a 2.83@xmath0 aperture radius , with a sky annulus from 2.83@xmath0 to 4.72@xmath0 on all sources visible within the subarray - equivalent field of view . we restricted our analysis to this region because everything outside the subarray field of view in the full frame images had already been discounted by a previous proper motion search for widely separated companions without psf subtraction @xcite . several sources detected in subarray images were found to have counterparts in the full - frame images , including sources near the edge of the subarray frame that we originally thought to be noise ( see sources at the edges of hd 160691 , hd 147513 , and scr j1845 - 6357 ) . some sources appearing in full - frame images were below the sensitivity of the subarray images . as expected , sources detected near the target star in subarray were overwhelmed by psf residual noise in the full - frame images , notably hd 147513 - 2 , scr j1845 - 6357 - 2 , and gj 876 - 7 . figure [ hd160691 ] shows a comparison of objects seen in the subarray field of view between subarray and full - frame images for one parent star . taking the lowest - error photometry in each band from either subarray or full frame , and assuming all of these sources to be at the same distance as their parent star , we plotted their [ 3.6]@xmath2[4.5 ] color versus 4.5 @xmath1 m absolute magnitude ( see figure [ hrdiagram ] ) to compare with model substellar - mass objects of various temperatures @xcite and previously detected l and t dwarfs @xcite . we calculated sensitivity limits for each of our targets as a function of projected distance from the central star . our photometric sensitivity at any location within an image is limited by the local noise level around that point . the noise level within the frame is generally highest at the location of the target star and decreases radially outward , but with azimuthal deviations due to psf spikes and other features . we generated azimuthally averaged sensitivity curves , giving the noise - equivalent flux density ( nefd ) for given radial distances from the target star ( figures [ sensitivity ] and [ sensitivity2 ] ) . we calculated the nefd in each point of the psf - subtracted image as the total rms noise flux over an aperture with diameter equal to the fwhm of the stellar psf ( 1.66@xmath0 for 3.6 @xmath1 m , 1.72@xmath0 for 4.5 @xmath1 m ) . overlaid on these plots are the estimated magnitudes of planets with different masses , derived from non - irradiated extrasolar planet models with a given age from @xcite , scaled to the distance of the primary star . the only models available from @xcite for ages comparable to our objects are 1 , 3 , and 5 gyr , so we chose the model age closest to the age of each star . for ages greater than 5 gyr , we used the 5 gyr model , though this underestimates the mass of potential companions . the age we used for the model is listed with the star s estimated true age above each plot . also available were the bt - settl models by @xcite , which have a greater range of available ages , but much higher lower limit of @xmath61 m@xmath3 for all but the youngest isochrones , making them insufficient to match our sensitivity . we compared the burrows and allard models for the same object age and found they give comparable magnitudes for a given object mass . for the smallest bt - settl object available for the 10 gyr isochrone , 63 m@xmath3 , the difference in 4.5 @xmath1 m model magnitude from 5 to 10 gyr is 0.42 . less - massive objects should have lost more of their primordial heat by the time they reach 5 gyr old , thus they would be expected to dim by less than this amount over the following 5 gyr , so we feel confident that we are not significantly overestimating our mass sensitivity for objects older than 5 gyr . our best sensitivity occurs for nearby , young systems like scr j1845 - 6357 , where we detected no @xmath625 m@xmath3 objects at projected separations between 25 - 80 au from its parent , and gj 876 , for which our limit is of a @xmath626 m@xmath3 object at projected separations from 50 - 100 au . for the older , more distant stars like @xmath46 and , 55 cnc , and 51 peg , we detected no objects greater than 25 m@xmath3 between projected separations of roughly 200 - 300 au , with the caution that our mass model sensitivity is overestimated for systems older than 5 gyr . our lower - limit sensitivity around the remaining stars varies from 7 to 15 m@xmath3 in the range of projected separations of 50 to 250 au . 47 uma has three previously confirmed rv planets of roughly jupiter mass @xcite . the outermost , 47 uma d , has the highest eccentricity of the three ( e=0.16@xmath63 ) . this suggested eccentricity level is difficult to explain without an external perturber . our subarray observations revealed 47 uma-1 , a source that appears in 4.5 @xmath1 m only , and has upper limit colors ( [ 3.6]-[4.5]@xmath64 and m[4.5]=14.7 ) comparable to a brown dwarf at the distance of the primary . if the object is a bound companion , it would be located at a projected separation of @xmath10300 au ( 22@xmath0 ) from 47 uma . if it is not bound , its color indicates it may be a background red giant or galaxy ( @xcite ; @xcite ) . we found archival full frame observations of 47 uma , performed apr 20 , 2004 as part of pid 347513231 . 47 uma-1 was detected in the full - frame images but is unfortunately directly on top of a strong pulldown artifact in both bands . full frame photometry confirms this object to be more than a magnitude red in [ 3.6]-[4.5 ] . assuming it to be at the same distance as 47 uma gives it [ 3.6]-[4.5 ] vs. m[4.5 ] coordinates matching a t5 dwarf , with a model mass of @xmath65 ( see figure [ hrdiagram ] ) . another full - frame observation of 47 uma ( 4.5 @xmath1 m only ) was made dec 23 , 2008 ( pid 332319740 ) , so we attempted to verify common proper motion . we measured the proper motion of 118 sources surrounding 47 uma to compare to the motion of source 1 and the expected position change of the primary over the 4.67 years between the two observations ( figure [ propermotion ] ) . the measured position change of 47 uma-1 is more than 3@xmath66 from that expected of 47 uma , making it unlikely to be a co - moving companion . the other source in the same frame , 47 uma-2 , was affected too much by artifacts in both full frame archival epochs to obtain reliable photometry or proper motion . undetected objects at projected separations between 200 - 350 au are constrained to have a model dependent mass less than 15 m@xmath3 , with the caveat that we used the 5 gyr model for this 7.4-gyr - old star . the four known rv planets of the hd 160691 system all have [email protected] @xcite . @xcite noted that this system has a 2/1 period commensurability , which could indicate mean motion resonance , which in turn could explain the eccentricities of the system without invoking an external body . we found one source , hd 160691 - 8 , with only an upper limit color , potentially similar to a @xmath10t7 dwarf with mass near 25 m@xmath3 . if a bound companion , it would be located at a projected separation of @xmath10300 au ( 21@xmath0 ) from hd 160691 . in the full - frame images taken sept 9 , 2004 ( pid 240828495 ) , this source lies directly on top of psf spike residuals ( see figure [ hd160691 ] ) , so we were not able to calculate its full - frame photometry . we found a model - dependent upper mass limit of 25 m@xmath3 for unseen objects at a projected distance between 200 and 400 au from the primary . of the four known rv planets around gj 581 , the outermost two , gj 581d and gj 581e , have e=0.205 and e=0.32 , respectively @xcite . @xcite claims that the existence of planet d is questionable but confirms b , c , and e. planet e is the least massive of the four ( respectively , they are 0.05 m@xmath3 , 0.017 m@xmath3 , 0.019 m@xmath3 , and 0.0061 m@xmath3 ) , so it may be likely that some earlier scattering event led to its current orbit . if gj 581 - 3 is a bound object , its 4.5 @xmath1 m flux implies it would be a y dwarf , with model mass of 7 - 10 m@xmath3 , located at a projected separation of @xmath10150 au ( 23@xmath0 ) from gj 581 . we found no objects with mass @xmath68 between projected separations of 75 - 150 au from the primary . our mass limits for this star are underestimates : our oldest mass model was 5 gyr and this star has an age of 8.5 gyr . gj 581 had no available archival full - frame observations . gj 86 has one marginally detected source in the very corner ( 26@xmath0 from center ) of the 4.5 @xmath1 m subarray frame . this source is not detected at 3.6 @xmath1 m , due to strong psf - subtraction residuals caused by the irac filter ghost . " nevertheless , the limiting sensitivity at that location ( [ 3.6]@xmath6214.94 ) implies a color [ 3.6]-[4.5]@xmath620.9 . if the source is a bound companion at the distance of the primary , its 4.5 @xmath1 m magnitude and color limit would be consistent with a t7 dwarf however , the high level of noise , even in the 4.5 @xmath1 m frame , suggests the source to be a spurious detection . we attempted to measure full - frame photometry of gj 86 - 1 , but both its wcs - coordinate location and where it would be if it shared common proper motion with gj 86 are overwhelmed by the pulldown in the full frame . the only source detected near hd 69830 is cut off at the edge of the frame . its estimated photometry puts it at the [ 3.6]-[4.5 ] color boundary between stars and t - dwarfs ( figure [ hrdiagram ] ) . there were no full - frame archival observations of this object in order to obtain better irac photometry , but it does have 2mass photometry available . hd 69830 has visibly moved between the 2mass epoch and our observations taken in nov 2007 , but hd 69830 - 1 has not , so they do not share common proper motion . @xcite reported a candidate m companion around hd 69830 , with relative coordinates ( pointing angle , separation ) of -152.5@xmath69 , 9.99@xmath0 and k=15.46 . assuming its k magnitude to be similar to [ 3.6 ] , this object is just below our sensitivity at that position ( see figure [ sensitivity ] ) . using the combined photometry from our subarray images and the available archival full - frame images , we found four potentially interesting candidates : one with matching brown dwarf colors and three objects with compatible [ 3.6]@xmath2[4.5 ] color lower limits . in absence of common proper motion confirmation , we are unable to determine if these candidates are true companions , or unrelated background quasars or mass - losing giants ( see @xcite ; @xcite for typical colors of background red sources ) . this ambiguity could , in principle , be resolved by acquiring jhk photometry ( see @xcite ) , albeit new near - ir data capable of detecting these objects would also likely provide accurate astrometry , sufficient to test for common proper motion . we presented details of our psf- and artifact - subtraction procedure , in which we were able to automatically co - align images within 0.0027@xmath0 for purposes of psf subtraction and effective removal of electronic artifacts . as anticipated , subarray observations allowed a smaller inner working angle than full - frame observations of the same field of view , which were dominated by psf subtraction residuals . for all stars , we calculated upper limits on unseen t- or y - dwarfs within the mass and projected semi - major axis ranges given in figures [ sensitivity ] and [ sensitivity2 ] . outside of 10@xmath0 from most sources , our sensitivity is in the range of 10 m@xmath3 . our best sensitivity is for close , young stars like gj 849 ( 5 m@xmath3 ) and scrj 1849 - 6357 ( @xmath70 ) . our worst sensitivity ( 25 m@xmath3 ) was for @xmath46 and due to its brightness : artifacts overwhelmed the image . we could not confirm the existence of any new wide planetary or brown dwarf companions around the stars in our sample . we have ruled out a section of parameter space , but there is a degeneracy for perturbers between mass and semi - major axis . it is still possible for less - massive companions closer than our inner working angle to exist ( ongoing systematic searches with new ground - based adaptive optics systems will explore the remaining parameter space ) , or there could be some other reason for the measured high eccentricities of planets in those systems . follow up observations are necessary for hd 160691 - 8 , gj 86 - 1 , and gj 581 - 3 to obtain 3.6 @xmath1 m photometry and check for common proper motion . if the candidates are confirmed , numerical simulations will be necessary to determine whether they can account for the high eccentricities of planets in their respective systems . thanks to brett kail for helping to brainstorm ideas on image co - alignment . this work is 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 . support for this work was provided by nasa through an award issued by jpl / caltech . , m.m . , and j.c . were partially supported by the u.s . national science foundation under award no . j.c . received support from the south carolina space grant consortium and the research corporation for science advancement ( award no . 21026 ) . patten , b. m. , luhman , k. l. , hora , j. l. , et al . 2005 , protostars and planets v , proceedings of the conference held october 24 - 28 , 2005 , in hilton waikoloa village , hawaii . lpi contribution no . 1286 , p.8042 . lcccccccccccc gj 876 & 1 & 22:53:18.216 & -14:15:32.80 & [email protected]@xmath0 & [email protected]@xmath69 & & & & & & [email protected] & @xmath71 + gj 876 & 2 & 22:53:18.254 & -14:16:04.80 & [email protected]@xmath0 & [email protected]@xmath69 & & & & & [email protected] & [email protected] & [email protected] + gj 876 & 3 & 22:53:16.760 & -14:16:14.49 & [email protected]@xmath0 & [email protected]@xmath69 & & & & & & [email protected] & [email protected] + gj 876 & 4 & 22:53:16.049 & -14:16:07.01 & [email protected]@xmath0 & [email protected]@xmath69 & & & & & & [email protected] & [email protected] + gj 876 & 5 & 22:53:16.296 & -14:15:45.50 & [email protected]@xmath0 & [email protected]@xmath69 & & & & @xmath72 & @xmath73 & [email protected] & [email protected] + hd 147513 & 1 & 16:24:02.914 & -39:11:46.43 & [email protected]@xmath0 & [email protected]@xmath69 & 13.06 & 12.84 & 13.20 & [email protected] & [email protected] & [email protected] & [email protected] + hd 147513 & 2 & 16:24:01.805 & -39:11:23.93 & [email protected]@xmath0 & [email protected]@xmath69 & & & & [email protected] & [email protected] & & + hd 147513 & 3 & 16:23:59.686 & -39:11:22.02 & [email protected]@xmath0 & [email protected]@xmath69 & 13.10 & 13.18 & 14.31 & [email protected] & [email protected] & [email protected] & [email protected] + hd 147513 & 4 & 16:24:02.338 & -39:11:47.15 & [email protected]@xmath0 & [email protected]@xmath69 & & & & [email protected] & [email protected] & & + hd 147513 & 5 & 16:24:00.295 & -39:11:55.39 & [email protected]@xmath0 & [email protected]@xmath69 & & & & & & [email protected] & [email protected] + hd 147513 & 6 & 16:24:01.641 & -39:11:16.38 & [email protected]@xmath0 & [email protected]@xmath69 & & & & @xmath74 & @xmath75 & [email protected] & [email protected] + hd 147513 & 7 & 16:24:03.070 & -39:11:36.31 & [email protected]@xmath0 & [email protected]@xmath69 & & & & & & [email protected] & [email protected] + hd 147513 & 8 & 16:24:02.031 & -39:11:59.26 & [email protected]@xmath0 & [email protected]@xmath69 & 13.57 & 15.18 & 14.80 & & & [email protected] & [email protected] + hd 147513 & 9 & 16:24:03.240 & -39:11:16.37 & [email protected]@xmath0 & [email protected]@xmath69 & & & & [email protected] & @xmath76 & [email protected] & [email protected] + 47 uma & 1 & 10:59:26.441 & + 40:26:05.10 & [email protected]@xmath0 & [email protected]@xmath69 & & & & @xmath77 & [email protected] & [email protected] & [email protected] + 47 uma & 2 & 10:59:28.980 & + 40:26:04.88 & [email protected]@xmath0 & [email protected]@xmath69 & & & & [email protected] & [email protected] & @xmath78 & [email protected] + hd 160691 & 1 & 17:44:09.617 & -51:49:42.35 & [email protected]@xmath0 & [email protected]@xmath69 & 12.83 & 13.07 & 14.54 & [email protected] & [email protected] & [email protected] & [email protected] + hd 160691 & 2 & 17:44:09.403 & -51:50:20.54 & [email protected]@xmath0 & [email protected]@xmath69 & & & & [email protected] & [email protected] & [email protected] & [email protected] + hd 160691 & 3 & 17:44:07.754 & -51:50:16.37 & [email protected]@xmath0 & [email protected]@xmath69 & & & & [email protected] & [email protected] & [email protected] & [email protected] + hd 160691 & 4 & 17:44:06.403 & -51:50:09.53 & [email protected]@xmath0 & [email protected]@xmath69 & & & & & & [email protected] & [email protected] + hd 160691 & 5 & 17:44:06.667 & -51:50:19.21 & [email protected]@xmath0 & [email protected]@xmath69 & 12.63 & 12.43 & 12.77 & & & [email protected] & [email protected] + hd 160691 & 6 & 17:44:10.234 & -51:50:20.17 & [email protected]@xmath0 & [email protected]@xmath69 & & & & & & [email protected] & [email protected] + hd 160691 & 7 & 17:44:08.602 & -51:49:44.83 & [email protected]@xmath0 & [email protected]@xmath69 & & & & [email protected] & @xmath79 & & + hd 160691 & 8 & 17:44:10.481 & -51:50:15.79 & [email protected]@xmath0 & [email protected]@xmath69 & & & & @xmath80 & [email protected] & & + scr j1845 - 6357 & 1 & 18:45:06.809 & -63:57:59.44 & [email protected]@xmath0 & [email protected]@xmath69 & & & & [email protected] & [email protected] & [email protected] & [email protected] + scr j1845 - 6357 & 2 & 18:45:10.488 & -63:57:36.86 & [email protected]@xmath0 & [email protected]@xmath69 & & & & [email protected] & [email protected] & [email protected] & [email protected] + scr j1845 - 6357 & 3 & 18:45:06.698 & -63:57:55.51 & [email protected]@xmath0 & [email protected]@xmath69 & & & & [email protected] & [email protected] & [email protected] & [email protected] + scr j1845 - 6357 & 4 & 18:45:08.546 & -63:57:30.38 & [email protected]@xmath0 & [email protected]@xmath69 & & & & [email protected] & [email protected] & [email protected] & [email protected] + scr j1845 - 6357 & 5 & 18:45:05.128 & -63:57:56.62 & [email protected]@xmath0 & [email protected]@xmath69 & & & & & & [email protected] & [email protected] + gj 86 & 1 & 02:10:29.273 & -50:48:58.00 & [email protected]@xmath0 & [email protected]@xmath69 & & & & @xmath81 & [email protected] & & + @xmath46 and & 1 & 01:36:49.806 & + 41:24:12.36 & [email protected]@xmath0 & [email protected]@xmath69 & & & & & & [email protected] & @xmath82 + @xmath46 and & 2 & 01:36:46.562 & + 41:23:55.39 & [email protected]@xmath0 & [email protected]@xmath69 & & & & & & [email protected] & + gj 581 & 1 & 15:19:25.370 & -07:43:34.50 & [email protected]@xmath0 & [email protected]@xmath69 & 14.75 & 14.12 & 13.87 & [email protected] & [email protected] & n / a & n / a + gj 581 & 2 & 15:19:25.109 & -07:43:13.73 & [email protected]@xmath0 & [email protected]@xmath69 & & & & [email protected] & @xmath83 & n / a & n / a + gj 581 & 3 & 15:19:26.614 & -07:43:44.26 & [email protected]@xmath0 & [email protected]@xmath69 & & & & @xmath84 & [email protected] & n / a & n / a + gj 581 & 4 & 15:19:26.796 & -07:43:45.19 & [email protected]@xmath0 & [email protected]@xmath69 & & & & & @xmath85 & n / a & n / a + gj 849 & 1 & 22:09:42.319 & -04:38:25.98 & [email protected]@xmath0 & [email protected]@xmath69 & & & & [email protected] & [email protected] & n / a & n / a + gj 436 & 1 & 11:42:11.213 & + 26:42:32.33 & [email protected]@xmath0 & [email protected]@xmath69 & & & & [email protected] & [email protected] & n / a & n / a + hd 69830 & 1 & 08:18:22.421 & -12:38:13.85 & [email protected]@xmath0 & [email protected]@xmath69 & 13.93 & 13.32 & 13.18 & [email protected] & [email protected] & n / a &

the extraordinarily rapid evolution of the born - again giant sakurai s object following discovery in 1996 has been investigated . the evolution can be traced both in a continued cooling of the stellar surface and dramatic changes in chemical composition on a timescale of a mere few months . the abundance alterations are the results of the mixing and nuclear reactions which have ensued due to the final he - shell flash which occurred during the descent along the white dwarf cooling track . the observed changes in the h and li abundances can be explained by ingestion and burning of the h - rich envelope and li - production through the cameron - fowler mechanism . the rapidly increasing abundances of the light-elements ( including sc ) is consistent with current-processing by neutrons released from the concomitantly producedc . however , the possibility that the-elements have previously been synthesized during the agb - phase and only mixed to the surface in connection with the final he - shell flash in the pre - white dwarf can not be convincingly ruled out either . since sakurai s object shows substantial abundance similarities with the rcrb stars and has recently undergone rcrb - like visual fading events , the `` birth '' of an rcrb star may have been witnessed for the first time ever . sakurai s object thus lends strong support for the suggestion that at least some of the rcrb stars have been formed through a final he - shell flash in a post - agb star .

astro-ph9811208

normally stellar evolution proceeds on time - scales of millions or billions of years . notable exceptions are e.g. supernovae , though such cataclysmic events only mark the end point of the evolution as normal stars . witnessing stellar evolution in real time without a simultaneous complete destruction of the star is thus very rare . `` born - again '' giants do , however , offer such a remarkable opportunity , which offers important windows through which stellar nucleosynthesis and evolution can be glimpsed . on its initial descent of the white dwarf cooling track , a star may be ballooned back to supergiant dimensions by a final he - shell flash which is triggered by the compressional heating of the stellar interior . rather than fading steadily to oblivion as a white dwarf the star makes a second appearance as a luminous giant , a `` born - again '' giant . theoretical estimates suggest that about 10% of all low- and intermediate mass stars going through a planetary nebula phase will experience such a final he - shell flash ( renzini 1990 ; iben et al . the observational rarity of such events is the result of the very short life - time as a born - again giant typically 100 - 1000 years , depending mainly on the stellar envelope mass before it completes its loop in the hertzsprung - russell diagram and once again starts contracting to become a white dwarf , this time for good . to date , only three stars are believed to have been observed going through a born - again phase : v605aql ( nova aql 1919 ) , fgsge and the recently discovered sakurai s object ( v4334sgr ) ( nakano et al . 1996 ) . a further handful of stars ( e.g. abell 30 and abell 78 ) presently in a seemingly second stage as planetary nebulae may be identified as having recently been born - again giants . furthermore , the rcrb variables could also possibly be born - again giants ( asplund et al . 1997b ; asplund et al . 1998 ; lambert et al . 1998 ) ; their h - deficiency can be explained if the final he - shell flash occurs after the outer h - burning shell has been extinguished , in which case the convection zone due to the he - flash may ingest and burn the h - rich envelope . both v605aql , fgsge and sakurai s object have shown a remarkably fast stellar evolution . from having been an o star at the end of last century ( van genderen & gautschy 1995 ) , fgsge has continued to brighten in the visual while cooling , presumably the result of the expansion following the final he - shell flash ( see kipper 1996 for a recent review ) . furthermore , an enrichment of @xmath0-process elements which took place within seven years has been reported ( langer et al . 1974 ) . lately , fgsge has also started to show rcrb - like variability ( papousek 1992 ) , which may classify the star as a `` new - born '' rcrb star , in particular since there are hints that the star may be h - deficient ( gonzalez et al . 1998 ) . after discovery in 1996 , sakurai s object has shown an exceedingly rapid evolution , even out - pacing fgsge ; whether or not it is evolving faster than v605aql is difficult to judge considering the scarceness of data from the outburst in 1919 but the evolutionary speed for v605aql may have been comparable ( clayton & de marco 1997 ) . like fgsge sakurai s object has cooled significantly ( asplund et al . 1997b ; duerbeck et al . more impressively though , its chemical composition seems to have been altered drastically on a time - scale of a mere few months ( asplund et al . fgsge s title as the `` fastest evolving star ever identified '' ( van genderen & gautschy 1995 ) is certainly called in doubt by sakurai s object . sakurai s object offers a unique opportunity to study stellar evolution and attendant nucleosynthesis in real time . several papers has been devoted to the star and its remarkable evolution ( e.g. asplund et al . 1997b ; duerbeck & benetti 1996 ; duerbeck et al . 1997 ; eyres et al . 1998 ; jacoby et al . 1998 ; kimeswenger & kerber 1998 ; kipper & klochkova 1997 ; shetrone & keane 1998 ) . however , the published abundance analyses of the star have relied on very different model atmospheres and spectral features , making a direct comparison of the results difficult . therefore we have re - analysed the available published ( asplund et al . 1997b ; kipper & klochkova 1997 ; shetrone & keane 1998 ) and unpublished ( d. pollacco and g. wallerstein ) spectra in a homogeneous way , using the same model atmospheres and whenever possible the same spectral lines to avoid as far as possible systematic differences . the resultant map of the evolution of the composition offers important clues to the star s immediate past history . some of the observations on which the present study is based have previously been used by asplund et al . ( 1997b ) , kipper & klochkova ( 1997 ) and shetrone & keane ( 1998 ) for abundance analyses . for completeness we repeat the observational details here , together with the relevant information on the previously unpublished spectra . the spectra of asplund et al . ( 1997b ) cover 3700 - 10150@xmath2 at a resolving power of about 30000 and were obtained with the 2.7 m telescope at mcdonald observatory on may 5 and 6 and on october 7 , 1996 . another spectrum was acquired with the mcdonald 2.1 m telescope on may 9 , 1996 , which covers the region 5720 to 7200 at a resolving power of about 60000 ; the three may spectra have been combined to a single line list . shetrone & keane ( 1998 ) observed sakurai s object on april 20 , 1996 , with ctio 4 m telescope with a resolving power of about 14000 . the spectrum covers the region 3800 - 7500@xmath2 with s / n ranging from 100 to 200 , depending on the wavelength region . in addition g. wallerstein ( unpublished research ) obtained a spectrum of the star with the ctio 4 m telescope at a resolving power of about 12500 on april 25 . the two april spectra have here been combined to one line list though greater weight has been given to the april 20 spectrum due to its higher quality . a spectrum obtained with the william herschel 4.2 m telescope ( pollacco , unpublished research ) was acquired on june 4 , 1996 , at a resolving power of 45000 and s / n of about 70 . the spectrum covers the wavelength region 3800 - 8000 . finally , kipper & klochkova ( 1997 ) obtained spectra of the star on july 3 , 1996 , with the echelle spectrometer on the russian 6 m telescope , providing a spectral resolving power of about 20000 and s / n of around 75 . the wavelength coverage was 5000 - 8000 . in addition , other spectra obtained by us or kindly provided by various colleagues have been attempted for abundance analyses , but in all cases the resolution proved insufficient for reliable results , and they are therefore not included in the present study . with the exceptions of a few spectral regions for which spectral synthesis was carried out ( see below ) , the abundance analysis is based on lines for which equivalent widths could be measured readily . the line lists for the april , june and july spectra were modelled following the lines used by asplund et al . ( 1997b ) for the may and october observations whenever the spectral coverage allowed . the lines are thus in general not the same in the present study as in the original works by shetrone & keane ( 1998 ) and kipper & klochkova ( 1997 ) . spectral synthesis was applied to a few wavelength regions of particular interest : h@xmath3 , h@xmath4 , the c@xmath5 swan 0 - 1 and 1 - 0 band - heads , the lid doublet at 6707 , the blended cii lines at 6578.05@xmath2 and 6582.88@xmath2 and the he i d@xmath6 line at 5876 , which is blended by two ci lines . the adopted atomic data were the same as in lambert et al . the @xmath7-values for the ci lines present in the above synthesized regions were decreased according to the magnitude of the `` carbon problem '' ( typically 0.7dex ) of the individual spectra , see sect . [ s : param ] . in order to derive accurate stellar parameters a range of different @xmath8 and log@xmath9 criteria was utilized : ionization balance ( mainly ni / nii , sii / siii , cri / crii , and fei / feii ) , excitation balance ( oi , fei and feii ) , and line strengths of sensitive spectral features ( c@xmath5 , cii and hei lines ) . the extended line wings of h@xmath3 and h@xmath4 allowed a determination of the stellar h abundance , but the profiles also provided additional information on the adopted @xmath8 and log@xmath9 ( fig . [ f : hbeta ] ) . the microturbulence parameter @xmath10 was estimated mainly from ci , tiii , fei , feii and yii lines of different strengths ; to within the uncertainties all species indicated the same @xmath10 . the derived temperatures from @xmath11 colours ( duerbeck et al . 1997 ) assuming a reddening of @xmath12 ( n.k . rao , private communication ) provided another temperature estimate which was in very good agreement with the other criteria ( fig . [ f : sakteff ] ) . line - blanketed , hydrogen - deficient model atmospheres similar to those described by asplund et al . ( 1997a ) but with a range of hydrogen abundances especially constructed for the present analysis formed the basis of the investigation . since the present study of the may and october 1996 spectra is based on the same observations , spectral lines and model atmospheres as in asplund et al . ( 1997b ) , the previously derived stellar parameters for may ( @xmath13k , log@xmath14 , @xmath15kms@xmath16 ) and october ( @xmath17k , log@xmath18 , @xmath19kms@xmath16 ) did not warrant a re - consideration . within the framework of the present analysis the april spectrum is characterized by @xmath20k , log@xmath21 and @xmath22kms@xmath16 . the june spectrum indicates @xmath23k , log@xmath24 and @xmath25kms@xmath16 . similarly the july spectrum is adequately described using @xmath26k , log@xmath27 and @xmath25kms@xmath16 . the slight differences compared with the original analyses of kipper & klochkova ( 1997 ) and shetrone & keane ( 1998 ) are not alarming considering the very different adopted model atmospheres as well as the use of different @xmath8-log@xmath9 criteria . the quoted uncertainties in the derived parameters represent the discrepancies between the various indicators ; additional systematic errors are not accounted for . however , since the same criteria has been used , the systematic errors are likely very similar for the different observing epochs , and thus will not compromise our conclusions regarding the very rapid evolution of the star . lcccc epoch & @xmath8 & log@xmath9 & @xmath10 & c / he + & [ k ] & [ cgs ] & [ km s@xmath16 ] & + + april 20 - 25 & @xmath29 & @xmath30 & @xmath31 & 10% + may 5 - 9 & @xmath32 & @xmath33 & @xmath34 & 10% + june 4 & @xmath35 & @xmath36 & @xmath37 & 10% + july 3 & @xmath38 & @xmath39 & @xmath37 & 10% + october 7 & @xmath40 & @xmath39 & @xmath41 & 10% + for all model atmospheres except for a few test cases , a c / he ratio ( by number ) of 10% has been adopted , following comparison of the cii and hei line strengths in the april and may spectra . it should be emphasized , however , that these lines do not provide a very accurate estimate of the c / he ratio in cool h - deficient stars such as sakurai s object due to the sensitivity of the lines to the adopted stellar parameters and the significant contribution of blending ci lines with their unexplained `` carbon problem '' ( see below ) to the hei 5876@xmath2 triplet . the june and july spectra are consistent with the adopted c / he ratio of 10% , though a better agreement between the cii and hei lines would be obtained with a slightly higher ratio . the @xmath8 in october was too low to enable a determination of the c / he ratio through the high - excitation hei and cii lines . in the h - deficient and c - rich photospheres of sakurai s object and the rcrb stars , the continuous opacity at the flux - carrying wavelengths is no longer provided by hydrogen but by photoionization of highly excited levels of ci . furthermore , carbon is by far the most important electron donor in the line - forming regions . the atmospheric structure is therefore essentially completely determined by c ( asplund et al . since the photospheric c abundance is derived from lines originating from levels with only slightly lower excitation energy than those providing the continuous opacity , the strengths of ci lines in these stars should be essentially independent on both the c abundance and the adopted @xmath8 and log@xmath9 , which is also confirmed by observations and theoretical calculations ( lambert et al . 1998 ) . therefore , the theoretical line strengths of weak ci lines should agree with observations since there are no free parameters to tune in order to achieve agreement as in normal analyses . this is unfortunately neither the case for the rcrb stars ( gustafsson & asplund 1996 ; asplund et al . 1997a ; lambert et al . 1998 ) nor for sakurai s object ( asplund et al . 1997b ; kipper & klochkova 1997 ) ; all analyses reveal a significant discrepancy between the predicted and observed line strengths to the extent that the derived abundance is about 0.6dex smaller than the input c abundance for the model atmospheres . the explanation for this `` carbon problem '' is still unknown , but it may be related to inappropriate assumptions for the construction of the model atmospheres ( 1-d , static , flux - constant atmospheres in lte ) ( lambert et al . 1998 ) . sakurai s object also shows the same discrepancy by on average 0.7dex . as will be illustrated further below , the abundances of most elements have remained the same throughout 1996 , which suggests that the c / he ratio has not changed ; an increasing c abundance would manifest itself as a decrease in derived abundances for all elements when assuming a constant c / he in the analysis unless for some reason x / c remained the same for all elements except for he . fortunately , various tests ( lambert et al . 1998 ) suggest that although absolute abundances can be severely affected , abundance ratios such as [ x / fe ] are largely immune to the carbon problem because of the similar sensitivity to the atmospheric structure of most elements ; changing c / he from 10% to 1% introduces differences in [ x / fe ] by @xmath42dex ( asplund et al . therefore , conclusions regarding the evolutionary history of sakurai s object and the rcrb stars can still be drawn , in spite of the unsolved dilemma caused by the carbon problem . lcccccccc element & sun@xmath43 & & rcrb & v854cen@xmath44 + & & april 20 - 25 & may 5 - 9 & june 4 & july 3 & october 7 & majority@xmath45 & + + h & 12.00 & 10.0 & 9.7 & 9.7 & 9.6 & 9.0 & @xmath46 & 9.9 + he & 10.93 & 11.4@xmath47 & 11.4@xmath47 & 11.4@xmath47 & 11.4@xmath47 & 11.4@xmath47 & 11.5@xmath48 & 11.4@xmath48 + li & 3.31 & 3.6 & 3.6 & 3.6 & 4.0 & 4.2 & @xmath49 & @xmath50 + c & 8.52 & 9.7@xmath51 & 9.7@xmath51 & 9.6@xmath51 & @xmath52 & 9.8@xmath53 & 8.9@xmath54 & @xmath55 + n & 7.92 & @xmath56 & @xmath57 & @xmath58 & @xmath59 & @xmath60 & 8.6 & 7.8 + o & 8.83 & @xmath61 & @xmath62 & @xmath63 & @xmath64 & @xmath65 & 8.2 & 8.9 + ne & 8.08 & @xmath65 & @xmath66 & @xmath62 & @xmath62 & & & + na & 6.33 & @xmath67 & @xmath68 & @xmath69 & @xmath70 & @xmath71 & 6.1 & 6.4 + mg & 7.58 & @xmath72 & @xmath73 & @xmath74 & @xmath74 & @xmath75 & 6.4 & 6.2 + al & 6.47 & @xmath76 & @xmath70 & @xmath75 & @xmath77 & 6.3 & 6.0 & 5.7 + si & 7.55 & @xmath78 & @xmath79 & @xmath79 & @xmath80 & @xmath81 & 7.1 & 7.0 + p & 5.45 & 6.2 & @xmath82 & @xmath83 & 6.3 & & & + s & 7.33 & @xmath71 & @xmath67 & @xmath76 & @xmath68 & @xmath84 & 6.9 & 6.4 + k & 5.12 & & @xmath85 & 4.7 & @xmath86 & @xmath87 & & + ca & 6.36 & @xmath88 & @xmath89 & @xmath90 & @xmath91 & @xmath92 & 5.4 & 5.1 + sc & 3.17 & @xmath93 & @xmath93 & @xmath94 & @xmath95 & @xmath96 & & 2.9 + ti & 5.02 & @xmath97 & @xmath98 & @xmath99 & @xmath100 & @xmath101 & & 4.1 + cr & 5.67 & @xmath102 & @xmath103 & @xmath104 & @xmath105 & @xmath106 & & 4.2 + fe & 7.50 & @xmath107 & @xmath107 & @xmath107 & @xmath70 & @xmath77 & 6.5 & 6.0 + ni & 6.25 & @xmath108 & @xmath83 & @xmath109 & @xmath110 & @xmath111 & 5.9 & 5.9 + cu & 4.21 & & @xmath112 & @xmath113 & @xmath114 & @xmath86 & & + zn & 4.60 & @xmath115 & @xmath105 & 4.9 & 5.1 & 5.4 & 4.3 & 4.4 + rb & 2.60 & & @xmath116 & 4.2 & & 4.6 & & + sr & 2.97 & @xmath117 & @xmath118 & @xmath119 & & @xmath120 & & 2.2 + y & 2.24 & @xmath121 & @xmath122 & @xmath122 & @xmath123 & @xmath99 & 2.1 & 2.2 + zr & 2.60 & @xmath124 & @xmath125 & @xmath126 & @xmath95 & @xmath127 & & 2.1 + ba & 2.13 & @xmath128 & @xmath129 & @xmath129 & @xmath130 & @xmath131 & 1.6 & 1.3 + la & 1.17 & & @xmath132 & & 1.3 & 1.5 & & 0.4 + from grevesse & sauval ( 1998 ) . for li the meteoritic value has been adopted . from lambert et al . the abundances are the mean from a sample of 14 stars ; the four minority rcrb stars are not included here . from asplund et al . ( 1998 ) . input c / he ratio for model atmospheres : c / he=10% estimated for sakurai s object from the april and may spectra and v854cen and 1% assumed for rcrb stars . spectroscopically determined ci abundance , which differs from the input model atmosphere c abundance , see text . with the adopted stellar parameters described in sect . [ s : param ] , the resulting abundances under the assumption of lte as observed throughout 1996 are listed in table [ t : abund ] . to within the uncertainties the two april and the three may spectra give identical abundances . the abundances for may and october have been re - analysed for the present study , which has led to a few minor adjustments compared with the original study ( asplund et al . 1997b ) : the abundances of ne , k , fe , cu and zn have all increased by 0.1dex for may , as a result of inclusion of additional lines , improved selection of lines and adoption of better atomic data . from an inspection of table [ t : abund ] two important conclusions are immediately obvious . first , the abundances of most elements ( he , c , n , o , ne , na , mg , al , si , p , s , k , ca , fe , ni , cu and la ) have remained the same within the uncertainties ( about 0.3dex ) throughout 1996 . in most cases the total variation is only about 0.2dex , which is remarkable considering the intrinsic uncertainties in the analyses . given the good agreement , it is worthwhile emphasizing , that in all cases the adopted stellar parameters have been determined before considering the absolute abundances and their implications , and no a posteriori fine - tuning has been allowed in order to improve the consistency . second , several elements show abundance alterations which significantly exceed the statistical scatter . this is true for h ( decrease by 1.0dex ) , li ( @xmath133dex ) , sc ( @xmath134dex ) , zn ( @xmath135dex ) , rb ( @xmath136dex ) and y ( @xmath137dex ) , and probably also ti ( @xmath133dex ) , cr ( @xmath133dex ) , sr ( @xmath135dex , though uncertain due to only based on very strong lines ) and zr ( @xmath138dex ) . there is also some indication that the heavier @xmath0-element ba increased steadily in 1996 , though such a conclusion is not indisputable . unfortunately , trustworthy abundances of other heavy @xmath0-elements throughout 1996 which could verify such a finding are lacking . the slightly larger variations ( in total 0.4dex ) than the typical scatter for si , s and ca are less likely to be signs of real abundance variations , since no definite trend with time is apparent and only one of the epochs show a disparate value . we therefore attribute these variations to the uncertainties in the analyses . for june and july the h abundance may be overestimated by @xmath139dex compared with the other epochs , since the abundances are derived solely from h@xmath3 as the spectra did not cover h@xmath4 . for the other times the greatest weight has been given to h@xmath4 , which in general needs slightly lower abundance than h@xmath3 for an acceptable fit to the extended line wings in sakurai s object . considering the drastic abundance variations apparent in sakurai s object , it is natural to ask whether perhaps they can be blamed on erroneous stellar parameters or inappropriate approximations in the analysis . it is important to realize that all abundance alterations can not be simultaneously annuled by _ any _ change in the adopted stellar parameters , regardless of how large the variation of the parameters . the required @xmath140k to invalidate most of the abundance changes would be clearly incompatible with the various @xmath8-log@xmath9 criteria and furthermore would only introduce other severe modifications of the stellar abundances , which would be less easily explainable within the context of a born - again giant . most of the elements showing an increasing or decreasing abundance trend are not more sensitive than the elements whose abundance has remained constant in the same time - span . the observation of a definite trend in abundance over time for all elements with a suspected abundance change , further supports the finding of a changing chemical composition . likewise , the consistency between the different epochs for most elements also speaks against the stellar parameters being seriously in error . many of the spectral lines of the species with changing chemical content are by necessity relatively strong , which could cause problems in their interpretation as abundances . the effects of hyperfine and isotope splittings have been investigated for a few li , sc , y and ba lines but found to be insignificant except for the liid doublet ; the li abundances in table [ t : abund ] have been derived from such synthesis . since the current analysis is based on the lte approximation , departures from lte may introduce spurious results which could be interpreted as abundance variations . we , however , find it unlikely that such departures may be responsible for all of the observed drastic variations ( in some cases 1.0dex ) , since ( 1 ) in most cases the abundances are derived from lines of the dominant species in the atmosphere ( exceptions : li , zn and rb ) , ( 2 ) the line scatter of the relevant elements is similar or smaller than for the other elements , ( 3 ) in general the abundances are based on more lines than for the typical elements , ( 4 ) other elements with similar atomic structures , which thus should experience similar departures from lte , do not show any abundance variations , and ( 5 ) it seems less likely that the relatively minor differences in stellar parameters between the four observing epochs could introduce _ differential _ nlte effects of @xmath141dex . naturally though , it can not be excluded that nlte effects may be accountable for parts of the observed abundance alterations for some elements , e.g. for ti and cr ( see below ) . as will be further discussed below , the abundance variations can be naturally explained as a result of the mixing and nucleosynthesis which has occurred following the final he - shell flash h - burning , li - production and @xmath0-processing which furthers supports the conclusion that the changes are real . the elements which should not participate in any nuclear reactions to any significant extent by the he - shell flash do not show any abundance variations , thereby making the whole abundance pattern consistent with expectations . in the absence of any plausible alternative explanation , we are thus forced to the conclusion that the observed changes in chemical composition are most likely real . the abundance alterations first noted by asplund et al . ( 1997b ) are thus confirmed by the present , more detailed analysis covering significantly more observational data . the current chemical composition of sakurai s object is clearly not pristine . the very evolved nature of the star is evidenced by the h - deficiency and high abundances of heavy elements , not characteristic of neither disk dwarfs nor halo dwarfs ( edvardsson et al . sakurai s object seem to have undergone h - burning , followed by he - burning and a second phase of cno - cycling , as well as associated nuclear reactions such as li - production and @xmath0-processing . sakurai s object is slightly metal - poor . with a c / he ratio of 10% the fe mass fraction is 0.2dex ( mean of the four observing epochs ) below solar , but had instead the input carbon abundance been used the mass fraction would be 0.7dex lower as a result of the carbon problem . however , only [ mg / fe]@xmath142 , [ ca / fe]@xmath143 , [ ti / fe]@xmath144 and [ cr / fe]@xmath145 ( mean of april , may , june and july for ti and cr ) are consistent with such an origin while all other elements seem to have been modified throughout the evolutionary history of sakurai s object , with the possible exceptions of si ( [ si / fe]@xmath146 ) and s ( [ s / fe]@xmath147 ) ; also ti and cr seem currently to be modified ( sect . [ s : abundchanges ] ) . the most obvious property is , of course , the presence of h - burning material due to the h - deficiency and high he - content . furthermore , the high [ na / fe]@xmath148 and [ al / fe]@xmath149 abundance ratios can probably be explained by proton - captures on @xmath150ne and @xmath151 mg . the significant c and o ( [ o / fe]@xmath148 ) enhancements require he - burning through the 3@xmath3-process followed by subsequent @xmath3-captures . the overabundant n ( [ n / fe]=2.0 ) can not be accounted for by cno - cycling of the initial cno nuclei but requires a second phase of h - burning in which c and o produced by he - burning are converted to n. such a conclusion is supported by the very high [ ne / fe]@xmath152 ; ne has presumably been synthesized from @xmath3-captures on products of he - burning and cno - cycling ( since mg is not overabundant , ne has more likely been formed from @xmath153n seeds rather than @xmath154o nuclei ) . also the low @xmath155c/@xmath1c ratio in sakurai s object strongly suggests a second stage of cno - cycling following he - burning , since the @xmath1c abundance should be very low after @xmath3-processing ; the observed @xmath155c/@xmath1c ratio ( @xmath156c@xmath157c@xmath158 ) encompasses the equilibrium value of about 3.5 from cno - cycling . the high @xmath1c / n ratio requires though that the proton supply was exhausted before conversion of the available c to n was completed . the si and s abundances may also require nuclear processing , as in other h - deficient stars ( lambert et al . 1998 ) , but the channels and conditions under which this would have occurred are still unclear . the observed li has most likely been synthesized from the cameron - fowler ( 1971 ) be transport mechanism . sakurai s object is strongly enhanced in @xmath0-process elements , in particular the light @xmath0-elements . also the very high abundance ratios [ k / fe]=0.8 , [ sc / fe]=1.6 ( october ) , [ ni / fe]=0.9 , [ cu / fe]=1.8 and [ zn / fe]=1.7 ( october ) are likely explained by neutron captures of lighter seed nuclei ( presumably mainly s , ar , ca and fe ) , as previously suggested for rcrb stars ( asplund et al . 1998 ; lambert et al . 1998 ) and fgsge ( gonzalez et al . 1998 ) . very likely @xmath1c is the neutron source ; the alternative reaction @xmath150ne(@xmath3,n)@xmath151 mg which ignites at higher temperatures , seems to be ruled out since the required neutron exposure ( @xmath159 neutrons per fe seed nucleus ) is not consistent with the observed mg / fe ( @xmath160 ) ratio . all the elements ni - la could be synthesized with a _ single _ neutron exposure @xmath161mb@xmath16 ; the sr / rb ratio indicates a neutron density of @xmath162@xmath163 ( malaney 1987 ) . an exponential exposure of neutrons provides less good agreement . the fact that also the abundances of ni , cu and zn are well explained with a single neutron exposure indicates that most of the envelope has been exposed to neutrons and little dilution with gas which has not suffered @xmath0-processing has occurred . though [ ba / fe]=0.6 is not atypical for agb - stars , many of the light @xmath0-elements have very extreme ratios ( e.g. [ sr / fe]=3.0 and [ y / fe]=2.2 in may ) not observed in agb - stars or post - agb stars evolving towards the planetary nebula - phase for the first time , which may suggest that these elements have been synthesized following departure from the agb . indeed , both the li and the ( light ) @xmath0-element abundances seem to have increased dramatically throughout 1996 , due to mixing and processing which was initiated in connection with the final he - shell flash . though most elemental abundances remained the same throughout 1996 , a few elements are distinguished by their changing content within only a few months . these variations may be the result of either mixing of a previously modified chemical composition or current nuclear processing . the mixing time - scale is determined by the convective time - scale which is on the order of a couple of months for sakurai s object . the nuclear burning time - scale is comparable since it is determined by convection to bring fresh protons to h - burning temperatures . thus , both processes may in principle be able to explain the extremely rapid evolution of sakurai s object . rather than reflecting current nucleosynthesis in the star , the elemental abundance variations may simply be due to mixing of the surface layers with gas from the interior . if the initial envelope abundances are significantly different from those of the interior , a changing chemical content may be observed without invoking present nuclear processing . such a depth - dependent chemical composition may either be the result of previous nucleosynthesis ( which must have occurred _ during _ the agb - phase ) , or through dust - gas separation ( which must have occurred _ after _ departure from the agb ) . pure mixing of the envelope with the h - depleted gas of the interior could in principle explain the decreasing h - content . however , some amount of h - burning is expected after the final he - shell flash on theoretical grounds ( renzini 1979 ) , since the flash should have caused an extensive convection zone ingesting the still h - rich envelope , which would bring the gas to h - burning temperatures . though mixing may explain part of the diminishing h abundance , it is not possible to invoke mixing to explain the increasing li content . since sakurai s object is h - deficient and h - burning necessarily destroys any present amount of the much more fragile li , the observed high li abundance can not have been inherited from a previous agb - phase . the observed li / h ratio ( @xmath164 ) far exceeds the value in even the most li - rich agb - stars ( @xmath165 ) . the observed changes of the light @xmath0-elements may be the consequences of mixing with gas exposed to neutrons in between the thermal pulses on the agb now brought to the surface . unfortunately , the exact nature of the last flashes immediately before departing from the agb is poorly known , but a preferential production of the light @xmath0-elements as observed is not inconceivable . whether the very extreme [ @xmath0/fe ] ratios ( e.g. [ rb / fe]=2.9 , [ sr / fe]=3.3 and [ y / fe]=2.9 in october ) could be accomplished remains to be shown , however . as noted in sect . [ s : nucleo ] the abundance distribution of the heavy elements is better described with a single neutron exposure rather than an exponential exposure as expected for thermal pulses in agb - stars , which may indicate that the @xmath0-elements are not inherited from the previous agb - phase . the fact that the heavy element abundances in sakurai s object even before the observed abundance alterations in 1996 are quite different from those observed in post - agb stars evolving towards the planetary nebula - regime for the first time , also seems to support this conclusion . the observed abundance increases in ti and cr are more problematic to explain as a result of nucleosynthesis , whether from a previous phase or presently occurring . in principle , ti and cr could be synthesized through @xmath0-processing , as is the case with sc . both ti and cr are , however , preceded by elements ( sc and v respectively ) which have significantly lower abundances . the neutron capture cross - sections ( beer et al . 1992 ) do not suggest a predominant build - up of ti and cr at the expense of the surrounding elements . ( though zn succeeds the less abundant cu for october 1996 , the calculations of malaney ( 1987 ) show that a single neutron exposure may produce such an abundance distribution due to the differences in cross - sections . ) furthermore , to explain the high cr abundance for october , the ca abundance would have had to decrease if it were the seed , which is not observed . conceivably , repeated @xmath3-captures could explain the ti and cr abundance increases , in particular in the light of the high [ ne / fe ] , [ si / fe ] and [ s / fe ] ratios , but it would seem to require unreasonably high temperatures . furthermore , ca is not significantly enhanced . in the light of a lacking plausible nucleosynthetic explanation , it is not precluded that a combination of errors ( e.g. nlte effects , unaccountable blends ) may have conspired to produce spurious results , especially since the interpretation of an abundance alteration of ti and cr is largely controlled by the october spectrum , while the [ ti / fe ] and [ cr / fe ] ratios for the other dates are representative of slightly metal - poor stars . for the moment , we are unable to find a nucleosynthetic explanation for the observed trends in ti and cr . we note that a similar problem of observed high [ ti / fe ] and [ cr / fe ] ratios also exists for some rcrb stars ( lambert et al . 1998 ) . alternatively , mixing may cause variations not because the interior has undergone previous phases of nucleosynthesis , but because the surface layers have experienced dust fractionation , as e.g. suggested for post - agb stars and fgsge ( blcker & schnberner 1997 ) . such a suggestion may seem able to explain the observed increases in sc , ti , cr and y , since those elements can be heavily depleted in post - agb stars ( e.g. gonzalez et al . 1997 ) . however , the scenario is unlikely to be the main solution to the observed variations . elements such as al , ca , fe and ni should then also show similar or stronger trends which are not observed , and the increase in zn abundance is directly at odds with its inability to condense into dust ( cardelli 1994 ) . also , the variations in the h and li content cause severe problems for the suggestion . furthermore , dust depletion can not explain the abundance ratios relative to fe during the later times , in particular the extreme @xmath0-element ratios ( e.g. [ rb / fe]=2.9 , [ sr / fe]=3.3 and [ y / fe]=2.9 in october ) , when the abundance pattern should be less affected by the actions of dust fractionation . it is also hard to understand why exactly those elements which are likely to be synthesized in connection with a final he - shell flash ( h , li , and the @xmath0-elements ) should be particularly modified while no other elements show similar trends . we therefore find such an explanation rather contrived . similar arguments can be constructed against the suggestion of dust fractionation being responsible for the observed variations in the abundances of the @xmath0-elements in fgsge ( langer et al . 1974 ; blcker & schnberner 1997 ) . the most notable abundance alteration in 1996 is the decreasing h - content by a factor of 10 , due to the ingestion of the h - rich envelope in connection with the he - flash . since most of the luminosity is now probably provided by h - burning rather than he - burning ( renzini 1990 ) , the diminishing h - content is in fact the nuclear fuel of the star being consumed . not only is the generated energy sufficient to provide the high luminosity of the star , the released energy is also causing the expansion of the star to giant dimensions . assuming a mass of @xmath166m@xmath167 of the ingested material and an initial h content of log@xmath168 ( april ) , the liberated energy over six months produces a luminosity which exceeds the classical eddington limit by about a factor of 6 ( see also sect . [ s : tefflum ] ) . the generated energy during these months is less than the estimated binding energy of the stellar envelope outside the he - burning shell ( assuming the shell being located at a stellar radius of @xmath169 cm and a mass of the convective shell of @xmath170m@xmath167 , renzini 1990 ) . however , assuming instead an initial solar h - abundance prior to the he - shell flash in the ingested material is sufficient to provide the required energy for lifting the outer layers to produce a born - again giant . these simple - minded estimates naturally assume that the diminishing h - content only reflects the effects of h - burning rather than mixing of the envelope with previously h - depleted gas from the interior . the increased li abundance is most easily explained by li - production through the cameron - fowler mechanism . since h - burning also destroys present li , the observed li can not have been generated in a previous agb - phase but must be currently produced in the star , simultaneous with the consumption of h. furthermore , since the required @xmath171he , which is produced during the main - sequence phase , is eradicated during he - burning , the li - production must proceed in an environment not previously exposed to he - burning . the high @xmath1c abundance provides an efficient neutron source for the @xmath0-processing through @xmath1c(@xmath172o . the abundance alterations of the light @xmath0-process elements ( zn , rb , sr , y and zr , as well as sc ) can be interpreted as being due to current synthesis in the star ; it is less clear whether this is also true for the heavy @xmath0-elements though there are indications that ba may have increased in 1996 . as discussed above , the abundance distribution of the heavy elements suggests that the variations are not due to pure mixing , but rather to ongoing @xmath0-processing . if this interpretation is correct , however , it may pose a serious challenge for any theoretical modelling . the ignition of cn - cycling due to initiated flash - driven envelope convection at a temperature of @xmath173k should cause a splitting of the shell into two convective zones , the lower one burning he at the bottom , while the upper experiencing cn - cycling . as described above , li can be produced in the upper zone ( `` hot - bottom - burning '' ) , but the @xmath1c produced there does not release neutrons unless the temperature is significantly higher ( @xmath174k ) , i.e. in conditions typical of the lower convection zone . a possible solution may be that h - burning has essentially run to completion during the peak he - burning phase , in which case the two convection zones would once again connect , and thereby bringing down @xmath1c to the relevant high temperatures for neutron release to occur . no doubt , the final he - shell flash in sakurai s object was a highly dynamical phenomenon with a very short timescale ( @xmath175months ) , which may well have been severely inhomogeneous , similar to what is observed in simulations of supernovae . such asymmetric mixing may perhaps explain the simultaneous li - production and @xmath0-processing . it is even possible that the surface layers may not have been chemically homogeneous during 1996 and further variations in the observed abundances of h , li , @xmath1c and the @xmath0-elements could be expected . as discussed in detail above , the increase in ti and cr abundances presents a problem when interpreting the variations as reflecting nuclear processing . the most likely explanation for the observed changes in chemical composition of sakurai s object is thus the direct consequences of the mixing _ and _ nuclear reactions which has occurred following the he - shell flash : ingestion and burning of the h - rich envelope through ( interrupted ) cn - cycling , li production from the simultaneously ingested @xmath171he , and @xmath0-processing from neutrons liberated through @xmath3-captures on the available @xmath1c , which in turn has been produced from the concomitant proton captures on @xmath155c . in fact , the abundance alterations are in excellent accordance with theoretical expectations ( renzini 1990 ) , but the exact details how in particular the @xmath0-processing would have proceeded is less clear . however , we are presently not able to convincingly rule out the possibility that the @xmath0-elements have been synthesized in between previous he - shell flashes on the agb and are now being brought to the surface purely through dredge - up . such mixing though is not possible to invoke to explain the li - enrichment or all of the diminishing h - content , which requires current nucleosynthesis . in any way , the observed changes can only be understood if the final he - shell flash occurred _ after _ the h - burning shell had been extinguished in the post - agb star , since otherwise the entropy barrier from h - burning would prevent ingestion of the h - rich envelope and no substantial mixing or nucleosynthesis besides he - burning would occur . the observed variations of ti and cr are , however , not easily explained within this framework . sakurai s object has evolved rapidly in effective temperature following discovery , as evident both from the spectroscopic analyses presented here for 1996 and from @xmath176 photometry ( duerbeck et al . 1997 ) covering 1996 and early 1997 , as illustrated in fig . [ f : sakteff ] . the photometric temperatures have been derived using similar h - deficient model atmospheres as those applied here , which shows the good agreement between spectroscopic and photometric temperature estimates in 1996 . comparing the analysed spectra of sakurai s object in 1996 reveals clear signatures of a decreasing @xmath8 which was already noted by asplund et al . ( 1997b ) : from april to october pronounced c@xmath5 and cn bands developed ( fig . [ f : c2 ] ) and the h lines became weaker as apparent from fig . [ f : hbeta ] ( also due to the diminishing h - content ) , an evolution which has continued in 1997 ( kerber et al . 1997 ) and 1998 ( fig . [ f : h1 ] , though the recent development of heavy line - blanketing in the visual region makes the weakening of the h - lines less apparent ) . in the beginning of 1998 the stellar spectrum resembles a carbon star ( figs . [ f : c2 ] and [ f : h1 ] ) . the latter does not , however , imply similarly low @xmath8 as in carbon stars since the h - deficiency and c - enhancement make molecular features containing c very prominent ( asplund et al . rather @xmath8 is more likely around 5000 - 6000k , though a detailed abundance analysis of these very crowded spectra will not be an altogether easy task , as obvious from figs . [ f : c2 ] and [ f : h1 ] . this rapid cooling is the result of the stellar expansion caused by the drastically increased energy release after the final he - shell flash . a similar phenomenon is well - known to occur in novae in which the photosphere expands at roughly constant luminosity down to about 6000k , when the expanding shell becomes optically thin . we would thus expect the luminosity to have remained roughly constant during the expansion . from @xmath176 monitoring of sakurai s object , duerbeck et al . ( 1997 ) find evidence for a luminosity increase of about 30% . it is , however , not unlikely that this conclusion may be compromised by adoption of inappropriate bolometric corrections for the star ; considering the h - deficiency and its peculiar composition the line - blanketing will be very different compared with normal supergiants of similar temperatures . a constant luminosity can therefore in our opinion not be ruled out ; clearly more detailed calculations of bolometric corrections for the star are needed . our stellar parameters offer information on the luminosity : l @xmath177 according to standard relations . fig.[f : param ] shows three loci of constant l in the log @xmath9 versus @xmath8 plane together with the derived values . when taken at face value , the spectroscopic parameters suggest an evolution with a decreasing luminosity ; however , the parameters are consistent with a constant luminosity and perhaps also with the modest increase suggested by duerbeck et al . ( 1997 ) within the errors of measurement . spectroscopic estimates of luminosity are dependent on the assumption about the stellar mass but the primary uncertainty may arise from the basic assumptions underlying the construction of the model atmospheres . as pointed out by asplund et al . ( 1997b ) , the assumption of hydrostatic equilibrium for the star may be inappropriate . in particular , an underestimation of the surface gravity will result from neglecting the hydrodynamics due to , for example , an overall atmospheric expansion or turbulent pressure ( gustafsson et al . 1975 ) . the assumption of constant luminosity and the observed decrease in @xmath8 require e.g. the expansion velocity to have _ increased _ during 1996 . this conclusion follows directly from the second derivative of the luminosity @xmath178 : @xmath179 due to the linear decrease in @xmath8 with time in 1996 , @xmath180 and thus the expansion velocity must have increased : @xmath181 . duerbeck et al . ( 1997 ) arrived at the same conclusion based on an analysis of photometric data . the observed trend of a decreasing @xmath10 also suggests that hydrostatic equilibrium may have been inapplicable directly following discovery . furthermore , for both april and may the spectroscopically derived parameters are located at the classical eddington luminosity for electron scattering ( asplund 1998 ) . thus , we can not at present rule out an evolution at constant luminosity from the spectroscopy . assuming a stellar mass of 0.8m@xmath167 ( see below ) , the spectroscopically derived parameters for october indicate a stellar luminosity of @xmath182l@xmath167 , which is in very good accordance with predictions for born - again giants ( iben & macdonald 1995 ; blcker & schnberner 1997 ) . this corresponds to a distance of about 7kpc ( using the above - mentioned @xmath183 ) , which agrees with previously estimated distances of 5.5 - 8kpc ( duerbeck & benetti 1996 ; duerbeck et al . 1997 ) but is in sharp contrast to the short distance of 1.1kpc claimed by kimeswenger & kerber ( 1998 ) ( see also jacoby et al . adopting the spectroscopic parameters of the earlier dates would require a larger distance . a distance of only 1.1kpc would imply log@xmath184 for october , a value well outside the acceptable uncertainties in the spectroscopic analysis . the very rapid evolution of sakurai s object following discovery is significantly faster than for fgsge ( van genderen & gautschy 1995 ) though seemingly similar to that of v605aql though available data is scarce ( lundmark 1921 ; clayton & de marco 1997 ) . fgsge has throughout the last century brightened visually while cooling , while v605aql reached peak brightness within only two years , before starting to show rcrb - type fadings and turning too faint for followed monitoring ( seitter 1987 ) . sakurai s object is currently evolving on a time - scale of a few years ; the star probably started to brighten in 1994 ( duerbeck et al . it is of interest to compare these short evolutionary time - scales with theoretical predictions . according to renzini ( 1990 ) the time - scale for burning the ingested h - envelope ( which generates more luminosity than the initial he - burning which triggered h - burning ) after the he - shell flash is @xmath185 where @xmath186 denotes the energy release from h - burning ( cno - cycling ) , @xmath187 the amount of ingested h , and @xmath188 the luminosity plateau during the post - agb evolution ; the numerical factor takes into account that the h - burning proceeds at a luminosity roughly 40 times higher than @xmath188 . for a typical case of a 0.6m@xmath167 post - agb star , @xmath189 will be about 25years ( renzini 1990 ) , which is in good agreement with the only published numerical calculation of a final he - shell flash event by iben & macdonald ( 1995 ) , in which the subsequent brightening occurred over 17years furthermore , the theoretical time - scale is in agreement with the observed time - scale for h - burning in sakurai s object ( sect . [ s : abundchanges ] ) . different evolutionary time - scales are expected with different stellar masses @xmath190 , since @xmath187 is a very sensitive function of @xmath190 ( e.g. blcker & schnberner 1997 ) . the very rapid evolution of sakurai s object and v605aql can thus presumably be accomplished with a slightly higher @xmath190 and therefore smaller @xmath187 . the main difference in evolutionary time - scales between fgsge on one hand and sakurai s object and v605aql on the other is therefore probably due to different @xmath190 . the life - time of the quiescent born - again giant phase may be about a factor of 40 longer than @xmath189 , since the generated energy will be radiated away at a luminosity not very different from @xmath188 ( renzini 1990 ) . in reality , the life - time will be shorter since ( 1 ) the star can be found only part of the time in the giant regime and ( 2 ) due to the effects of mass - loss . typical life - time as a bright giant of 100 - 1000years can thus be expected , which is not incompatible with the observational evidence ; the visual brightness of v605aql has decreased greatly since outburst in 1919 but the bolometric luminosity has probably remained roughly the same ( harrison 1996 ; clayton & de marco 1997 ) . judging from their similarities in rise - time and visual variability at peak brightness , it is not a bold speculation that sakurai s object will soon also form an optically thick surrounding dust cloud , which will cause a visual fading it may not recover from ; the development of an ir excess ( eyres et al . 1998 ) is certainly suggestive of such a prospect . the very rapid brightening of v605aql ( and now sakurai s object ) may perhaps have endowed a dynamical ejection of the outer envelope ( asplund 1998 ) which subsequently underwent dust condensation , while the more gentle brightening of fgsge did not provide the necessary conditions for formation of a completely obscuring dust shell but only the irregular dust condensations typical of rcrb stars . sakurai s object bears several striking similarities with the rcrb stars ( asplund et al . 1997b ; lambert et al . 1998 ) , in particular with the rcrb star v854cen ( asplund et al . 1998 ) ; from an abundance perspective sakurai s object can be classified as an rcrb star . recently , sakurai s object has also been reported to show visual variability ( liller et al . 1998 ) : in february 1998 the star had faded by almost two magnitudes which by april it had recovered from , a behaviour characteristic of the rcrb stars . the irregular variability has since then continued ( h. duerbeck , private communication ) . thus , sakurai s object fulfills the two defining properties of the rcrb stars : a distinct h - deficiency and unpredictable visual dimming events , presumably due to dust formation episodes in the vicinity of the stars . for the first time , the `` birth '' of an rcrb star has therefore been witnessed , proving the `` final flash '' scenario ( renzini 1979 ) as a viable channel for the formation of rcrb stars , though perhaps not the only option . the rcrb stars can be put in two groups by chemical composition : a homogeneous majority class and a relatively heterogeneous minority class ( lambert et al . 1998 ) , which are distinguished by a lower metallicity ( here : fe / c ) and several extreme abundance ratios , in particular [ si / fe ] and [ s / fe ] . sakurai s object , however , presently has traits of both groups . it is , therefore , not possible for the moment to identify which , if either , of the two rcrb groups evolved from a star experiencing a final flash . sakurai s object lacks the extreme [ si / fe ] and [ s / fe ] ratios of the minority , but has e.g. their low fe / c ratio and high [ na / fe ] ratio . furthermore , in no minority member has li previously been detected , which is a feature in four of the majority stars ( uwcen , rcrb , rznor and sutau ) . lithium , however , could conceivably be destroyed as sakurai s object continues to evolve . sakurai s object resembles most closely the rcrb star v854cen , whose status as a majority or minority member is also unclear ( asplund et al . the only property of sakurai s object not known to be shared with the rcrb stars is the high @xmath1c content ( asplund et al . possibly , the high @xmath1c abundance is a transient feature . it should be noted too that even rough estimates of the @xmath1c abundance are not available for most rcrb stars . additional support for the final flash scenario comes from fgsge and v605aql . currently fgsge is showing typical rcrb - like variability ( papousek 1992 ) and there are hints that it may be h - deficient ( gonzalez et al . 1998 ) . unfortunately the published abundance analyses ( e.g. kipper & kipper 1993 ; gonzalez et al . 1998 ) of fgsge differ significantly , even in [ x / fe ] , which prevents a detailed comparison with sakurai s object and the rcrb stars , in particular since a normal h - abundance has been assumed for the analyses . fgsge is , however , also significantly enhanced in the heavy @xmath0-elements and not only in the lighter @xmath0-elements , as is the case in sakurai s object . during outburst , the spectrum of v605aql resembled closely those of cool rcrb stars ( lundmark 1921 ; ludendorff 1922 ; clayton & de marco 1997 ) , though detailed information on its chemical composition is lacking . the situation is therefore slightly unfortunate . all of the identified born - again giants sakurai s object , fgsge and v605aql show similarities with the rcrb stars , clearly suggesting the final flash scenario being a possible route to forming rcrb stars . however , none of the three stars are readily identified with neither the majority nor the minority group . it is tempting to tentatively identify at least the minority stars as descendants of a final flash , because of their more heterogeneous nature and their possibly higher c / he ratios ( asplund et al . 1997b ; asplund et al . 1998 ; lambert et al . 1998 ) , which are in better agreement with theoretical predictions ( iben & macdonald 1995 ; schnberner 1996 ) . by applying occam s razor , one could speculate that _ all _ rcrb stars have indeed been formed through a final he - shell flash in a cooling pre - wd . until the abundance differences between the stars can be explained either by their earlier nucleosynthetic evolution or through the work of dust - gas separation ( lambert et al . 1998 ) , this conclusion seems premature . here we , therefore , are happy to conclude that at least one rcrb star sakurai s object is a final flash object . given the striking similarities in terms of chemical composition between v854cen and sakurai s object ( asplund et al . 1998 ) it is very likely that at least these two stars share the same evolutionary background . an analysis of the available published and unpublished spectra of the born - again giant sakurai s object from 1996 confirms the extremely rapid evolution of the star previously noted by asplund et al . ( 1997b ) . throughout 1996 sakurai s object cooled significantly by about 1000k , which is also obvious from photometry ( duerbeck et al . 1997 ) ; the cooling has also continued until the present ( 1998 ) . such a cooling is presumably the result of the expansion of the photosphere following the drastically increased luminosity due to the he - shell flash . even more spectacular , the chemical composition of sakurai s object shows definite signs of having been significantly altered within only a few months ( from april to october 1996 ) . again , these changes are most likely interpreted as being caused by the mixing and nuclear reactions which have ensued as a result of the final he - shell flash : ingestion and burning of the h - rich envelope , li - production through the cameron - fowler mechanism , and @xmath0-processing of the light @xmath0-elements ( including sc ) . to our knowledge , sakurai s object represents the fastest case of stellar evolution ever observed when disregarding complete stellar disruptions such as supernovae . sakurai s object shows strong abundance similarities with the rcrb stars , not only a distinct h - deficiency and c - enhancement . in all respects , sakurai s object would be classified as an rcrb star judging from a compositional perspective ( lambert et al . 1998 ) . since sakurai s object also recently has started showing irregular fading episodes typical of the rcrb stars ( liller et al . 1998 ) , sakurai s object indeed seem to have evolved into being an rcrb star . thus , for the first time ever , the `` birth '' of an rcrb star may have been witnessed , which lends strong support for the final flash scenario as a probable channel for forming at least some of the rcrb stars . the exact relation between sakurai s object and the majority and minority classes of the rcrb stars needs , however , to be clarified before one draws any conclusions on how many of the rcrb stars have been formed through a final he - shell flash in a post - agb star on the wd cooling track . further studies of the future evolution of sakurai s object is clearly needed . judging from its rapid metamorphosis already since discovery in 1996 , it is not unlikely that the star will provide astronomers with even more surprises and spectacular changes of appearance in the years to come . in this respect , both a close photometric monitoring of the visual variability and spectroscopic studies of the changes in chemical composition after 1996 are of great importance . the latter will , however , require a brave soul in order to analyse the terrifying richness of atomic and molecular lines , many of which are yet unidentified lines presumably due to @xmath0-elements . as long as the ir excess is not too disturbing ( eyres et al . 1998 ) , it may be advantageous to use high - resolution near - ir spectrum for such abundance analyses . furthermore , we encourage studies on the chemical composition of fgsge , in particular regarding its h - content , and the rcrb star uaqr , which shows very pronounced @xmath0-element enhancements similar to fgsge and sakurai s object ( bond et al . 1979 ) , in order to assess better the relationship between born - again giants and the rcrb stars . finally and most urgently , the theoretical modelling of the final he - shell flash events needs to be extended to a range of stellar masses and initial conditions with a detailed study of the mixing processes and including the relevant nuclear reactions such as cno - cycling , he - burning , li - production and @xmath0-processing , which will likely provide strong constraints on the models . admittedly , such modelling is a herculean task considering the dynamical nature of the event , but is unfortunately required in order to fully understand the rapid evolution of sakurai s object and related stars . we are grateful to g. gonzalez , m. keane , v. klochkova and g. wallerstein for obtaining spectra of sakurai s object . the referee h. duerbeck provided us with useful suggestions . ma acknowledges financial support connected with a post - doctoral fellowship at nordita . dll s research has been supported in part by the u.s . national science foundation ( grant nsf 9618414 ) and the robert a. welch foundation of houston , texas . the financial support tk has received through the estonian science foundation grant 3166 is acknowledged . asplund m. , 1998 , a&a 330 , 641 asplund m. , gustafsson b. , kiselman d. , eriksson k. , 1997a , a&a 318 , 521 asplund m. , gustafsson b. , lambert d.l . , rao n.k . , 1997b , a&a 321 , l17 asplund m. , gustafsson b. , rao n.k . , lambert d.l . , 1998 , a&a 332 , 651 beer h. , voss f. , winters r.r . , 1992 , apjs 80 , 403 blcker t. , schnberner d. , 1997 , a&a 324 , 991 bond h.e , luck r.e . , newman m.j . , 1979 , apj 233 , 205 cameron a.g.w . , fowler w.a . , 1971 , apj 164 , 111 cardelli j.a . , 1994 , science 265 , 209 clayton g.c . , de marco o. , 1997 , aj 114 , 2679 duerbeck h.w . , benetti s. , 1996 , apj 468 , l111 duerbeck h.w . , benetti s. , gautschy a. , et al . , 1997 , aj 114 , 1657 edvardsson b. , andersen j. , gustafsson b. , et al . , 1993 , a&a 275 , 101 eyres s.p.s . , evans a. , geballe t.r . , et al . , 1998 , mnras 298 , l37 gonzalez g. , lambert d.l . , giridhar s. , 1997 , apj 481 , 452 gonzalez g. , lambert d.l . , wallerstein g. , et al . , 1998 , apjs 114 , 133 grevesse n. , sauval a.j . , 1998 , space sci . , in press gustafsson b. , asplund m. , 1996 , in : hydrogen - deficient stars , jeffery c.s . , heber u. ( eds . ) . series vol . gustafsson b. , bell r.a . , eriksson k. , nordlund . , 1975 , a&a 42 , 407 harrison t.e . , 1996 , pasp 108 , 1112 iben i.jr . , macdonald j. , 1995 . in : white dwarfs , koester d. , werner k. ( eds . ) . springer , berlin , p. 48 iben i.jr . , tutukov a.v . , yungelson l.r . , 1996 , apj 456 , 750 jacoby g.h . , de marco o. , sawyer d.g . , 1998 , aj , 116 , 1367 kerber f. , gratl h. , roth m. , 1997 , iac circ . 6601 kimeswenger s. , kerber f. , 1998 , a&a 330 , l41 kipper t. , 1996 , in : hydrogen - deficient stars , jeffery c.s . , heber u. ( eds . ) . series vol . 96 , p. 329 kipper t. , kipper m. , 1993 , a&a 276 , 389 kipper t. , klochkova v.g . , 1997 , a&a 324 , l65 lambert d.l . , rao n.k . , gustafsson b. , asplund m. , 1998 , submitted to a&a langer g.e . , kraft r.p . , anderson k.s . , 1974 , apj 189 , 509 liller w. , janson m. , duerbeck h.w . , van genderen a. , 1998 , iau circ . 6825 ludendorff m.h . , 1922 , an 217 , 167 lundmark k. , 1921 , pasp 33 , 314 malaney r.a . , 1987 , apj 321 , 832 nakano s. , sakurai y. , hazen m. , et al . , 1996 , iau circ . 6322 papousek j. , 1992 , iau circ . 5604 renzini a. , 1979 , in : stars and star systems , westerlund b.e . reidel , dordrecht , p. 155 renzini a. , 1990 , in : confrontation between stellar pulsation and evolution , cacciari c. , clementini g. ( eds . ) . series vol . 11 , p. 549 schnberner d. , 1996 , in : hydrogen - deficient stars , jeffery c.s . , heber u. ( eds . ) . series vol . 96 , p. 433 seitter w.c . , 1987 , messenger 50 , 14 shetrone m.d . , keane m. , 1998 , in : faint blue stars , davis - philip a.g . l. davis press , schenectady , in press van genderen a.m. , gautschy a. , 1995 , a&a 294 , 453

in this article , we establish a relation between the values of a frieze of type and some values of an-tiling associated with a particular quiver of type . this relation allows us to compute , independently of each other , all the cluster variables in the cluster algebra associated with a quiver of type * mathematics subject classification ( 2010 ) * : 13f60 , 16g20 . * keywords * : cluster algebras , representations of quivers .

1405.1023

cluster algebras were introduced by s. fomin and a. zelevinsky in [ fz02 , fz03 ] . they are a class of commutative algebras which was shown to be connected to various areas of mathematics like , for instance , combinatorics , lie theory , poisson geometry , teichmller theory , mathematical physics and representation theory of algebras . a cluster algebra is generated by a set of variables , called _ cluster variables _ , obtained recursively by a combinatorial process known as _ mutation _ starting from a set of _ initial _ cluster variables . explicit computation of cluster variables is difficult and has been extensively studied , see [ ars10 , adss11 , ad11 , ar12 , bmrrt06 , bmr08 ] . in order to compute cluster variables , one may use _ friezes _ , which were introduced by coxeter [ c71 ] and studied by conway and coxeter [ cc73-i , ii ] . various relationships are known between friezes and cluster algebras , see [ cc06 , m11 , pr08 , ars10 , bm12 , d10 , ad11 , adss12 , ar12 ] . the present work is motivated by the use of friezes to compute cluster variables and is inspired by the results in [ ars10 , ar12 , ma13 ] giving an explicit formula as a product of @xmath6 matrices for all cluster variables in coefficient - free cluster algebras of types @xmath7 and @xmath8 , thus explaining at the same time the laurent phenomenon and positivity . our objective here is to show that similar techniques can be used for computing cluster variables in coefficient - free cluster algebras of type @xmath9 . in this paper , we establish a relation between the values of a frieze of type @xmath9 and some values of an @xmath1-tiling @xmath2 associated with a particular quiver of type @xmath7 . for this we associate with each quiver of type @xmath0 a particular quiver of type @xmath3 . this correspondence allows us to obtain an algorithm for computing cluster variables of a cluster algebra of type @xmath0 independently of each other , that is without iteration . the article is organized as follows . in section @xmath10 , we recall some basic notions on @xmath1-tilings associated with a quiver of type @xmath7 . in section @xmath11 , we set preliminaries on friezes of type @xmath9 and establish a correspondence between a frieze of type @xmath9 and an @xmath1-tiling associated with a particular quiver of type @xmath7 . in section @xmath12 , we give an algorithm to compute all the cluster variables in a cluster algebra associated with a quiver of type @xmath9 . in this section we recall some notions on @xmath1-tilings and apply them to a quiver of type @xmath7 according to [ ars10 ] . let @xmath13 be a field and @xmath14 the discrete plane . we call @xmath1-tiling of the plane a map @xmath15 such that for any @xmath16 @xmath17 + the following example represents an @xmath1-tiling with @xmath18 . + @xmath19 + the @xmath1-tiling is an extension to the whole plane of the frieze introduced by coxeter [ c71 ] and studied by conway and coxeter [ cc73-i , ii ] . let @xmath20 be a finite acyclic ( containing no oriented cycles ) quiver with @xmath21 the set of its points and @xmath22 the set of its arrows and @xmath13 a field . the translation quiver @xmath23 associated with @xmath20 ( see [ ass - viii.1.1 ] ) consists of two sets : the set of points @xmath24 and the set of arrows + @xmath25 let us define a frieze associated with the quiver @xmath20 . in the translation quiver @xmath23 ( see [ ass - viii.1.1 ] ) , let us replace the points @xmath26 by their images under a frieze function @xmath27 : @xmath28 defined for some initial values @xmath29 as follows : @xmath30 where the product is taken over the arrows ( see [ ars10 - 2 ] ) . the resulting translation quiver with values associated with its vertices is called a frieze . + in general we associate a _ boundary _ to an @xmath1-tiling . we call boundary a bi - infinite sequence @xmath31 with @xmath32 and @xmath33 for all @xmath34 . coefficients @xmath35 equal to one are usually omitted in the sequence representing a boundary . each boundary may be embedded into the euclidean plane in the following way : the @xmath36 ( or @xmath37 ) determine the horizontal ( or vertical , respectively ) segments of a discrete path , that is @xmath36 ( or @xmath37 ) corresponds to a segment of the form [ @xmath38 $ ] ( or @xmath39 $ ] , respectively ) in the plane . the variables @xmath40 become thus labels of the vertices of a discrete path . a boundary is called _ admissible _ if none of the sequences @xmath41 and @xmath42 is ultimately constant . given an admissible boundary @xmath43 embedded in the plane , let @xmath44 be a point in @xmath14 . then the _ word associated with _ @xmath44 is the portion of the boundary @xmath43 delimited by the horizontal and vertical projections of the point @xmath45 on the boundary @xmath43 . the following example shows an embedded boundary and how to associate a word with a point in @xmath14 . the word associated with the point @xmath46 is @xmath47 . + @xmath48\ar@{-}[r ] & c_{4}\ar@{-}[r ] & \ar@{.}[r ] & & \\ & & & & & c_{2}\ar@{-}[d ] & & & & & \\ & & & c_{-1}\ar@{-}[r]\ar@{-}[d ] & c_{0}\ar@{-}[r ] & c_{1 } & & & & \\ & & & c_{-2}\ar@{-}[d ] & & & p\ar@{.}[uuu]\ar@{.}[lll ] & & & \\ & & & \ar@{.}[d ] & & & & & & \\ & & & & & & & & & & & & & & & & \\ } $ ] if we set @xmath49 , for all @xmath50 then the word associated with the point @xmath46 can be written as follows : yxxyyx . + it is always possible to construct an @xmath1-tiling starting from an admissible boundary in the plane . the work [ ars10 ] provides a formula for a value in the tiling at the point @xmath51 . this formula is given in terms of the associated word and the following matrices : + @xmath52 , @xmath53 . + the following theorem from [ ars10-theorem @xmath12 ] allows us to compute the variable lying at the point @xmath51 . + given an admissible boundary @xmath43 , there exists a unique @xmath1-tiling @xmath2 of the plane extending the embedding of the boundary into the plane . for any point @xmath44 below the boundary , with an associated word @xmath55 where @xmath56 , @xmath57 , @xmath58 , the tiling @xmath2 is defined by the formula + @xmath59 . @xmath60 + now we give an application of the notion of @xmath1-tiling to a quiver of type @xmath7 according to [ ars10 ] . + consider an acyclic quiver @xmath61 of type @xmath62 , @xmath63 , such that its points are labelled by natural numbers modulo @xmath64 in clockwise orientation . let us form a word @xmath65 by associating a variable @xmath66 with arrows of @xmath61 as follows . let @xmath67 if the arrow is @xmath68 and @xmath69 if it is @xmath70 . let @xmath65 be such a word @xmath71 which encodes the orientation of arrows in the quiver @xmath61 . let @xmath72 be the extension of @xmath65 by considering indices @xmath73 and putting @xmath74 if @xmath75 equals @xmath76 modulo @xmath64 . then @xmath72 defines an admissible boundary associated with the quiver @xmath61 . according to [ ars10 ] , given some initial values @xmath77 the values of the tiling @xmath2 from theorem @xmath78 below the admissible boundary @xmath72 are values of the frieze associated with the quiver @xmath61 . these values are computed by applying the map @xmath2 . + we present in the following example an @xmath1-tiling for a quiver @xmath61 of type @xmath79 . + consider the following quiver @xmath61 of type @xmath79 : + + we have @xmath80 and the boundary associated with @xmath61 is @xmath81 . this gives us the following @xmath1-tiling : + @xmath82 the word corresponding to the values @xmath83 in the tiling is @xmath84 and then , according to theorem @xmath78 , + @xmath85 + an admissible boundary @xmath43 is said to be _ periodic _ if it is of the form + @xmath86 and in this case its period is @xmath64 . the @xmath1-tiling associated with this boundary has periodicity determined by the vector @xmath87 , where @xmath88 ( or @xmath89 ) is the number of @xmath36 ( or @xmath37 , respectively ) among @xmath90 . we have also @xmath91 . we call the finite sequence @xmath92 a generator of the boundary @xmath43 . note that when extending a generator @xmath92 to a boundary @xmath86 we glue adjacent copies of the generator in a way that there is only one occurrence of @xmath93 between @xmath94 and @xmath95 ( @xmath96 ) . the admissible boundary @xmath43 associated with the quiver @xmath61 of type @xmath97 is @xmath72 constructed above . its generator can be obtained by cutting the quiver @xmath61 at one of its points and reading the resulting quiver clockwise ( by reading the resulting quiver anticlockwise we obtain a generator of a boundary equivalent to @xmath43 ) . this operation creates an additional point of the quiver @xmath61 , the second copy of the point at which we cut . thus the quiver of type @xmath97 is transformed into a quiver of type @xmath98 . in example @xmath99 let us cut the quiver @xmath100 at its point @xmath101 . we obtain in this way a quiver of type @xmath102 ( reading clockwise ) labelled as follows : @xmath103 & { 2}\ar[r ] & 3\ar[r ] & 4 & 1\ar[l ] } $ ] . if in this quiver of type @xmath104 we denote by @xmath36 the arrows oriented toward the right and by @xmath37 those oriented toward the left then we get the generator @xmath80 . + a quiver @xmath61 of type @xmath97 has @xmath105 possibilities of cutting , we choose any one of them . the quiver underlying @xmath65 is the quiver of type @xmath106 . + we now recall the definition of a seed due to fomin and zelevinsky [ fz03 - 1.2 ] . + let @xmath107 be a quiver ( without loops or oriented two - cycles ) with @xmath64 points and @xmath108 a set of variables called cluster variables , such that a variable @xmath109 is associated with the point @xmath50 ( with @xmath110 ) of @xmath107 . the set @xmath111 is called a cluster and the pair @xmath112 is called a seed . + we can obtain other seeds by mutation ( see [ fz03 - 1.2 ] ) starting from the seed @xmath113 . the set of all cluster variables obtained by successive mutation generates an algebra over @xmath114 called _ cluster algebra _ which is denoted by @xmath115 . for cluster algebras , we refer to the papers [ fz02 ] , [ fz03 ] . we say that the type of a seed @xmath113 and the cluster algebra @xmath115 coincides with the type of quiver @xmath107 . if we associate a point @xmath76 of a quiver @xmath61 of type @xmath62 with the variable @xmath116 , these variables will correspond to the vertices of the boundary . then the variables of the @xmath1-tiling below the boundary are cluster variables of the cluster algebra @xmath117 . they are obtained by applying the formula of theorem @xmath78 ( see [ ars10 - 8 ] ; [ ar12 - 4.4 ] ) . + our aim in this paper is to compute cluster variables of a cluster algebra of type @xmath0 . in this section we establish a relation between values of a frieze of type @xmath9 and some values of an @xmath1-tiling associated with a particular quiver of type @xmath7 . to this end we propose a way to associate an admissible boundary with a quiver of type @xmath9 . consider a quiver @xmath4 whose underlying graph is of type @xmath0 . we agree to label the points of @xmath4 as follows : @xmath118 & & \\ \text { } & & 3 \ar@{-}[r]\ar@{-}[ul]\ar@{-}[dl ] & \ldots \ar@{-}[r ] & ( n-1 ) & \text { } \\ & 2 & & & & ( n+1)\ar@{-}[ul ] } \ ] ] + where a solid segment represents an arrow without its orientation . + the _ forks _ are the full sub - quivers of @xmath4 generated by the points @xmath119 and by the points @xmath120 , respectively . + we agree to call : + - _ fork arrows _ , the arrows of each fork , + - _ joint _ of each fork , the points @xmath11 or @xmath121 , respectively + - and _ fork vertices _ , the points @xmath101 and @xmath10 or @xmath122 and @xmath64 , respectively . + if we associate with each vertex @xmath50 of @xmath4 a variable @xmath109 then we get the seed @xmath123 whose underlying graph can be represented by the following diagram : @xmath124 & & \\ \text { } & & u_{3 } \ar@{-}[r]\ar@{-}[ul]\ar@{-}[dl ] & \ldots \ar@{-}[r ] & u_{n-1 } & \text { } .\\ & u_{2 } & & & & u_{n+1}\ar@{-}[ul ] } \ ] ] + let us define a frieze on the translation quiver @xmath125 by the function + @xmath27 : @xmath126 such that for @xmath127 we have : + @xmath30 with initial variables @xmath128 and the product taken over the arrows . then all the values of the frieze are cluster variables of @xmath129 ( see [ ad11 - 1.3 ] ) . + the study of seeds of type @xmath0 can be subdivided into different cases depending on the orientation of fork arrows . for each fork we have the following three cases : + - the fork is composed by two arrows leaving the joint , + - the fork is composed by two arrows entering the joint , + - the fork is composed by one arrow leaving the joint and another arrow entering the joint . + it is well - known that for two seeds @xmath130 and @xmath131 , which are mutation equivalent ( we refer to [ fz03 - 8 ] for the notion of mutation equivalent ) , the cluster algebras @xmath132 and @xmath133 associated with these seeds , respectively , coincide ( see [ fz03 - 1.2 ] ) . + our aim in this paper is to compute cluster variables of a cluster algebra of type @xmath0 independently of each other . the following lemma in [ ass - vii.5.2 ] allows us to reduce the study to one of the three different cases mentioned above . let @xmath134 and @xmath135 be two quivers having the same underlying graph @xmath136 . if @xmath136 is a tree then @xmath134 and @xmath135 are mutation equivalent . @xmath60 + in the following , @xmath137 denotes a seed of type @xmath0 with a quiver whose each fork is composed by two arrows both entering ( or leaving ) the joint . + we denote by @xmath138 the frieze associated with the seed @xmath137 . we define in the following the notion of _ modelled quiver _ which will help us to compute the cluster variables of a cluster algebra of type @xmath0 without using iteration . we call modelled quiver @xmath139 associated with @xmath137 , the translation quiver obtained from the frieze @xmath140 as follows : + @xmath141 by gluing in @xmath140 the arrows of each shifted copy of the forks , + @xmath142 by multiplying the values assigned to vertices of the fork corresponding to the arrows that were glued in step @xmath101 . namely the arrows obtained by gluing in step @xmath101 have @xmath143 or @xmath144 as the corresponding variables . + we give an example of a modelled quiver associated with a seed of type @xmath145 . @xmath146 & & u_{4}\ar[dl ] & & \\ \text{for the following seed of type $ \tilde{\mathbb{d}}_{4}$:}\quad & & u_{3 } & & & \text{,}\\ & u_{2}\ar[ur ] & & u_{5}\ar[ul ] } $ ] + the frieze @xmath140 has the form : + + and the modelled quiver @xmath147 has the form : + + all the squares of the form + in @xmath139 satisfy the relation @xmath148 called the uni - modular rule , with @xmath149 . the values on the upper and bottom extreme lines in @xmath147 are products of two cluster variables ( products created by the passage from @xmath138 to @xmath147 ) . note that the pairs of cluster variables forming these products ( in the case of both arrows of a fork entering or leaving the joint ) are given by fractions whose numerators are equal and denominators coincide up to exchanging @xmath150 and @xmath151 ( or @xmath152 and @xmath153 ) , which appear in denominators with exponent one ( see [ bmr09 - 1 ] ) . therefore , a product of these two variables is of the form @xmath154 for one fork and @xmath155 for the other fork , with @xmath156 . + then the two extreme lines in the modelled quiver @xmath147 contain the sequences of variables of the form @xmath157 for the horizontal line passing through the variable @xmath158 and of the form @xmath159 for the horizontal line passing through the variable @xmath160 . + as a consequence of the definition of a frieze , we have the following relations for all @xmath161 : @xmath162^{2 } = \displaystyle\frac{1}{(u_{1}u_{2})^{2}}\phi^{2}_{k}\phi^{2}_{k+1 } , \hspace{3 cm}\ ] ] @xmath163^{2 } = \displaystyle\frac{1}{(u_{n}u_{n+1})^{2}}\psi^{2}_{k}\psi^{2}_{k-1}. \hspace{3 cm}\ ] ] + let @xmath4 be a quiver of type @xmath0 and * @xmath164 * be the full sub - quiver of @xmath4 generated by all points except the points @xmath10 , @xmath64 . we agree to draw @xmath164 from the left to the right in such a way that the vertices appear in increasing order . note that @xmath164 is a quiver of type @xmath165 . this leads to a way to associate a quiver of type @xmath3 with a quiver of type @xmath0 , which , in turn , will allow us to associate an admissible boundary with a quiver of type @xmath0 . for a quiver @xmath166 of type @xmath167 ( drawn from left to right ) , let us denote by @xmath168 the transpose of @xmath166 , that is , the quiver obtained by redrawing @xmath166 from right to left . + with a quiver @xmath4 of type @xmath0 we associate the quiver @xmath169 of the form + + we know that the boundary @xmath170 associated with a quiver @xmath61 of type @xmath97 is periodic and its period @xmath105 corresponds to the length of a generator @xmath65 . for the quiver @xmath169 of type @xmath3 there is a choice of @xmath171 different generators for its associated boundary @xmath172 : one can cut at any of @xmath171 points of @xmath169 to obtain a generator . we are going to use only one of these generators to construct the admissible boundary @xmath173 associated with a quiver of type @xmath0 . the generator @xmath174 of the boundary @xmath172 is obtained by cutting the quiver @xmath169 at the point @xmath101 of its arrow @xmath175 & { 1 } } $ ] and reading the quiver clockwise . consider also a generator @xmath176 of an boundary equivalent to @xmath172 : @xmath176 is obtained by cutting @xmath169 at the point @xmath122 of its arrow @xmath175 & { n}}$ ] and reading the quiver anticlockwise ( see example @xmath177 ) . we get in this way two different generators associated with the quiver @xmath169 , that is , @xmath174 with the underlying quiver @xmath178 & \text{o}\ar[r ] & ^{t}{\sigma}\ar[r ] & \text{o}\ar[r ] & 1 } $ ] and @xmath176 with the underlying quiver @xmath179 & ^{t}{\sigma}\ar[l ] & \text{o}\ar[l ] & { \sigma}\ar[l]}.$ ] let us denote by @xmath180 the part of @xmath4 whose underlying quiver is @xmath164 . then we have @xmath181 and @xmath182 . now we are in a position to associate an admissible boundary with a quiver of type @xmath0 whose each fork consists of two arrows entering or leaving the joint . + let @xmath4 be a quiver of type @xmath0 whose each fork consists of two arrows both entering or both leaving the joint . the admissible boundary @xmath173 associated with @xmath4 is of the form @xmath183 , where @xmath174 is repeated infinitely many times on the left and @xmath176 is repeated infinitely many times on the right . the boundary @xmath173 is obtained by gluing to @xmath180 the generators @xmath184 periodically on the left and @xmath176 periodically on the right . note that @xmath180 is also contained in @xmath174 and @xmath176 . however the decomposition of the boundary of the form @xmath183 defines a distinguished occurrence of @xmath180 in @xmath173 . it is this occurrence of @xmath180 which we call the _ root _ of @xmath4 in @xmath173 . + we give an example showing how to associate an admissible boundary with a quiver @xmath4 of type @xmath0 . @xmath185 & & { 4}\ar[dl ] & & \\ \text{consider the following quiver $ q$ of type $ \tilde{\mathbb{d}}_{4}$:}\quad & & { 3 } & & & \text{,}\\ & { 2}\ar[ur ] & & { 5}\ar[ul ] } $ ] + @xmath186 & { 3 } & 4\ar[l ] & & \text{}\\ & & & & } $ ] @xmath187 & { 3 } & 1.\ar[l ] & & \text{}\\ & & & & } $ ] + the quiver @xmath169 of type @xmath188 associated with @xmath4 is : + + this gives us two generators @xmath174 and @xmath176 associated with @xmath169 such that their underlying quivers are respectively + @xmath189 & 3 & 4\ar[l]\ar[r ] & \text{o}\ar[r ] & 4\ar[r ] & 3 & 1\ar[r]\ar[l ] & \text{o}\ar[r ] & 1 & } $ ] and + @xmath190 & 4\ar[l]\ar[r ] & { 3 } & 1\ar[l ] & \text{o}\ar[l ] & 1\ar[r]\ar[l ] & 3 & 4.\ar[l ] & } $ ] + the generators @xmath174 and @xmath176 are obtained by cutting the quiver @xmath169 of type @xmath188 at its points @xmath101 and @xmath12 represented in bold characters on the picture of @xmath169 , respectively . + the boundary @xmath173 associated with @xmath4 is : @xmath191 from an admissible boundary and given initial values it is always possible to construct an @xmath1-tiling @xmath2 , therefore we construct an @xmath1-tiling @xmath2 from the boundary @xmath173 ( with all initial values equal one ) . + we now recall the notion of _ ray _ in a tiling @xmath2 from [ ars10 - 6.1 ] . this notion allows us to find the values of a frieze of type @xmath0 among values in the tiling @xmath2 below the boundary @xmath173 . + given a mapping @xmath15 , a point @xmath192 and a nonzero vector @xmath193 , we consider the sequence @xmath194 . such a sequence will be called a _ ray _ associated with @xmath2 . we call @xmath195 the _ origin _ of the ray and @xmath196 its _ directing vector_. the ray is _ horizontal _ if @xmath197 , _ vertical _ if @xmath198 and _ diagonal _ if @xmath199 . + let @xmath200 and consider the quiver @xmath4 of type @xmath145 of example @xmath177 . + the boundary @xmath173 associated with @xmath4 is @xmath201 . this gives us the following @xmath1-tiling below the boundary @xmath173 : + @xmath202 + the values represented in bold characters in the tiling are the vertical ( diagonal , horizontal ) rays with origin @xmath10 ( @xmath203 , @xmath11 respectively ) . + now we associate variables with the points of @xmath164 as follows : + @xmath141 with a vertex labelled by @xmath204 we associate the products @xmath158 , @xmath160 respectively , + @xmath142 with a vertex labelled by @xmath205 we associate the variable @xmath109 . starting with the quiver @xmath164 with associated variables as above , we can reconstruct the modelled quiver @xmath147 associated with the seed @xmath137 by using the uni - modular rule and formulas @xmath206 and @xmath207 of remark @xmath208 . if we associate also the vertex of @xmath169 labelled @xmath209 with the variable @xmath210 then the result of this association is a seed which will be denoted by @xmath211 . its underlying graph with variables is : + + since @xmath210 is not a cluster variable in @xmath137 , in the rest of this paper we evaluate @xmath212 . note that the generators @xmath213 and @xmath176 are obtained from @xmath169 and thus our association of variables with the points of @xmath169 naturally associates variables with the vertices of @xmath213 and @xmath176 . from now on , we denote by @xmath213 and @xmath176 the corresponding generators with variables . we denote by @xmath214 the admissible boundary @xmath173 with variables and the following theorem establishes a link between the frieze @xmath138 and the tiling associated with @xmath214 . let @xmath215 be an integer , @xmath137 a seed of type @xmath0 and @xmath147 the modelled quiver associated with @xmath137 . then the horizontal lines in @xmath147 are the diagonal rays in the @xmath1-tiling associated with @xmath214 such that the origin of each ray is a vertex of the root @xmath180 in @xmath216 . * proof * consider the admissible boundary @xmath217 by doing a translation of parentheses toward extremities , we can rewrite @xmath214 as follows : @xmath218 let us denote @xmath219 and @xmath220 . the following figure illustrates the admissible boundary @xmath214 in its form @xmath221 . @xmath222^{s ' } & & & \\ & & & & & & & & i ' \ar@{-}[u ] & & & & & & & & & \\ & & & & & & & & j'\ar@{.}[rrdd]\ar@{-}[u ] & & & & & & & \\ & & & & & & & & & & & & & & \\ & & & & & & & & & & j'_{k } & & \\ & & & & & & & & \\ & & & & & & & & & & & & & & & & & & \\ & & \ar@{.}[dlld]^{s } & i\ar@{-}[l]&j\ar@{.}[uuuuurrrr]^{\bar{\omega}}\ar@{.}[rrdd]\ar@{-}[l]&&&&&&\\ & & & & & & & & & & & & & & & & & \\ & & & & & & j_{k}&&&&&&&&&&&&&}\ ] ] + our aim is to prove that the horizontal lines in the modelled quiver @xmath147 coincide with the diagonal rays with origins at the vertices of the root @xmath180 in @xmath214 , that is , with the oblique band delimited by the diagonal rays with origins @xmath223 and @xmath224 ( inclusively ) in the figure above . to this end , we first calculate variables lying at the points @xmath225 and @xmath226 and prove that they have the form of the variables on the top and bottom horizontal lines of @xmath147 , see remark @xmath208 . + consider the transpose of @xmath89 , which is @xmath227 . by a translation of parentheses toward the right , we can rewrite @xmath228 as follows : @xmath229 this allows us to rewrite the boundary in the form : @xmath230 . then the embedding of @xmath214 in the plane gives the following scheme : + @xmath231\ar@{.}[ururr]^{^{t}s } & \\ & & & & & & & & & & & & & \\ & & & & & & b\ar@{-}[r ] & i\ar@{-}[r ] & j \ar@{.}[urrur]^{^{t}\lambda } & & \\ & & & & & & & & & & & & & \\ & & a\ar@{.}[rrrrrrrrr]\ar@{.}[rrrruu]^{\lambda}\ar@{.}[dll]^{s } & & & & & & & & & j_{k}\ar@{.}[llluu]&&\\ & & & & & & & & & & & & \\ & & & & & & & & & & & & & \\ & & & & & & & & & & & & & \\ } $ ] + let us write @xmath232 for the walk on the boundary from @xmath233 to @xmath234 and @xmath235 for that from @xmath223 to @xmath236 . then the word associated with @xmath225 is of the form @xmath237 . by construction of @xmath214 , the end point of the walk @xmath232 is the variable @xmath158 , and thus @xmath158 is also the starting vertex of @xmath235 . computing the tiling function @xmath2 from theorem @xmath78 for the point @xmath225 , gives : @xmath238 + where @xmath239 is a product of the variables in @xmath232 except the variable @xmath158 , which are also by transposition in @xmath235 , and @xmath240 arise from the product of matrices corresponding to the walk @xmath232 . the computation gives @xmath241^{2}.\hspace{3 cm}\ ] ] thus the value lying at the point @xmath225 on the diagonal ray with origin @xmath223 is a perfect square divided by @xmath158 . to show that the value lying at a point @xmath226 on the diagonal ray with origin @xmath224 is a perfect square divided by @xmath242 we perform a similar calculation using the following form of the boundary @xmath243 . this form is obtained by noticing that the transpose of @xmath244 is @xmath245 and by a translation of parentheses toward the left , we can rewrite @xmath246 as follows : + @xmath247 + since the tiling @xmath2 ( below @xmath214 ) satisfies the uni - modular rule and since @xmath180 with variables coincides with the quiver @xmath164 with variables then by virtue of remark @xmath248 it only remains to prove that the variables on the two diagonal rays with origins @xmath223 and @xmath224 satisfy relations of the form @xmath206 and @xmath207 from remark @xmath208 . + consider the following scheme which represents four adjacent points @xmath249 forming a square in the tiling below @xmath214 . @xmath250&d&&&&&&\\ & & & & & & & & & & & & & & & & & & & & & & & & \\ & & & & & & & & & & & & & & & & & & & & & & & & \\ & & & & & & & & & & & & & & c\ar@{.}[dddd]\ar@{.}[uuu ] & \\ & & & & & & & & & & & & & & \\ & & & & & & & & & b\ar@{-}[r ] & i\ar@{-}[r ] & j \ar@{.}[urrur]^{^{t}\lambda } & & \\ & & & & & & & & & & & & & & \\ & & \ar@{-}[d]&&&a\ar@{.}[lll]\ar@{.}[rrrrrrrrr]\ar@{.}[rrrruu]^{\lambda } & & & & & & & & & j_{k}\ar@{.}[llluu]&d_{k}\ar@{.}[uuuuuuu]&\\ & & b ' & & & & & & & & & & & & i_{k+1}\ar@{.}[llllllllllll ] & j_{k+1 } & & & & & & & & & & \\ & & & & & & & & & & & & & & & & & & & & & \\ } $ ] + the words associated with the points @xmath251 and @xmath252 are respectively of the form @xmath253 and @xmath254 where @xmath232 is the walk on the boundary from @xmath233 to @xmath234 and @xmath255 is that from @xmath256 to @xmath234 . + by definition of @xmath214 , the point @xmath257 is on the diagonal ray with origin @xmath258 . we need to prove that each value @xmath259 on this diagonal satisfies the relation : @xmath260^{2 } = t(j_{k})t(j_{k+1 } ) . \hspace{3 cm}\ ] ] to this end , we compute the values lying at the points @xmath251 and @xmath252 . + due to @xmath261 , we have : @xmath262^{2 } , \nonumber \\ t(j_{k+1 } ) & = & \displaystyle\frac{1}{u_{1}u_{2}}\left[\frac{\tilde{a}+\tilde{b}}{\tilde{\gamma}}\right]^{2}. \nonumber \end{aligned}\ ] ] for the two remaining variables , the application of the map @xmath2 gives : @xmath263 , \nonumber \\ t(i_{k+1 } ) & = & \displaystyle\frac{1}{(u_{1}u_{2})^{2}\gamma \tilde{\gamma}}\left ( \tilde{a},\tilde{b}\right ) \left ( \begin{array}{cccccc } u_{1}u_{2 } & & 1\\ 0 & & 1 \\ \end{array } \right ) \left ( \begin{array}{cccccc } 1 & & 1 \\ 0 & & u_{1}u_{2}\\ \end{array } \right)\left ( \begin{array}{cccccc } a \\ b \\ \end{array } \right ) \nonumber \\ & = & \displaystyle\frac{1}{u_{1}u_{2}\gamma\tilde{\gamma } } \left[\tilde{a}(a + 2b ) + b\tilde{b}\right ] . \nonumber\end{aligned}\ ] ] then we have @xmath264 . according to the uni - modular rule in the tiling @xmath265 , therefore we get @xmath266 + let us compute now @xmath267 + @xmath268 + 1 \nonumber \\ & = & \displaystyle\frac{1}{u_{1}u_{2}\gamma \tilde{\gamma } } \left[a(\tilde{a } + 2\tilde{b } ) + b\tilde{b } + u_{1}u_{2}\gamma \tilde{\gamma}\right ] \nonumber \\ & = & \displaystyle\frac{1}{u_{1}u_{2}\gamma \tilde{\gamma}}\left[a(\tilde{a } + 2\tilde{b } ) + b\tilde{b } + ( \tilde{a}b - a\tilde{b})\right ] \nonumber \\ & = & \displaystyle\frac{(a + b ) ( \tilde{a}+\tilde{b } ) } { u_{1}u_{2}\gamma \tilde{\gamma } } . \nonumber \end{aligned}\ ] ] then we have @xmath269^{2 } & = & \displaystyle\frac{(a + b)^{2}}{u_{1}u_{2}\gamma^{2 } } \times \displaystyle\frac{(\tilde{a } + \tilde{b})^{2}}{u_{1}u_{2 } \tilde{\gamma}^{2 } } \nonumber \\ & = & t(j_{k})t(j_{k+1 } ) . \nonumber \end{aligned}\ ] ] we prove in the same way that the sequence of the points @xmath270 lying on the diagonal ray with origin @xmath271 satisfies the relation : @xmath272^{2 } = t(j'_{k})t(j'_{k-1 } ) . \hspace{3 cm}\ ] ] all the squares in the oblique band delimited by the diagonal rays with origins @xmath223 and @xmath224 satisfy the uni - modular rule due to properties of the @xmath1-tiling @xmath2 . this and relations @xmath273 and @xmath274 , by virtue of remark @xmath248 , allows us to conclude that the diagonal rays with origins at the vertices of the root @xmath180 of the quiver @xmath4 in @xmath214 coincide with horizontal lines of the modelled quiver @xmath147 associated with the seed @xmath275 @xmath60 + we proved theorem @xmath276 for a seed @xmath137 whose associated quiver has forks composed by two arrows entering ( or leaving ) the joint . for a seed @xmath131 whose associated quiver @xmath135 has at least one fork consisting of one arrow entering and one arrow leaving the joint there are two possibilities . one possibility is to perform a mutation on one of the fork vertices and thus reduce this case to the one considered in the proof . another possibility is to work directly with the seed @xmath277 . in this case the given proof can easily be modified , namely one has to associate with the vertex labelled by @xmath278 ( or @xmath279 ) in @xmath280 the variable @xmath281 ( or @xmath282 ) and to consider the following underlying graph for @xmath280 with variables : + @xmath283 & u_{3 } \ar@{-}[r ] & u_{4 } \ldots & u_{n-1 } \ar@{-}[l ] & u_{n}u_{n+1 } \ar@{-}[l ] & } $ ] + or @xmath284 & u_{3 } \ar@{-}[r ] & u_{4 } \ldots & u_{n-1 } \ar@{-}[l ] & \displaystyle\left ( \frac{u_{n+1}(1+u_{n-1})}{u_{n}}\right ) \ar@{-}[l ] } $ ] , respectively . let @xmath285 be a seed with @xmath4 of type @xmath0 with forks composed by two arrows both entering or both leaving the joint . in this section we compute cluster variables of the cluster algebra @xmath286 of type @xmath0 by an explicit formula using matrix product ( from theorem @xmath78 ) and by using signed continuant polynomials . + there are two kinds of cluster variables of a cluster algebra @xmath286 of type @xmath287 . cluster variables of @xmath286 either are transjective and correspond to those lying in the translation quiver @xmath125 as embedded in the auslander - reiten quiver of the cluster category or are nontransjective and correspond to those lying on the tubes ( see [ ss - x.1 ] ) . + [ [ computation - of - transjective - cluster - variables - of - a - cluster - algebra - of - type - tildemathbbd_n ] ] computation of transjective cluster variables of a cluster algebra of type @xmath0 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ it is well - known , see [ ars10 ] and also [ ad11 ] that for @xmath288 , the values contained in the frieze @xmath138 give all transjective cluster variables of @xmath286 . therefore the values of the frieze at the points of the modelled quiver @xmath147 associated with the seed @xmath137 are either transjective cluster variables of @xmath286 or products of two transjective cluster variables of @xmath286 . in the @xmath1-tiling @xmath2 , the variables lying on diagonal rays with origins @xmath289 , respectively , which are vertices of the root @xmath180 in @xmath214 , are the transjective cluster variables . their computation is obtained by applying the formula of theorem @xmath78 . the variables lying on the diagonal rays with origins @xmath158 or @xmath160 , respectively , which are extreme vertices of the root @xmath180 in @xmath214 , are products of two transjective cluster variables ( products created by the passage from @xmath138 to @xmath147 ) . recall that the pairs of cluster variables forming these products are given by fractions whose numerators are equal and denominators coincide up to exchanging @xmath150 and @xmath151 ( or @xmath152 and @xmath153 ) , which appear in denominators with exponent one ( see [ bmr09 - 1 ] ) . therefore , a product of these two transjective variables is a perfect square divided by @xmath158 ( or @xmath160 , respectively ) . each value @xmath290 lying on the diagonal rays with origins @xmath158 or @xmath160 gives rise to two transjective cluster variables @xmath291 and @xmath196 as follows : + @xmath292 where + @xmath293^{\dfrac{1}{2}}$ ] and + @xmath294^{\dfrac{1}{2}}$ ] for the diagonal ray with origin @xmath158 or + @xmath295^{\dfrac{1}{2}}$ ] and + @xmath296^{\dfrac{1}{2}}$ ] for the diagonal ray with origin @xmath160 . @xmath60 + [ [ computation - of - nontransjective - cluster - variables - of - cluster - algebra - of - type - tildemathbbd ] ] computation of nontransjective cluster variables of cluster algebra of type @xmath9 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ according to theorem @xmath276 the correspondence between a frieze associated with a seed @xmath137 of type @xmath0 and an @xmath1-tiling associated with a particular seed of type @xmath3 , allows us to regard transjective variables of the cluster algebra @xmath286 of type @xmath0 as those of a particular type @xmath3 . this suggests the idea of looking for nontransjective variables of the cluster algebra @xmath286 also among cluster variables of the same cluster algebra of type @xmath3 . to this end we introduce a property of @xmath1-tilings from [ br10- formula @xmath274 ] . + we call @xmath302 the _ linearization coefficient _ of column @xmath303 . there is a similar property and definition for three successive rows @xmath304 and @xmath305 . + let @xmath306 be elements of some ring @xmath307 , the signed continuant polynomials ( see [ ar12 - 2.1 ] ) are defined recursively as follows : + @xmath308 setting @xmath309 and @xmath310 . it is known according to [ ar12-theorem @xmath311 that cluster variables of a cluster algebra of type @xmath7 either appear as elements of the corresponding @xmath1-tiling , or as continuant polynomials of the linearization coefficients of the @xmath1-tiling . we recall here the technique of computation of the nontransjective cluster variables of cluster algebra of type @xmath7 according to [ ar12 ] . we use the linearization coefficients of columns , the case of rows being analogous . we need a new concept of _ word _ associated with a set of successive columns . + let @xmath312 be a finite set of successive columns of the tiling @xmath2 associated with a boundary @xmath43 . we call _ word associated with the set @xmath313 _ , the portion of the boundary @xmath43 between its intersections with the columns @xmath314 augmented with one step to the left and one to the right . consider the scheme of example @xmath315 and let @xmath313 be the set of successive columns containing the variables from @xmath316 to @xmath93 . the word associated with this set @xmath313 of columns is : @xmath317 . + consider an @xmath1-tiling @xmath2 associated with some boundary @xmath43 with variables in @xmath13 . let @xmath318 be @xmath75 successive columns of the tiling @xmath2 , with linearization coefficients @xmath319 . let @xmath320 , @xmath56 , @xmath57 , @xmath58 be the word associated with this set of columns . then the signed continuant polynomial @xmath321 is equal to + @xmath322 . @xmath60 + we recall that in the case @xmath97 with @xmath88 clockwise oriented arrows and @xmath89 anti - clockwise oriented arrows , the nontransjective cluster variables are lying on two tubes of rank @xmath88 and @xmath89 ( see [ ss - xiii.2.2 ] ) . in case @xmath0 nontransjective variables lie on three tubes such that two are of rank @xmath10 and the third one is of rank @xmath323 ( see [ ss - xiii.2.2 ] ) . consider our quiver @xmath169 of type @xmath324 , with @xmath325 , which is of the form : + + the analysis of orientation of arrows of @xmath169 gives us the equation @xmath326 . from this and the equation @xmath325 we obtain @xmath327 and then @xmath328 . we shall prove that the tube of rank @xmath89 in a particular case @xmath3 corresponding to the seed @xmath211 with @xmath329 can be identified with the tube of rank @xmath323 in the case @xmath0 . according to [ d10-theorem @xmath330 , it is enough to know the variables lying on the mouth of a tube to determine all the variables lying on this tube . + we prove the following theorem which allows us to compute the @xmath323 nontransjective cluster variables lying on the mouth of the tube of rank @xmath323 in the case @xmath0 . + let @xmath331 be an integer , @xmath137 a seed of type @xmath0 and @xmath211 the seed of type @xmath3 associated with @xmath137 . the tube of rank @xmath332 containing nontransjective variables of the cluster algebra @xmath333 with @xmath329 coincides with the tube of rank @xmath323 containing nontransjective cluster variables of the cluster algebra @xmath286 of type @xmath0 . before the proof of theorem @xmath334 , we recall formula @xmath335 from [ adss12 - 1.3 ] , which we are going to use to compute variables lying on the mouths of tubes . for any locally finite quiver @xmath4 with variables ( here @xmath4 is a quiver of type @xmath0 ) , we define a family of matrices with coefficients in @xmath336 $ ] as follows : + for any arrow @xmath337 in @xmath134 we set : + @xmath338 and @xmath339 . + consider now , for all @xmath340 , a reduced walk @xmath341 from a vertex @xmath342 to a vertex @xmath343 in @xmath4 ( for the notion of reduced walk , we refer to [ ass - ii.1.1 ] ) , here @xmath344 , are arrows . + we define @xmath345,\ ] ] where @xmath346 . the empty product is equal to @xmath101 . @xmath349 \left ( \prod_{k=0}^{m}m(d_{k})v_{c}(k+1)\right ) \left [ \begin{array}{cccccc } 1 \\ 1 \\ \end{array } \right ] \qquad \qquad \qquad ( 6)\ ] ] where @xmath350 is the identity matrix by convention . + * proof of theorem 4.6 * it is enough to prove , according to [ d10 - 5.1 ] , that the variables lying on the mouth of the tube of rank @xmath332 in case @xmath3 corresponding to the seed @xmath211 with @xmath329 are all the @xmath323 nontransjective cluster variables lying on the mouth of the tube of rank @xmath323 in the case of the seed @xmath137 of type @xmath0 . according to [ ad11 - 1.3 ] , the nontransjective cluster variables of cluster algebra of type @xmath0 lying on the mouth of the tube of rank @xmath323 correspond to simple @xmath353-modules @xmath354 associated with points between the two joints of forks inclusively , that is with the points @xmath355 . then we obtain @xmath356 simple @xmath353-modules lying on the mouth of the tube . + to compute the cluster variables of the cluster algebra @xmath286 of type @xmath0 corresponding to these simple modules , we identify each point with a walk of length @xmath357 and apply formula @xmath358 . it is easy to see that the obtained variables are equal to the linearization coefficients of columns of the tiling @xmath2 passing through the vertices @xmath359 of the root @xmath180 for our orientation of @xmath4 . on the other hand we know [ ar12 ] that the linearization coefficients of rows and columns of the @xmath1-tiling @xmath2 below the boundary @xmath214 are the variables lying on the mouths of tubes of rank @xmath88 and @xmath89 , respectively , of the auslander - reiten quiver corresponding to @xmath169 of type @xmath3 . therefore we conclude that the @xmath356 cluster variables of cluster algebra of type @xmath0 obtained on the mouth of the tube of rank @xmath323 coincide with the @xmath356 variables on the mouth of the tube of rank @xmath332 in case @xmath3 and are set in the same order . + thus we have found @xmath356 out of @xmath323 cluster variables lying on the mouth of the tube in the case @xmath286 of type @xmath0 . to compute the @xmath323th cluster variable we use the existence of linear recurrence relations between the corresponding values of the frieze @xmath138 at a joint of a fork and its neighbours from [ ks11 - 6 ] . namely , we denote by @xmath360 the auslander - reiten translation ( see [ ass - iv.2.3 ] ) , @xmath361 and by @xmath362 ( note that all cluster variables lying on the mouth of a tube of rank @xmath323 are of this form ) the image by the caldero - chapoton map ( see [ cc06 ] ) of the @xmath353-module @xmath363 . corresponding to the sequence @xmath364 associated with the joint @xmath11 of the quiver @xmath4 , we get the following relation : @xmath365 rewriting this relation as follows : @xmath366 , one can see that each variable @xmath367 lying on the mouth of the tube is the linearization coefficient of a column passing through the variable @xmath364 in the @xmath1-tiling @xmath2 defined by the boundary @xmath214 . this allows us to conclude that the @xmath323th cluster variable also corresponds to a linearization coefficient in the @xmath1-tiling @xmath2 . thus the @xmath323 cluster variables of the cluster algebra @xmath286 of type @xmath0 lying on the mouth of the tube of rank @xmath323 are equal to the linearization coefficients of columns passing through the points in the tiling below the boundary @xmath214 which correspond to the joints of the forks in the quiver @xmath140 . ( recall [ ar12 ] that the sequence of linearization coefficients of the columns of the @xmath1-tiling corresponding to a quiver of type @xmath97 is periodic with the period @xmath89 . since in our case @xmath328 , there are only @xmath323 such coefficients . ) thus we identify the tube of rank @xmath332 in the case @xmath3 ( corresponding to the cluster algebra @xmath333 with @xmath329 ) with the tube of rank @xmath323 in case @xmath0 ( corresponding to the cluster algebra @xmath286 ) . @xmath60 + let @xmath368 be the signed continuant polynomial , @xmath369 with @xmath370 , be a variable of depth @xmath122 on the tube and @xmath371 be a variable lying on the mouth of the tube . according to [ ad11 - 3.5 ] the variables lying on a tube with rank @xmath372 are related by the following relation @xmath373 . + it remains now to determine the nontransjective cluster variables lying on the mouths of the two tubes of rank @xmath10 . + we do this by interpreting the results on quiver representations from [ ss - xiii-2.6-a and b ] in terms of cluster variables . this gives that the nontransjective cluster variables on the mouth of each tube of rank @xmath10 correspond to reduced walks in the seed @xmath137 as follows : + 1 . for the first tube 1 . the unique reduced walk from @xmath150 to @xmath153 2 . the unique reduced walk from @xmath151 to @xmath152 2 . for the second tube 1 . the unique reduced walk from @xmath150 to @xmath152 2 . the unique reduced walk from @xmath151 to @xmath153 consider the quiver @xmath4 of type @xmath145 of example @xmath177 . + @xmath374 & { 3 } & 4\ar[l ] } $ ] @xmath375 & { 3 } & 1\ar[l ] } $ ] . + the root @xmath180 of @xmath4 with variables is : @xmath376 . the boundary @xmath214 associated with @xmath4 is : @xmath377 , this gives the following @xmath1-tiling below the boundary @xmath214 : + @xmath378&u_{3}\ar@{-}[u]& ... &&\\ & & & & & & & & & & & 1\ar@{-}[u]&\displaystyle\frac{1+u_{3}}{u_{1}u_{2}}&&&&&\\ & & & & & & & & & & & u_{1}u_{2}\ar@{-}[u]&2+u_{3}& ... &&&&\\ & & & & & & & & & & u_{4}u_{5}\ar@{-}[r ] & u_{3 } \ar@{-}[u ] & \displaystyle\frac{(1+u_{3})^{2}}{u_{1}u_{2 } } & & \\ & & & & & & & & & & 1 \ar@{-}[u ] & \displaystyle\frac{1+u_{3}}{u_{4}u_{5 } } & & ... & \\ & & & & & & & & & & u_{4}u_{5}\ar@{-}[u]\ar@{.}[dddrrrr ] & 2+u_{3 } & & ... & & \\ & & & & & & & u_{1}u_{2}\ar@{-}[r]&1\ar@{-}[r ] & u_{1}u_{2}\ar@{-}[r]\ar@{.}[dddrrr ] & u_{3}\ar@{-}[u]&v_{3 } & & & & & & & \\ & & & & u_{4}u_{5}\ar@{-}[r]&1\ar@{-}[r]&u_{4}u_{5}\ar@{-}[r]&u_{3}\ar@{-}[u]&\displaystyle\frac{1+u_{3}}{u_{1}u_{2}}&2+u_{3}&v_{1}&v_{2 } & & ... & & \\ & u_{1}u_{2}\ar@{-}[r]&1\ar@{-}[r]&u_{1}u_{2}\ar@{-}[r]&u_{3}\ar@{-}[u ] & \displaystyle\frac{1+u_{3}}{u_{4}u_{5}}&2+u_{3 } & & ... & & & & & & \\ & u_{3}\ar@{.}[dl]\ar@{-}[u]&\displaystyle\frac{1+u_{3}}{u_{1}u_{2}}&2+u_{3}&& ... &&&& ... &&&&&&&&&\\ & & & & & & & & ... & & & ... & \\ } \ ] ] + the oblique band delimited by the diagonal rays with origins @xmath158 and @xmath379 corresponds , according to theorem @xmath276 , to the modelled quiver @xmath147 and contains the transjective cluster variables of cluster algebra of type @xmath145 . the variables @xmath380 and @xmath381 represent the products of two cluster variables on the bottom and upper lines in the modelled quiver @xmath147 and @xmath382 represents some cluster variable in @xmath147 . + the cluster variables in these three positions are computed as follows . + the words associated with @xmath380 , @xmath382 and @xmath381 are @xmath383 , + @xmath384 and @xmath385 . + then we have , by applying theorem @xmath78 , the following results : + @xmath386 @xmath380 and @xmath381 being placed on the bottom and upper lines in @xmath147 , we write these values in the form of a product of two transjective cluster variables as in corollary @xmath388 above . the other position corresponds to the following transjective cluster variable of type @xmath145 : + @xmath389 we compute thus all transjective cluster variables of cluster algebra of type @xmath145 by this method . + we compute now nontransjective cluster variables of cluster algebra of type @xmath145 . because @xmath390 , these cluster variables all lie in tubes of rank @xmath10 . for one of these tubes , say @xmath391 , we have two nontransjective cluster variables @xmath392 and @xmath393 on the mouth . these variables are given by the linearization coefficients of columns of the @xmath1-tiling passing through points corresponding to the joints of forks in @xmath138 . these cluster variables can also be obtained by applying the formula of theorem @xmath394 . we choose to use theorem @xmath394 to compute the linearization coefficient @xmath392 of the column passing through the vertex @xmath258 of the root @xmath180 and to use proposition @xmath395 to compute the linearization coefficient @xmath393 of the column passing through @xmath382 . + for the computation of @xmath392 we apply theorem @xmath394 . that is , first we need to determine the word associated with the column containing the vertex @xmath258 of the root @xmath180 . according to the embedding of the boundary @xmath214 in the plane , this word is : + @xmath396 applying the formula of theorem @xmath394 to this word we obtain : + @xmath397 we apply proposition @xmath395 to compute the linearization coefficient @xmath393 of the column passing through the variable @xmath382 . let us calculate @xmath393 at the vertex @xmath258 at the intersection of the column containing @xmath382 with the boundary @xmath214 . then we have : + @xmath398 = \frac { u_{1}u_{2}u_{4}u_{5 } + ( 1+u_{3})^{2 } } { u_{1}u_{2}u_{3 } } $ ] . + thus , @xmath392 and @xmath393 are nontransjective cluster variables lying on the mouth of the tube @xmath391 . + for the second tube @xmath399 , we associate the modules lying on the mouth with the following reduced walks from @xmath150 to @xmath400 and from @xmath151 to @xmath401 : @xmath402 and @xmath403 . by applying formula @xmath358 , we obtain the corresponding nontransjective cluster variables @xmath404 and @xmath405 : thus @xmath404 and @xmath405 are the nontransjective cluster variables lying on the mouth of @xmath399 . + for the third tube @xmath408 , we associate the modules lying on the mouth with the following reduced walks from @xmath150 to @xmath401 and from @xmath151 to @xmath400 : @xmath409 and @xmath410 . by applying formula @xmath358 , we obtain the corresponding nontransjective cluster variables @xmath411 and @xmath412 : acknowledgements . the author is grateful to ibrahim assem and vasilisa shramchenko for useful discussions and careful reading of the manuscript and also to the department of mathematics of the university of sherbrooke where this work was done . the author gratefully acknowledges support from cida ( canadian international development agency ) fellowship . i. assem , d. simson and a. skowroski , _ elements of the representation theory of associative algebras 1 : techniques of representation theory _ , london mathematical society student texts 65 , cambridge university press , ( 2006 ) . d. simson , a. skowroski , elements of the representation theory of associative algebras , volume 2 : tubes and concealed algebras of euclidean type , in:_london mathematical society student texts _ , vol . 71 , _ cambridge university press _ , 2007 .

the physical sizes of supernova remnants ( snrs ) in a number of nearby galaxies follow an approximately linear cumulative distribution , contrary to what is expected for decelerating shock fronts . this phenomenon has been variously attributed to observational selection effects , or to a majority of snrs being in `` free expansion '' , with shocks propagating at a constant velocity into a tenuous ambient medium . we compile multi - wavelength observations of the 77 known snrs in the magellanic clouds , and argue that they provide a fairly complete record of the sne that have exploded over the last kyr , with most of them now in the adiabatic , sedov phase of their expansions . the roughly linear cumulative distribution of sizes ( roughly uniform in a differential distribution ) can be understood to result from the combination of the deceleration during this phase , a transition to a radiation - loss - dominated phase at a radius that depends on the local gas density , and a probability distribution of densities in the interstellar medium varying approximately as . this explanation is supported by the observed powerlaw distributions , with index , of three independent tracers of density : neutral hydrogen column density , h surface brightness , and star - formation rate based on resolved stellar populations . in this picture , the observed cutoff at a radius of 30 pc in the snr size distribution is due to a minimum in the mean ambient gas density in the regions where supernovae ( sne ) explode . we show that m33 has a snr size distribution very similar to that of the magellanic clouds , suggesting these features , and their explanation , may be universal . in a companion paper ( maoz & badenes 2010 ) , we use our sample of snrs as an effective `` sn survey '' to calculate the sn rate and delay time distribution in the magellanic clouds . the hypothesis that most snrs are in free expansion , rather than in the sedov phase of their evolution , would result in sn rates that are in strong conflict with independent measurements , and with basic stellar evolution theory . [ firstpage ] supernovae : general supernovae : remnants galaxies : individual : lmc , smc

1003.3030

the ecology of galaxies is dominated by supernova ( sn ) explosions , which inject energy and enriched material into the interstellar medium and trigger the formation of the next generations of stars . many fundamental aspects of sne are still poorly understood , both for the core - collapse ( cc ) sn explosions that are thought to end the lives of massive ( @xmath4 ) stars and for the type - ia sne ( sne ia ) that are believed to be the thermonuclear combustions of co white dwarfs ( wds ) that approach the chandrasekhar mass . nevertheless , the two flavors of sne deposit a similar amount ( @xmath5 erg ) of kinetic energy into the surrounding medium , leaving behind supernova remnants ( snrs ) that remain visible for thousands of years . in recent times , the study of young snrs at x - ray wavelengths has emerged as a new way to explore the physics of cc and type ia sn explosions ( see * ? ? ? * and references therein ) , but most known snrs are too old to provide much information about the specific events that originated them ( see discussions in * ? ? ? * ; * ? ? ? * ) . in spite of this , much can be learned by studying the entire population of snrs , young and old , in nearby galaxies . because the timescales for snr evolution are short compared to most of the processes that affect the structure of galaxies , snr catalogues provide a clean record of the environments where sne explode , which can be used to put constraints on the properties of their progenitors ( e.g. * ? ? ? moreover , by considering the properties of the entire population of snrs in a galaxy together with the bulk properties of the gas they are expanding into , we can gain insights into the evolutionary phases of snrs @xcite , the structure of galaxies on scales comparable to the average snr size @xcite , and the cycles of matter and energy in the interstellar medium @xcite . in this paper and in a companion publication ( maoz & badenes 2010 , henceforth paper ii ) , we use the snr population in the magellanic clouds to explore some of these issues . the magellanic clouds ( mcs ) have the advantage of being close enough to study key aspects of their global structure in great detail , and they also harbor a large and extensively observed population of snrs . thus , they are the optimal setting to study the interplay between local density , star formation , sn explosions , and snr evolution on a galactic scale . our ultimate goal , and the focus of paper ii , is to derive the sn rate and delay time distribution ( i.e. , the sn rate as a function of time following a brief burst of star formation ) in the magellanic clouds . however , this can not be done without understanding first the relationship between the lifetime of snrs and the properties of their local environments . this is the subject of the present work . the evolution of snrs has been the subject of many theoretical studies ( e.g. , * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? because accurate ages are only known for a handful of young , often historical objects , any observational tests of these theoretical models must rely on snr size as a proxy for age . given that snrs of equal ages will have different sizes if they expand in different media , this necessarily brings the role of local density into the picture . previous works on the distribution of snr sizes initially focused on the milky way and the mcs ( e.g. * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ) , but more recent efforts have also explored other galaxies in the local group , including m31 @xcite , m33 @xcite , and m83 @xcite . with few exceptions , these studies gave little consideration to the bulk properties of the gas in the galaxies hosting the snrs . in many cases , their samples were also affected by issues of completeness and biases from working at a single wavelength . not surprisingly , these efforts have failed to produce a unified , physically motivated picture of the evolution of snrs in the interstellar medium . here , we propose a first approximation to the problem in the context of the magellanic clouds . this paper is organized as follows . in [ catalog ] , we present a compilation of multi - wavelength observations for the 77 known snrs in the magellanic clouds , and we argue that it provides a fairly complete record of all the sne that have exploded over the last @xmath0 kyr . in [ distribution ] , we examine the size distribution of snrs in both galaxies , and we find that the cumulative distribution is close to linear ( i.e. , the differential is close to uniform ) , within the uncertainties associated with the relatively small number of objects , up to a marked cutoff at a physical radius of @xmath6 30 pc . we also show that the snr size distribution in m33 has very similar properties , suggesting that these features might be widespread . in [ physics ] , we propose a physical model to explain this distribution , based on the assumption that most objects are in the sedov ( adiabatic ) stage of their evolution , and that they rapidly fade away once they transition to the radiative stage , at an age that depends on the local density . under these conditions , the uniform distribution of snr sizes requires that the gas density in the clouds have a probability distribution described by a power law with an index of @xmath71 ( i.e. , @xmath8 ) . in [ densityestimates ] , we test this requirement by examining the distribution of three independent density tracers in the clouds : neutral hydrogen column density , h@xmath3 surface brightness , and star - formation rate based on resolved stellar populations . we find that these tracers are indeed well described by powerlaws with a @xmath71 index . this lends credence to our model , and provides us with the crucial means to estimate the visibility time of snrs in different locations , which we review briefly in [ sec : disc ] , and more extensively in paper ii , where we use it to derive the sn rate and delay time distribution in the clouds . we conclude by summarizing our main results and outlining avenues for future work in [ sec : summary ] . the population of snrs in the mcs has been the object of extensive study for many decades . several catalogues have been compiled at different wavelengths , from the radio to the optical and x - ray . because snrs in the milky way and the mcs are usually discovered in the radio , it is the radio catalogues that often have the largest number of entries , but there is some confusion in the literature regarding the total number of snrs in the clouds . @xcite listed all the discrete radio sources in the parkes survey of the mcs , and found 32 snrs and 12 snr candidates in the lmc , and 12 snrs in the smc . according to @xcite , a revision of the parkes survey complemented with australian telecope compact array ( atca ) data yields 52 confirmed snrs and 20 candidates in the lmc , but these sources are not listed in their paper . instead , the authors present optical spectroscopy of 25 of the 52 confirmed lmc snrs . data from atca was also collected for the smc , where @xcite list 16 confirmed snrs . an optical catalogue of mc snrs is being assembled as part of the magellanic clouds emission line survey ( mcels , * ? ? ? * ) ; the most recent on - line version of this catalogue lists 31 snrs in the lmc and 11 in the smc . in the x - rays , @xcite published an atlas of _ rosat _ sources , which contained 31 lmc snrs . the most recent on - line version of this catalogue , complemented with optical , radio and infrared data , lists 38 confirmed snrs in the lmc . finally , @xcite present _ xmm - newton_observations of 13 smc snrs . taken separately , these catalogues are not adequate for our purposes , because they disagree on the classification of particular sources , and assign different names , sizes and positions to well - known objects . catalogues at different wavelengths can also be sensitive to snrs at different stages of their evolution , so a multi - wavelength compilation is the best way to obtain a global picture of the snr population . at the writing of this paper , there is still no unified , multi - wavelength catalogue of all the snrs in the mcs in the refereed literature , although the magellanic clouds supernova remnants ( mcsnr ) collaboration is assembling one @xcite . the current version of the mcsnr catalogue contains 48 confirmed objects in the lmc and 19 in the smc . we have merged this on - line catalogue with the ones listed in the previous paragraph , taking only the sources that are classified as confirmed snrs in at least one catalogue , and eliminating all duplicates and sources listed only as candidate snrs . this yields 54 confirmed snrs in the lmc and 23 in the smc , which we list in table 1 . in figure [ fig - himaps ] , we plot the positions of these 77 snrs on the hi maps of the mcs published by @xcite and @xcite . because we are interested mostly in reliable sizes ( to derive the histograms in [ distribution ] ) and positions ( to calculate the delay time distribution in paper ii ) , we take the centres and diameters of the snrs from high resolution x - ray observations ( _ chandra _ or _ xmm - newton _ ) , whenever possible . if no high - resolution x - ray data are available , we revert to the sizes and centers listed in the original catalogues , which were determined on a case - by - case basis , usually selecting the wavelength that provided the cleanest measurement ( see table 1 and references for details ) . we have found that high - resolution x - ray data provide the most reliable measurements for several x - ray bright and radio faint objects , which are often small and have boundaries that are hard to resolve clearly using single - dish radio data . for the bulk population , sizes determined at different wavelengths might disagree in individual cases , but this does not have a large impact on the final distribution of sizes , as long as the data sets are of reasonable quality ( e.g. , see figure 7 and related discussion in * ? ? ? * ) . 160mm@llccccccc snr & alternative & & diameter & wave- & refe- & radio flux & catalogue + & name & r.a . & decl . & ( arcsec ) & length & rence & density ( jy ) & entries + + j0448.4@xmath76660 & @xmath9 & 04h 48 m 22s & @xmath766d 59 m 52s & 220 & r & m & 0.04 & m + j0449.3@xmath76920 & @xmath9 & 04h 49 m 20s & @xmath769d 20 m 20s & 133 & r & m & 0.11 & m + j0450.2@xmath76922 & b0450@xmath76927 & 04h 50 m 15s & @xmath769d 22 m 12s & 210 & r & p & 0.27 & p + j0450.4@xmath77050 & b0450@xmath7709 & 04h 50 m 27s & @xmath770d 50 m 15s & 357 & x & w & 0.56 & wsfm + j0453.2@xmath76655 & n4 & 04h 53 m 14s & @xmath766d 55 m 13s & 252 & x & w & 0.11 & wpfm + j0453.6@xmath76829 & b0453@xmath7685 & 04h 53 m 38s & @xmath768d 29 m 27s & 120 & x & g03 & 0.19 & wfsm + j0453.9@xmath77000 & b0454@xmath77005 & 04h 53 m 52s & @xmath770d 00 m 13s & 420 & r & p & 0.26 & p + j0454.6@xmath76713 & n9 & 04h 54 m 33s & @xmath767d 13 m 13s & 177 & x & s06 & 0.06 & wfm + j0454.8@xmath76626 & n11l & 04h 54 m 49s & @xmath766d 25 m 32s & 87 & x & w & 0.11 & wpfsm + j0455.6@xmath76839 & n86 & 04h 55 m 37s & @xmath768d 38 m 47s & 348 & x & w & 0.26 & wpfsm + j0459.9@xmath77008 & n186d & 04h 59 m 55s & @xmath770d 07 m 52s & 150 & o & w & 0.07 & wpfsm + j0505.7@xmath76753 & dem l71 & 05h 05 m 42s & @xmath767d 52 m 39s & 72 & x & b07 & 0.01 & wsm + j0505.9@xmath76802 & n23 & 05h 05 m 55s & @xmath768d 01 m 47s & 111 & x & h06 & 0.35 & wpfsm + j0506.1@xmath76541 & @xmath9 & 05h 06 m 05s & @xmath765d 41 m 08s & 408 & r & m & 0.11 & m + j0506.8@xmath77026 & b0507@xmath77029 & 05h 06 m 50s & @xmath770d 25 m 53s & 330 & r & p & 0.36 & p + j0509.0@xmath76844 & n103b & 05h 08 m 59s & @xmath768d 43 m 35s & 28 & x & b07 & 0.51 & wpfsm + j0509.5@xmath76731 & [email protected] & 05h 09 m 31s & @xmath767d 31 m 17s & 29 & x & b07 & 0.08 & wsm + j0513.2@xmath76912 & dem l109 & 05h 13 m 14s & @xmath769d 12 m 20s & 215 & r & boj07 & 0.20 & wfsm + j0518.7@xmath76939 & n120 & 05h 18 m 41s & @xmath769d 39 m 12s & 134 & x & r08 & 0.35 & wfm + j0519.6@xmath76902 & b0519@xmath7690 & 05h 19 m 35s & @xmath769d 02 m 09s & 31 & x & b07 & 0.10 & wfsm + j0519.7@xmath76926 & b0520@xmath7694 & 05h 19 m 44s & @xmath769d 26 m 08s & 174 & o & w & 0.12 & wpfsm + j0521.6@xmath76543 & @xmath9 & 05h 21 m 39s & @xmath765d 43 m 07s & @xmath9 & r & m & 0.03 & m + j0523.1@xmath76753 & n44 & 05h 23 m 07s & @xmath767d 53 m 12s & 228 & r & k98 & 0.14 & wpfm + j0524.3@xmath76624 & dem l175a & 05h 24 m 20s & @xmath766d 24 m 23s & 234 & o & w & 0.11 & wpfsm + j0525.1@xmath76938 & n132d & 05h 25 m 04s & @xmath769d 38 m 24s & 114 & o & w & 3.71 & wpfsm + j0525.4@xmath76559 & n49b & 05h 25 m 25s & @xmath765d 59 m 19s & 168 & x & p03b & 0.32 & wpfsm + j0526.0@xmath76605 & n49 & 05h 26 m 00s & @xmath766d 04 m 57s & 84 & x & p03 & 1.19 & wpfsm + j0527.6@xmath76912 & b0528@xmath7692 & 05h 27 m 39s & @xmath769d 12 m 04s & 147 & o & w & 0.05 & wpfsm + j0527.9@xmath76550 & dem l204 & 05h 27 m 54s & @xmath765d 49 m 38s & 303 & o & w & 0.14 & wpfsm + j0527.9@xmath76714 & b0528@xmath76716 & 05h 27 m 56s & @xmath767d 13 m 40s & 196 & r & m & 0.10 & fsm + j0528.1@xmath77038 & b0528@xmath77038 & 05h 28 m 03s & @xmath770d 37 m 40s & 60 & r & p & 0.28 & p + j0529.1@xmath76833 & dem l203 & 05h 29 m 05s & @xmath768d 32 m 30s & 667 & r & m & 0.24 & m + j0529.9@xmath76701 & dem l214 & 05h 29 m 51s & @xmath767d 01 m 05s & 100 & r & p & 0.06 & pm + j0530.7@xmath77008 & dem l218 & 05h 30 m 40s & @xmath770d 07 m 30s & 213 & r & m & 0.06 & m + j0531.9@xmath77100 & n206 & 05h 31 m 56s & @xmath771d 00 m 19s & 192 & o & w & 0.33 & wfsm + j0532.5@xmath76732 & b0532@xmath7675 & 05h 32 m 30s & @xmath767d 31 m 33s & 252 & x & w & 0.16 & wsm + j0534.0@xmath76955 & b0534@xmath7699 & 05h 34 m 02s & @xmath769d 55 m 03s & 114 & x & h03 & 0.08 & wpfsm + j0534.3@xmath77033 & dem l238 & 05h 34 m 18s & @xmath770d 33 m 26s & 180 & x & b06 & 0.06 & wpfsm + j0535.5@xmath76916 & snr1987a & 05h 35 m 28s & @xmath769d 16 m 11s & 2 & r & n08 & 0.05 & wm + j0535.7@xmath76602 & n63a & 05h 35 m 44s & @xmath766d 02 m 14s & 66 & x & w03 & 1.43 & wpfsm + j0535.8@xmath76918 & honeycomb & 05h 35 m 46s & @xmath769d 18 m 02s & 102 & x & w & 0.30 & wfsm + j0536.1@xmath76735 & dem l241 & 05h 36 m 03s & @xmath767d 34 m 36s & 135 & x & ba06 & 0.29 & wpfsm + j0536.1@xmath77039 & dem l249 & 05h 36 m 07s & @xmath770d 38 m 37s & 180 & x & b06 & 0.05 & wpfsm + j0536.2@xmath76912 & b0536@xmath76914 & 05h 36 m 09s & @xmath769d 11 m 53s & 480 & r & p & 4.26 & pf + 160mm@llccccccc + j0537.4@xmath76628 & dem l256 & 05h 37 m 27s & @xmath766d 27 m 50s & 204 & r & m & 0.06 & m + j0537.6@xmath76920 & b0538@xmath76922 & 05h 37 m 37s & @xmath769d 20 m 23s & 169 & o & s & 1.01 & fs + j0537.8@xmath76910 & n157b & 05h 37 m 46s & @xmath769d 10 m 28s & 102 & x & c06 & 2.64 & wfsm + j0540.0@xmath76944 & n159 & 05h 39 m 59s & @xmath769d 44 m 02s & 78 & x & w00 & 1.90 & wm + j0540.2@xmath76920 & b0540@xmath7693 & 05h 40 m 11s & @xmath769d 19 m 55s & 60 & x & h01 & 0.88 & wfsm + j0543.1@xmath76858 & dem l299 & 05h 43 m 08s & @xmath768d 58 m 18s & 318 & x & w & 0.21 & wsm + j0547.0@xmath76943 & dem l316b & 05h 46 m 59s & @xmath769d 42 m 50s & 84 & x & w05 & 0.52 & wfsm + j0547.4@xmath76941 & dem l316a & 05h 47 m 22s & @xmath769d 41 m 26s & 56 & x & w05 & 0.33 & wfm + j0547.8@xmath77025 & b0548@xmath7704 & 05h 47 m 49s & @xmath770d 24 m 54s & 102 & x & h03 & 0.05 & wfsm + j0550.5@xmath76823 & @xmath9 & 05h 50 m 30s & @xmath768d 22 m 40s & 312 & r & m & 0.64 & m + + j0040.9@xmath77337 & dem s5 & 00h 40 m 55s & @xmath773d 36 m 55s & 121 & r & m & 0.19 & m + j0046.6@xmath77309 & dem s32 & 00h 46 m 39s & @xmath773d 08 m 39s & 136 & x & h & 0.07 & hmf05 + j0047.2@xmath77308 & ikt2 & 00h 47 m 12s & @xmath773d 08 m 26s & 66 & x & h & 0.46 & hmf05 + j0047.5@xmath77306 & b0045@xmath7733 & 00h 47 m 29s & @xmath773d 06 m 01s & 180 & r & f05 & 0.14 & f05 + j0047.7@xmath77310 & hfpk419 & 00h 47 m 41s & @xmath773d 09 m 30s & 90 & x & h & 0.14 & hm + j0047.8@xmath77317 & ns21 & 00h 47 m 48s & @xmath773d 17 m 27s & 76 & r & f05 & 0.03 & f05 + j0048.1@xmath77309 & ns19 & 00h 48 m 06s & @xmath773d 08 m 43s & 79 & r & f05 & 0.08 & f05 + j0048.4@xmath77319 & ikt4 & 00h 48 m 25s & @xmath773d 19 m 24s & 84 & x & h & 0.15 & hmf05 + j0049.1@xmath77314 & ikt5 & 00h 49 m 07s & @xmath773d 14 m 05s & 116 & x & h & 0.28 & hmf05 + j0051.1@xmath77321 & ikt6 & 00h 51 m 07s & @xmath773d 21 m 26s & 144 & x & h05 & 0.10 & hmf05 + j0051.9@xmath77310 & ikt7 & 00h 51 m 54s & @xmath773d 10 m 24s & 97 & x & m & 0.00 & m + j0052.6@xmath77238 & b0050@xmath7728 & 00h 52 m 33s & @xmath772d 37 m 35s & 323 & r & f05 & 0.22 & mf05 + j0058.3@xmath77218 & ikt16 & 00h 58 m 16s & @xmath772d 18 m 05s & 200 & x & h & 0.07 & hmf05 + j0059.4@xmath77210 & ikt18 & 00h 59 m 25s & @xmath772d 10 m 10s & 158 & x & h & 0.37 & hmf05 + j0100.3@xmath77134 & dem s108 & 01h 00 m 21s & @xmath771h 33 m 40s & 149 & r & f05 & 0.21 & mf05 + j0103.2@xmath77209 & ikt21 & 01h 03 m 13s & @xmath772d 08 m 59s & 62 & x & h & 0.10 & hmf05 + j0103.5@xmath77247 & hfpk334 & 01h 03 m 30s & @xmath772d 47 m 20s & 86 & r & m & 0.05 & m + j0104.0@xmath77202 & b0102@xmath77219 & 01h 04 m 02s & @xmath772d 01 m 48s & 44 & x & g00 & 0.28 & hmf05 + j0105.1@xmath77223 & ikt23 & 01h 05 m 04s & @xmath772d 22 m 56s & 170 & x & p03c & 0.09 & hmf05 + j0105.4@xmath77209 & dem s128 & 01h 05 m 23s & @xmath772d 09 m 26s & 124 & x & h & 0.05 & hmf05 + j0105.6@xmath77204 & dem s130 & 01h 05 m 39s & @xmath772d 03 m 41s & 79 & r & f05 & 0.05 & f05 + j0106.2@xmath77205 & ikt25 & 01h 06 m 14s & @xmath772d 05 m 18s & 110 & x & h & 0.01 & hmf05 + j0114.0@xmath77317 & n83c & 01h 14 m 00s & @xmath773d 17 m 08s & 17 & r & m & 0.02 & m + _ catalogues : _ * f * : @xcite . catalogue of discrete radio sources in the mcs . data from parkes telescope . * f95 * : @xcite . catalogue of discrete radio sources in the lmc , with fluxes at 1.40 , 2.45 , 4.75 , 4.85 and 8.55 ghz . data from parkes telescope . * f02 * : @xcite . catalogue of discrete radio sources in the smc , with fluxes at 1.42 , 2.37 , 4.80 , and 8.64 ghz . data from atca and parkes telescope . * f05 * : @xcite . catalogue of radio snrs in the smc . data from atca . * h * : @xcite . catalogue of x - ray bright snrs in the smc . data from _ xmm - newton_. * m * : mcsnr on - line database , ( r. williams et al . ) http://hoth.ccssc.org / mcsnr/. * p * : @xcite . optical spectroscopy of radio snrs in the lmc . radio data from atca , optical data from siding spring observatory and south african astronomical observatory . * s * : mcels on - line catalogue ( c. smith et al . ) , http://www.ctio.noao.edu/mcels/snrs/snrcat.html . optical data from ctio . * w * : @xcite . x - ray atlas from _ rosat _ , also including radio and optical observations . available on - line at http://www.astro.illinois.edu/projects/atlas/index.html . _ other references : _ b06 : @xcite ; ba06 : @xcite ; b07 : @xcite ; boj07 : @xcite ; c06 : @xcite ; g00 : @xcite;g03 : @xcite ; h01 : @xcite ; h03 : @xcite ; h05 : @xcite ; h06 : @xcite ; k98 : @xcite ; n08 : @xcite ; p03 : @xcite ; p03b : @xcite ; p03c : @xcite ; r08 : @xcite ; s06 : @xcite ; w00 : @xcite ; w03 : @xcite ; w05 : @xcite . with only one exception ( snr j0051.9@xmath77310 in the smc , see table 1 ) , all the snrs in the mcs are detected in the radio , and their flux densities can be found either in the mcsnr catalogue or in the general multiband radio source catalogues of @xcite and @xcite . we have listed these flux densities in table 1 , at 1.4 ghz when available , and at higher frequencies when there was no 1.4 ghz detection . in figure [ fig - radiofluxes ] , we plot the radio flux densities as a function of snr diameter , adopting the fiducial distances of 50 kpc to the lmc @xcite and 60 kpc to the smc @xcite to convert angular sizes to linear dimensions ( we will use these distances for the remainder of the paper ) . the flux densities of the smc snrs have been multiplied by a factor 1.44 for comparison to the lmc snrs . remarkably , there appears to be a `` floor '' in snr flux density at @xmath10 mjy , with only a few objects being fainter than this limit . limits for the selection of candidate snrs in both galaxies , accounting for the flux limit per beam and the number of beams per object of given diameter ( see * ? ? ? * ; * ? ? ? the smc limit has also been scaled by 1.44 . there is a real paucity of snrs below a flux density of @xmath10 mjy at 50 kpc , corresponding to a luminosity of @xmath11 w hz@xmath12 . [ fig - radiofluxes],scaledwidth=32.0% ] the degree of completeness of our sample is hard to estimate , because we include objects that were found using a very hetrogeneous mix of selection criteria , but a number of factors indicate that we should not be missing a large number of snrs . first , the floor at 50 mjy is much brighter than the nominal sensitivity of the atca / parkes radio surveys , which at 1.4 ghz is around 0.3 mjy per beam in the lmc @xcite , and 1.8 mjy per beam in the smc @xcite , with beam sizes of @xmath13 and @xmath14 , respectively . even requiring at least a @xmath15 detection for the selection of snr candidates , and allowing for the fact that large snrs should be harder to identify at the same level of significance , the vast majority of our objects are safely above any conservative detection limits in the radio ( see figure [ fig - radiofluxes ] ) . second , the systematic searches for snrs on the atca / parkes surveys by @xcite and @xcite have not uncovered a substantial population of previously unknown snrs between the radio floor and the detection limits . we note that the total number of confirmed snrs reported ( but not listed ) by @xcite in the lmc is very close to our own count ( 52 vs. 54 ) . third , the existence of a substantial population of radio detected snrs that are masquerading as other kinds of sources is unlikely in the presence high quality multi - wavelength data . small , radio - faint snrs that might be mistaken for background sources are often young objects with bright x - ray emission like snrs j0509.5@xmath76731 ( [email protected] ) or j0505.7@xmath76753 ( dem l71 ) . older , more extended objects that might be confused with surrounding or nearby hii regions are usually identified at optical wavelengths by diagnostic quantities like the [ s ii]/h@xmath3 ratio , which can be used to distinguish shocked nebulae from photoionized gas @xcite . a related concern is that some snrs might be `` lost '' inside superbubbles formed by several nested sn explosions @xcite . in practice , snrs evolving inside low - density cavities do not disappear they just become fainter and evolve more slowly , at least until the blast wave reaches the cavity edges . the largest , most prominent superbubble complex in the lmc , 30 doradus , harbors several well - studied snrs , including j0537.8@xmath76910 ( n157b ) @xcite , and many other objects listed in table 1 are associated with superbubbles . as long as the surveys are sensitive enough to find faint snrs , this should not be a major issue . in the absence of a systematic re - analysis of all the available data with a set of homogeneous criteria to identify snrs in both clouds , we can not guarantee that our sample is @xmath16 complete . such an analysis is outside the scope of the present work , but all the evidence seems to indicate that our compilation should not be missing a large number of objects , and it should include few , if any , spurious ones . we conclude that our list of mc snrs is at least _ fairly _ complete ( which is enough for our goals ) , and that the paucity of snrs below @xmath10 mjy is probably real , and not due to any observational selection effects . this suggests that any mc snrs with flux densities between the radio floor and the detection limits must fade relatively quickly . as we shall discuss in [ sec : disc ] , the results of paper ii can be used to argue independently that the snr sample we have compiled has a high degree of completeness . figure [ sizedist ] shows the cumulative and differential histograms of remnant sizes in the lmc and in the smc . in both clouds , the cumulative distributions are roughly linear in remnant size , i.e. , with an equal number of remnants per equal size bin in the differential distributions , up to a cutoff at a physical radius @xmath17 pc . because the data points in a cumulative histogram are not independent of one another , these curves can not be fitted in the usual way ( using the @xmath18 statistic ) . a robust , quantitative estimate for the slope of the distributions can be obtained by performing instead a maximum - likelihood fit , as follows ( see * ? ? ? * for a similar treatment applied to a different problem ) . suppose that a particular model predicts a size distribution @xmath19 , which integrates to @xmath20 , where @xmath21 is the total number of remnants up to @xmath22 . if we bin our data into many small bins between @xmath23 and @xmath22 , each of width @xmath24 , most bins will have zero snrs , and some will have one snr . given the model , the poisson probability of finding @xmath25 remnants in the @xmath26th bin , for which the model predicts @xmath27 remnants , is @xmath28 we will define the likelihood of a given model as the product of these probabilities . the logarithm of the likelihood , considering that @xmath25 always equals either 0 or 1 , simplifies to @xmath29 - n , \label{loglike}\ ] ] where the summation is only over the specific data values of the snr radii . a power law of index @xmath3 , having the proper normalization , will have the form @xmath30 inserting in eq . [ loglike ] , differentiating @xmath31 with respect to @xmath3 , and equating to zero to find the maximum gives the maximum likelihood solution , @xmath32 with an uncertainty on @xmath3 of @xmath33 this procedure yields a maximum likelihood index of @xmath34 for the lmc , and @xmath35 for the smc . thus , the lmc appears to indeed have a snr size distribution that is close to uniform . in the smc , the best fit is intermediate between a flat distribution and one that rises linearly with radius , but given the smaller number of snrs , it is consistent with both slopes . from figure [ sizedist ] , it appears that the steeper slope in the smc is driven by the small - radius side of the distribution . indeed , if we fit separately the first 8 points and the following 14 points , the maximum likelihood solution is @xmath36 at small radii , and @xmath37 thereafter . this result confirms the visual impression , but formally it is still consistent with a slope close to zero at all radii at the @xmath38 level . we conclude that the snr size distribution in both clouds is consistent with being roughly uniform , although there are hints for a deviation at small radii in the smc . it is possible that a deficit at small radii is also present in the lmc distribution ( see figure [ sizedist ] ) , but the poisson errors are too large to claim that the data require it . the linear cumulative distribution of snr sizes in the mcs ( and also the milky way ) has been previously noted and discussed by @xcite , @xcite , @xcite , @xcite , @xcite , @xcite , @xcite , and most recently by @xcite . several of these papers also pointed out cutoffs in the distribution . all of these papers considered smaller snr samples , often based on much shallower radio data and , with few exceptions , did not include multi - wavelength observations . some of these authors interpreted the observed size distribution as evidence that most mc snrs are in their `` free expansion '' phase , during which the shock velocity is constant , and inferred that these snrs expand into an extremely low - density medium . alternatively , @xcite and @xcite warned that the observed distribution was the result of selection effects ; most objects they discussed had been selected in x - rays , and the x - ray flux limits then led to the exclusion of larger and fainter remnants , and their faint radio counterparts . our present compilation is bigger , it incorporates the most recent multi - wavelength data , it extends to larger sizes , and most importantly , it is sensitive to radio flux densities two orders of magnitude below the observed luminosity floor . with these data we now confirm the luminosity floor , the uniform size distribution , and the cutoff at @xmath39 pc in the magellanic cloud snrs . these features of the snr size distribution are also present in other galaxies . due to its proximity and face - on orientation , m33 has probably the best snr sample outside of the mcs . the most recent catalogue of m33 snrs has been published by @xcite , and it includes data in the radio , optical , and x - rays ( from the chasem33 survey by _ the distribution of snr sizes given in table 3 of @xcite , which we plot in figure [ sizedistm33 ] , has the same broad characteristics we find in the clouds , but backed by much better statistics ( 137 objects ) . the cutoff at a physical radius of @xmath40 pc is very clear , and the distribution appears remarkably uniform for radii between @xmath41 pc and the cutoff . in this size range , our method to estimate the maximum likelihood index yields @xmath42 . graphically , a linear function with the correct normalization reproduces the cumulative histogram quite well ( see figure [ sizedistm33 ] ) . a significant deficit of snrs at radii below 10 pc is also obvious in the data . @xcite warn that a few small ( young ) snrs might have escaped detection ( see their section 9 ) , but given the numbers involved , most of the deficit may be real . this indicates a deviation at small diameters from the uniform distribution along the lines of the one we noted in the mcs . the available data seem to indicate that the distribution of snr sizes in m33 and the magellanic clouds is indeed close to uniform between @xmath43 pc and a sharp cutoff at @xmath44 pc . in the case of m33 , this conclusion needs to be validated by a careful assessment of the completeness of the snr sample , although from the discussion in 7.3 of @xcite it seems unlikely that a large number of m33 snrs have escaped detection . unfortunately , snr samples of this quality are hard to obtain for other galaxies , so it is difficult to verify how widespread these features of the snr size distribution are in reality . we will return briefly to this issue in [ sec : summary ] . in the following section , we argue that a uniform size distribution , including a deficit of snrs at small radii and a sharp cutoff at a certain radius , arises naturally from the physics of snr evolution and the distribution of densities in the interstellar medium . the evolution of snrs is usually separated into three phases ( e.g. * ? ? ? in the first , free - expansion , phase , the mass swept up by the expanding shock is small compared to the ejected mass , and the shock velocity is constant with time . in the sedov - taylor phase , the swept - up mass is larger than the ejecta mass , and the expansion decelerates . however , the cooling time of the shocked gas is still longer than the age of the remnant , and the evolution is approximately adiabatic . once the cooling time becomes comparable to the age , the snr quickly loses energy radiatively , entering the radiation - loss - dominated snowplough phase , after which it slows down , breaks up , and merges with the interstellar medium . quantitative estimates show that typical snrs should be in their sedov - taylor phases for ages between a few and a few tens of kyrs , and for sizes of order a few to a few tens of parsecs ( e.g. * ? ? ? * ; * ? ? ? * ; * ? ? ? thus , most of the mc snrs in our sample should be in the sedov - taylor phase of their expansions , with radii growing as @xmath45 where @xmath46 is the kinetic energy of the explosion , @xmath47 is the ambient gas density , and @xmath48 is the time . the shock velocity therefore decreases as @xmath49 or equivalently expressed in terms of @xmath50 rather than @xmath48 , @xmath51 reasonably assuming a constant sn rate in the lmc over the past few kyrs , @xmath52 , the naively expected size distribution of snrs in their sedov phase is @xmath53 where we have used eq . [ vsimrhor ] to substitute for @xmath54 . this is contrary to the observed nearly uniform ( @xmath55 ) distribution . as noted above , this has been interpreted before either as evidence for free expansion , or for selection effects , where fading of the older ( and hence larger ) remnants takes them below the detection limits , and culls the pileup in the distribution expected at large sizes for a decelerating population . both explanations are unlikely for our mc snr sample . we propose , instead , that the uniform size distribution arises as a result of the transition from the sedov phase to the radiative phase ( see also * ? ? ? * ; * ? ? ? the radius of this transition depends on ambient density , with snrs evolving in dense environments becoming radiative sooner than snrs in low - density regions . when coupled with the distribution of densities in the clouds , this leads to fewer and fewer sites at which large - radius sedov - phase snrs can exist . we show this as follows . the cooling time of the shocked gas depends on the density as @xmath56 where @xmath57 is the cooling function at temperature @xmath58 . following , e.g. , @xcite , @xcite , or @xcite , @xmath57 can be approximated with piecewise power laws , with a dependence of @xmath59 in the temperature range of relevance for the shocked gas , around @xmath60 . by relating the temperature to the shock velocity @xmath61 , @xmath62 ( where @xmath63 is the proton mass ) , and equating @xmath64 to the age @xmath48 as expressed in eq . [ sedovv ] , one obtains that the transition radius , @xmath65 , scales as ( e.g. * ? ? ? * ) @xmath66 a remnant expanding beyond this radius , at the given ambient density , will enter the radiative phase and quickly fade from view . for a fairly large range of plausible cooling function dependences , e.g. , indices @xmath67 of @xmath68 to @xmath69 , @xmath70 depends on density as @xmath71 to @xmath72 . conversely , the maximum ambient density that will permit a sedov - phase snr of radius @xmath50 is @xmath73 where @xmath74 is likely in the range @xmath75 to @xmath76 . the expected size distribution of remnants will now be @xmath77 where @xmath78 is the distribution of gas densities in a given galaxy , and @xmath79 is the minimum density in that galaxy , on the size scales relevant for snrs . let us parametrize the density distribution as a power law of index @xmath80 , @xmath81 with @xmath82 . then , substituting in eq . [ dndr ] , integrating , and expressing @xmath83 in terms of @xmath50 using eq . [ rhomaxvsr ] , gives a remnant size distribution @xmath84 a uniform size distribution , @xmath85 , is obtained if @xmath86 ( for @xmath87 ) or @xmath88 ( for @xmath89 ) . in other words , the observed , roughly uniform , snr size distribution implies a powerlaw distribution of densities with index @xmath90 . we note that , because of the dependence of @xmath70 on @xmath64 , @xmath70 can depend on the metallicity of the gas , @xmath91 . assuming a cooling function proportional to @xmath91 , eq . [ rmaxwithepsilon ] , in addition to the dependence on @xmath47 and @xmath46 , will depend on @xmath91 as @xmath92 however , in dwarf galaxies like the mcs , where @xmath91 does not change much from one location to another , this will be a small effect . for example , if @xmath93 , changing @xmath91 by a factor 5 would only change @xmath70 by @xmath94 . the existence of a cutoff in the distribution beyond some size is naturally expected in the above picture , given the existence of some minimum density @xmath95 in the regions of the galaxies where sne explode , and in view of the decrease of @xmath83 with radius . at some radius , @xmath83 will equal @xmath95 and the integral in eq . [ dndr ] will therefore become zero . in other words , there will be nowhere in the mcs a region with a density low enough to permit a snr of that size that is still in its bright sedov phase . a deficit of snrs at small radii , as observed in m33 and the mcs , is also expected in this scenario . before the onset of the sedov stage , faster shock velocities will lead to fewer objects observed in the bins with the smallest radii . this onset happens at ages ( sizes ) that depend on the details of the ejecta structure , as well as the ambient density ( see 7 in * ? ? ? * ) , but for most snrs it should occur around a few hundred years ( a few pc ) , which is consistent with the deficits that we have observed in m33 and the mcs . this regime affects only a small number of objects in the mcs , and does not impact any of the arguments made above , so we will ignore it for the remainder of the paper . we have also ignored the deviations from the standard evolutionary picture that can be introduced by the shape of the circumstellar medium excavated by the sn progenitors . @xcite showed that most type ia snrs with known ages have sizes that are consistent with an interaction with a uniform ambient medium , but no such study has been done for cc snrs . the fast stellar outflows expected from the more massive cc sn progenitors will modify the sizes of a few individual snrs at certain stages of their evolution ( e.g. , * ? ? ? * ; * ? ? ? * ) , but most of the objects that we consider here are too large to be expanding in even the most extreme wind - blown cavities . as long as the bulk of the snrs in the sample spend most of their lifetimes in the sedov stage , this should not affect our scenario . similarly , the fact that some snrs evolve inside superbubbles ( e.g. * ? ? ? * ) is naturally incorporated into our picture superbubbles merely become one more of the factors driving the density distribution in the interstellar medium . incidentally , the size distribution of superbubbles also relates to the properties of the interstellar medium , as shown by @xcite for several nearby galaxies , including the smc . as we have seen , a uniform snr size distribution can be understood as the result of sedov expansion , combined with a transition to the radiative phase at an age that depends on the local density , provided that the density of the gas in the interstellar medium follows a distribution close to a power law with an index of @xmath96 , @xmath97 . in this section , we test this hypothesis by examining three indirect tracers of gas density in the magellanic clouds : hi column density ; star - formation rate ( sfr ) based on resolved stellar populations ; and h@xmath3 emission - line surface brightness . these tracers are well suited for our goals because they are valid over a wide range of densities , and the necessary data are available from public surveys that cover the whole extent of the clouds , as described in detail below . we have taken the surface brightness of hi 21 cm line emission in the mcs from the maps of @xcite and @xcite , which combine single - dish parkes and aperture - synthesis atca data to probe both small and large scales in the lmc and the smc , respectively . the 21 cm emission is optically thin , so the surface brightness is directly proportional to the hi column density . since the lmc possesses a fairly face - on ( inclination @xmath98 , * ? ? ? * ) , well - ordered hi disk , the column density should , in turn , be roughly proportional to the volume density @xmath47 . @xcite report that the hi column density distribution of individual `` clouds '' of neutral hydrogen in the lmc follows a log - normal form , rather than a power law . however , their figure 13 suggests that , above a low cutoff of @xmath99 , the distribution does behave as a power law of slope @xmath2 , over at least an order of magnitude . to re - examine this , in figure [ hihist ] we show the differential distribution of hi column density in the individual beam - sized pixels of the @xcite lmc map ( as opposed to the cumulative plot for clouds shown in * ? ? ? we see that the hi column in the lmc does follow an index @xmath96 power law fairly well , between a column of @xmath100 and @xmath101 . the observed low cutoff in the column density is unavoidable because of the integration through the disk and over the beam size ( every line of sight is basically sampling the densest regions at that point ) . in regions with low density , the h atoms might be ionized , as in the `` warm ionized '' phase of the interstellar medium @xcite , so the tracer may become less reliable there . it is quite plausible that , in the regions where sne actually explode , the underlying distribution of densities also reaches a minimum as we recall , a minimum density is required in order to reproduce the observed upper cutoff in snr size but this does not necessarily correspond with the lower threshold in the hi distribution . to illustrate this , we have calculated the mean hi columns in each of the spatial `` cells '' defined by @xcite and @xcite that contain snrs ( see [ sec : sfr ] below for a description of the cells ) , which we display with the horizontal rulers in figure [ hihist ] . we note that , in the lmc , these cells have average column values between @xmath102 ( close to , but higher than the low hi cutoff ) and @xmath103 , although most snrs appear clustered around @xmath104 . that intersect the data at a column of @xmath105 . the column density values for the cells that contain the snrs in both galaxies have been represented by tick marks on the horizontal rulers.,scaledwidth=32.0% ] as shown in figure [ hihist ] , the distribution of hi column densities in the smc is flat over the same range , although the rise and fall from the plateau happen at the same densities as the rise and fall of the powerlaw in the lmc . while we do not know the reason for this , we speculate that it may be related to an smc geometry that is elongated along our line of sight , and the integration effect that results . the actual `` depth '' of the smc , whether just a few kpc or as much as 20 kpc , is debated @xcite , but is likely at least a few times larger than that of the nearly face - on lmc . such an integration effect would explain why the smc snrs are found at hi column densities that are a factor @xmath64 higher than the lmc snrs ( see horizontal rulers in figure [ hihist ] ) . two other tracers of density , examined below , do suggest a @xmath96 power law in the smc , as well as in the lmc , and do not display this offset in snr ambient densities . as a second way to estimate the local gas densities and test the powerlaw density distribution hypothesis , we use star formation rates ( sfrs ) calculated from the star - formation history ( sfh ) maps of the mcs published by @xcite and @xcite , which are based on resolved stellar populations . the maps were elaborated using four - band ( _ u , b , v _ , and _ i _ ) photometry from the magellanic clouds photometric survey @xcite , which has a limiting magnitude between 20 and 21 in _ v _ , depending on the local degree of crowding in the images . in each cloud , the data were divided into regions or `` cells '' with enough stars to produce color - magnitude diagrams of sufficient quality , which were then fed into the starfish code @xcite to derive the local sfh for each cell . for the lmc , @xcite divided more than 20 million stars into spatial cells encompassing the central 8@xmath106 of the galaxy ( see their figure 4 ) . in total , they produced 1376 cells for the lmc , most of them @xmath107 in size , and about 50 cells in regions of lower stellar density with a larger size ( @xmath108 ) . for the smc , @xcite divided over 6 million stars into 351 @xmath107 cells , leaving out two areas that are contaminated by galactic globular clusters in the foreground ( see their figure 3 ) . @xcite and @xcite provide the sfh for each of these cells , which we use in paper ii to derive the delay time distribution of sne in the clouds , but for our present purposes we are only interested in the current sfr , which we have obtained by averaging the star formation in each cell over the last 35 myr . the @xcite law relates star formation to gas mass column , @xmath109 , roughly as @xmath110 . if gas density is distributed as @xmath111 , and @xmath112 , then sfr will be distributed as a power law of index @xmath113 . coincidentally , if @xmath114 , the sfr distribution will also have an index @xmath96 powerlaw distribution . @xcite gives an update of the schmidt law , relating star formation rate surface density , @xmath115 , to gas mass column @xmath116 , as @xmath117 as emphasized by @xcite , the schmidt law has a threshold at a given mass column , below which the star - formation rate falls steeply . the exact value of this threshold is disputed : @xcite finds @xmath118 , which corresponds to @xmath119 , but higher and lower threshold values have also been reported in the literature ( see * ? ? ? * for an updated discussion of this issue ) . from eq . [ schmidtlaw ] with @xmath120 , the threshold mass column corresponds to a sfr of @xmath121 , or @xmath122 per @xmath123 cell in the @xcite maps of the lmc . figure [ sfhist ] shows that , above a value of @xmath124 , the sfr is indeed distributed as a power law of index @xmath96 , over more than one order of magnitude in sfr , for both the lmc and the smc . to meaningfully compare the sfrs in cells with different physical areas , we have counted the few @xmath125 cells in the lmc as four @xmath123 cells with the same sfr , and we have scaled down the sfrs of all the @xmath123 cells in the smc by a factor of 1.44 , to account for the fact that this galaxy is 20% more distant than the lmc . that intersect the data at a sfr of @xmath126 . the sfr values for the cells that contain the snrs in both galaxies are represented by tick marks on the horizontal rulers.,scaledwidth=32.0% ] rather than measuring sfr via the resolved stellar populations of the mcs , we can trace it by means of h@xmath3 emission , which gives us a third probe of gas density . h@xmath3 is principally powered by photoionisation from o - type stars , whose numbers are proportional to the sfr . the schmidt law again connects this sfr to the gas mass column . we have examined the continuum - subtracted h@xmath3 emission maps of the lmc and smc from the shassa survey of @xcite . figure [ hahist ] shows that both the lmc and the smc have distributions of h@xmath3 surface brightness that follow power laws of index @xmath2 , this time over 2 orders of magnitude for the smc , and almost 3 for the lmc . as before , through the schmidt law , this implies a power law with index @xmath96 for the gas density distribution over 1 - 2 orders of magnitude in density . surface brightness is independent of distance for such nearby galaxies , and can therefore be compared directly between the lmc and smc . surface brightness in the lmc ( red ) and the smc ( blue ) . dotted lines are power laws of index @xmath96 that intersects the data at @xmath127 dr pixel@xmath12 . the flux values for the cells that contain the snrs in both galaxies are represented by tick marks on the horizontal rulers.,scaledwidth=32.0% ] the density tracers that we have examined here are not free of biases , but they should provide a fairly accurate picture of the density distribution in the interstellar medium of the mcs . specific regions within the clouds , like the lmc bar or 30 dor , will probably have density distributions that are different from the simple powerlaws we have found here ; our analysis only applies to the bulk , statistical properties of the gas on galactic scales . several theoretical studies have adressed the probability distribution of densities in the interstellar medium : see @xcite , @xcite , @xcite , and @xcite . the consensus of these studies is that , despite local inhomogeneities and deviations , a combination of galaxy - wide and local mechanisms conspire to produce a lognormal distribution of densities in the gas for a wide range of conditions . at the higher densities where most stars are formed and most sne explode , the tail of the lognormal distribution is well reproduced by a power law with an index @xmath2 ( see figure 19 in * ? ? ? this is consistent with our findings for the mcs . the derivation in [ physics ] ignores many subtleties in what is clearly a complex problem , involving different physical processes and temporal scales . in this context , the fact that the distribution of gas densities in the mcs seems to behave like a power law with an index @xmath128 does not _ prove _ that our proposed picture for the evolution of snrs is correct . however , our simple ( but physically motivated ) scenario appears to explain the available observations quite well , and is consistent with what we know about the bulk properties of the interstellar gas in the mcs . furthermore , there are indirect ways to test that the basic scenario is sound . in paper ii , we use the compilation of snr observations from table 1 and the sfh maps of @xcite and @xcite to derive the sn rate and delay time distribution in the magellanic clouds . some key results from that exercise , which we describe briefly here , serve as plausibility arguments for our picture of snr evolution . a basic ingredient in the derivation of sn rates is the `` visibility time '' of snrs ( i.e. , the time during which a snr would be visible , if it were there ) . we have identified this time with the transition to the radiative phase in [ physics ] , but without assigning any specific value to it . this can not be done using theoretical arguments alone , because many of the necessary parameters are not known with sufficient accuracy , like the distribution of sn kinetic energies or the normalization of the cooling function ( see eq . [ rmaxwithepsilon ] ) . the visibility time can not be calibrated using individual objects either , because the only snrs that have reliable ages from light echoes or historical observations are much younger than the old snrs in the sedov stage that form the bulk of our sample . in paper ii , this conundrum is solved by imposing a condition on the delay time distribution : that the vast majority of stars more massive than @xmath129 explode as core collapse sne , with very few `` silent '' events that collapse directly to form a black hole without much ejection of material and hence without leaving a snr behind . if this condition holds , an absolute value for the snr visibility time can be obtained by equating the time - integrated sn rate in the temporal bin of the delay time distribution that is associated with core collapse sne ( in the binning used in paper ii , that is all sne with delays shorter than 35 myr ) to the number of massive stars per total stellar mass formed . this procedure yields snr visibility times that depend on the tracer used to determine the local density , and vary between @xmath130 kyr ( for the hi density tracer ) and @xmath131 kyr ( for the sfr - based density tracer see table 2 in paper ii ) . these ages are for regions where the local density corresponds to the mean value of the tracer , @xmath132 and @xmath133 , respectively ( see figures [ hihist ] and [ sfhist ] ) ; for values 10 times lower than the mean , the visibility times could be a factor 3 to 4 longer ( see paper ii for details ) . this is in good agreement with the maximum ages of snr shells inferred by snr - pulsar associations in the milky way , which are @xmath134 kyr @xcite . another key result from paper ii is the total rate of sne ( cc and type ia ) in the magellanic clouds . the rate depends again on the density tracer used to estimate the visibility time , and varies between @xmath135 ( for hi ) and @xmath136 ( using sfr as the tracer ) . these rates , and the visibility times quoted in the previous paragraph , are derived assuming that our snr sample is complete and our evolution scenario is correct . as explained in paper ii , the rates , when normalized per unit stellar mass , are typical of dwarf irregular galaxies , of which the mcs are prototypes . the rates are also in loose agreement with the historical records , which show 2 sne in the mcs in the last 400 yr ( i.e. , roughly @xmath137 , see discussion in 5.3 of * ? ? ? * ) , although this number of events is too small to put strong constraints on the sn rate . adopting the assumption that most snrs are actually in free expansion and not in the sedov stage , as has been invoked by a number of authors to explain the observed uniform size distribution ( see [ distribution ] ) , would lead to inconsistencies . undecelerated sn ejecta reach velocities in excess of 10000 km s@xmath12 , so we would expect the shock velocities in freely expanding snrs to be in the range @xmath138 to 10000 km s@xmath12 . at these velocities , the cutoff radius of @xmath139 pc would translate to maximum ages of 3 to 6 kyr . the 77 snrs in the mcs would then imply a sn explosion in the mcs once every 40 to 80 years , on average , or even more frequently if our snr sample is incomplete . finally , if the visibility time were fixed at 6 kyr , for instance , this would correspond to a cc - sn yield ( i.e. , the number of cc sne per stellar mass formed ) of @xmath140 ( see paper ii ) . to produce so many cc - sne all stars with initial mass above @xmath141 in a standard imf would have to explode , which is in direct contradiction to stellar evolution theory , and to the semi - empirical initial - final mass relation for wds @xcite . in this paper , we have proposed a physically motivated scenario for the evolution of snrs in the interstellar medium of galaxies , and we have applied it to the sample of snrs in the magellanic clouds . we compiled multi - wavelength observations of the 77 known snrs in the mcs from the existing literature ( table 1 and figure [ fig - himaps ] ) . we verified that this compilation is fairly complete , and that the size distribution of snrs is approximately flat , within the allowed uncertainties , up to a cutoff at @xmath40 pc , as discussed by other authors before . we noted that these features are also present in the larger snr sample assembled by @xcite for m33 . in our model , the flat size distribution can be explained if most snrs are in the sedov , decelerating , stage of their expansion , quickly fading below detection as soon as they reach the radiative stage at an age ( size ) that depends on the local density . under these circumstances , a flat distribution of snr sizes arises naturally if the probability distribution of densities in the gas follows a power law with index @xmath96 . to test this hypothesis , we have examined the distributions of three different density tracers in the clouds : hi column density , h@xmath3 flux , and sfr obtained from resolved stellar populations . we have verified that these tracers do indeed indicate a density distribution that follows a power law with index @xmath96 , over a wide range of densities , and that this agrees with our theoretical understanding of the dynamics of the interstellar medium . our scenario implies that snrs will remain visible for different times at different locations in the magellanic clouds , depending on the local density . this visibility time is a key ingredient in the calculation of sn rates and sn delay time distributions , which we derive for the magellanic clouds in paper ii . the absolute value of the visibility time can not be determined from theoretical arguments alone , but we have used the results from paper ii to verify that the range of values we obtain ( @xmath142 to @xmath143 kyr for regions of mean density , depending on the tracer ) is consistent with the existing limits for the maximum ages of snrs , and with basic stellar evolution theory . it would be interesting to extend our analysis of snr sizes and interstellar medium densities to other galaxies in the local group . on the one hand , this would allow us to test the validity of our snr evolution scenario in different settings . on the other hand , the techniques developed here and in paper ii would yield more accurate sn rates and more detailed delay time distributions with larger snr samples , provided the sfhs in the host galaxies could be determined with enough spatial and temporal resolution . a growing number of nearby galaxies have published snr catalogues , but samples of the quality of the one we have assembled here for the magellanic clouds or the one published by @xcite for m33 are hard to obtain . interstellar extinction and distance uncertainties plague the milky way snrs , and the radio detection limits become comparable to the fluxes of the fainter magellanic cloud snrs for distances beyond several hundred kpc ( see * ? ? ? * ) . with a judicious investment of observing time , however , reasonably complete multi - wavelength catalogues of snrs in a few well - suited nearby galaxies could become available in the near future . we are grateful to jos luis prieto , who first suggested using resolved stellar populations around snrs to constrain the properties of sn progenitors . we want to thank the referee for a careful reading of the manuscript , and dennis zaritsky for many helpful discussions about several details of the sfh maps of the mcs . we acknowledge useful interactions with laura chomiuk , bryan gaensler , avishay gal - yam , dave green , jack hughes , amiel sternberg , marten van kerkwijk , jacco vink , and eli waxman . thanks the benoziyo center for astrophysics for support at the weizmann institute of science . d.m . acknowledges support by the israel science foundation and by the dfg through german - israeli project cooperation grant ste1869/[email protected]/15@xmath71 . this research has made use of nasa s astrophysics data system ( ads ) bibliographic services , as well as the nasa / ipac extragalactic database ( ned ) . c. , leiton r. , pizarro s. , 2000 , in astronomical society of the pacific conference series , vol . 221 , stars , gas and dust in galaxies : exploring the links , d. alloin , k. olsen , & g. galaz , ed . , pp .

using the transfer matrix in the angular - momentum space we investigate the impact of trigonal warping on magnetotransport and scaling properties of a ballistic bilayer graphene in the corbino geometry . although the conductivity at the charge - neutrality point and zero magnetic field exhibits a one - parameter scaling , the shot - noise characteristics , quantified by the fano factor and the third charge - transfer cumulant , remain pseudodiffusive . this shows that the pseudodiffusive transport regime in bilayer graphene is not related to the universal value of the conductivity but can be identified by higher charge - transfer cumulants . for corbino disks with larger radii ratios the conductivity is suppressed by the trigonal warping , mainly because the symmetry reduction amplifies backscattering for normal modes corresponding to angular - momentum eigenvalues . weak magnetic fields enhance the conductivity , reaching the maximal value near the crossover field^{-1}$ ] , where ( ) is the nearest - neighbor intra- ( inter- ) layer hopping integral , is the skew - interlayer hopping integral , and ( ) is the outer ( inner ) disk radius . for magnetic fields we observe quasiperiodic conductance oscillations characterized by the decreasing mean value , where . the conductivity , as well as higher charge - transfer cumulants , show beating patterns with an envelope period proportional to . this constitutes a qualitative difference between the high - field ( ) magnetotransport in the case ( earlier discussed in ref . ) and in the case , providing a finite - system analog of the lifshitz transition . = 1

1511.04705

bilayer graphene ( blg ) rises as a top - tier candidate material for the upcoming carbon - based electronics @xcite either due to a tunable band gap @xcite or due to topologically - protected quantum channels along domain walls @xcite , recently proposed to host nonlocal einstein - podolsky - rosen pairs @xcite . low - energy physics of blg is mainly governed by microscopic parameters describing the coupling between the two layers @xcite , some of which are still far from being precisely determined . in the most common bernal stacking , the leading tight - binding parameters : the intralayer hopping integral between nearest neighbors @xmath16ev , and the nearest neighbor interlayer hopping integral @xmath17ev , are both link to the basic blg band - structure characteristics , which are the fermi velocity @xmath18 and the electron effective mass @xmath19 where @xmath20 nm is the lattice constant in a single layer and @xmath21 is the free electron mass . for the next - nearest neighbor ( or _ skew _ ) interlayer hopping integral @xmath6 , the corresponding characteristic is the lifshitz energy @xmath22 which can be defined as a value of the electrochemical potential below which the fermi surface splits into a four - element manifold . @xmath23 is difficult to be directly determined in the experiment ; values of @xmath6 obtained from the infrared spectroscopy covers the range from @xmath24ev @xcite up to @xmath25ev @xcite . on the other hand , in blg even a tiny band structure modification near the dirac point due to @xmath15 may have a significant impact on the minimal conductivity @xcite . for @xmath14 , both the mode - matching analysis @xcite and the kubo formalism @xcite lead to @xmath26 where @xmath27 denotes the universal conductivity of a monolayer @xcite . for @xmath15 , similar theoretical considerations show the conductivity @xmath28 is no longer universal but size - dependent , and monotonically grows with the system size @xmath29 . depending on the crystallographic orientation of a sample , the conductivity approaches @xmath30 @xcite . the transport anisotropy appears as the secondary dirac cones , present for @xmath15 , break the rotational invariance of the dispersion relation @xcite . in principle , the experimental study of @xmath28 for clean bilayer samples should be sufficient to determine @xmath6 @xcite , also in the presence of small interaction - induced energy gap @xcite . ( for instance , the value of @xmath31 reported by mayorov _ et al . _ @xcite coincides with the above - mentioned large-@xmath29 predictions . ) unfortunately , device - to - device conductance fluctuations in real disordered systems may put the effectiveness of such a procedure in question . for this reason , new phenomena , unveiling the influence of @xmath6 on transport characteristics which can be measured in a single - device setup , are desired . in this paper , we consider the corbino geometry , in which a disk - shaped sample is surrounded from both interior and exterior sides with metallic leads [ see fig . [ setcoraux](a ) ] . such a choice is motivated by the absence of boundary effects , and irrelevance of the sample crystallographic orientation . it is worth to stress that anisotropic aspects of quantum transport are still present in such a system as the rotational symmetry is intrinsically broken due to trigonal warping . in effect , the total angular momentum is not conserved , and the standard mode - matching method can not be applied as in simpler systems studied earlier ( namely , the monolayer disk @xcite and the blg disk with @xmath14 @xcite ) . here , we overcome this difficulty by developing a numerical transfer - matrix approach in the angular momentum space . contrary to real - space discretization applied in some related works @xcite , our approach takes benefit from the sparsity of transfer matrix in such a representation allowing highly - efficient ( albeit conceptually simple ) transport calculations for physical disk diameters exceeding @xmath32 m . recently developed linear - scaling approach employing the time evolution scheme @xcite seems to be a possible counterpart but its efficient adaptation for calculating higher charge - transfer cumulants might be difficult . the paper is organized as follows . in sec . ii we present details of the system considered , the effective dirac equation for low - energy excitations , and provide a description of our numerical procedure . iii focuses on the size dependence of the conductivity @xmath33 , the fano factor @xmath34 , and the third charge - transfer cumulant for blg corbino disk at zero magnetic field . in sec . iv , the effects of perpendicular magnetic field on the above - mentioned transport characteristics are discussed . also in sec . iv , the numerical results for the disk are compared with analytical ones , obtained for an artificial system [ a rectangular device with periodic boundary conditions , see fig . [ setcoraux](b ) ] , for which the mode - matching analysis is possible , and the universal magnetoconductance characteristics , relevant for large system sizes and high magnetic fields , are identified . the conclusions are given in sec . v. systems discussed in the paper ( schematic ) . ( a ) the corbino disk in bernal - stacked bilayer graphene . the current flows through the disk - shaped area with the inner radius @xmath35 and the outer radius @xmath36 in a perpendicular magnetic field @xmath37 . the coordinate system and the electrostatic potential @xmath38 ( with @xmath39 ) are also shown . the leads ( white areas ) are modeled as infinitely - doped graphene regions ( @xmath40 ) . ( b ) an artificial ( nanotube - like ) system formed of a blg strip of the width @xmath41 , contacted by two electrodes at a distance @xmath29 in uniform field @xmath42 , upon applying periodic boundary conditions . ] the corbino disk in bernal - stacked blg is depicted schematically in fig . [ setcoraux](a ) . we start our analysis from the four - band low - energy hamiltonian for @xmath43 valley @xcite , which is given by @xmath44 where @xmath45 , with the gauge - invariant momenta @xmath46 ( @xmath47 ) and the symmetric gauge @xmath48 corresponding to the uniform magnetic field parallel to @xmath49-axis . we have further defined the dimensionless parameter @xmath50 , and introduced the polar coordinates @xmath51 . the potential energy @xmath38 depends only on @xmath39 , and the remaining symbols are the same as in eqs . ( [ vfedef][elidef ] ) . as mentioned earlier , the available values of @xmath6 following from different experimental @xcite and computational @xcite approaches are far from being consistent . magnetotransport through blg disk with @xmath14 were discussed in analytical terms in ref . @xcite . in this paper we take the values of @xmath6 varying from @xmath52ev up to @xmath53ev in order to investigate numerically how it affects the system behavior . for the disk area , @xmath54 , we set @xmath55 and the effective dirac equation @xmath56 ( with @xmath57 the fermi energy ) can be written as @xmath58 where @xmath59 @xmath60 , @xmath61 , and the magnetic length @xmath62 . in the absence of trigonal warping ( @xmath63 ) the system possesses a cylindrical symmetry and the effective hamiltonian ( [ hameffu ] ) commutes with the total angular momentum operator @xcite @xmath64 where @xmath65 is the @xmath66 identity matrix , and @xmath67 is one of the pauli matrices . in such a case , the wavefunctions are products of angular and radial parts @xmath68^{t}\left(r\right)$ ] with @xmath69 being an integer angular - momentum quantum number . in the presence of trigonal warping ( @xmath70 ) the cylindrical symmetry is broken , and the wavefunctions do not correspond directly to eigenstates of @xmath71 . a generic workaround has been developed for systems with symmetry - breaking potentials ( or impurities ) , where one can still express wavefunctions as linear combinations of eigenfunctions of an ideal system @xcite . souma and suzuki @xcite considered quantum transport through corbino disks in two - dimensional electron gas and showed that the effects of impurities can be studied numerically , starting from truncated wavefunctions in the basis of angular - momentum eigenstates . here we adapt this method for blg corbino disk , as the term proportional to @xmath72 in the hamiltonian ( [ hameffu ] ) can be regarded as a peculiar type of a symmetry - breaking potential . a general solution of the dirac equation ( [ eq : dirac ] ) can be written as an infinite linear combination of angular momentum eigenfunctions , namely @xmath73 with arbitrary amplitudes @xmath74 , @xmath75 . multiplying the dirac equation ( [ eq : dirac ] ) by the factor @xmath76 ( with @xmath77 an arbitrary integer ) and integrating over the polar angle @xmath78 , we obtain the system of equations @xmath79 where @xmath80 notice that the terms proportional to @xmath72 correspond to the mode mixing due to trigonal warping . the @xmath81 dihedral symmetry of these terms ( coinciding with the blg lattice symmetry ) results in the fact that equation for @xmath82 is coupled only to @xmath83 and @xmath84 , which tremendously simplifies the numerical integration . ( [ eq : uklrow ] ) and ( [ eq : glr ] ) , along with the mode - matching conditions for @xmath85 and @xmath86 ( we model the leads as heavily - doped blg areas ) , allows us to construct a transfer matrix ( see appendix a for details ) which can be utilized in the landauer - bttiker formalism in order to calculate the conductivity and other charge - transfer characteristics . typically , the convergence is reached for the wavefunction truncated by taking @xmath87 in eq . ( [ eq : uklrow ] ) , with @xmath88 depending on the system size , the strength of the trigonal warping , and the applied field ( with the upper value corresponding to @xmath89 m , @xmath90ev , and @xmath91 t ) . other computational aspects are also described in appendix a. a particularly intriguing property , arising from the earlier theoretical study of ballistic transport in blg with skew interlayer hoppings ( @xmath15 ) , is the one - parameter scaling @xcite . in the absence of disorder and electron - electron interactions the scaling function @xmath92 which plays a central role in conceptual understanding of the metal insulator transition @xcite , reproduces the scenario predicted for disordered dirac systems with coulomb interaction @xcite . here , the discussion is complemented by calculating the fano factor @xmath0 quantifying the shot - noise power , and the factor @xmath1 quantifying the third charge transfer cumulant . employing the landauer - bttiker formula for the linear - response regime one can write @xcite @xmath93}{\mbox{tr}\,\mathbf{t } } , \label{landff } \\ \mathcal{r } & = & \frac{\mbox{tr}\left[\,\mathbf{t}\left(\mathbf{1}-\mathbf{t}\right)\left(\mathbf{1}-2\mathbf{t}\right)\,\right]}{\mbox{tr}\,\mathbf{t } } , \label{landrr}\end{aligned}\ ] ] where @xmath94 is the conductance quantum ( with the factor @xmath95 following from the spin and valley degeneracies ) , the dimensionless prefactor ( @xmath96 ) in eq . ( [ landsig ] ) is equal to @xmath97 for the corbino geometry of fig . [ setcoraux](a ) or @xmath98 for the rectangular geometry @xcite , and @xmath99 , with @xmath100 the transmission matrix determined via the transfer matrix ( see appendix a ) . it was shown in refs . @xcite and @xcite that the conductivity of ballistic blg with @xmath15 scales with the system size . for a rectangular geometry , we still get @xmath101 for small systems , while for the larger ones ( @xmath102 nm ) the conductivity can be rationalized as @xmath103,\ ] ] with typical parameters @xmath104 and @xmath105 depending both on @xmath6 and the sample orientation @xcite . the approximating formula given by eqs . ( [ eq : asymp ] ) and ( [ asymparam ] ) applies for generic crystallographic orientation of the sample . however , for some particular orientations one obtains different asymptotic behaviors , including a slower power - law convergence to @xmath106 with the exponent @xmath107 if the current is passed precisely along an armchair direction @xcite , or the oscillating conductivity with the lower bound of @xmath108 if the current is passed precisely along a zigzag direction @xcite . such issues are absent in the corbino geometry , allowing one to focus on the universal scaling properties of the material . minimal conductivity of unbiased blg corbino disk with the radii ratio @xmath109 as a function of the radii difference @xmath110 , specified in the units of @xmath111 [ see eqs . ( [ elldef],[ellval ] ) in the main text ] , for different values of @xmath6 . triangles , squares and circles represent the data obtained numerically for @xmath90ev , @xmath112ev , and @xmath113ev ( respectively ) , with the lines ( dot - dashed , solid and dotted ) depicting the approximating eq . ( [ eq : asymp ] ) with best - fitted parameters listed in table [ tab : comparison - of - pseudodiffusive ] . the inset in the top panel presents the scaling function @xmath114 [ see eq . ( [ betadef ] ) ] , with @xmath115 , extracted numerically from @xmath116 datasets . ( notice that the best - fitted lines almost overlap each other . ) the bottom panel is a zoom in , for smaller radii differences , which allows to depict the region where the actual value of the conductivity may deviate from eq . ( [ eq : asymp ] ) . we further notice that only selected datapoints from the bottom panel are shown in the top panel for clarity . ] same as fig . [ fig : minimal - conductivity - of ] but for the radii ratio @xmath117 . for the parameters in eq . ( [ eq : asymp ] ) corresponding to the lines depicted , see table [ tab : comparison - of - pseudodiffusive - r3o2 ] . ] . [ tab : comparison - of - pseudodiffusive ] least - squares fitted parameters in eq . ( [ eq : asymp ] ) corresponding to the lines in fig . [ fig : minimal - conductivity - of ] . the last column gives the values of @xmath118 , such that the function given by eq . ( [ eq : asymp ] ) matches the actual conductivity with @xmath119 accuracy . [ cols="^,^,^,^,^,^",options="header " , ] the fano factor @xmath0 [ red solid lines ] and the third charge transfer cumulant @xmath1 [ blue dashed lines ] for the corbino disk with @xmath109 as functions of @xmath110 . skew - interlayer hopping @xmath6 is varied between the panels . for large system size , where the fabry - perot oscillations become negligible , the numerical results obtained in the presence of trigonal warping approach the pseudodiffusive values of @xmath120 and @xmath121 [ horizontal lines ] . ] same as fig . [ fig : dependence - of - the ] but for @xmath117 . ] in the numerical calculations presented in this section , we chose radii ratios @xmath109 and @xmath117 , for which the number of nonzero transmission eigenvalues @xmath122 is relatively large even at the charge - neutrality point , allowing one to expect some remainders of the pseudodiffusive behavior known for a monolayer @xcite . it is also worth to mention that such radii ratios are close to that of real corbino device in a monolayer @xcite . a multimode character of the charge transport , combined with a mode mixing due to the trigonal warping , and with a necessity to study large systems in attempt to demonstrate a one - parameter scaling , provides us with an excellent test case to investigate computational aspects of the numerical approach presented in section iib . ( for the details , see appendix a. ) zero - magnetic field results are presented in figs . [ fig : minimal - conductivity - of ] , [ fig : minimal - conductivity - r3o2 ] , [ fig : dependence - of - the ] , and [ fig : dependence - of - r3o2 ] . in order to present the data for different @xmath6 on a single plot , we have defined the length @xmath111 , related to the distance between primary and secondary dirac points in quasimomentum space @xmath123 leading to @xmath124 briefly speaking , the actual conductivity reaches a close - to - asymptotic behavior , described by eq . ( [ eq : asymp ] ) with @xmath115 , for radii differences lying in a relatively narrow interval of @xmath125 ( notice that varying the skew - interlayer hopping from @xmath113ev to @xmath90ev is equivalent to changing the parameter @xmath111 by a factor of @xmath126 between the datasets ) . for larger @xmath29 , lines in figs . [ fig : minimal - conductivity - of ] and [ fig : minimal - conductivity - r3o2 ] , corresponding to eq . ( [ eq : asymp ] ) with the least - squares fitted parameters @xmath106 , @xmath105 , and @xmath127 listed in tables [ tab : comparison - of - pseudodiffusive ] and [ tab : comparison - of - pseudodiffusive - r3o2 ] , are closely followed by the datapoints obtained numerically for @xmath113ev , @xmath128ev , and @xmath53ev . for each case , the specific value of @xmath129 ; i.e , the radii difference above which the approximating eq . ( [ eq : asymp ] ) matches the actual conductivity with an accuracy better than @xmath119 ; is also given in table [ tab : comparison - of - pseudodiffusive ] or [ tab : comparison - of - pseudodiffusive - r3o2 ] . for smaller @xmath29 , in particular for @xmath130 [ see bottom panels in figs . [ fig : minimal - conductivity - of ] and [ fig : minimal - conductivity - r3o2 ] ] , the conductivity becomes nonuniversal ( both parameter- and geometry - dependent @xcite ) approaching @xmath65 for @xmath131 . although the conductivity strongly deviates from the pseudodiffusive value @xmath65 ( even for the lowest considered value of @xmath113ev ) , the shot - noise power and the third charge - transfer cumulant are close to their pseudodiffusive values , i.e. , @xmath132 and @xmath133 , which are usually approached for @xmath134 ( see figs . [ fig : dependence - of - the ] and [ fig : dependence - of - r3o2 ] ) . the approximating eq . ( [ eq : asymp ] ) leads , via eq . ( [ betadef ] ) , to the scaling function of the form @xmath135 in turn , the parametres @xmath127 and @xmath136 defines the position and the slope coefficient at the attractive fixed point ( @xmath137 , @xmath138 ) of the renormalization group flow @xcite . the scaling functions @xmath114 [ see eq . ( [ betadef ] ) ] , calculated numerically for the corbino disks with @xmath109 and @xmath117 ( see insets in figs . [ fig : minimal - conductivity - of ] and [ fig : minimal - conductivity - r3o2 ] , respectively ) , @xmath113ev , @xmath128ev , and @xmath53ev , suggest that one - parameter scaling is universal with respect to the strength of the trigonal warping . these numerical results coincide with the corresponding analysis of ref . @xcite for a rectangular sample . it was pointed out that generic mlg billiard with ( at least ) one narrow opening shows a crossover to the so - called _ quantum - tunneling _ transport regime @xcite , in which charge - transfer characteristics are governed by a limited number of quantum channels , with transmission probabilities showing a power - law decay with the system size . in particular , for an _ undoped _ corbino disk the conductance @xmath139 at zero field reads @xmath140 } \simeq { } \frac{8r_{\rm i}}{r_{\rm o}},\ ] ] where the asymptotic form applies for @xmath141 , and represents contributions from the two channels with angular - momentum quantum numbers @xmath142 and @xmath143 . for blg billiards , the conformal mapping technique employed in ref . @xcite can not be utilized even in the absence of trigonal warping , and the existence of a quantum - tunneling regime is not obvious . for blg corbino disk the conductance , for @xmath14 and at the charge - neutrality point , can be written as @xcite @xmath144 } , \label{tmpm0dsk}\end{aligned}\ ] ] where the transmission probabilities @xmath145 ( with @xmath75 being the angular - momentum quantum number ) correspond to eigenvalues of the matrix @xmath146 in eqs . ( [ landsig][landrr ] ) , and we have further defined @xmath147 + \frac { r_{\rm o}^2\!-\!r_{\rm i}^2 } { 4l_\perp^2 } \sinh\left[\ln\left(\frac{r_{\rm o}}{r_{\rm i}}\right)\right ] . \label{upsidef}\end{aligned}\ ] ] we focus now on the system behavior for @xmath148 @xcite . in such a parameter range , eqs . ( [ caladef],@xmath149[upsidef ] ) lead to @xmath150}{\ln\left(r_{\rm o}/r_{\rm i}\right)}.\ ] ] for any integer @xmath151 the conductance , analyzed as a function of @xmath152 , reaches a local maximum with @xmath153 , following from the presence of two ballistic channels with @xmath154 [ see eqs . ( [ gblg0dsk ] ) and ( [ tmpm0dsk ] ) ] , occurring at @xmath155 similarly , eq . ( [ maxgblg ] ) with half - odd integer @xmath156 approximates a local conductance minimum ( as @xmath157 ) with @xmath158 ; i.e. , _ twice _ as large as the mlg disk conductance , see eq . ( [ gmlgdisk ] ) ; and four dominant channels characterized by @xmath159 . in turn , a quantum - tunneling regime reappears periodically when varying @xmath152 , near any local conductance minimum . the number of well - pronounced minima can roughly be estimated as @xmath160 $ ] , as for @xmath161 the system enters a multimode pseudodiffusive transport regime . on the other hand , for @xmath162 , where the threshold value corresponds to the last conductance maximum following from eq . ( [ maxgblg ] ) with @xmath163 , charge transport is governed by two equivalent channels , with angular momenta @xmath2 and @xmath164 monotonically decaying with increasing @xmath152 . this suggests the system may re - enter a quantum - tunneling limit for @xmath165 . in fact , for a fixed @xmath166 and @xmath167 we have @xmath168 , and such a limit can be regarded as an additional conductance minimum . the asymptotic form of eq . ( [ gblg0dsk ] ) then reads @xmath169 although qualitative features of the quantum - tunneling regime for the corbino geometry are reproduced [ in particular , @xmath170 , the asymptotic conductance is elevated by a large factor of @xmath171 in comparison to the mlg disk case @xcite . radii - ratio dependence of the conductance @xmath172 ( top panel ) and the effective number of transmission channels ( bottom panel ) at a fixed @xmath173 nm and varying @xmath6 ( specified for each line ) . vertical lines mark the values of @xmath174 ( left ) , @xmath175 ( middle ) , and @xmath176 ( right ) following from eq . ( [ calaappr ] ) . the corresponding results for two decoupled mlg disks ( @xmath177 ) are also shown . ] in fig . [ gblgzerofi ] , we compare the conductance ( in the top panel ) and the effective number of transmission channels ( in the bottom panel ) @xmath178 with the index @xmath179 accounting for angular - momenta and layer degrees of freedom , following from eqs . ( [ gblg0dsk][upsidef ] ) [ red lines ] for @xmath14 ; as well as for the case of decoupled mlg disks ( @xmath177 ) [ blue lines ] ; with the numerical results obtained from eqs . ( [ landsig ] ) and ( [ landff ] ) for @xmath113ev , @xmath128ev , and @xmath53ev [ remaining lines ] . the inner radius is fixed at @xmath180 , and the outer radius is varied in the range of @xmath181 . the effects of trigonal warping are visible for all radii ratios considered , and become particularly significant when approaching @xmath182 , corresponding to @xmath176 [ following from eq . ( [ calaappr ] ) ] . in such a parameter range , @xmath183 systematically decreases , whereas @xmath184 systematically increases when enlarging @xmath6 . we also notice that fabry - perot resonances , corresponding to integer @xmath185 , are visible for @xmath15 , indicating the contribution from secondary dirac points . for smaller radii ratios , including @xmath186 ( corresponding to @xmath175 ) and @xmath187 ( corresponding to @xmath174 ) [ see vertical lines in fig . [ gblgzerofi ] ] , the system is close to the pseudodiffusive charge - transport regime . in the @xmath14 case , the conductance minimum is shifted from @xmath186 to @xmath188 due to the influence of transmission channels with higher angular momenta . ( we further notice that the effective number of channels has a local maximum @xmath189 at @xmath190 , where it also precisely matches the value for two decoupled mlg disks , in good agreement with predictions for a quantum - tunneling regime reported earlier in this subsection . ) the trigonal warping noticeably enhances the conductance for @xmath191 ; i.e. , the effect is opposite then for larger @xmath152 ; with some exception for the smallest considered @xmath113ev and @xmath192 , as the disk diameter @xmath193 in such a case . the evolution of @xmath183 and @xmath184 with the trigonal warping strength , illustrated in fig . [ gblgzerofi ] , clearly shows that the role of two transmission channels with angular momenta @xmath2 , prominent for @xmath14 and large @xmath152 , is strongly suppressed for @xmath15 , indicating the gradual crossover to a quantum tunneling regime characterized by @xmath194 . we attribute it to the fact that a @xmath195-wave symmetry of normal modes in leads with @xmath196 does not match the @xmath81 dihedral symmetry of the low - energy hamiltonian for @xmath15 . moreover , for the two largest considered values of @xmath112ev and @xmath53ev , both @xmath183 and @xmath184 are noticeably amplified in comparison to the relevant characteristics for two decoupled mlg disks , signalling the role of quantum states close to the secondary dirac points is important when the crossover to a quantum - tunneling regime occurs . our discussion of the magnetotransport characteristics starts from pointing out that the influence of a uniform magnetic field @xmath42 @xcite on analytical results given by eqs . ( [ gmlgdisk ] ) and ( [ gblg0dsk ] ) can be expressed in a very compact form : namely , it is sufficient to replace @xmath197 and @xmath69 in arguments of hyperbolic cosine by @xcite @xmath198 with @xmath199 denoting the flux piercing the disk area ( @xmath200 ) and the basic period of magnetoconductance oscillations ( @xmath201 ) . all the considered charge - transfer characteristics @xmath33 ( [ landsig ] ) , @xmath34 ( [ landff ] ) , and @xmath202 ( [ landrr ] ) , are predicted theoretically to show periodic oscillations with the magnetic flux @xmath200 [ see eq . ( [ phddef ] ) ] piercing a graphene - based corbino disk @xcite , provided the fermi energy corresponds to the dirac point or to any other landau level ( ll ) . the oscillations appear due to a limited number of transmission channels for @xmath141 , and show a formal analogy with similar effects discussed for a nanotube in a magnetic field applied along the axis @xcite . one should notice , however , that the oscillation period @xmath201 [ see eq . ( [ ph0def ] ) ] for a disk in a uniform , perpendicular field , corresponds to a physical field of @xmath203mt for a typical @xmath204 m disk ( @xmath205 nm ) , while for a @xmath206 nm diameter nanotube in axial field the period , given by the standard aharonov - bohm flux quantum @xmath207 , corresponds to @xmath208 t . similar effects were also considered for graphene disks with strain - induced pseudomagnetic fields @xcite and with the spin - orbit coupling @xcite , extending the list of different theoretical proposals for producing a valley polarization with graphene - based nanostructures @xcite and related systems @xcite . for mlg disks , the oscillations magnitude depends only on the radii ratio @xmath152 . for the conductivity , we have @xmath209 if @xmath210 . for blg disks , the mode - matching analysis for the @xmath14 case @xcite unveils an interference between the two transmission channels for each angular momentum eigenvalue , following from the coupling between the layers quantified by @xmath5 . in turn , the oscillations magnitude depends also on the physical system size . for instance , conductivity oscillations are predicted to vanish ( @xmath211 ) for @xmath212 where @xmath213 . ( [ condvani ] ) with _ even _ @xmath214 [ equivalent to half - odd integer @xmath215 in eq . ( [ maxgblg ] ) ] gives the condition for maximal oscillations , with the magnitude @xmath216 , same in the limit of decoupled layers ( @xmath217 ) . if the fermi energy @xmath57 is close but not precisely adjusted to the dirac point , the oscillations in both mlg and blg disks are still predicted to appear in the limited range of magnetic fluxes , namely @xmath218 with @xmath219 , away from which the conductivity is strongly suppressed . ( [ phidmax ] ) can be rewritten to obtain the corresponding energy range for a given field @xmath220 the limits given by eqs . ( [ phidmax ] ) and ( [ efermax ] ) essentially applies to higher charge - transfer cumulants as well , albeit the dimensionless characteristics @xmath34 and @xmath202 were recently found to show stable , quasiperiodic oscillations in the high source - drain voltage limit @xcite . later this section , we employ the numerical procedure described in section ii in order to find out how the magnetotransport characteristics of blg disks are affected by the trigonal warping ( @xmath15 ) . before discussing magnetotransport of the corbino disk , it is instructive to consider a simpler artificial system depicted schematically in fig . [ setcoraux](b ) . a blg strip of width @xmath41 , contacted by the electrodes at a distance @xmath29 , in a uniform field @xmath42 , and with periodic boundary conditions in the transverse direction , was earlier discussed in the @xmath221 limit @xcite , in which the pseudodiffusive charge transport is predicted to appear near lls . here we primarily focus on the @xmath222 range ( a nanotube - like geometry ) which do not seem to have a direct physical analogue , but can be treated in analytical terms and possesses a discrete spectrum of transmission channels closely resembling the situation in the corbino disk . the wavefunctions for a rectangular sample are presented in appendix b. each spinor component can be written as a product of the exponential function and the airy function , with their arguments scaling as @xmath223 for high fields ; see eq . ( [ phiksamp ] ) . in turn , taking the asymptotic form of the airy function @xmath224 , we find the conductivity in the high - field limit can roughly be approximated by @xmath225\right\ } , \label{eq : gapprox } \end{gathered}\ ] ] where @xmath226 , @xmath227 and @xmath228 denotes the real part of @xmath49 . in the absence of trigonal warping , eq . ( [ eq : gapprox ] ) is replaced by exact expression for the conductivity , namely @xmath229^{-1 } \cos\left(\frac{\pi{}ne}{h}lwb\right)\right\},\end{gathered}\ ] ] where @xmath230 $ ] . a fourier series on the right - hand side represents periodic ( approximately sinusoidal for moderate aspect ratios @xmath231 ) magnetoconductance oscillations with the period @xmath232 @xcite . for @xmath15 , the approximating expression given by eqs . ( [ eq : gapprox],[eq : gammak ] ) is particularly convenient when extracting the beating frequency which governs high - field magnetotransport characteristics of the system . it can be shown that the presence of two relevant transmission eigenvalues for each momentum @xmath233 in eq . ( [ eq : gapprox ] ) follows directly from the fact that the lowest ll in blg has an additional twofold degeneracy absent for higher lls @xcite . we define the two field - dependent effective system sizes determining transmission probabilities in eq . ( [ eq : gapprox ] ) , namely @xmath234 and the corresponding momentum - quantization shifts @xmath235 where @xmath236 the above reasoning holds true for high magnetic fields and arbitrary @xmath237 ratio . for instance , the conductivity in the @xmath221 limit can be approximated by @xmath238 restoring the @xmath14 value ( @xmath239 ) for @xmath240 . conductivity of the rectangular blg sample at the charge - neutrality point as a function of the magnetic field @xmath42 . the sample length is fixed at @xmath241 nm , the width is @xmath242 , and the skew - interlayer hopping integral @xmath6 is varied between the panels . red solid lines : exact numerical results obtained from the mode - matching analysis via eq . ( [ landsig ] ) . blue solid lines : results from eqs . ( [ eq : gapprox],[eq : gammak ] ) for @xmath243 . dashed lines : results from eq . ( [ eq : gapprox2 ] ) for @xmath244 . ] in fig . [ reclowfi ] we compare exact numerical results obtained from eq . ( [ landsig ] ) for @xmath245 and different values of skew - interlayer hopping @xmath6 [ red solid lines ] with corresponding results following from the approximating eqs . ( [ eq : gapprox],[eq : gammak ] ) [ blue solid lines ] and eq . ( [ eq : gapprox2 ] ) [ dashed lines ] . for low magnetic fields , the conductivity monotonically grows with increasing @xmath42 ( for any @xmath15 ) , up to the maximal value ( at @xmath246 ) which may exceed @xmath247 for larger @xmath6-s . for @xmath248 , the effect of trigonal warping on the conductivity is gradually suppressed , leading to @xmath249 , in agreement with the approximating eq . ( [ eq : gapprox2 ] ) . the value of @xmath250 is related to the quasimomentum shifts given by eq . ( [ delkpm ] ) . for a finite aspect ratio @xmath237 , maximal conductivity appears when the average shift is of the same order of magnitude as the distance between primary and secondary dirac points [ given by eq . ( [ elldef ] ) ] , namely @xmath251 what can be rewritten as @xmath252 where we have defined the crossover _ lifshitz field _ it is also visible in the top panel of fig . [ reclowfi ] that the approximating eqs . ( [ eq : gapprox],[eq : gammak ] ) [ blue dashed lines ] reproduces the peak position ( albeit not the maximal conductivity ) with a good accuracy for @xmath90ev . the conductivity maximum at @xmath254 can be regarded as the first effect of the trigonal warping , appearing for samples with finite aspect ratios , but still well - visible for @xmath255 . magnetoconductance oscillations for the rectangular blg sample of the length @xmath256 nm and the width @xmath257 at the charge - neutrality point . top panel : the conductivity as a function of magnetic field for three values of the skew - interlayer hopping integral @xmath6 ( specified for each line ) . the lines for @xmath90ev and @xmath112ev are shifted by @xmath258 and @xmath143 for clarity . bottom panel : consecutive periods of the beating envelope extracted from the data shown in the top panel , for @xmath113ev ( circles ) , @xmath112ev ( squares ) , and @xmath90ev ( diamonds ) . solid lines correspond to the approximating eq . ( [ eq : envper ] ) . ] the second effect , present in systems with @xmath222 , is the emergence of beating patterns . in the top panel of fig . [ rectbeats ] we display the conductivity , as a function of magnetic field , for @xmath259 . quasiperiodic beatings are characterized by the field - dependent envelope period , which can be approximated by @xmath260 while the period of internal oscillations remains the same as in eq . ( [ eq : gexact ] ) . the comparison between the actual envelope periods refined from the numerical data [ datapoints ] and the values following from eq . ( [ eq : envper ] ) [ lines ] is provided in the bottom panel of fig . [ rectbeats ] . for high magnetic fields , eq . ( [ eq : envper ] ) leads to @xmath261 . in turn , the @xmath14 behavior characterized by single - frequency , sinusoidal magnetoconductance oscillations with a size - dependent amplitude [ see eq . ( [ eq : gexact ] ) ] can not be restored for any @xmath15 in the high - field limit . we interpret this effect as a finite - system analog of the zero - temperature conductance instability ( the parameter - driven lifshitz transition ) in bulk blg samples @xcite . conductivity @xmath33 [ @xmath149specified in the units of @xmath262 , the fano factor @xmath0 , and the @xmath1 factor for blg corbino disk with @xmath263 , @xmath264 , and different skew - interlayer hopping integrals @xmath6 ( specified for each panel ) . vertical lines correspond to @xmath265 [ see eq . ( [ bpeakcorb ] ) ] , the horizontal lines mark the pseudodiffusive values @xmath266 and @xmath267 , the dashed lines represent the approximating eq . ( [ eq : przewbc ] ) . ] charge - transfer characteristics @xmath33 , @xmath0 , and @xmath1 for blg disk with @xmath180 , @xmath268 . each of the main panels compares the numerical results for @xmath14 ( blue line ) and @xmath90ev ( red line ) . the vertical grey line in the top panel marks the value of @xmath253 obtained from eq . ( [ bpeakcorb ] ) for @xmath90ev . the inset : period of the beating envelope for three different values of @xmath90ev ( circles ) , @xmath128ev ( orange squares ) , @xmath52ev ( triangles ) . solid lines correspond to eq . ( [ eq : okrcor ] ) . ] our numerical study of magnetotransport through blg corbino disks focuses on two different systems : first one , characterized by @xmath263 and @xmath264 is in the pseudodiffusive charge - transport regime , whereas the second one , with @xmath180 and @xmath268 , shows the beating patterns . in the latter case , the parameters are chosen such that the magnetoconductance oscillations vanish for @xmath14 ( see ref . @xcite ) , in order to illustrate the role of trigonal warping more clearly . the numerical results for @xmath33 , @xmath34 , and @xmath202 are presented , as functions of magnetic field @xmath42 , in figs . [ pdiffdisk ] and [ qbeatdisk ] . in the pseudodiffusive transport regime ( see fig . [ pdiffdisk ] ) some irregular fluctuations , visible for all the discussed charge - transfer characteristics ( and all three values of @xmath6 ) , are suppressed for magnetic fields @xmath269 t . we attribute these fluctuations rather to a fabry - perot interference than to the angular momentum quantization . a striking feature of the data presented in fig . [ pdiffdisk ] is that a distinct conductance peak appears at each panel near the field @xmath270 closely resembling the phenomena described above for a rectangular sample . [ notice that the rightmost equality in eq . ( [ bpeakrect ] ) and eq . ( [ bpeakcorb ] ) are equivalent provided that @xmath115 ] . above the crossover field @xmath253 , the conductivity can be approximated by the formula @xmath272 which is visualized with dashed lines in fig . [ pdiffdisk ] , and can be regarded as a version of eq . ( [ eq : gapprox2 ] ) for blg disk . for the disk with larger radii ratio ( see fig . [ qbeatdisk ] ) all three magnetotransport characteristics exhibit quasiperiodic beating patterns for @xmath90ev . this is not the case for @xmath14 , when each characteristic shows approximately sinusoidal , single - frequency , oscillations with a constant amplitude . as the system no longer possesses a rotational symmetry in the presence of trigonal warping , the magnetotransport can not be simply rationalized by defining two effective system sizes in analogy to eq . ( [ lenpmdef ] ) . comparing to the rectangular system case ( which can be regarded , due to the periodic boundary conditions , as a nanotube - like , i.e. , possessing the rotational symmetry also for @xmath15 ) illustrated in fig . [ rectbeats ] , the beatings are slightly less regular now , clustering in the groups of three . this certain feature of the data displayed in fig . [ qbeatdisk ] suggests the presence of four , rather than two , quasiperiodic components determining the transmission probabilities . it is worth to stress here that the main features of the magnetotransport characteristics still resemble the rectangular ( or a nanotube - like ) system case . in particular , we found that the beating - envelope period can now be approximated by @xmath273 see the inset in fig . [ qbeatdisk ] . on the other hand , the period of the internal oscillations , @xmath274 $ ] , remains the same as the basic period for mlg disks or blg disks with @xmath14 @xcite . also , a high - field behavior of the conductivity averaged over consecutive intervals , each one of the @xmath275 width , can be approximated by eq . ( [ eq : przewbc ] ) . we have investigated , by means of numerical transfer - matrix approach in the angular - momentum space , the effects of the skew - interlayer hopping integrals ( the trigonal warping ) on selected transport characteristics of bilayer - graphene ( blg ) corbino disks . additionally , the analytical mode - matching for an artificial ( nanotube - like ) system , formed of a blg strip upon applying the periodic boundary conditions , was briefly presented and the analogies between these two systems were put forward . if the fermi energy is close to the charge - neutrality point , both the scaling behavior at zero magnetic field ( which would require a comparison between devices of different sizes in an experimental study ) and the single - device magnetotransport discussion unveils several phenomena , in which transport characteristics , such as the conductivity , the fano factor , and the third charge - transfer cumulant , are noticeably affected by the trigonal warping . in the pseudodiffusive transport regime , corresponding to the disk radii ratios @xmath276 , the conductivity shows a one - parameter scaling , in agreement with predictions of ref . @xcite for a rectangular sample . in the corbino geometry , however , the role a crystallographic orientation is eliminated , and the zero - field minimal conductivity can be approximated by @xmath277,\ ] ] where @xmath278 depends only on the skew - interlayer hopping @xmath6 , and varies from @xmath279 nm for @xmath90ev to @xmath280 nm for @xmath113ev . in the uniform magnetic field @xmath42 , the conductivity increases reaching the maximal value @xmath281 near the so - called lifshitz field @xmath253 , for which the magnetic length follows the relation @xmath282 where @xmath283 is defined by @xmath6 and other microscopic parameters : the lattice spacing @xmath284 as well as the nearest - neighbor intra- and interlayer hoppings @xmath4 and @xmath5 . above @xmath253 , the conductivity gradually decreases showing the asymptotic behavior @xmath285 [ see eq . ( [ eq : przewbc ] ) ] . the second and third charge transfer cumulants stay close to their pseudodiffusive values ( @xmath266 , @xmath267 ) when varying the system size or the magnetic field . in the opposite , quantum - tunneling regime ( corresponding to @xmath286 ) , the charge - transfer characteristics are also sensitive to @xmath6 . at zero field and @xmath14 , the transport is governed by two quantum channels with angular momenta @xmath2 . for @xmath15 , the backscattering is enhanced for these channels , as the related wavefunctions no longer match the symmetry of the low - energy hamiltonian . at high magnetic fields , all the charge - transfer characteristics show quasiperiodic beating patterns , with the envelope period @xmath287 [ see eq . ( [ eq : okrcor ] ) ] . most remarkably , the beating patterns , triggered by the trigonal - warping ( @xmath15 ) , remain well - pronounced in the landau quantization regime . ( unlike the average conductivity enhancement , which is usually eliminated by a few tesla field . ) it seems this finite - system version of the lifshitz transition can be related to numerous phenomena appearing in different branches of physics , starting from semiconducting heterostructures @xcite , via strongly - correlated electron systems @xcite , to neutrino physics @xcite , in which scattering the particles between quantum states with different effective masses leads to oscillations in relevant counting statistics , although this time the interference occurs between the evanescent waves . we stress here that finding the lifshitz field @xmath253 , via the asymptotic behavior of the conductivity , or via the beating period , may allow one at least in principle to determine the value of @xmath6 from a single - device magnetotransport measurement . apart from the possible verification of tight - binding parameters in blg hamiltonian , we believe the effects we describe , when confirmed experimentally , will provide a thorough insight into the interplay between massless- and massive - chiral states ruling the quantum transport through blg devices near the charge - neutrality point . for instance , the conductivity enhancement for @xmath288 may dominate the signatures of interaction - related magnetic catalysis phenomenon @xcite ( particularly in finite - size systems ) , and one should precisely distinguish single- and many - body aspects when searching for this intriguing phenomenon in blg . as we have focused on clean ballistic systems , several factors which may modify the transport properties of graphene - based devices , including the disorder @xcite , lattice defects @xcite , or magnetic impurities @xcite , are beyond the scope of this work . some experimental @xcite and numerical @xcite findings suggest that charge - transfer characteristics in the pseudodiffusive transport regime are quite robust against such factors . for the opposite , quantum - tunneling regime , we put forward the following reasoning : in the presence of trigonal warping , the rotational symmetry supposed in earlier studies of mlg @xcite or blg @xcite disks , no longer applies . in spite of this fact , the basic oscillation period [ @xmath289 , in terms of magnetic flux piercing the disk area ] remains unaltered , allowing one to believe that oscillations and beating patterns would appear in a more general situation as well . we thank stephan roche for the correspondence . the work was supported by the national science centre of poland ( ncn ) via grant nos . 2014/14/e / st3/00256 ( ar ) and 2014/15/n / st3/03761 ( gr ) . computations were partly performed using the pl - grid infrastructure . a version of the transfer - matrix approach utilized in this paper is chosen such that the differential equation for transfer matrix can be directly derived from eqs . ( [ eq : uklrow ] ) and ( [ eq : glr ] ) [ see section iib ] for the wavefunctions , and solved analytically in the absence of trigonal warping ( @xmath290 ) . the numerical procedure for @xmath70 is described below . first , a basis set for the transfer matrix is constructed starting from a general angular - momentum eigenstate in the disk area ( @xmath291 ) , corresponding to a given angular - momentum quantum number @xmath292 , a radial part of which can be written as a linear combination of four spinor functions @xmath293 . the coefficients @xmath294 are arbitrary amplitudes and @xmath295 is a normalized spinor function . ( the normalization is carried out in such a way that the total radial current remains constant . ) the wavefunction @xmath296 can be represented as @xmath297 where @xmath298 is the @xmath299 matrix with elements @xmath300_{m , n}=\phi_{m}^{n , l}\left(r\right)$ ] , and @xmath301 . next , the radial part of the actual wavefunction [ see eq . ( [ eq : canm ] ) in the main text ] , describing the system in the presence of trigonal warping , is truncated by the linear combination of limited number ( @xmath302 ) of basis functions , each given by eq . ( [ eq : wf ] ) , corresponding to angular - momentum quantum numbers @xmath303 . namely , we define @xmath304^{t},\ ] ] with @xmath305 where @xmath306^{t}$ ] , and the @xmath307 matrix @xmath308 is to be specified later in this appendix . the relation between wavefunctions at different radii , say @xmath309 and @xmath8 , can be expressed within the propagator @xmath310 as follows @xmath311 substituting eq . ( [ eq : prop ] ) into eq . ( [ eq : uklrow ] ) in the main text , we obtain @xmath312 with the boundary condition @xmath313 where @xmath314 denotes the @xmath315 identity matrix . the sparse matrix @xmath316 in eq . ( [ eq : rroz ] ) has nonzero elements @xmath317_{m , n}$ ] directly following from eq . ( [ eq : uklrow ] ) . defining @xmath318 , \nonumber \\ \alpha_m & = \left\lfloor \left(m-1\right)/\left(2m+1\right)\right\rfloor,\end{aligned}\ ] ] where @xmath319 denotes the largest integer smaller or equal to @xmath320 , one can write down @xmath321_{m , n } & = \left [ -g(l_m , r)\delta_{m , n}+i\nu{}t\delta_{m , n-3}+2\nu{}g(l_m\!-\!2,r)\delta_{m , n-6m-6 } \right]\delta_{\alpha_m,0 } + \left [ g(l_m\!-\!1,r)\delta_{m , n } - it\delta_{m , n-2m-1 } \right]\delta_{\alpha_m,1 } \nonumber \\ & + \left [ g(l_m , r)\delta_{m , n}+i\nu{}t\delta_{m , n+3}-2\nu{}g(l_m\!+\!2,r)\delta_{m , n+2m+4 } \right]\delta_{\alpha_m,2 } - \left [ g(l_m\!+\!1,r)\delta_{m , n}+it\delta_{m , n+6m+3 } \right]\delta_{\alpha_m,3 } , \end{aligned}\ ] ] where @xmath322 , the symbol @xmath323 denotes the kronecker delta of @xmath324 and @xmath325 , @xmath326 , and @xmath327 is given by eq . ( [ eq : glr ] ) in the main text . the angular - momentum cutoff @xmath328 is chosen to be large enough to reach the convergence of the charge - transfer characteristics . we observe that the desired relative precision of @xmath329 requires @xmath328 growing approximately linearly with the system size @xmath110 , the magnetic field @xmath42 , and the skew - interlayer hopping @xmath6 . for instance , in our numerical examples with @xmath90ev , the number of modes varies from @xmath330 for @xmath331 nm and @xmath332 , up to @xmath333 for @xmath334 m and @xmath335 t . it is also worth to mention , that in the magnetic field @xmath336 efficient computation requires the angular momentum quantum numbers are varied in a range @xmath337 where @xmath338 . the numerical integration of eq . ( [ eq : rroz ] ) , with the boundary condition given by eq . ( [ eq : bcuri ] ) and the matrix @xmath316 given by eqs . ( [ eq : lmalpm],[eq : armn ] ) , was carried out by employing the @xmath95-th order explicit runge - kutta method with a fixed step size @xcite . floating - point arithmetic , with up to @xmath339 decimal digits , was used to guarantee the numerical stability when inverting the blocks of the resulting transfer matrix for the whole system ( see below ) . a procedure , described in the above , brings us to the propagator for the disk area @xmath340 . writing down the standard mode - matching conditions for wavefunctions in the leads and in the disk area @xmath341 together with eq . ( [ eq : prop ] ) for @xmath86 , gives us @xmath342 in order to find the transfer matrix for the whole system , we choose the wavefunctions in the leads such that @xmath343 where the vector @xmath344 ( @xmath345 ) contains @xmath346 amplitudes for normal modes in the outer ( inner ) lead . taking the limit of infinite doping in the leads one can disregard the parameter @xmath72 , and write down @xmath347 for @xmath348 or @xmath349 , where @xmath350 @xmath351 denotes the kronecker product of the matrices @xmath352 and @xmath353 , and we have further skipped the physically - irrelevant constant phase . in turn , the matrices @xmath354 defining the wavefunction via eqs . ( [ eq : phir1234 ] ) and ( [ eq : phiimijr ] ) can be found as blocks of the matrix @xmath355 . the transfer matrix thus reads @xmath356 where the rightmost equality maps the matrix blocks of @xmath357 onto the elements of the scattering matrix : @xmath100 , @xmath358 the transmission and reflection matrices for a wavefunction incoming from the inner lead , and @xmath359 , @xmath360 the transmission and reflection matrices for a wavefunction incoming from the outer lead @xcite . in this appendix we present the wavefunctions utilized in sec . ivb to discuss the magnetotransport through a rectangular blg sample at the charge - neutrality point . the low - energy hamiltonian has a general form as given by eq . ( [ hameffu ] ) in the main text , but the potential energy now depends only on the @xmath320-coordinate @xmath361 and we choose the landau gauge @xmath362 . subsequently , the solution of the dirac equation @xmath56 , corresponding to a given transverse wavenumber @xmath233 , can be written as @xmath363 . ( for the sample width @xmath41 and the periodic boundary conditions along @xmath364-direction , we have @xmath226 . ) using the compact notation : @xmath365 , @xmath366^{2/3}$ ] [ see eq . ( [ eq : gammak ] ) in the main text ] , and @xmath367 $ ] , one can write down a zero - energy wavefunction for the sample area ( @xmath368 ) as a combination of four linearly - independent spinors @xmath369 with @xmath370 being arbitrary coefficients . the spinor components in eq . ( [ phiksamp ] ) are given by @xmath371,\nonumber \\ \psi_{2}^{1 } & = & e^{\chi}\left[\,i\tau\text{bi}'\left(\gamma\right)-(\nu/2)\text{bi}\left(\gamma\right)\,\right],\nonumber \\ \psi_{1}^{2 } & = & e^{-\chi^{*}}\text{ai}\left(\gamma^{*}\right),\nonumber \\ \psi_{2}^{2 } & = & e^{-\chi^{*}}\text{bi}\left(\gamma^{*}\right),\nonumber \\ \psi_{1}^{3 } & = & e^{-\chi^{*}}\left[\,i\tau^{*}\text{ai}'\left(\gamma^{*}\right)-(\nu/2)\text{ai}\left(\gamma^{*}\right)\,\right],\nonumber \\ \psi_{2}^{3 } & = & e^{-\chi^{*}}\left[\,i\tau^{*}\text{bi}'\left(\gamma^{*}\right)-(\nu/2)\text{bi}\left(\gamma^{*}\right)\,\right],\nonumber \\ \psi_{1}^{4 } & = & e^{\chi}\text{ai}\left(\gamma\right),\nonumber \\ \psi_{2}^{4 } & = & e^{\chi}\text{bi}\left(\gamma\right ) , \label{psi14rec}\end{aligned}\ ] ] where @xmath372 and @xmath373 are the airy functions @xcite , and we have further defined @xmath374{-2i\nu t^{-2}}$ ] . p. avouris , z. chen , and v. perebeinos , nature nanotechnol . * 2 * , 605 ( 2007 ) ; g. fiori , f. bonaccorso , g. iannaccone , t. palacios , d. neumaier , a. seabaugh , s.k . banerjee , and l. colombo , _ ibid . _ * 9 * , 768 ( 2014 ) . a. schroer , p.g . silvestrov , and p. recher , phys . b * 92 * , 241404 ( 2015 ) . l. m. malard , j. nilsson , d. c. elias , j. c. brant , f. plentz , e. s. alves , a. h. castro neto , and m. a. pimenta , phys . b 76 201401(r ) ( 2007 ) . m. titov , europhys . lett . * 79 * , 17004 ( 2007 ) ; j.h . bardarson , j. tworzydo , p.w . brouwer , and c.w.j . beenakker , phys . lett . * 99 * , 106801 ( 2007 ) ; p. san - jose , e. prada , and d.s . golubev , phys . rev . b * 76 * , 195445 ( 2007 ) . we further notice a very slow convergence of the actual @xmath379 obtained from eqs . ( [ gblg0dsk][upsidef ] ) to the asymptotic form in eq . ( [ gblgasym ] ) . in particular , for @xmath380 a @xmath381 deviation is still observed for @xmath382 . in comparison , for @xmath383 a better agreement with the asymptotic form in eq . ( [ gmlgdisk ] ) occurs for @xmath384 . p. recher , b. trauzettel , a. rycerz , ya.m . blanter , c.w.j . beenakker , and a.f . morpurgo , phys . b * 76 * , 235404 ( 2007 ) ; p. recher , j. nilsson , g. burkard , and bjrn trauzettel , _ ibid . _ * 79 * , 085407 ( 2009 ) . notice that the conductivity given by eq . ( [ eq : gexact ] ) oscillates as a function of magnetic flux @xmath385 with an unusual period of @xmath386 , signalling the fractional carrier charge @xmath387 ; see : r. jackiw , phys . d * 29 * , 2375 ( 1984 ) ; c .- y . hou , c. chamon , and m. mudry , phys . rev . lett . * 98 * , 186809 ( 2007 ) . y. s. gui , c. r. becker , j. liu , v. daumer , v. hock , h. buhmann , and l. w. molenkamp , europhys . lett . * 65 * , 393 ( 2004 ) ; f. nichele , a. n. pal , r. winkler , c. gerl , w. wegscheider , t. ihn , and k. ensslin , phys . rev . b * 89 * , 081306(r ) ( 2014 ) . the integration step needs to be shrunken approximately linearly with the system size and magnetic field , in order to reach the desired relative precision ( @xmath329 ) for the output charge - transfer cumulants . for instance , if @xmath331 nm , the step varies from @xmath388 m ( @xmath332 ) down to @xmath389 m ( @xmath335 t ) ; for @xmath334 m and @xmath332 the desired precision is achieved at @xmath390 m .

for stochastic systems with nonvanishing noise , i.e. , at the desired state the noise port does not vanish , it is impossible to achieve the global stability of the desired state in the sense of probability . this bad property also leads to the loss of stochastic passivity at the desired state if a radially unbounded lyapunov function is expected as the storage function . to characterize a certain ( globally ) stable behavior for such a class of systems , the stochastic asymptotic weak stability is proposed in this paper which suggests the transition measure of the state to be convergent and the ergodicity . by defining stochastic weak passivity that admits stochastic passivity only outside a ball centered around the desired state but not in the whole state space , we develop stochastic weak passivity theorems to ensure that the stochastic systems with nonvanishing noise can be globally locally stabilized in weak sense through negative feedback law . applications are shown to stochastic linear systems and a nonlinear process system , and some simulation are made on the latter further . stochastic differential systems , transition measure , ergodicity , stochastic weak passivity , asymptotic weak stability , stabilization 60h10 , 62e20 , 70k20 , 93c10 , 93d15 , 93e15

1602.07406

stochastic phenomena have emerged universally in many physical systems due to noise , disturbance and uncertainty . the unpredictability to them leads to it a great challenge to stabilize a stochastic system . during the past decades , the stabilization of nonlinear stochastic systems had constituted one of central problems in stochastic process control both theoretically and practically . a great deal of methods emerge as the times require , among which stochastic passivity based control is a popular one . rooting in the passivity theory @xcite and the stochastic version of lyapunov theorem @xcite , the stochastic passivity theory @xcite was developed for stabilization and control of nonlinear stochastic systems . by means of state feedback laws , the asymptotic stabilization in probability can be achieved for a stochastic affine system provided some rank conditions are fulfilled and the unforced stochastic affine system is lyapunov stable in probability @xcite . following this study , lin et . al . @xcite explored the relationship between a stochastic passive system and the corresponding zero - output system , and further established the global stabilization results . parallelizing to the development of stochastic passivity in theory , satoh et . al . @xcite applied this methodology to port - hamiltonian systems , and the solutions for stabilization of a large class of nonlinear stochastic systems are thus available . there are also some reports that stochastic passivity is applied to @xmath1 filtering problem @xcite and controlling stochastic mechanical systems @xcite . despite the large success achieved , stochastic passivity based control seems to only work under the condition that the noise vanishes at the stationary solution ( very often being at the origin ) if a radially unbounded lyapunov function is expected as the storage function . this means that if a stochastic system has nonzero noise port at the stationary solution or has persistent noise port , such a method may be out of action . one of the aims of this paper is to derive the necessary conditions that a stochastic system is stochastically passive , and further give the sufficient conditions to say a stochastic system losing stochastic passivity . equivalently , we prove that there does not exist a radially unbounded lyapunov function rendering the stochastic system to be globally asymptotically stable in probability provided the noise does not vanish at the desired state . the ubiquitousness of such a class of systems in the mechanical and biological fields motivates us to define a kind of novel stability , termed as stochastic asymptotic weak stability , to characterize a certain ( globally ) stable behavior for them . the stochastic asymptotic weak stability requests the system state to be convergent in distribution and ergodic . the former means the state to evolve within a small region around the desired state in a large probability while the latter ensures that the state evolution almost always take place within this region . on the face of it , the stochastic asymptotic weak stability is somewhat similar to the concept of stochastic bounded stability proposed in in that a stochastic system with persistent noise is considered for the same purpose . that concept also means that the state will evolve within a bounded region around a desired state with a large probability which depends on the region radius . especially when the region radius goes infinite , the probability will be one . however , there is evident difference between these two kinds of stability . stochastic bounded stability can not characterize the ergodicity of the state . namely , once the trajectory of the state runs out of the bounded region with a small probability , the coming evolution will take place in a larger bounded region to reach a new " stochastic bounded stability with a larger probability . in addition , the stochastic asymptotic weak stability is different from stochastic noise - to - state and input - to - state stability too . the latter two kinds of stability also serves for characterizing the stable behavior of stochastic systems with nonvanishing noise . they describe the convergence of the expectation of the state , for which the transition measure is controlled by defining a particular function . comparatively speaking , they say nothing about the ergodicity of the state , and do not mean either that the state must evolve within a small region around the desired state . therefore , the stochastic asymptotic weak stability is able to provide more details on characterizing the stable " evolution of the state . in the concept of stochastic asymptotic weak stability , the convergence in distribution describes the evolution trend of the probability distribution of the stochastic system under consideration . as one may know , for a stochastic system the probability density function satisfies the fokker - planck equation @xcite . hence , a usual way to achieve convergence in distribution often starts from analyzing the properties of the solutions of the fokker - planck equation , including the existence , uniqueness and convergence . based on this equation , zhu et al . @xcite studied the exact stationary solution of distribution density function for stochastic hamiltonian systems . liberzon et . @xcite developed a feedback controller to stabilize in distribution a class of nonlinear stochastic systems for which the steady - state distribution density function can be solved from the fokker - planck equation . in addition , probability analysis is another way to serve for achieving the weak stability . zakai @xcite presented a lyapunov criterion to suggest the existence of stationary probability distribution and the convergence of transition probability measure for stochastic systems with globally lipschitzian coefficients . stettner @xcite pointed out that the strongly feller and irreducible process are stable in distribution . khasminskii @xcite constructed a markov chain to analyze the convergence of the probability distribution , and further obtained the markov process to be convergent in distribution @xcite if it is mix sufficiently well " in an open domain and the recurrent time is finite . the conditions that renders the recurrent time to be finite give us large inspiration on developing the stabilizing ways in weak sense for stochastic systems with nonvanishing noise . in this paper , we will show that the recurrent property of a stochastic system is highly relevant to the stochastic passivity behavior . based on this comparison , we define the stochastic passivity not in the whole state space , but only outside a ball centered around the desired state , which is labeled as stochastic weak passivity in the context . within the framework of stochastic weak passivity , we do not need to care whether the noise port of a stochastic system vanishes at the desired state or not . therefore , it is suited to handle the stabilization issue of stochastic differential systems with nonvanishing noise . further , we link the stochastic weak passivity with the stochastic asymptotic weak stability , and develop stabilizing controllers using the stochastic weak passivity to achieve the asymptotic weak stability of stochastic systems . the sufficient conditions for global and local asymptotic stabilization in weak sense are provided by means of negative feedback laws , respectively . the rest of the paper is organized as follows . section @xmath2 presents some preliminaries on stochastic passivity . in section @xmath3 , the loss of stochastic passivity is analyzed and the problem of interest is formulated . in section @xmath4 , we propose the framework of stochastic weak passivity theory and make a link between stochastic weak passivity and asymptotic weak stability . some basic concepts and the main results ( expressed as two stochastic weak passivity theorem and one refined version ) for stabilizing stochastic systems in weak sense are given in this section . section @xmath5 illustrates the efficiency of the stochastic weak passivity theory through two application examples . finally , section @xmath6 concludes this paper and makes a prospect of future research . in this section , we will give a birds - eye view of mathematical systems theory related to stochastic differential systems . we begin with a stochastic differential equation written in the sense of it @xmath7 where @xmath8 , @xmath9 , @xmath10 and @xmath11 are locally lipschitz continuous functions , and @xmath12 is a standard wiener process defined on a complete probability space . assume @xmath13 to be the stochastic process solution and @xmath14 to be the equilibrium solution ( if exists ) of eq . ( [ stochasticequation ] ) , then we have the transition measure of @xmath13 , denoted by @xmath15 , is a function from @xmath16 to @xmath17 $ ] such that @xmath18 where @xmath19 is the @xmath20-algebra of borel sets in @xmath21 , @xmath22 is a borel subset , and @xmath23 denotes the probability function . let @xmath24 be a measure defined on a borel space @xmath19 , then @xmath24 is an probability invariant measure for a stochastic system of eq . ( [ stochasticequation ] ) if @xmath25 and @xmath26 the equilibrium solution @xmath14 of eq . ( [ stochasticequation ] ) is + @xmath27 stable in probability if @xmath28 + @xmath29 locally asymptotically stable in probability if @xmath30 + @xmath31 globally asymptotically stable in probability if @xmath32 in order to analyze the stability of stochastic systems , the stochastic version of the second lyapunov theorem and passivity theorem were proposed in succession . [ stochasticlyapunov ] if there exists a positive definite @xmath33 function @xmath34 with respect to @xmath35 such that @xmath36\leq 0,~\forall~ \bm{x}\in\mathbb{d}\ ] ] then the equilibrium solution @xmath14 of eq . ( [ stochasticequation ] ) is stable in probability , where @xmath37 is a bounded open neighborhood of @xmath14 and @xmath38 $ ] is the infinitesimal generator of the solution of eq . ( [ stochasticequation ] ) , calculated through @xmath39=\frac{\partial ( \cdot)}{\partial \bm{x}}\bm{f}+\frac{1}{2}\mathrm{tr}\left\{\frac{\partial^{2 } ( \cdot)}{\partial \bm{x}^{2}}\bm{h}\bm{h}^\top\right\}\ ] ] if the equality in eq . ( [ lv ] ) holds if and only if @xmath40 , then @xmath14 is locally asymptotically stable in probability . further , if @xmath41 , @xmath42 ( often said that the lyapunov function @xmath34 is radially unbounded ) and @xmath43=0\leftrightarrow \bm{x}=\bm{x}^*$ ] , then @xmath14 is globally asymptotically stable in probability . the stochastic passivity theorem is not handed directly from the literature , but it may be obtained immediately from the definition of stochastic passivity . [ defstochasticpassivity ] an input - output stochastic differential system in the sense of it @xmath44 is said to be stochastically passive if there exists a positive semi - definite @xmath45 function @xmath46 such that @xmath47\leq \bm{u}^\top\bm{y } , ~\forall~\bm{x}\in\mathbb{r}^{n}\ ] ] where @xmath48 is the state , @xmath49 the input , @xmath50 the output , the drift term @xmath51 , the diffusion term @xmath52 and @xmath53 all satisfy the condition of local lipschitz continuity , and @xmath54 , @xmath55 share the same meaning with those in eq . ( [ stochasticequation ] ) . the nonnegative real function @xmath46 is called the storage function , the state where @xmath56 is the stochastic passive state and the inner product @xmath57 is called the supply rate . the negative feedback connection of two stochastic passive systems is stochastically passive . a stochastic passive system with a positive definite storage function is stable in probability if a stochastic passive controller with a positive definite storage function is connected in negative feedback . based on _ result _ 1 , the whole negative feedback connection is stochastically passive . as long as the input of the stochastic passive system ( labeled by the subscript @xmath58 " ) is manipulated according to @xmath62 , then @xmath63 , which means @xmath64\leq 0 $ ] . here , the operator @xmath65 is the stochastic passive controller ( labeled by the subscript @xmath2 " ) defined by @xmath66 . the stability in probability of @xmath13 is immediately from _ theorem _ @xmath58 , so is that of @xmath67 . deterministic passive systems are a kind of special cases of stochastic passive systems . therefore , the frequently - used passive controllers @xcite , such as pid controller , model predictive controller , etc . , can all serve for stabilizing the stochastic passive systems in probability . this section contributes to elaborating that stochastic passivity will vanish either in some stochastic systems or when some control problems are addressed , and further to formulating the problem of interest . as can be known from _ definition _ [ defstochasticpassivity ] , a key point to capture stochastic passivity lies in finding a storage function . we will derive the necessary condition for stochastic passivity in the following , and then get the sufficient condition to say the loss of stochastic passivity . for this purpose , we go back to the stochastic differential equation of eq . ( [ stochasticequation ] ) . [ vanish1]if a stochastic differential equation given by eq . ( [ stochasticequation ] ) has a global solution , then it must be not stable in probability at those states that result in the nonzero diffusion term . let the set of all states result in the nonzero diffusion term be given by @xmath68 for any @xmath69 and @xmath70 , it is expected that @xmath71 must be not true . towards this purpose , we assume @xmath72 for simplicity but without loss of generality ( which means @xmath73 ) , and further construct a real - valued function @xmath74 in the form of @xmath75 based on this function , a positive definite , twice continuously differentiable and bounded real function mapping @xmath76 to @xmath77 is defined by @xmath78 clearly , @xmath79 $ ] and , moreover , @xmath80 in the @xmath81-neighborhood of @xmath82 . in order to finish the proof , we impose the infinitesimal generator @xmath38 $ ] on @xmath83 , and are only concerned about the result at @xmath84 . from eq . ( [ infinitesimalgenerator ] ) , we have @xmath85(\mathbbold{0}_{n})=\frac{12}{23}\text{tr}\big\{\bm{h}(\mathbbold{0}_{n})\bm{h}^\top(\mathbbold{0}_{n})\big\}>0.\ ] ] on the other hand , from the definition of @xmath38 $ ] @xcite we get @xmath85(\mathbbold{0}_{n } ) = \lim_{t\to 0 } \frac{\text{e}^{\mathbbold{0}_{n}}\big[u\big(\bm{x}(t)\big)\big]-u(\mathbbold{0}_{n})}{t}\ ] ] where @xmath82 appearing in @xmath86 $ ] indicates that the initial condition is @xmath87 . since the stochastic differential equation ( [ stochasticequation ] ) has a global solution , there exist a time @xmath88 and a constant @xmath89 so that @xmath90=c\tau\textgreater 0 $ ] . also , since @xmath91&= & \text{e}^{\mathbbold{0}_{n}}\big[~u\big(\bm{x}(\tau)\big)~\big|~u\big(\bm{x}(\tau)\big)<{\epsilon^2}~\big ] \text{p}\big(u\big(\bm{x}(\tau)\big)<{\epsilon^2}~\big|~\bm{x}(0)=\mathbbold{0}_{n}\big ) \notag \\ & + & \text{e}^{\mathbbold{0}_{n}}\big[~u\big(\bm{x}(\tau)\big)~\big| ~u\big(\bm{x}(\tau)\big)\geq{\epsilon^2}~\big ] \text{p}\big(u\big(\bm{x}(\tau)\big)\geq{\epsilon^2}~\big|~\bm{x}(0)=\mathbbold{0}_{n}\big)\notag\end{aligned}\ ] ] and @xmath92\textless \epsilon^2,~\text{p}\big(u\big(\bm{x}(\tau)\big)<{\epsilon^2}~\big|~\bm{x}(0)=\mathbbold{0}_{n}\big)\leq 1\ ] ] together with the fact that @xmath93\rightarrow\text{e}^{\mathbbold{0}_{n}}\big[~u\big(\bm{x}(\tau)\big)~\big| ~u\big(\bm{x}(\tau)\big)\geq{\epsilon^2}~\big]\textless 1\ ] ] where @xmath94 is any positive number , we have @xmath95 we set @xmath94 to be sufficiently small so that @xmath96 where @xmath97 . from the definition , the leftmost term in the above inequality can be calculated by @xmath98 where @xmath99 is a stopping time . then there exists at least one point , denoted by @xmath100 , on the surface of the ball @xmath101 such that @xmath102 note that eq . ( [ stochasticequation ] ) is autonomous , therefore @xmath103 namely , for @xmath104 there always exist a @xmath100 to make the above inequality be true . clearly , the above inequality suggests that eq . ( [ stochasticequation ] ) must be not stable in probability at @xmath84 . it is straightforward to write the inverse negative proposition of _ theorem _ [ vanish1 ] as a corollary . [ vanish2 ] for a stochastic differential equation in the form of ( [ stochasticequation ] ) with a global solution , if it is stable in probability at a desired state @xmath105 ( may be not the equilibrium @xmath14 ) , then @xmath105 must belong to @xmath106 which is defined by @xmath107 note that the above result depends on the condition that the stochastic differential equation ( [ stochasticequation ] ) has a global solution . however , under the condition of local lipschitz continuity , eq . ( [ stochasticequation ] ) has a unique solution only before explosion time . based on this result , we will reveal that there is no explosion for some stochastic passive systems , so it must have a global solution . to this task , attention is turned to the non - explosion condition of a stochastic differential equation proposed by narita @xcite . given a stochastic differential equation represented by eq . ( [ stochasticequation ] ) , if for @xmath108 , there exist two positive numbers @xmath109 and @xmath110 , and a scalar function @xmath111\times \mathbb{r}^{n};\mathbb{r}\big)$ ] such that @xmath112\leq c_t\ ] ] holds for all @xmath113 and @xmath114 , and moreover , @xmath115 then the solutions of eq . ( [ stochasticequation ] ) are of non - explosion , i.e. , the explosion time beginning at any @xmath116 and @xmath117 , denoted by @xmath118 , satisfying @xmath119 in the following , that _ lemma _ @xmath58 is applied to a stochastic passive system yields for a stochastic differential system @xmath120 governed by eq . ( [ stochasticsystem ] ) , if there exists a radially unbounded lyapuonv function so that @xmath120 is stochastically passive , then the unforced version of eq . ( [ stochasticsystem ] ) has a global solution . assume @xmath34 to be the radially unbounded lyapuonv function that suggests @xmath120 to be stochastically passive , then by designating the zero controller to @xmath120 , i.e. , @xmath63 , we have @xmath121\leq 0 $ ] . it is naturally to observe that @xmath34 satisfies eq . ( [ non - explosion2 ] ) . note that the state evolution of this unforced version of @xmath120 is just the same as eq . ( [ stochasticequation ] ) . hence , the solutions of @xmath120 are of non - explosion based on _ lemma _ @xmath58 . namely , eq . ( [ stochasticsystem ] ) has a global solution . from _ proposition _ @xmath58 , one can know that some stochastic passive systems must have a global solution without force . combining this result with _ @xmath58 , we get the necessary condition for saying @xmath120 to be of stochastic passivity , which is expressed as follows . if there exists a radially unbounded lyapunov function that can render a stochastic differential system @xmath120 described by eq . ( [ stochasticsystem ] ) to be stochastically passive , then the unforced diffusion term must vanish at the stochastic passive state . from _ theorem _ @xmath58 of stochastic lyapunov theorem , @xmath120 is stable in probability at the stochastic passive state with the zero controller . this together with _ proposition _ @xmath58 and _ corollary _ @xmath58 yields the result to be true . we further express the inverse negative proposition of _ theorem _ @xmath3 to get the sufficient condition for loss of stochastic passivity . [ vanish3 ] if the unforced diffusion term @xmath122 in a stochastic differential system @xmath120 in the form of ( [ stochasticsystem ] ) does not vanish at any state @xmath123 , then there does not exist any radially unbounded lyapunov function to ensure @xmath120 to be stochastically passive . _ corollary _ [ vanish3 ] implies that stochastic passivity will lose when the desired state @xmath105 makes @xmath124 and the storage function is expected to be a radially unbounded lyapunov function , so it is impossible for one to use stochastic passivity theory , and further , stochastic lyapunov theorem to analyze the globally asymptotical stability of @xmath120 at @xmath105 in the sense of probability . the above analysis reveals that when @xmath124 , stochastic passivity will fail to capture the globally asymptotical stability ( in the probability sense ) of a stochastic differential system @xmath120 at the desired state @xmath105 , often set as the equilibrium state @xmath14 ( if exists ) in many control problems . in fact , the nonzero diffusion term is frequently encountered in many real stochastic systems , such as chemical reaction networks , tracking systems , etc . one case is that the noise is persistent in quite a few stochastic systems , which means that for @xmath125 and @xmath126 and thus @xmath14 does not exist at all ; the other case is that some special control purposes are served for , such as the desired state @xmath105 being not @xmath14 so that @xmath124 , even if @xmath14 exists . apparently , the nonzero diffusion term in real stochastic systems restricts greatly the applications of stochastic passivity theory , a powerful tool for stabilization . however , what is even worse is that it may lead to the system under consideration being not stable at all in probability at the desired state , as stated in _ theorem _ these two awkward situations motivate us to find a new solution for stabilizing those stochastic systems with nonzero diffusion term at the desired state . on the one hand , it is impossible to stabilize some stochastic systems at any state in probability , on the other hand , the excellent performance of stochastic passivity is hoped to be used . thus , we take a hack at the next best way to address the current control problem , including seeking the convergence in distribution and ergodicity instead of the convergence in probability , and finding the stochastic passivity behavior only outside a certain neighborhood of the desired state instead of in the whole state domain . the objective in this section is to present the theory of stochastic weak passivity with which some stochastic systems with nonzero diffusion term can be analyzed concerning the convergence of the transition measure and ergodicity . this theoretical framework includes some basic concepts related to stochastic weak passivity , properties of invariant measure , and results for stabilization which are parallel to those appearing in the stochastic passivity theory . we firstly give the definitions of convergence in distribution and of ergodicity . assume a stochastic differential equation described by eq . ( [ stochasticequation ] ) to have an invariant measure @xmath24 . if there exists a subset of @xmath76 , denoted by @xmath127 , such that for any borel subset @xmath128 with zero @xmath24-measure boundary the equation @xmath129 is true , then the stochastic process is said to be locally convergent in distribution . if @xmath130 , then the convergence in distribution is globally . if for any borel subset @xmath131 the state @xmath13 satisfies @xmath132 where @xmath133 " represents almost surely " and @xmath134 then the stochastic process is said to be locally ergodic . especially , when @xmath130 , the ergodicity is global . here , analogous to the definition of stability in probability , we also distinguish the local and global notations to emphasize the importance of the initial condition . in the control view point , the convergence of the transition measure and ergodicity both describe certain senses of stable behaviors for stochastic systems . the former means that the distribution of the state will converge to an invariant measure as time goes infinite . therefore , as long as the invariant measure is shaped to fasten on a small region around the desired state , then the state will evolve within this region with a large probability , i.e. , not to deviate from the desired point too far with a large probability . the latter implies that the state evolution almost always take place within the mentioned region . even if the trajectory sometimes run from the region , it will come back into the region immediately . clearly , the convergence of the transition measure and ergodicity reveal that the state of a stochastic system almost always evolves near the desired state if the invariant measure is assigned properly . we define this behavior as stochastic asymptotic weak stability . a stochastic differential equation of eq . ( [ stochasticequation ] ) is of local@xmath0global stochastic asymptotic weak stability if its distribution locally@xmath0globally converges to an invariant measure and its process is of local@xmath0global ergodicity . next , we define the stochastic weak passivity that serves for stabilizing a stochastic differential system in weak sense . note that the loss of stochastic passivity mainly originates from the nonzero diffusion term at the desired state , which further results in some unexpected behaviors appearing around it . thus , a naive idea is to give up the stochastic passivity near the desired state , but only to suggest it outside a neighborhood of the desired state . [ defstochasticweakpassivity ] a stochastic differential system @xmath120 , as described by eq . ( [ stochasticsystem ] ) , is said to be of stochastic weak passivity if there exist a @xmath135 function @xmath34 , i.e. , the storage function , such that for @xmath136 and @xmath137 the following inequality holds @xmath138\leq \bm{u}^\top\bm{y}\ ] ] where the state @xmath139 is the sole minimum point for @xmath140 and @xmath141 is called the stochastic passive radius . similar to the concept of the strict passivity , we may further define strict stochastic weak passivity . [ defstrictstochasticweakpassivity ] consider a stochastic weak passive system . suppose that there exists a positive constant @xmath142 such that for @xmath143 and @xmath144 @xmath138\leq \bm{u}^\top\bm{y}-\delta\|\boldsymbol{\xi}\|_2^{2}\ ] ] the system is * strictly state stochastic weak passive if @xmath145 . * strictly input stochastic weak passive if @xmath146 . * strictly output stochastic weak passive if @xmath147 . definition @xmath148 reveals that the stochastic asymptotic weak stability is concerned with the convergence in distribution of the state and ergodic behavior . however , for a stochastic system , unlike its equilibrium it is not quite obvious to know something about its invariant measure , such as the existence , uniqueness , etc . we separate this subsection to analyze the properties of the invariant measure of the stochastic differential equation under consideration . in fact , it is not a new research issue to analyze the properties of the invariant measure of a stochastic system @xcite . a sufficient condition to say it convergent in distribution was reported as follows . if a right markov process on @xmath21 is strongly feller , i.e. , @xmath149 the transition semigroup @xmath150 transforms bounded borel functions into @xmath151 , and moreover @xmath152 is irreducible , i.e. , @xmath153 and any open set @xmath154 there is @xmath155 , then any probability measure converges to the invariant measure ( if exists ) . moreover the invariant measure ( if exists ) is equivalent to each transition measure @xmath156 , @xmath157 , @xmath158 . this theorem provides a solution to capture the convergence in distribution for a right markov process . however , it is not easy to verify the conditions of strongly feller " and irreducible " in practical applications . as an alternative , khasminskii @xcite proposed a more practical way to say a stochastic system to be convergent in distribution , which works if a markov process is mix sufficiently well " in an open domain @xmath159 and the recurrent time is finite ( cf . _ theorems _ @xmath160 , @xmath161 and _ corollary _ @xmath162 in @xcite ) . here , we will combine this practical way with zakai s work @xcite , and give a lyapunov criterion to say stochastic asymptotic weak stability . however , the drift and diffusion terms of the stochastic system are set to have local lipschitz continuity instead of global lipschitz continuity in @xcite . [ lem finite mean recurrent time ] for a stochastic differential equation ( [ stochasticequation ] ) having a global solution @xmath13 , if there exist a function @xmath163 , a state @xmath164 , and two positive numbers @xmath165 and @xmath166 such that @xmath167<-k , ~\forall~ \|\bm{x}(t)- \bm{x}_{\tilde{r}}\|_{2}\geq \tilde{r}\ ] ] then for all @xmath168 the first passage time from @xmath169 to the sphere @xmath170 , denoted by @xmath171 , satisfies @xmath172\leq\frac{v\big(\bm{x}(0)\big)}{k}\ ] ] at the time of @xmath173 , by dynkin s formula we have @xmath174&= & v\big(\bm{x}(0)\big)+\text{e}\left[\int_{0}^{t\wedge\tau}\mathcal{l}\big[v\big(\bm{x}(s)\big)\big]ds\right]\\ & \leq & v\big(\bm{x}(0)\big)-\text{e}\big[\int_{0}^{t\wedge\tau}kds\big]\end{aligned}\ ] ] note that @xmath175 , so @xmath176\leq\frac{v\left(\bm{x}(0)\right)}{k}$ ] . the inequality ( [ ineqfmrt ] ) naturally holds due to the monotone convergence . [ weakstability ] for a stochastic equation in the form of ( [ stochasticequation ] ) , if there exists a nonnegative function @xmath163 satisfying the following conditions : * @xmath177 ; * @xmath178<-k$ ] ; * @xmath179 , if @xmath180 , then @xmath181 . then there is a unique finite invariant measure @xmath24 such that for any borel subset @xmath128 with zero @xmath24-measure boundary @xmath182 and for any borel subset @xmath131 @xmath183 i.e. , eq . ( [ stochasticequation ] ) being globally asymptotical stable in weak sense . according to _ lemmas _ @xmath58 and @xmath2 , the first two conditions could suggest that eq . ( [ stochasticequation ] ) has a unique global solution and for any initial state @xmath184 satisfying @xmath185 , we have @xmath186\leq\frac{v\left(\tilde{\bm{x}}\right)}{k}\ ] ] hence , for any compact subset @xmath187 we get @xmath188\leq \sup_{\tilde{\bm{x}}\in\mathcal{k}}\frac{v\big(\tilde{\bm{x}}\big)}{k}<\infty.\ ] ] further based on the strong maximum principle for solutions of elliptic equations , the third condition implies the system to be irreducible ( cf . _ lemma 4.1 _ in ) , which combining the above inequality suggests that an ergodic markov chain can be induced for this stochastic process by constructing a circle . the ergoic property of the markov chain will ensure that there exists a sole invariant measure to which the transition measure converges ( cf . _ theorems _ @xmath160 and @xmath161 , and _ corollary _ @xmath162 in ) , and the ergodicity of the system under consideration is true ( cf . _ theorem _ @xmath148 in ) . namely , eqs . and hold . the theorem provides a lyapunov function based method to address the issues of the existence and uniqueness of the invariant measure together with the convergence of the transition probability measure and ergodicity for a stochastic differential equation , so it can be associated with the lyapunov stability theory conveniently . there are two differences between the above theorem and the corresponding result in @xcite . one is that the non - singularity of @xmath189 is not necessary in the whole state space but only holds in an open ball ( @xmath180 ) . the latter is believed to be achieved more easily in practice . the other is that the storage function must be radially unbounded here . in fact , this is not a necessary condition , which can be removed if the drift term and diffusion term in the stochastic equation are assumed to be globally lipschitz continuous . for a stochastic asymptotic weak stable system , to ensure the state to evolve within a small region around the desired point , the invariant measure needs to be assignable or at least partially shaped by the control to fasten on this region . in the sequel , we will prove that the invariant measure can be shaped purposefully by controlling the change rates of the nonnegative function @xmath34 and the radius @xmath165 of the ball @xmath190 . [ lem stop time estimation1 ] for a stochastic differential equation admitting a global solution @xmath13 , if @xmath191 @xmath163 and @xmath192 such that @xmath193 and @xmath194 then for any @xmath195 we have @xmath196\geq \frac{\left(v_{2}-v_{1}\right)^{2}}{2 c v_{2 } } ; ~\mathrm{(2)~ e}\left[\tau_{2i-1}-\tau_{2i-2}\right]\leq \frac{v_{2}-v_{1}}{k};~~\mathrm{and}\ ] ] @xmath197 \leq \frac{2 c v_{2}}{2 c v_{2}+k\left(v_{2}-v_{1}\right)}\ ] ] where @xmath198 satisfying @xmath199 , @xmath200 represents the first time at which the state hits the region @xmath201 after @xmath202 , @xmath203 is the first time at which the trajectory reaches the surface of @xmath204 after @xmath200 , and @xmath205 means the initial time . \(1 ) according to dynkin s formula , for any @xmath195 @xmath206 & = & v\left(\bm{x}(\tau_{2i-1})\right ) + \text{e } \left[\int_{\tau_{2i-1}}^{t\wedge\tau_{2i } } \mathcal{l}v\left(\bm{x}(s)\right ) ds\right ] \\ & \leq&v\left(\bm{x}(\tau_{2i-1})\right ) + \text{e}\left [ \int_{\tau_{2i-1}}^{t\wedge\tau_{2i } } c ds\right]\\ & \leq&v_{1 } + c\left(t-{\tau_{2i-1}}\right)\end{aligned}\ ] ] also , since @xmath206&=&\text{e } \left[v\left(\bm{x}(t)\right)\right]\text{p}\left(\tau_{2i}\textgreater t\right)+\text{e } \left[v\left(\bm{x}(\tau_{2i})\right)\right]\text{p}\left(\tau_{2i}\leq t\right)\geq v_{2 } \text{p}\left(\tau_{2i}\leq t\right)\\\end{aligned}\ ] ] we have @xmath207}{v_{2 } } \leq\frac{v_{1}+c\left(t-{\tau_{2i-1}}\right)}{v_{2}}\ ] ] therefore , we get @xmath208 & = & \int_{0}^{\infty } \text{p}\left(\tau_{2i}-\tau_{2i-1 } > s \right ) ds \geq \int_{0}^{\frac{v_{2}-v_{1}}{c } } \text{p}\left(\tau_{2i } > s+\tau_{2i-1 } \right ) ds\\ & \geq & \int_{0}^{\frac{v_{2}-v_{1}}{c } } 1-\frac{v_{1}+cs}{v_{2}}ds=\frac{\left(v_{2}-v_{1}\right)^{2}}{2 c v_{2}}\end{aligned}\ ] ] \(2 ) on the other side , for any @xmath209 we have @xmath210 & = & v\left(\bm{x}(\tau_{2i-2})\right ) + \text{e } \left[\int_{\tau_{2i-2}}^{t\wedge\tau_{2i-1 } } \mathcal{l}v\left(\bm{x}(s)\right ) ds\right ] \\ & \leq&v\left(\bm{x}(\tau_{2i-2})\right ) + \text{e}\left [ \int_{\tau_{2i-2}}^{t\wedge\tau_{2i-1 } } -k ds\right]\\ & = & v_{2 } -k\text{e}\left[t\wedge\tau_{2i-1}-{\tau_{2i-2}}\right]\end{aligned}\ ] ] i.e. , @xmath211\leq\frac{v_{2}-\text{e } \left[v\left(\bm{x}(t\wedge\tau_{2i-2})\right)\right]}{k}\leq\frac{v_{2}-v_{1}}{k}\ ] ] by monotone convergence theorem , the inequality @xmath212\leq\frac{v_{2}-v_{1}}{k}$ ] is true . \(3 ) based on the results of ( 1 ) and ( 2 ) , we have that for any @xmath213 @xmath214 \geq \frac{k \left(v_{2}-v_{1}\right)}{2 c v_{2}}.\ ] ] besides , eq . and the result ( 2 ) imply there re almost surely infinite many @xmath215 , so the notations @xmath216 " and @xmath217 " in the following are not in vain . applying fatou s lemma yields @xmath218 \geq \frac{k \left(v_{2}-v_{1}\right)}{2 c v_{2}}\ ] ] let @xmath219 , utilizing which we have @xmath220 hence , @xmath221\\ & = & \liminf_{t\to\infty}\frac{1}{1+\frac{\sum_{i=1}^{i(t ) } ( \tau_{2i}-\tau_{2i-1})}{\sum_{i=1}^{i(t ) } ( \tau_{2i+1}-\tau_{2i})}}\end{aligned}\ ] ] by taking the expectation of both sides and further combining eq . , we will get the result of ( 3 ) . for a stochastic equation in the form of , if there exists a nonnegative function @xmath163 satisfying the following conditions : * @xmath177 ; * @xmath178<-k$ ] ; * @xmath179 , if @xmath180 , then @xmath181 ; then for any borel subset @xmath131 satisfying @xmath222 , we have @xmath223 where @xmath224 $ ] , @xmath225 and @xmath226 . if @xmath227 , then the result holds true automatically . so we only prove the result in the case of @xmath228 which implies @xmath229 . from @xmath177 and @xmath226 , we obtain @xmath230 therefore , when @xmath231 , @xmath121<-k$ ] . let @xmath232 and @xmath233 mentioned in _ lemma 3 _ be chosen as @xmath234 and @xmath235 . then , by ergodicity in _ theorem _ , eq . ( ) is achieved . according to _ theorem _ and eq . we have @xmath236 & = & \liminf_{t\to\infty } \frac{1}{t}\int_{0}^{t}\mathbbold{1}_{\left\{v\left(\bm{x}(t)\right)\geq v_{\mathbbold{b}}\right\ } } dt\\ & = & \pi\left(v\left(\bm{x}\right)\geq v_{\mathbbold{b}}\right)\end{aligned}\ ] ] by applying the result ( 3 ) of _ lemma 3 _ we can get @xmath237 note that @xmath238 , we thus have @xmath239 [ rem shape the invariant measure ] the above theorem reveals that the invariant measure can be shaped by controlling the change rates @xmath166 and @xmath240 of @xmath34 together with the ball radius @xmath165 that takes effect through affecting @xmath241 in terms of @xmath226 . when @xmath165 is fixed , the invariant measure will become larger if @xmath166 increases and@xmath0or @xmath240 decreases . the larger the invariant measure is , the greater possibility the trajectory of the state fastens on the region near the desired state , i.e. , the more stable the stochastic system is in the weak sense . now , we are able to tackle the problem of stabilizing the stochastic systems in weak sense based on the stochastic weak passivity . we name the main result as stochastic weak passivity theorems in the context . here , concern is only given to the implicit negative proportional controller @xmath242 for the purpose of stabilization , where @xmath243 is a positive definite matrix with suitable dimension . [ stochasticweakpassivitytheorem ] for a stochastic differential system @xmath120 in the form of eq . ( [ stochasticsystem ] ) , assume that there exists a radially unbounded storage function @xmath244 suggesting it to be stochastically weakly passive with the stochastic passive radius @xmath245 and the desired state @xmath105 . also , we suppose that there exists a negative proportional controller @xmath246 connected with @xmath120 in feedback so that * @xmath247 be bounded away from zero when @xmath248 ; * @xmath179 , rank(@xmath249 when @xmath250 . then there exists an unique finite invariant measure @xmath24 , and moreover , for any borel subset @xmath128 with zero @xmath24-measure boundary @xmath251 and for any borel subset @xmath252 @xmath253 that is to say @xmath120 being globally asymptotical stable in weak sense . from the given conditions and the definition of @xmath38 $ ] in eq . , for any state outside the ball @xmath254 we have @xmath255\leq \bm{y}^{\top}\bm{u } = -\bm{y}^{\top}(\bm{x},\bm{u})\bm{k}\bm{y}(\bm{x},\bm{u})\leq-\lambda_{\min}\|\bm{y}\big(\bm{x},\bm{u}(\bm{x})\big)\|_{2}^{2}\ ] ] where @xmath256 is the minimum eigenvalue of @xmath243 . since @xmath247 is bounded away from zero when @xmath257 , we could find a positive number @xmath166 such that for any @xmath248 @xmath258\textless -k\ ] ] combining it with the non - singularity of @xmath259 when @xmath250 yields the results ( based on _ theorem _ @xmath5 ) . note that the above theorem works when the output norm is bounded way from zero outside a ball . in fact , this condition is not necessary . it may be replaced by setting the system to be strictly state stochastic weak passive . [ stochasticweakpassivitytheorem2 ] consider a strictly state stochastic weak passive system under the conditions of radially unbounded @xmath34 , @xmath245 , @xmath105 and @xmath142 . suppose that there exists a negative proportional controller @xmath246 connecting with the system in feedback so that * @xmath179 , rank(@xmath249 when @xmath250 . then there exists an unique finite invariant measure @xmath24 , and moreover , for any borel subset @xmath128 with zero @xmath24-measure boundary @xmath251 and for any borel subset @xmath252 @xmath260 that is to say @xmath120 being globally asymptotical stable in weak sense . from the known conditions and the definition of @xmath38 $ ] in eq . , we have at @xmath254 @xmath255\leq \bm{y}^{\top}\bm{u}-\delta\|\bm{x}-\bm{x}^{\dag}\|_2^{2 } = -\bm{y}^{\top}(\bm{x},\bm{u})\bm{k}\bm{y}(\bm{x},\bm{u})-\delta\|\bm{x}-\bm{x}^{\dag}\|_2^{2 } \leq-\delta r^{2}\ ] ] i.e. , @xmath261\textless -k$ ] . further applications of _ theorem _ @xmath5 yield the results immediately . clearly , the change rate @xmath166 of the energy function @xmath34 is closely dependent on the the feedback gain matrix @xmath243 . when @xmath243 is designed to be stronger , @xmath166 will be larger . this can lead to the increase of the invariant measure @xmath262 , and further the state evolution taking place within a more intensive region around the desired state @xmath105 . therefore , it is an effective way for shaping the invariant measure @xmath262 purposefully to strengthen the feedback gain matrix @xmath243 . stochastic weak passivity theorems suggest sufficient conditions to stabilize a stochastic differential system in weak sense . although some items are difficult to realize , such as the non - singularity of @xmath259 when @xmath250 , some ones are relatively weak , e.g. , the energy function @xmath34 only requires to be positive semi - definite , and the simple proportional controller is qualified . to weaken the harsh condition on the non - singularity of @xmath259 in the stochastic weak passivity theorem , we might separate those linear independent rows from @xmath263 to construct a new diffusion term @xmath264 . clearly , it is much easier to realize @xmath265 to be full - rank than @xmath259 to be nonsingular . for this purpose , we first define a transformation . a homeomorphism @xmath266 , expressed as @xmath267 is called a decomposition transformation of system @xmath268 if it can transform the stochastic differential system @xmath268 equipped with eq . ( [ stochasticsystem ] ) into two subsystems : one is a stochastic differential system @xmath269 , the other is a deterministic system @xmath270 . here , @xmath271 and @xmath272 . these two subsystems are written respectively as @xmath273 where the drift term @xmath274 and diffusion term @xmath275 in @xmath269 are both locally lipschitz continuous . the diffusion term @xmath276 in @xmath269 can be extracted from @xmath263 to the greatest extent according to the rank so that the rank of @xmath276 may reach @xmath277 . @xmath270 is obviously a fixed point , and is certainly stable . thus , the stabilization of @xmath268 may be realized by stabilizing @xmath269 . [ rem explaination of how to find decompation transformation ] in practice , the defined decomposition transformation is not difficult to be constructed . for many nonlinear stochastic systems , the state evolution really takes place in an invariant manifold , denoted by @xmath278 , but not in the whole state space @xmath76 . for example , the state of a chemical reaction network will evolve in a positive stoichiometric compatibility class , which is a subset of @xmath76 . therefore , an immediate idea to construct the decomposition transformation is to decompose the state space @xmath76 into the invariant manifold @xmath278 and its orthogonal complement @xmath279 . the projection from @xmath280 to @xmath281 leads to the stochastic subsystem @xmath269 while the projection from @xmath280 to @xmath282 induces the fixed point subsystem @xmath270 . utilizing the decomposition transformation , we can give the refined stochastic weak passivity theorem . [ transformestatemanifold ] for a stochastic differential system ( [ stochasticsystem ] ) if there exists a decomposition transformation @xmath283 transform the system into two subsystems @xmath269 ( [ subsystem1 ] ) and @xmath270 ( [ subsystem2 ] ) , then the state @xmath13 staring from any initial state @xmath169 satisfies @xmath284 since @xmath285 , also since @xmath286 , the result is true . it is clear that @xmath287 can be any element in @xmath288 . this in turn means that for any @xmath289 there is a manifold @xmath290 defined . for simplicity of notation , we identify this manifold by @xmath291 . [ refinedswpt ] assume that the system @xmath268 can be decomposed into two subsystems @xmath269 ( [ subsystem1 ] ) and @xmath270 ( [ subsystem2 ] ) using the decomposition transformation @xmath283 . further , suppose that there exist a radially unbounded storage function @xmath292 suggesting @xmath269 to be stochastic weak passivity with the stochastic passive radius @xmath245 and the desired state @xmath293 , and a negative implicit proportional controller @xmath294 so that * @xmath295 is bounded away from @xmath296 when @xmath297 ; or the system is strictly state stochastic weak passive . * @xmath298 when @xmath299 . then there exist an invariant measure @xmath24 and a corresponding manifold @xmath300 such that for any borel subset @xmath128 with zero @xmath24-measure boundary @xmath301 and for any borel subset @xmath252 @xmath302 when the controller @xmath303 is connected with @xmath269 in feedback . that is to say @xmath268 being locally asymptotically weakly stable . first , we consider the existence of invariant measure under the transformed coordinate @xmath304 . for @xmath269 connected by the controller @xmath303 in feedback , by stochastic weak passivity theorems ( _ theorem _ [ stochasticweakpassivitytheorem ] and _ theorem _ [ stochasticweakpassivitytheorem2 ] ) there exists an unique finite invariant measure @xmath305 so that for any borel subset @xmath306 with boundary @xmath307 , when @xmath308 we have @xmath309 and for any borel subset @xmath310 @xmath311 for @xmath270 , define @xmath312 be a measure on ( @xmath313,@xmath314 ) that satisfies @xmath315 where @xmath316 is a fixed point in @xmath313 and @xmath317 is a borel subset of @xmath313 , then for all @xmath54 and the initial condition @xmath318 we have @xmath319 consider the product measure of @xmath305 and @xmath312 , denoted by @xmath320 , i.e. , @xmath321 note that the existence and uniqueness of @xmath320 are guaranteed by hahn - kolmogorov theorem and @xmath20-finite property , respectively . imposing @xmath320 on the set @xmath322 yields @xmath323 where @xmath324 is the complementary set of @xmath325 in @xmath288 . next , we discuss the convergency of @xmath320 under the transformed coordinate @xmath304 . let the boundary of @xmath335 be @xmath307 , then for @xmath336 there exist two sequences of points @xmath337 ( @xmath338 ) and @xmath339 ( @xmath340 ) such that @xmath341 hence , for any point @xmath342 in @xmath343 , there exist two sequences of points @xmath344 and @xmath345 such that @xmath346 further let @xmath347 denote the boundary of @xmath128 , then we have @xmath348 and @xmath349 assume @xmath350 , i.e. , @xmath308 , with which we get for @xmath351 @xmath352 this complete the proof of the convergence of @xmath353 . finally , we consider the existence and convergence of invariant measure under original coordinate . let a measure @xmath24 satisfy @xmath354 for all @xmath355 . we have @xmath356 which means @xmath24 is invariant . further let @xmath357 denote the boundary of @xmath358 , then @xmath359 due to the bicontinuity of @xmath360 . thus , if we assume @xmath361 , then @xmath362 hence , for @xmath363 @xmath364 which shows the convergence of the transition measure . similarly , we can prove the local ergodicity of the process . a point should be noted that the current invariant measure is no longer unique . it is closely dependent on the initial condition @xmath169 . hence , the refined stochastic weak passivity theorem actually suggests the conditions of local asymptotic weak stability for a stochastic differential system . in this section , the stochastic weak passivity theory is applied to linear systems and a nonlinear process system . consider a representative linear time - invariant system described by @xmath365 where @xmath366 and @xmath367 are constant matrices with suitable dimensions . for simplicity let @xmath82 be the desired state . since the noise port @xmath367 , there does not exist any lyapunov function that could suggest this linear system to be globally stable in the sense of probability . however , it is possible for this system to reach stochastic asymptotic weak stability . for a linear system described by eq . ( [ linearsystem ] ) , if there exists a positive definite matrix @xmath368 with suitable dimension such that * @xmath369 ; * @xmath370 is negative definite . then the system is stochastically weak passive . let @xmath371 , then @xmath372 & = & \bm{x}^\top\bm{d}(\bm{ax}+\bm{bu})+\frac{1}{2}\text{tr}\{\bm{d}\boldsymbol{\sigma}\boldsymbol{\sigma}^\top\ } \\ & = & \bm{x}^{\top}\bm{d}\bm{bu}+\bm{x}^{\top}\bm{da}\bm{x } + \frac{1}{2}\text{tr}\left\{\bm{d}\boldsymbol{\sigma}\boldsymbol{\sigma}^\top\right\}\\ & = & \bm{y}^{\top}\bm{u}+\frac{1}{2}\bm{x}^{\top}(\bm{da}+\bm{a}^\top\bm{d})\bm{x } + \frac{1}{2}\text{tr}\left\{\bm{d}\boldsymbol{\sigma}\boldsymbol{\sigma}^\top\right\}\end{aligned}\ ] ] assume @xmath373 to be the maximum eigenvalue of @xmath370 . since the matrix is negative definite , we have @xmath374 and @xmath375 . note that @xmath376 . hence , as long as @xmath377 , we get @xmath372&\leq&\bm{y}^{\top}\bm{u}\end{aligned}\ ] ] which completes the proof . clearly , the used lyapunov function @xmath371 is radially unbounded , so if the noise port @xmath378 is full row rank and the measurement matrix @xmath379 is full column rank , then all conditions required in _ theorem _ [ stochasticweakpassivitytheorem ] are true . namely , any negative proportional controller @xmath246 can globally stabilize this class of stochastic linear systems in weak sense . next , we manage to analyze the stochastic weak passivity of a nonlinear process system and further stabilize it based on the refined stochastic weak passivity theorem . consider a continuous stirred tank reactor ( cstr ) in which a first - order chemical reaction takes place @xmath380 the pure component @xmath381 with a fixed concentration @xmath382 is fed at the inlet of the reactor while the fluid mixture of @xmath381 and @xmath383 is released from the outlet of the reactor . for simplicity this process is assumed to be isothermal and isometric . furthermore , the volume flow rates of inflow and outflow are regulated to be the same . in addition , the chemical reaction goes on with a unstable reaction rate coefficient . denote the components concentrations of @xmath381 and @xmath383 by @xmath384 and @xmath385 , respectively , the volume flow rate of inflow or outflow by @xmath386 , the reaction rate coefficient by @xmath166 and the disturbance on @xmath166 by @xmath20 ( note that @xmath387 and @xmath388 ) , then we have the dynamical equation @xmath389\dd t -\sigma{x}_{1}\dd \omega \\ x_{2}&= & \left[k{x}_{1}-{x}_{2}q\right ] \dd t + \sigma{x}_{1}\dd \omega \end{array } \right.\ ] ] construct the input - output pair as @xmath390 where @xmath391 is the desired concentration of @xmath381 constrained by @xmath392 . then the input - output representation of the stochastic cstr model ( [ cstrmodel ] ) can be written as @xmath393 apparently , at the desired state the diffusion term will not vanish . therefore , it is impossible to find a radially unbounded lyapunov function so that the stochastic crtr system is stable at the desired point in the sense of probability . according to the conservation law , it is easy to get that the state @xmath394 only evolves in the manifold @xmath395 . therefore , based on _ remark _ the decomposition transformation can be defined as @xmath396 through which the system is decomposed into @xmath397\dd t -\sigma\bar{{x}}_{1}\dd \omega\\ { y}_1&=&\left(\bar{{x}}_{1}-{x}_{1}^{\dag}\right)\left(c_1^\text{in}-\bar{{x}}_{1}\right ) \end{array } \right.\end{aligned}\ ] ] and @xmath398 note the facts that @xmath399 and the volume of inflow or outflow is finite within a finite interval , i.e. , @xmath400 , so we get @xmath401 which means the subsystem @xmath269 following @xmath402 hence , the vector field of @xmath270 equals zero and @xmath283 is a decomposition transformation of the cstr process system . construct the lyapunov function to be @xmath403 , then we have @xmath404&=&\left(\bar{{x}}_{1}-{x}_{1}^{\dag}\right ) \left[-k\bar{{x}}_{1}+ \frac{k{x}_{1}^{\dag}\left(c_1^\text{in}-\bar{{x}}_{1}\right)}{c_1^\text{in}-{x}_{1}^{\dag } } + \left(c_1^\text{in}-\bar{{x}}_{1}\right)u\right]+\frac{1}{2}\sigma^{2}\bar{{x}}_{1}^{2}\\ & = & yu-\frac{kc_1^\text{in}}{\left(c_1^\text{in}-{x}_{1}^{\dag}\right)}\left(\bar{{x}}_{1}-{x}_{1}^{\dag}\right)^{2 } + \frac{1}{2}\sigma^{2}\bar{{x}}_{1}^{2}\end{aligned}\ ] ] let @xmath405 , then the above equation changes to be @xmath404 & = & yu-2\delta\left(\bar{{x}}_{1}-{x}_{1}^{\dag}\right)^{2}+\frac{1}{2}\sigma^{2}\bar{{x}}_{1}^{2}\end{aligned}\ ] ] hence , for any @xmath406 where @xmath407 we have @xmath408 & \leq & yu -\delta\|\bar{\bm{x}}_{1}-\bm{x}_{1}^{\dag}\|^{2}\end{aligned}\ ] ] this means that the subsystem @xmath270 is strictly state stochastic weak passive with respect to the radially unbounded storage function @xmath409 . note that the stochastic passive radius @xmath245 is quite small due to @xmath388 . additionally , there exists a positive constant @xmath94 such that @xmath410 and for any @xmath411 the noise port @xmath412 is nonsingular . thus , based on _ theorem _ [ refinedswpt ] , the system can be locally stochastically asymptotically weakly stable under any negative proportional controller . for @xmath384 and @xmath413 for @xmath385.,scaledwidth=100.0% ] to better exhibit the stochastic asymptotic weak stability , some simulations are made on the above cstr process with the initial state designated as @xmath414 , the desired state as @xmath415 , and the disturbance as @xmath416 . the other parameters are @xmath417 mole / m@xmath418/s , @xmath419 m@xmath418/s and @xmath420 mole / m@xmath418 . @xmath58 shows the time evolution of the state @xmath421 without control and the corresponding invariant probability density functions . the ranges , @xmath422 for @xmath384 and @xmath413 for @xmath385 , bounded by two dotted lines , respectively , represent the areas in which the state evolves with probability @xmath423 . when the controller @xmath424 is implemented on this process , the state evolution and the invariant probability density functions will fasten on the region around the desired state more . shown in fig . @xmath2 are the results . we also use two dotted lines to bound the areas in which the state evolves with @xmath423 . now , they change to be @xmath425 for @xmath384 and @xmath426 for @xmath385 , which is clearly more intensive around the desired state . this information can be also observed from fig . @xmath2(b ) in which the invariant probability density function changes thinner " around the desired state than the corresponding one appearing in fig . @xmath58(b ) . the convergent behaviors of the invariant probability density functions under the control are exhibited in fig . @xmath3 . to observe more detailed convergent process , we only exhibit the simulation from @xmath427 s to @xmath428 s. as can be seen in fig . @xmath3 , the initial probability density functions , at @xmath427 , deviate the desired state too much , but they will converge to the invariant probability density functions that fasten on the small region around the desired state as the controller is put into force . implemented . the bounded areas in ( a ) are @xmath425 for @xmath384 and @xmath426 for @xmath385.,scaledwidth=100.0% ] and ( b ) for @xmath383 with the controller @xmath424 implemented.,scaledwidth=100.0% ] this work has presented a theoretical framework of stochastic weak passivity serving for stabilizing the stochastic differential systems with nonvanishing noise . the main contributions include : i ) deriving the necessary conditions to say a stochastic system stochastically passive or the sufficient conditions that a stochastic system must lose stochastic passivity ; ii ) proving that it is impossible for some stochastic systems to be stabilized in probability ; iii ) defining a new concept of stochastic weak passivity to serve for those systems losing stochastic passivity , which captures the stochastic passivity of the system not in the whole state space but only outside a ball centered around the desired state ; iv ) associating stochastic weak passivity to asymptotic weak stability of systems , and further providing the sufficient conditions for global and local asymptotic weak stabilization of nonlinear stochastic differential systems by means of negative feedback laws . the stochastic weak passivity provides an alternative way to stabilize the transition measure as well as capturing the ergodicity of the stochastic differential systems with nonvanishing noise . however , there is still a large room for this method to be improved or expanded . an important issue is that the whole theoretical framework works under the assumption that the stochastic term @xmath55 is a standard wiener process . the motivation of making such an assumption is that the current concept is developed based on the stochastic passivity . for the latter , the noise term is assumed as a standard wiener process . however , the standard wiener process is just a kind of ideal noise , and is used mainly for the simplicity of analysis . as far as many practical systems are concerned , this ideal noise is not accurate enough to represent the internal modeling uncertainty . therefore , it is interesting but challenging to use other stochastic processes instead of the standard wiener process for developing stochastic weak passivity . towards this task , the infinitesimal generator @xmath429}$ ] of eq . needs to be redefined accordingly . in addition , other possible points of future research include : i ) weakening the condition of nonsigularity of the diffusion matrix @xmath430 ; ii ) applying the stochastic weak passivity theory to some special stochastic differential systems , such as stochastic affine systems , thermodynamic process systems , and drive the development of these fields in control techniques ; iii ) developing the determinist version of stochastic weak passivity .